Hartshorne Solutions

楊丕業 Piye Yang

[email protected]

June 9, 2019

(2019.6.9.)
Sorry for my decision to stop updating the amendments. I think I will not update this pdf of
solutions anymore. I got no offer for a Ph.D. program this year. Maybe that’s because I have no
talent and luck to study mathematics. So, that’s it. I give up. Best wishes for everyone. Thanks to
my friends who helped me correct my answers, and some guys on Internet who sent me e‑mails
and encouraged me. But, bye‑bye!

(Original)
This document is the solution of Hartshorne’s Algebraic Geometry by me during I was learning
the AG. During these exercises, I mainly referred to Algebraic Geometry and Étale Cohomology Theory
by Prof. Fu Lei (扶磊), Algebraic Geometry and Arithmetic Curves by Prof. Qing Liu (刘青), Prof. Aise
Johan de Jongʹs Stack Project, Andrew Egbertʹs solution I googled, and all related information in
website Stackexchange and Mathoverflow. I also need to thank my friends Zio Khái, Tsjiong ʔit‑
Miaeng, Kwèi Tsjièng‑Biaeng, Wiang Pyin, Ghrà Meng‑Dzjin, Lý Bwai‑Kon and Mràeng Dzjieng.
They helped me a lot when I did these exercises. By the way, ils peuvent également être la mesure
pour la recherche de petit ami. Sed bedaŭrinde ili estas ĉiuj malgejaj, so kann jede Dame, die die
Absicht von ihnen hat, mich mailen.
At first sheang‑je prá ĉiujn solvojn 都 사 hàlái 主要 shyh kràutok mīʹself 淆 μαθηʹ d dzjyghòu
tay yō‑i sheang‑tangnjien originally, 需要 sjý‑triò kak‑ 청초 tsair‑hao. Nyan‑I‑上‑t dzjy‑ghòu 跟
Biu lao‑shy 淆tɕí AG, tā njiàng ngǒ kaj Bwai‑Kon 在 1. Okt. chitiền tzyyjii 淆‑i‑淆 Harshorne‑t
dih‑1‑jang béngtshiá bǎ dei‑miuk 做‑1‑做 ankaŭ, ǘ‑shì ngǒ dziòu 做‑l‑1‑hah. Hartshorne‑t ghónta
liannshyi‑tyi dèi‑ǘ ʃo‑ʃin‑ʃa lai‑shuo ghón mal‑(hao‑zù), 在 char topʹàn dī dzjyghòu 也 huāshyàn
ghónta neng‑dzratàu‑d topʹàn dōu yeou sakʰɯgo or deubuò tre serioze. Gleichzeitig, ngǒ 還在
shanq Looijenga‑giáoþụ‑d AG‑course, 怹lao‑d khwà xiān gǎng‑l 1‑kà 淆‑ki d variety theory, kī‑
mạt d dzjyghòu wǒ damsim 怹 yaw kháu ki‑mạt kusi (pitkiàeng 頭‑1‑nen d shyr‑how 怹låo‑d
kimạt kháusjý 9.a.m. kara gogogoji mate shiken 습니다, dziòu 做‑l‑做 dì‑1‑jang‑d dei‑miuk again,
siáng‑triò njiokwá 要把 jychyan ne‑trouve‑pas‑topʹàn‑d dei‑miuk shiee hà‑lái, 就 đảthành‑l LATEX.
Nyan‑1‑上‑d shyr‑how tshom‑kra‑l le séminaire du Biu sen‑sei, gaang 怹‑d nà‑pwón AG‑livre,
hence jiá shiangtang‑ǘ 淆‑l Hartshorneʹd dì‑2‑tsjiang dzen‑5‑tset tek nội‑iong. Nyan‑1‑hà‑d shyr‑
how 월발‑dì jiódé dzìký một 淆doong, một‑năng baa Biu laoshy nà‑tàu kanzen‑dàisriù‑d lingo

1
trwyénhrwà‑wei jiiher‑d sysiáng, jiá một‑pháp iòng kiígha‑d drikkwan lai sɿkháu moidai. Ǘ‑dzjié
在 shyo♂jaang tek kiàn♂ngyè 下 dziòu khaiʂʅ thatriò Bwai­Kon mann­mann 做 Hartshorneʹd tyimuh,
swi‑njien tha 후래 betray‑l ŋǒ.
Kangdzai 也 gǎng‑guò, Hartshorne jehbeen‑sjio‑d deimiuk 對 ʃo‑ʃin‑ʃa laisjwiet 很 mal‑facile
tsuò. Tā baa EGA‑SGA1‑que‑d nwài‑jiong nōʃɯkʰɯ 成‑l 1‑been shū, but kràeng‑ta kràeng‑drióngʹiàu
kràeng‑ióutshiù d dongshi 怹 dawshyh tuo tshjwit‑dz‑l shyityi lý, keeyii sjwiet 是 très mwot‑ióu
tset‑tshàu l. 而且, ȵiokʷá jyyshyh tímù píkràu mal‑facila tekghrwàj yeejiow 算‑l, la teksto du livre
裏面 ghrwan ĉiam jín‑jiòng deimiuk‑t jyegwo, pwiit 1‑pen‑tsuòdei‑1‑pen‑淆‑d huah 근본 淆‑bù‑l,
1‑pen‑tsuòdei‑1‑pen‑淆‑d huah 근본 做‑bú‑ghrákhiò. Đis kréndrik 不是 mwot‑tset‑tshàu‑d wen‑
ntyi, kréndrik keeyii sjwiet 是 sehr muómràiP l. Tsàitsjiásjwiet, Hartshorne jehbeen‑sjio‑t ghòu‑2‑
jang xiě‑der lwàntshitprettsau‑t, 如果 tsjintek sheang 淆 curves and surfaces tek ney‑yong jiá‑ióu
ACGH‑d Geometry of Algebraic Curves kaj Arnaud Beauville‑t Complex Algebraic Surfaces keeyii 看,
hair pwiitnjio kuoh‑사 1‑hà biùliok‑t 3‑tsjiang nèiyóng, tsjìsjiáu 把 shiangjiau lýlwòn duō‑사‑tém,
or giùthéi jiehshaw 1‑hah étale topology‑t tungsei, shyityi lýmièn 連 elliptic curves d kypwón‑
gwiin douyaw swán, 사々 chingchuu 又 pwiitghwàj tzeenyanq. Sjwietbraek‑l, Hartshorne jehbeen
sjio genbeen 不 sjiekghop ʃoŋakʰɯ. 真 ʹiàu tshrio‑淆 AG tekghrwàj kiàn‑ngì yonq Biu laoshy nàp‑
wón AG book, njiokwá 迺 pwiit‑jiòng kè gaang s tháulwònpran d 話.
Tswàjghòu, ŋǒ 還是 ʹiàu gaanshieh Hartshorne honnêtement: īnuèi 在 사 tsjiàphien darann‑d
references d dzjyghòu ngǎ shyrtzay njín‑pwiit‑triù 要 thuódzau, thus mwotbrènpwiap jiow dzáu‑
l 1‑thàu pīnyīn ghèithuòng based on the Kwáng‑Wiìn‑ʹyimghèi. Cette Gedanke ŋá pwónkhwa
shyrhow 就 yeou‑l, 最後 khiakdzjié Hartshorne 치‑háu‑l mia procrastination.
I am not entirely sure that all my solutions are correct. So if anybody finds any mistakes in
my solutions, please e‑mails me. My e‑mail address is under my name at the beginning of this
document, you can see it even if you cannot understand the meaning of the above gibberish.
最後，做了一點微小的工作，謝謝大家！

Yang Pi‑Yeh 2 Hartshorne Solutions

Content Content

Content
1 Varieties 5
1.1 Affine Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Projective Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Rational Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Nonsingular Varieties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Nonsingular Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.7 Intersections in Projective Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.8 What is Algebraic Geometry? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

2 Schemes 19
2.1 Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3 First Properties of Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Separated and Proper Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.5 Sheaves of Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.6 Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.7 Projective Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.8 Differentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.9 Formal Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3 Cohomology 59
3.1 Derived Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.2 Cohomology of Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3 Cohomology of Noetherian Affine Scheme . . . . . . . . . . . . . . . . . . . . . . . 61
3.4 Čech Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.5 The Cohomology of Projective Space . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.6 Ext Groups and Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.7 The Serre Duality Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.8 Higher Direct Images of Sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.9 Flat Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.10 Smooth Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.11 The Theorem on Formal Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.12 The Semicontinuity Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

4 Curves 84
4.1 Riemann‑Roch Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.2 Hurwitzʹs Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.3 Embeddings in Projective Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.4 Elliptic Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
4.5 The Canonical Embedding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.6 Classification of Curves in P3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

Yang Pi‑Yeh 3 Hartshorne Solutions

Content Content

5 Surfaces 100
5.1 Geometry on a Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.2 Ruled Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5.3 Monoidal Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.4 The Cubic Surface in P3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.5 Birational Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.6 Classification of Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

Appendix 116
A Intersection Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
B Transcendental Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
C The Weil Conjectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

Yang Pi‑Yeh 4 Hartshorne Solutions

1 Varieties

1 Varieties
1.1 Affine Varieties
Solution 1.1.1. (a) A(Y) = k[x, y]/(x2 − y) = k[x, x2 ]  k[x].
(b) A(Z) = k[x, y]/(xy − 1) = k[x, 1x ]  k(T ). Then we may assume x = f (T ) and 1x = g(T ) both polynomial,
and
1
1 = x · = f (T )g(T )
x
So deg( f g) = 0, i.e. deg( f ) = deg(g) = 0, which makes a contradiction.
(c)Assume that p = ax2 + bxy + cy2 + dx + ey + f . Then we will consider the following conditions.
√ √ √ √
(1) If b2 = 4ac, we have ax2 +bxy+cy2 = ( ax+ cy)2 . If we write t = ax+ cy, we have p = t2 + d̃t+ ẽy+ f ,
where ẽ might be zero. We take t as t − d̃2 , then p = t2 + ẽy + f˜ = t2 + s, where s = ẽy + f˜.
If s = 0 or a constant, then we know that p is line or lines, which is not a conic. If s involved a variable,
then
A(W) = k[t, s]/(t2 + s)  k[t].
√ √
(2) If b2 , 4ac, similarly we can change the coordinate as p = t2 +s2 +g. If g = 0, p = (t+ −1s)(t− −1s) =
0 is two lines. If g , 0, we may assume h = −1 by changing coordinate, and
√ √
A(W) = k[t, s]/(t2 + s2 − 1) = k[t + −1s, t − −1s]/(t2 + s2 − 1) = k[x, y]/(xy − 1) = A(Z).

Solution 1.1.2 (The Twisted Cubic Curve). Clearly I(Y) = (y−x2 , z−x3 ), and A(Y) = k[x, y, z]/I(Y) = k[x, x2 , x3 ] 
k[x]. So dim Y = dim A(Y) = 1.

Solution 1.1.3. If x , 0, then xz−x = 0 means z = 1, so x2 −yz = x2 −y. Then k[x, y, z]/(z−1, x2 −y) = k[x, x2 ]  k[x].
So we can denote X1 as Z(xz − x, x2 − yz, z − 1), then X1 is irreducible and X1 $ Y.
If x = 0, then xz − x = 0 means nothing, and x2 − yz = 0 means yz = 0. Then Z(yz) = Z(y) ∪ Z(z). Then
we may denote X2 = Z(x, y) and X3 = Z(x, z), and we have Y ∩ Z(x) = X2 ∪ X3 . And obviously X2 and X3 are
irreducible.

Solution 1.1.4. Just need to find a closed set in A2 which cannot be treated as product of two closed set in
A1 . Define X as the zeros of x = y in A2 . Then if X = Y × Z for some closed sets Y and Z in A1 , we may find
a , b ∈ A1 , and (a, a), (b, b) ∈ A2 , but (a, b) and (b, a) are not in X, which makes a contradiction.

Solution 1.1.5. If B is a finitely generated algebra over k, it can be written as k[x1 , . . . , xn ]/a for some integer
n and ideal a ∈ k[x1 , . . . , xn ]. Since B has no nilpotent, then a is radical. Denoting Y = Z(a), we have A/I(Y) =
√
A/ a = A/a = B.
Conversely, If B = k[x1 , . . . , xn ]/I(Y) for some n and Y, then B is clearly finitely generated. Since I(Y) is
radical, B is reduced.

Solution 1.1.6. (a) Firstly, X = Ȳ ∪ (X − Y). Since X is irreducible, we have X = Ȳ or X = X − Y. Since Y is

non‑empty, we have X = Ȳ, i.e. Y is dense in X. Secondly, if Y = (Y ∩ Y1 ) ∪ (Y ∩ Y2 ) for some closed Y1 and Y2 ,
we know that X = Y1 ∪ Y2 ∪ (X − Y). Similarly, X = Y1 or X = Y2 or X = X − Y. Hence X = Y1 or X = Y2 , i.e. Y is
irreducible.
(b) If Ȳ = Y1 ∪ Y2 for some closed Y1 and Y2 , then we have Y = (Y ∩ Y1 ) ∪ (Y ∩ Y2 ). Since Y is irreducible,
we have Y = Y ∩ Y1 or Y = Y ∩ Y2 . We may assume Y = Y ∩ Y1 , then Y ⊂ Y1 , i.e. Ȳ ⊂ Y1 . Hence Ȳ = Y1 , which
means Ȳ is irreducible.

Solution 1.1.7. (a) The (i⇔iii) and (ii⇔iv) is trivial. Then we only need to prove the (i⇔ii).

Yang Pi‑Yeh 5 Hartshorne Solutions

1 Varieties 1.1 Affine Varieties

(i⇒ii): If {Xα }α∈I is a set of closed subset of X, we can pick some X1 in it. If X1 is minimal, it’s proved. If
not, we may find some X2 ⊂ X1 , then we do the same discuss of X2 and so on. Then we may find a chain
X1 ⊃ X2 ⊃ . . .. Since X is noetherian, we have some Xn = Xn+1 = . . ., then Xn is minimal.
(ii⇒i): If X1 ⊃ X2 ⊃ . . ., then {Xi } is a set of closed subset of X, then {Xi } has minimal one, namely Xn . Then
Xn ⊃ Xn+1 ⊃ . . . Xn implies Xn = Xn+1 = . . ., i.e. X is noetherian.
S
(b) If X has an open covering {Xα }α∈I , then we may define Y = { f initecovering Xi |Xi ∈ {Xα }}. Since Y is a set of
open subsets of X, then Y has a maximal element, i.e. U = X1 ∪ . . . ∪ Xn . Since U is maximal, if U $ X, we can
find some X 0 ∈ {Xα }α∈I such that X 0 ∩ (X − U) , ∅. Then U 0 = X1 ∪ . . . ∪ Xn ∪ X 0 % U, which is contradict with
the maximality of U. Hence U = X, i.e. X has a finite covering.
(c) If Y ⊂ X is a subset, and Y ∩ U0 ⊂ Y ∩ U1 ⊃ . . . is an ascending chain of open sets in Y for some Ui upen
in X. Since X is noetherian, we have Un = Un+1 = . . ., then Y ∩ Un = Y ∩ Un+1 = . . ., then Y is noetherian.
(d) We firstly prove that X has a discrete topology. For any closed subset Y ⊂ X, we know X − Y is open.
Then for any y ∈ Y and z ∈ X −Y, we have some open sets Uyz and Vyz such that y ∈ Uyz , z ∈ Vyz and Uyz ∩Vyz = ∅.
S
We may assume Vyz ∈ X − Y, since we may change Vyz as Vyz ∩ (X − Y). Then X − Y = z∈X−Y Vyz for this fixed
S T
y, then this covering has a finite subcovering, i.e. X − Y = i Vyzi . We may define an open set Uy = i Uyzi ,
S
then we have Uy ∩ (X − Y) = 0, i.e. Uy ⊂ Y. Then Y = y∈Y Uy is an open covering, hence it has a finite
subcovering, i.e. Y is an union of finite open set, namely Y is open. Hence every closed set is open, X has a
discrete topology.
Secondly we will prove X is finite. Clearly if X is not finite, we have an infinite ascending chain {x1 } ⊂
{x1 , x2 } ⊂ . . ., which is contradict with that X is noetherian.

Solution 1.1.8. We may assume H = I( f ), and Z is an irreducible component of Y ∩ H. Denote f¯ as the image
of f under A  A/I(Y), then Y * H means ( f ) * I(Y), i.e. f < I(Y), hence f¯ is not a zero‑divisor. Since we
have prime ideal p in A(Y) containing f¯, then if f¯ is a unit, then p = (1), which makes a contradiction. Thus
Y ∩ H , ∅.
Since I(Z) is prime ideal in A, then the image of I(Z) under A → A(Y), as we denote as p, is a prime, and
contains f¯. By irreducibility of Z, we have that p is minimal over ( f¯). Then by Hauptidealsatz, height p = 1,
i.e. dim z = dim A(Y)/p = dim A(Y) − height p = r − 1.
S
Solution 1.1.9. If Z(a) = Yi , then we only need to prove that height I(Yi ) ≤ r, where we know that I(Yi ) is
the minimal prime ideal over a. Then by Krull’s height lemma, we have height I(Yi ) ≤ r. where we used that
a is generated by r elements.

Solution 1.1.10. (a) Any chain in Y is also a chain in X, trivial.

(b) By (a), we just need to prove that dim X ≤ sup dim Ui . If X0 ⊂ X1 ⊂ . . . ⊂ Xn is a chain in X, where
n = dim X. So X0 is just a point, then there must have some Ui such that X0 ⊂ Ui . Thus for every X j , we
have X j ∩ Ui , ∅. Then X j ∩ Ui is irreducible and dense in X j . So X0 ∩ Ui ⊂ . . . Xn ∩ Ui is a chain in Ui , then
dim X ≤ dim Ui ≤ sup dim Ui .
(c) X = {0, 1} with open sets ∅, {1}, {0, 1}. Then dim X = 2, but dim{1} = 1.
(d) If Y , X, we know any chain Y0 ⊂ Y1 ⊂ . . . ⊂ Yn in Y can be expressed as Y0 ⊂ . . . ⊂ Yn ⊂ X in X, which
is contracted with that dim Y = dim X.
(e) Take X = Z, and closed set of X are all finite subsets. Then {0} ⊂ {0, 1} ⊂ . . . means that dim X = ∞. And
every descent closed chain of X has a finite beginning, thus the chain if finite.

Solution 1.1.11. Firstly we will prove that dim Y = 1. As we can define a homomorphism ϕ : A1 → Y,
s 7→ (s3 , s4 , s5 ). If ϕ(s) = ϕ(t), then clearly s = t, i.e. ϕ is a bijection. Thus dim Y = dim A1 = 1.
Secondly we will prove that I(Y) cannot be generated by two elements. If f ∈ A, and f (t3 , t4 , t5 ) = 0, we
P
may assume f = ai jk xi y j zk . Then X
0= ai jk t3i+4 j+5k ,

Yang Pi‑Yeh 6 Hartshorne Solutions

1 Varieties 1.2 Projective Varieties

P
hence 3i+4 j+5k=s ai jk = 0. So all polynomial satisfying this make up I(Y). When s = 5, we have a001 = 0. When
s = 10, we have a210 + a002 = 0. So by calculation, y2 − xz, x3 − yz, x2 y − z2 ∈ I(Y), and they cannot be generated
by two elements.

Solution 1.1.12. Let f = (x2 − 1)2 + y2 . Then f is irreducible in R[x, y]. But Z( f ) is two points.

1.2 Projective Varieties

Solution 1.2.1. If Z(a) = ∅, we know that Z(a) in An+1 is ∅ or {0}. Both cases are trivial. If Z(a) , ∅, we know in
An+1 , Z(a) is a cone of projective Z(a). If f (p) = 0 for all p ∈projective Z(a), then in affine Z(a), f (λp) = 0 for all
p ∈affine Z(a) and λ ∈ k. Thus f (p) = 0 for all Z(a) − {0}. But when f is homogeneous and Z(a) , ∅ in Pn , f has
no constant term, thus f (0) = 0 in An+1 , which means f is vanishing in Z(a). Thus by affine Nullstellensatz,
we have f q ∈ a for some q, which means the homogeneous Nullstellensatz.
√
Solution 1.2.2. (i⇒ii): Z(a) = ∅, then in An+1 we have Z(a) = ∅ or {0}. Then in the first case, a = S , i.e. a = S .
√
In the second case we have a = I({0}) = S + .
√ √
(ii⇒iii): If a = S , then a = S , then trivial. If a = S + , then for every xi , we have xiri ∈ a. Thus taking
r = max{ri }, we know that xir ∈ a, i.e. S r(n+1) ⊂ a by pigeonhole principle.
(iii⇒i): If a ⊃ S d ⊃ {x0d , . . . , xnd }, then Z(a) ⊂ Z(x0d , . . . , xnd ) = ∅.

Solution 1.2.3. (a) If p ∈ Z(T 2 ), then T 2 vanishes on p, i.e. T 1 vanishes on p, hence p ∈ Z(T 1 ).
(b) If f ∈ I(Y2 ), then f vanishes on Y2 , i.e. vanishes on Y1 , hence f ∈ I(Y1 ).
(c) By (b), we have I(Y1 ∪ Y2 ) ⊂ I(Y1 ) ∩ I(Y2 ). Conversely, if f ∈ I(Y1 ) ∩ I(Y2 ), then f vanishes on both Y1
and Y2 , i.e. f vanishes on Y1 ∪ Y2 . Hence I(Y1 ) ∩ I(Y2 ) ⊂ I(Y1 ∪ Y2 ).
√
(d) If Z(a) , ∅, then we know that Z(a) in An+1 is not only {0}, i.e. I(Z(a)) = a.
(e)Obviously, Y ⊂ Z(I(Y)), and Z(I(Y)) is closed, then Ȳ ⊂ Z(I(Y)). Conversely, if W ⊃ Y is closed and
W = Z(a), we have a ⊂ I(Z(a)) ⊂ I(Y), i.e. W ⊃ Z(I(Y)). Hence we have Z(I(Y)) = Ȳ.

Solution 1.2.4. (a) If Y is closed, then Y = Ȳ = Z(I(Y)). Conversely, if a is a radical homogeneous ideal, and if
√ √
Z(a) , ∅. Then I(Z(a)) = a = a. If Z(a) = ∅, then a = S or S + . But a , S + by hypothesis, then I(Z(a)) = a = a.
(b) (⇒): If f, g ∈ I(Y), then Z( f g) = Z( f ) ∪ Z(g) ⊃ Y. Then Y = (Y ∩ Z( f )) ∪ (Y ∩ Z(g)). Since Y is irreducible,
we may assume Y = Y ∩ Z( f ), i.e. Y ⊂ Z( f ), then f ∈ I(Y).
(⇐): If Y = Y1 ∪ Y2 , then I(Y) = I(Y1 ) ∩ I(Y2 ). Since I(Y) is prime, we may assume I(Y) = I(Y1 ), i.e. Y = Y1 .
(c) Since I(Pn ) = (0) is prime, obviously.

Solution 1.2.5. (a) If Y1 ⊃ Y2 ⊃ . . . is a descent chain in Pn , then I(Y1 ) ⊂ I(Y2 ) ⊂ . . . in S is an ascent chain,
hence stable.
(b) Write S as all irreducible projective varieties which cannot be written as union of finite irreducible
varieties in Pn , then S has a minimal element since Pn is noetherian, namely Y. Y is reducible, thus Y = Y1 ∪ Y2
for some variety Y1 and Y2 such that Y1 $ Y and Y2 $ Y. Then Y1 , Y2 < S . So Y1 , Y2 can be written as a union of
finite irreducible varieties, so does Y, which makes a contradiction.

Solution 1.2.6. Consider ϕi : Ui → An in (2.2), then ϕi (Ui ∩ Y) is an affine variety. Then if xi ∈ I(Y), then xi = 0
in S (Y). So S (Y) xi is trivial. If xi < I(Y), we know that S (Y) xi consists of all elements xft for some homogeneous
i
polynomial f and positive integer t. Then (S (Y) xi ) consists of all f ( xx0i , . . . xi−1
xi
, xi+1
xi
, . . . , xxni ). So we can define an
α : (S (Y) xi )0 → A(Yi ) and α−1 ( f ) = xi− deg f f . Thus A(Yi )  (S (Y) xi )0 . Then

S (Y) xi = (S (Y) xi )0 [xi , xi−1 ]  A(Yi )[xi , xi−1 ].

Thus dim S (Y) = deg K(S (Y)) = deg K(S (Y) xi ) = deg K(S (Y) xi )0 + 1 = 1 + deg K(A(Yi )) = 1 + dim Y ∩ Ui = 1 + dim Y.

Yang Pi‑Yeh 7 Hartshorne Solutions

1 Varieties 1.2 Projective Varieties

Solution 1.2.7. (a) S (Pn ) = S , then by dim S = n + 1 we have dim Pn = n.

(b) Denoting Yi = Y ∩ Ui , we have Ȳi = Ȳ ∩ Ui , then dim Y = dim Yi = dim Ȳi = dim Ȳ.

Solution 1.2.8. (⇐): If Y = Z( f ), then S (Y) = S /( f ). By Hauptidealsatz we have height ( f ) = 1, then dim S (Y) =
dim S − height ( f ) = n, hence dim Y = n − 1.
(⇒): If dim Y = n − 1, then dim S (Y) = n, and I(Y) is prime. Then height I(Y) = dim S − dim S (Y) = 1, i.e.
I(Y) has a single non‑constant generator g, then I(Y) = (g), i.e. Y = Z(g).

Solution 1.2.9 (Projective Closure of Affine Variety). (a) If f ∈ I(Ȳ), we may define g = f (1, x1 , . . . , xn ), i.e.
f = β(g), which means g ∈ I(Y). Thus I(Ȳ) ⊂ β(I(Y)). Conversely, if g ∈ I(Y), then trivially f = β(g) ∈ I(Ȳ).
(b) In 1.1.2.(a) we have proved that I(Y) = (y − x2 , z − x3 ). Then we may denote f1 = y − x2 , f2 = z − x3 .
Clearly, Ȳ = {(s3 , s2 t, st2 , t3 } ⊂ P3 . Then x3 = zw2 , y3 = z2 w. We have that I(Ȳ) ⊃ (z2 w − y3 ). But β( f1 ) = yw − x2 ,
β( f2 ) = zw2 − x3 , and z2 w − y3 < (yw − x2 , zw2 − x3 ).

Solution 1.2.10 (The Cone Over a Projective Variety). (a) Clearly C(Y) is also the zero of I(Y), thus an algebraic
set in An+1 , and ideal is also I(Y).
(b) (⇒) If Y = Y1 ∪Y2 , we know that C(Y) = C(Y1 ∪Y2 ) = C(Y1 )∪C(Y2 ). If C(Y) = C(Y1 ) then θ(C(Y)) = θ(C(Y1 )),
i.e. Y = Y1 .
(⇐) If Y is irreducible, then I(Y) is prime. So C(Y) is irreducible.
(c) dim C(Y) = dim S (Y) = dim Y + 1.

Solution 1.2.11 (Linear Varieties in Pn ). (a) (⇒) If I(Y) = ( f1 , . . . , fr ), then every point in Y satisfies fi , i.e. Y is
the intersection of hyperplane Z( fi ).
(⇐) If Y is the intersection of Y1 , . . . , Yr , then we have I(Y) = (Z(Y1 ), . . . , Z(Yr )), and Z(Yi ) is an ideal gener‑
ated by a single linear polynomial.
(b) Write Z(Yi ) = ( fi ). Since f1 , . . . , f r are linear independent. Then height I(Y) = r, which means dim Y =
n − r.
(c) If r + s − n ≥ 0, we may consider the cone C(Y) and C(Z) in An+1 , and dim C(Y) = r + 1, dim C(Z) = s + 1.
Then we have dim(C(Y) ∩ C(Z)) ≥ r + s + 1 − n ≥ 1. Then C(Y) ∩ C(Z) , ∅ with dimension ≥ 1, i.e C(Y) ∩ C(Z)
is not only {0}, hence Y ∩ Z , ∅.

Solution 1.2.12 (The d‑Uple Embedding). (a) Since the image of θ is all polynomial in k[x0d , . . . , xnd ], which is
a domain. Thus kernal of θ is prime.
(b) If k ∈ ker φ, then f (M0 , . . . , MN ) = 0. Thus Imρd ⊂ Z(a). If f ∈ I(Im(ρd )), we know f (x) = 0 for all
x ∈ Im(ρd ), i.e. f (M0 , . . . , MN ) = 0. Thus I(Im(ρd )) ⊂ ker φ.
(c) Clearly ρd is injective. And in (b) we proved that Imρd = Z(a). Thus ρd is homomorphism.
(d) All monomials in degree 3 in 2 variables are x03 , x02 x1 , x0 x12 , x13 . Thus ρd (x0 , x1 ) = (x03 , x02 x1 , x0 x12 , x13 ).

Solution 1.2.13. If a curve in P2 defined by f (x, y, z) = 0, maps into Y, then we know that f 2 can be treated as a
polynomial of x2 , y2 , z2 , xy, yz, zx, which we denote as g(x2 , y2 , z2 , xy, yz, zx). Thus Z = ρ2 (Z( f )) = Y ∩Z(g) = Y ∩V.

Solution 1.2.14 (The Segre Embedding). Write points of PN as ci j as order to write ψ as ci j = ai b j . Then the
equations of Imψ is ci j · ckl = cil · ck j for all i, j, k, l. Hence the image of ψ is a subvariety of PN .

Solution 1.2.15 (The Quadric Surface). (a) We know ψ : P1 × P1 → P3 , (a0 , a1 , b0 , b1 ) 7→ (a0 b0 , a1 b1 , a0 b1 , a1 b0 ).

Then by 1.2.14. we know that the equation of Imψ is just xy − zw = 0.
(b) {Lt } are given by all ψ(P1 × {p} for all p ∈ P1 . And similarly {Mt } are given by all ψ({p} × P1 ).
(c) Denote the curve x − y = 0 in Q as Y, thus the same as 1.1.4.

Yang Pi‑Yeh 8 Hartshorne Solutions

1 Varieties 1.3 Morphisms

Solution 1.2.16. (a) If p = (x, y, z, w) ∈ Q1 ∩ Q2 , then P satisfies x2 = yw and xy = zw. So y2 w = x2 y = xzw, i.e.
x = w = 0 or y2 = xz, which is a union of a line and a cubic curve.
(b) If p = (x, y, z) ∈ C ∩ L. Then p satisfies x2 − yz = 0 and y = 0. So x = 0 and z = 1. Thus p has only one
solution, i.e. I(p) = (x, y). Moreover, I(C) = (x2 − yz), I(L) = (y), and I(C) + I(L) = (x2 , y) , I(p).

Solution 1.2.17 (Complex Intersections). (a) We treat a as a ideal in k[x0 , . . . , xn ], then it defines a Ya in An+1
as the cone of Y. By 1.1.9., we have that dim Ya ≥ n + 1 − q, i.e. dim Y = dim Ya − 1 ≥ n − q.
(b) If I(Y) = ( f1 , . . . , fr ), we may denote Yi = Z( fi ). Firstly, if p ∈ Y, we know that I(Y) on p is 0. Then
T T
fi (p) = 0, i.e. p ∈ Z( fi ). Secondly, if p ∈ Z( fi ), clearly we have fi (p) = 0, i.e. p ∈ Y.
(c) Let Y be all (s3 , s2 t, st2 , t3 ) ∈ P3 . We have I(Y) = (xy − zw, zw2 − x3 , xz − y2 ). And let H1 = Z(xy − zw), and
H2 = Z(zw2 − x3 ), we have Y = H1 ∩ H2 .

1.3 Morphisms
Solution 1.3.1. (a) In 1.1.1.(c) we have proved that for any conic W in A2 , we have A(W)  A(Y) or A(Z), where
Y  A1 and Z  A1 − {0}. Then by corollary 3.7., trivially.
Q
(b) If Y is a proper open subset of A1 , we may assume Y = A1 − {y1 , . . . , yn }. Defining Z = Z(x · (y − yi ) − 1),
Q
we have Y  Z. Then A(Z) = k[x, y]/(x · (y − yi ) − 1) = k[y, Q(y−y 1
i)
1
]. But Q(y−yi)
< k[y]  A(A1 ).
(c) Since the automorphism group of P acts transitively on any sets of 3 points which are not on a line, we
2

may assume that the conic contains (0, 0, 1), (0, 1, 0), (1, 0, 0), i.e. the conic is of the form like axy + byz + czx = 0
for some a, b, c. So changing x 7→ acx , y 7→ aby , z 7→ bcz , we have xy + yz + zx = 0, which is the image of 2‑uple
embedding of P1 , i.e. all conics are isomorphic to P1 .
(d) We will prove that any two curves in P2 have nonempty intersection in 1.3.7., but it’s clearly wrong
in A2 .
(e) Denote this variety by Y, then O(Y) = k. But if Yhas more than one point. Then we have a function
wich can differ those two points, hence more than only k. Thus Y has only one point.

Solution 1.3.2. (a) We have ϕ−1 (x, y) = t = yx , and ϕ−1 (0, 0) = 0. Then clearly ϕ is bijection and bicontinuous.
But ϕ−1 cannot be written as a polynomial of x and y, thus cannot be an isomorphism.
(b) If injection of ϕ follows from the definition and the character p of the field. And the surjection of ϕ
follows from the perfectness of k, but this inverse function cannot be an isomorphism.

Solution 1.3.3. (a) If f is regular at ϕ(P), then f is regular at a neighborhood ϕ(P) ∈ U. So f ◦ ϕ is regular at
neighborhood P ∈ ϕ−1 (U), which means ϕ induces a homomorphism ϕ∗ : Oϕ(P),Y → PP,X .
(b) (⇒) If ϕ is an isomorphism, it clearly a homeomorphism. So ϕ and ϕ−1 induce ϕ∗ and ϕ∗ −1 like in (a),
hence a isomorphism.
(⇐) Take ψi : (x1 , . . . , xn ) 7→ xi , then ϕ−1 is defined by all ψi ◦ (ϕ∗ )−1 .
(c) If ϕ∗P ( f ) = 0, then f |ϕ(x)∩V = 0. So by density of ϕ, we have f = 0.

Solution 1.3.4. Just construct the inverse, for x0 , 0,

ρ−1
d : ρd (P ) → P
n n

(M0 , . . . , MN ) 7→ (Mid−1 ,0 , Mid−1 ,1 , . . . , Mid−1 ,n ) for some i such that there are not all 0

where Mid−1 , j is the term corresponding to xid−1 xi . Since there must exist an i such that Mid−1 ,0 , Mid−1 ,1 , . . . , Mid−1 ,n
are not all zero, and for i , j satisfying this, we must have (M jd−1 ,0 , M jd−1 ,1 , . . . , M jd−1 ,n ) = (M jd−1 ,0 , M jd−1 ,1 , . . . , M jd−1 ,n )
because they are the same up to a nonzero scalar (xi /x j )d−1 , this map is well‑defined. And clearly on each affine
piece ρ−1
d (xi , 0) this is a morphism. So ρd is an isomorphism. Then ρd is regular at x0 = 1.

Yang Pi‑Yeh 9 Hartshorne Solutions

1 Varieties 1.3 Morphisms

Solution 1.3.5. STEP 1. We prove that Pn − H is affine, where H is a hyperplane.

Under some coordinates changing, we may assume that H = (x0 = 0), then trivial.
STEP 2. By STEP 1 and 1.3.4., we know that for any homogenious polynomial h of degree d, then Pn is
isomorphic to ρd (Pn ) ⊂ PN . So Pn − H is isomorphic to ρd (Pn − H) = ρd (Pn ) ∩ ρd (H). Since H is defined by (h),
we know that ρd (H) is a hyperplane, then ρd (Pn − H) is isomorphic to an affine variety.

Solution 1.3.6. If h = gf ∈ O(X), and f, g are coprime, then g can only vanish at (0, 0). But dim Z(g) = 1, and
dim{(0, 0)} = 0, so g has no zeros, i.e. g =constantneq0. Hence O = k[x, y]. Since O(A2 ) = k[x, y], we have
id ∈ Hom(O(X), O(A2 )). Then id ∈ Hom(A2 , X). But id((0, 0)) < X, which makes a contradiction.

Solution 1.3.7. (a) If X, Y are two curves in P2 , we have dim X + dim Y − 2 = 0 ≥ 0. Then by 1.2.11.(c), X, Y has
non‑empty intersection.
(b) If Y ∩ H = ∅, then Y ⊂ Pn − H. And Pn − H is affine, thus Y is an affine variety. Then by 1.3.1.(e), Y is
only one point. But Y has dimensiongeq1, which makes a contradiction.

Solution 1.3.8. Write X = Pn − (Hi ∩ H j ). Then if f ∈ O(X), we have f = gh for some homogeneous polynomial
with same degree, and h has only zeros in Hi ∩ H j , i.e. dim Z(h) ≤ dim Hi ∩ H j ≤ n − 2. But Z(h) must be a
hypersurface and has dimension n − 1, which makes a contradiction.

Solution 1.3.9. Clearly S (X) = k[x, y], S (Y) = k[x, y, z]/(xy − z2 )  k[x, y] ⊕ k[x, y]. And [S (Y) : S (X)] = 2, thus
S (X)  S (Y).

Solution 1.3.10 (Subvarieties). If X 0 = X ∩ U and Y 0 = Y ∩ V for some open sets U and V, since ϕ : X → Y is
a morphism, we know that for all regular function f : Y → k, f ◦ ϕ is regular. If g : Y 0 → k is regular, since V
is open, we know that g = hf for some regular function on Y and h on Y has only zeros on Y − Y 0 . Then since
ϕ(X 0 ) ⊂ Y 0 , we know that h has no zeros on ϕ(X 0 ), i.e. g ◦ ϕ|X0 is regular. Hence ϕ|X0 is a morphism.

Solution 1.3.11. This problem is local, hence we may assume X is affine. Since the subvarieties containing
P correspond to prime ideals of A(X) contained in mP , and those prime ideals correspond to prime ideals of
the ring OP , so clearly.

Solution 1.3.12. Firstly we may assume that X is affine. By 3.2.(c) the proof is finished. For general case,
we may assume X ∈ Pn , and denote p ∈ Xi = X ∩ Ui . Then we have dim O p (Xi ) = dim Xi = dim X. Clearly,
O p (X) ⊂ O p (Xi ). Inversely, if f ∈ O p (Xi ), f is regular on a neighbourhood p ∈ U ⊂ Xi . But U in X is also an
open set, so f ∈ O p (X), i.e. O p (X) = O p (Xi ). Thus dim O p = dim X.

Solution 1.3.13 (The Local Ring of a Subvariety). Obviously OY,X is a ring, then we only need to prove that
mY = {all function in OY,X which vanishes on Y} is the unique maximal ideal in OY,X , i.e. mY equals to the
Jacobian of OY,X . If f ∈ mY , and g ∈ OY,X , we know that 1 + f g , 0 on Y. So 1+1f g is well‑defined on Y, hence it’s
regular on an open set U which contains Y. Thus (1 + f g) has inverse on OY,X .
Moreover, dim X = dim k[x]/I(Y) + height I(Y) = dim k[Y] + height I(Y) = dim Y + dim OY,X .

Solution 1.3.14 (Projection from a Point). (a) Under some coordinates changing, we may assume that P =
(1, 0, . . . , 0) and Pn = (x0 = 0) ⊂ Pn+1 , then ϕ is (x0 , x1 , . . . , xn ) 7→ (0, x1 , . . . , xn ), hence a morphism.
(b) ϕ : (t3 , t2 u, tu2 , u3 ) 7→ (t3 , t2 u, 0, u3 ), i.e. on P2 , it is (t3 , t2 u, u3 ) which has equation y3 = x2 z.

Solution 1.3.15 (Products of Affine Varieties). (a) If X × Y = Z1 ∪ Z2 for two closed subsets Z1 and Z2 , we may
define Xi = {x ∈ X, {x} × Y ⊂ Ui } for i = 1, 2. For any point x ∈ X − X1 , {x} × Y is closed in X × Y, hence there
exists a open subset containing {x} × Y and disjoint to Z1 . So there exists an open subset in X containing x
and disjoint to X1 , i.e. X1 is closed. And similarly we know X2 is also closed. Moreover, for any x ∈ X, since
Y  {x}Y ⊂ Z1 ∪ Z2 , by the irreducibility of Y we know {x}Y must be contained in at least one of Z1 and Z2 ,

Yang Pi‑Yeh 10 Hartshorne Solutions

1 Varieties 1.3 Morphisms

so x ∈ X1 or x ∈ X2 . So X = X1 ∪ X2 . Since X is irreducible, we may assume X = X1 , i.e. X × Y = U1 , hence

irreducible.
(b) We can define ϕ : A(X)⊗k A(Y) → A(X×Y), ( f ⊗g) 7→ f g. Then we need to prove this ϕ is an isomorphism.
For the surjectivity, since the coordinate functions x1 , . . . , xm , y1 , . . . , yn ∈ A(X × Y) all have preimages, then
every function in A(X × Y) is a polynomial of these coordinates, hence it must have a preimage in A(X) ⊗k A(Y).
P
For the injectivity, if 0 , ki=1 fi ⊗ gi ∈ ker ϕ with a minimal k, we know f1 (x0 ) , 0 for some x0 ∈ X. since
Pk −1 Pk
i=1 fi (x0 )gi (y) = 0 for any y ∈ Y, we know g1 (y) = − f1 (x0 ) i=2 fi (x0 ) ⊗ gi (y). So this contradicts with the
minimality of k, hence ϕ is injective.
(c) (i) If f : X → k is regular, we know f ◦ ρX : X × Y → k is just polynomial of X, hence regular. So
ρX : X × Y → X is a morphism. And so does ρX : X × Y → Y.
(ii) If ϕX : Z → X, ϕY : Z → Y. By product, we have a unique morphism ϕX × ϕY : Z → X × Y.
(d) If X0 $ . . . $ Xn and Y0 $ . . . $ Ym is chains in X and Y, then we have X0 × Y0 $ . . . $ Xn × Y0 $
Xn × Y1 $ . . . $ Xn × Ym is a chain in X × Y. Hence dim X × Y ≥ dim X + dim Y. Conversely, we clearly have
dim X × Y ≤ dim X + dim Y.

Solution 1.3.16 (Products of Quasi‑Projective Varieties). (a) We just need to prove that Pn × Pm is a projective
variety. Take ϕ : Pn ×Pm → Pnm+n+m , ((x0 , . . . , xn ), (y0 , . . . , ym )) 7→ (x0 y0 , x0 y1 , . . . , x0 ym , x1 y0 , . . . , xn ym ). Then clearly
ϕ is isomorphic between Pn × Pm and ϕ(Pn × Pm ). And ϕ(Pn × Pm ) is define in Pnm+n+m by zi j zkl = zil zk j for every
i, j, k, l, where zi j is the term corresponding to the term xi y j . Thus Pn × Pm is a projective variety.
(b) Similar with (a).
(c) If ϕX : Z → X and ϕY : Z → Y, we can define (ϕX , ϕY ) : Z → X × Y, z 7→ (ϕX (Z), ϕY (z)), where we treat the
X × Y as a variety in Pnm+n+m . Then we have Z → X × Y.

Solution 1.3.17 (Normal Varieties). (a) Since every conic in P2 is isomorphic to P1 , it is smooth, hence normal.
(b) Since Jacobian matrix (y, x, −w, −z) has rank 1 on all points of Q1 , then Q1 is smooth, hence normal.
At the cone point of Q2 , since local ring is k[x, y, z]/(z2 − xy) and xy is square‑free, it is integrally closed,
hence normal. At other points, similarly by Jacobian we know that is smooth, hence normal.
(c) At the cusp, it is obviously singular, hence not normal.
(d) Y is normal ⇔ every O p is integrally closed ⇔ O(Y) is integrally closed, where the second arrow is
implied by Atiyah 5.12.
(e) Take A as th integral closure of k[Y] in k(Y). Then we just need to find a Ȳ to satisfy k[Ȳ] = A, i.e.
k[Ȳ] is finitely generated over k without zero‑divisors. Then we just need to prove that k[Ȳ]/k[Y] is finitely
generated.
By Noether normalization, there exists B ⊂ k[Y] such that B ⊃ k[T 1 , . . . , T r ] and k[Y] is integral over B.
Then k(T 1 , . . . , T r ) ⊂ k(Y) but
k(T 1 , . . . , T r ) ⊃ B ⊂ k[Y] ⊂ A ⊂ k(Y).

Thus A is the integral closure of B in k(Y), and k(Y) is a finite field extension of k(T 1 , . . . , T r ). Then B is integral
closed, thus A/B is finitely generated, hence A/k[Y].

Solution 1.3.18 (Projective Normal Varieties). (a) If Y is projectively normal, then S [Y] is integrally closed.
So every localization of S [Y] is integrally closed, i.e. Y is normal.
(b) Since Jacobian matrix is  
 w −z −y x
  ,
2xz −3y2 x2 0
it has rank n everywhere, so Y is normal. But S (Y) = S /(xw − yz, x2 z − y3 ) is not integral closed.
(c) We have ϕ : P1 → Y, (t, u) 7→ (t4 , t3 u, tu3 , u4 ) and ψ : Y → P1 , (x, y, z, w) → (x, y). Then ϕ ◦ ψ = id and
ψ ◦ ϕ = id, hence isomorphic.

Yang Pi‑Yeh 11 Hartshorne Solutions

1 Varieties 1.4 Rational Maps

Solution 1.3.19 (Automorphisms of An ). If ϕ is isomorphism, there exists a ψ such that ψ◦ϕ = id and ϕ◦ψ = id.
∂ fi
If xi = xi ( f1 , . . . , fn ) are the equations of ψ, we have ( ∂x j
) · ( ∂∂xf ij ) = I, i.e. we get the determinant J · J x/ f = 1 for
two polynomial J and J x/ f . Then J =non‑zero constant.

Solution 1.3.20. (a) If f is regular on Y − P, then f = hg for two polynomials and h|Y,p , 0. Since dim Y ≥ 2, the
zeros of h is empty or has at least dimension 1 on Y, hence h is no zeros, i.e. h =nonzero constant. So hg can
define on the whole Y, i.e. f extends to a regular function on Y.
(b) If dim Y = 1, we just take f = x−p
1
.

Solution 1.3.21 (Group Varieties). (a) For any regular function A1 → k, we have f ◦ µ : A2 → k is also a
regular function obviously. Hence it is a group variety.
(b) For any regular function A1 − {0} → k, we have f ◦ µ : (A1 − {0})2 → k is also a regular function.
(c) For any f ∈ Hom(X, G), we can define a morphism − f : X → G, x 7→ − f (x). And for any f, g ∈
Hom(X, G), we can define f + g : X → G, x 7→ f (x) + g(x). Then clearly − f and f + g are both in Hom(X, G),
hence Hom(X, G) has a group structure from the structure on G.
(d) Define ϕ : Hom(X, Ga ) → O(X), ( f : X → Ga ) 7→ f . ϕ is clearly a bijection, and for every f, g ∈
Hom(X, Ga ), we have f + g : X → Ga , x 7→ f (x) + g(x), so ϕ( f + g) = ϕ( f ) + ϕ(g), hence an isomorphism.
(e) Define ϕ : Hom(X, Gm ) → O(X), ( f : X → Gm ) 7→ f . Then for every f ∈ Hom(X, Gm ), we have f , 0,
i.e. 1f ∈ O(X), hence Imϕ = O × (X). For every f, g ∈ Hom(X, Gm ), we have f g : X → Gm , x 7→ f (x)g(x), then
ϕ( f g) = ϕ( f )varphi(g), hence an isomorphism.

1.4 Rational Maps

Solution 1.4.1. Define h on U ∪ V as 


 x∈U
 f (x),
h(x) = 


 g(x), x∈V

Since f |U∩V = g|U∩V , h is well‑defined. Then we need to prove that h is regular. Thus if f = ff21 , g = gg12 , for some
f1 , f2 ∈ k[U] and g1 , g2 ∈ k[V]. On the U ∩ V\( f2 g2 = 0), we have ff12 = ff12 gg22 = ff22 gg12 = gg12 . Then h is regular on
U\V, V\U and U ∩ V, hence it is regular.

Solution 1.4.2. A map is regular iff it is regular in a neighbourhood of every point, so by 1.4.1. and Zorn’s
lemma, there exists a largest subset U of X such that f is regular on U.

Solution 1.4.3. (a) Clearly f is defined on the open subset x0 , 0, which is isomorphic to A1 . And f is the
projection of P2 → A1 of its second coordinate.
(b) We may define ϕ : P2 → P1 , (x0 , x1 , x2 ) 7→ (x0 , x1), which is defined on the open set P2 − {(0, 0, 1)}. So ϕ
is just the projection of first two coordinates.

Solution 1.4.4. (a) By 1.3.1.(c), every conic in P2 is isomorphic to P1 , hence birational equivalent to P1 .
(b) We can define ϕ : A1 → (y2 = x3 ), t 7→ (t2 , t3 ), and ψ : (y2 = x3 ) → A1 , (x, y) 7→ yx . Then (y2 = x3 ) is
birational equivalent to A1 , and A1 is birational equivalent to P1 .
(c) Clearly, ϕ : Y → P1 , (x, y, z) 7→ (x, y), and ψ : P1 → Y, (x0 , x1 ) 7→ (x12 x0 − x13 , x13 − x02 x1 , x03 ). Then Y is
birational equivalent to P1 .

Solution 1.4.5. Since ϕ : Q → P2 , (x, y, z, w) 7→ (x, y, z), and ψ : P2 → Q, (x0 , x1 , x2 ) 7→ (x0 x2 , x1 x2 , x22 , x0 x1 ), hence
birational equivalent. But not isomorphic because P1 × P1 is not isomorphic to P2 .

Solution 1.4.6 (Plane Cremona Transformations). (a) Define U = (xyz , 0). Since ϕ2 : U → U, (x, y, z) 7→
(x2 yz, xy2 z, xyz2 ) = (x, y, z), hence ϕ is birational.
(b) Just take U = V = (xyz , 0).
(c) On U → V, it is just (x, y, z) 7→ ( 1x , 1y , 1z ).

Yang Pi‑Yeh 12 Hartshorne Solutions

1 Varieties 1.5 Nonsingular Varieties

Solution 1.4.7. We may assume that X, Y are affine since it is a local problem, and P = Q = 0. Then A(X)m0 
A(X)m0 induced k(X)  k(Y), hence a birational equivalence between X and Y. Under the map of images gfii
of xi in k(Y). Since xi is not invertible in A(X)m0 , we have gi (xi ) , 0, fi (xi ) = 0, then ( gf11 (0), . . . , gfnn (0)) = 0, i.e.
P 7→ Q. Then the others is under corollary 4.5.

Solution 1.4.8. (a) Firstly, k is algebraic closed, then |k| is infinite, i.e. |An | = |kn | = |k|. Then |Pn | = |An ∪ An−1 ∪
. . . ∪ A0 | = (n − 1)|k| + 1 = |k|. Then for general X, if X ⊂ Pn for some n, we have |X| ≤ |Pn | = |k|. And for
some open affine U ⊂ X which dim U ≥ 1, so by normalisation there exists a finite surjection U → Adim U , so
|U| ≥ |Adim U | ≥ |k|. Then |k| ≤ |U| ≤ |X| ≤ |Pn | ≤ |k|, hence |X| = |k|.
(b) Any two curves over k has same cardinality, then they are homeomorphic under cofinite topology.

Solution 1.4.9. Since X has dimension r, we may assume xx10 , . . . , xx0r ∈ k(X) is algebraic independent. And we
P P
have some f ∈ k(X) such that ( xx10 , . . . , xx0r , f ) generates k(X), and f = x0−1 ni=r+1 ai xi . Define H1 = ( ai xi = 0) and
H2 = (xn+1 = 0), and P = (0, . . . , 0, 1, 0, . . . , 0), where 1 is the r + 2‑th term. By construction, f is taken to xxr+1 0
under transformation H1 → H2 , then we may assume P < X, and xx10 , . . . , xx0r is the transcendence basis of k(X),
and xxr+1
0
= f.
Define H2 = (xr+2 = 0) and π : X → X 0 is the projection about P and H2 . Then by this construction,
k(X)  k(X 0 ), hence birational.

Solution 1.4.10. Easily we know Ȳ = Z(y22 − x1 y21 , y32 − x2 y31 ) ⊂ A2 × P1 . Supposing x1x2 , 0, x1 y2 = x2 y1 and
x13 = x22 , we have y2 = x2x1y1 , y22 = x1 y21 and y32 = x2 y31 . So Ȳ ⊃ φ−1 (Y − {0}).
If (0, 0, y1 , y2 ) ∈ Ȳ, then y2 = 0, i.e. (0, 0, 1, 0) ∈ Ȳ.

1.5 Nonsingular Varieties

Solution 1.5.1. (a) Tacnode. (b) Node. (c) Cusp. (d) Triple Point.

Solution 1.5.2. (a) Pinch Point. (b) Conical double point. (c) Double line.

Solution 1.5.3 (Multiplicities). (a) Clearly, µP (Y) = 1 ⇔ f (P) = 0 and ∇ f P , 0 ⇔ P is nonsingular.

(b) µ(0,0) (a) = µ(0,0) (b) = µ(0,0) (c) = 2, µ(0,0) (d) = 3.

Solution 1.5.4 (Intersection Multiplicity). (a) We may assume that P = (0, 0). Firstly, take a neighbourhood
U of P such that (Y ∩ Z) ∩ U is only point T . Then we may take IP ⊂ A(U), by Nullstellensatz, IPr ⊂ ( f, g)
for some r > 0. So in OP , we have mrP ⊂ ( f, g). Thus we just need to show OP /mrP has finite length. Since
OP /mrP ⊃ mP /mrP ⊃ . . . ⊃ mr−1 P /mP ⊃ 1, and (mP /mP )/(mP /mP )  mP /mP has finite dimension, then O p /mP has
r i r i+1 r i i+1 r

finite length.
Secondly, denote the length of OP /( f, g) = l, m = µP (Y) and n = µQ (Y). We may assume m ≤ n. Since
there exists a chain of length mn in k[x, y]/( f, g), this chain induces a chain in OP /( f, g) = k[x, y](x,y) /( f, g). Thus
l ≥ mn.
(b) If L is not in the tangent cone of Y at P (only finite lines in the tangent cone), we may assume P = (0, 0),
and L = (y = 0). Then f = fm +higher terms and fm = xm + y · (polynomial of degree m − 1). So OP (y, f ) =
k[x, y](x,y) /(y, xm + y(. . .)) = k[x, y](x,y) /(y, xm )  k[x]/xm has length m, so (L · Y)P = m = µP (Y).
(c) Under some coordinate changing, we may assume L∩Y does not happen in x2 = 0 in P2 . So this problem
will be considered in A2 , and L ∩ Y does not in infinity. We may assume that L = (y = 0), and Y = ( f = 0)
for some f ∈ k[x, y]. And Y ∩ L does not in infinity means that the highest order of f is fd = xd + y · ((d − 1)‑
dimensional terms). Then points in Y ∩ L corresponds to the zero of fd (x, 0), i.e. the multiplicity of P = (α, 0)
is the length of OP (y, f )  k[x, y](x−a,y) /(y, f )  k[x, y](x−a) /( fd (x, 0)). Thus this module has length which equals
P P
to the multiplicity ma of a at fd (x, 0). Hence p (Y, L) p = a ma = d.

Yang Pi‑Yeh 13 Hartshorne Solutions

1 Varieties 1.5 Nonsingular Varieties

Solution 1.5.5. If char(k) = 0 or char(k) - d, we just need to take f = x0d + x1d + x2d . If char(k)|d, we just need to
f = x0d−1 x1 + x1d−1 x2 + x2d−1 x0 .

Solution 1.5.6 (Blowing Up Curve Singularities). (a) If Y is given by f = y2 − x3 + x4 + y4 , for coordinates

x, y, t, u in A2 × P1 and xt − yu = 0, we may consider U = (t = 1). Then φ−1 (Y) ∩ U is given by f = 0 and x = yu,
i.e. 


 x = yu




y2 (1 + u4 y2 + y2 − u3 y) = 0

Denote W = φ−1 (Y) ∩ U. Then W ∩ φ−1 (Y − {0}) satisfies x = yu and g = 1 + u4 y2 + y2 − u3 y = 0, i.e. I(Ỹ ∩ U) =
(x − yu, g). Since Ỹ ∩ U ∩ E = ∅, we have φ on Ỹ − (Ỹ ∩ E) is isomorphic. Thus Ỹ ∩ U is nonsingular.
Denote U 0 = (u = 1), then similarly we have I(Ỹ ∩ U 0 ) = (h, y − xt) for h = t2 + x2 + t4 x2 − x. Since
Ỹ ∩ U 0 ∩ E = (0, 0, 0, 1), out of this point the Ỹ is nonsingular. And at this point, its Jacobian is
   
2x(1 + t4 ) − 1 0 2t + 4t3 x2   −1 0 0
   =  
−t 1 −x x=y=t=0 0 1 0

has rank 2, hence nonsingular.

If Y is given by f = xy − (x6 + y6 ), we may take U = (t = 1). Then φ−1 (Y) ∩ U is given by x = uy and
uy − y6 (1 + u6 ). Similarly we have that Ỹ ∩ U is given by
2




 x = yu




u − y4 (1 + u6 ) = 0

Then Ỹ ∩ U ∩ E = (0, 0, 1, 0). And its Jacobian is

   
0 −4y3 (1 + u6 ) 1 − 6x4 u5   0 0 1
   =  
1 −u −y x=y=u=0 1 0 0

has rank 2. Thus Ỹ ∩ U is nonsingular.

Take U 0 = (u = 1). Similarly φ−1 (Y) ∩ U 0 is given by tx = y and tx2 − x6 (1 + t6 ) and Ỹ ∩ U 0 is given by y = x
 
and tx2 − x6 (1 + t6 ). Then Ỹ ∩ U 0 ∩ E = (0, 0, 0, 1). And its Jacobian 00 01 10 has rank 2. Thus Ỹ ∩ U 0 is also
nonsingular.
(b) We may assume P = (0, 0) and Y is given by f = f2 +higher terms, where f2 is homogeneous polynomial
in degree 2. Under some coordinate changing, we may assume that f2 = xy. So similarly with (a) we have
Ỹ ∩ U ∩ E = (0, 0, 1, 0) and Ỹ ∩ U 0 ∩ E = (0, 0, 0, 1), and both two points are nonsingular.
(c) If Y is given by f = x2 −(x4 +y4 ), we know in U = (t = 1), φ−1 (Y)∩U is given by x = yu and y2 u2 −y4 (1+u4 ),
hence Ỹ ∩ U is given by x − yu = 0 and g = u2 − y2 (1 + u4 ) = 0. Then Ỹ ∩ U ∩ E = (0, 0, 1, 0) and the Jacobian of
this point is
   
0 −2y(1 + u4 ) 2u − 4y2 u3   0 0 0
   =  
1 −u −y x=y=u=0 1 0 0
which has rank 1, hence node.
(d) On U = (t = 1), φ−1 (Y) ∩ U is given by x = yu and x(x2 − u3 ) = 0, i.e. gives a cusp. Then one more
blow‑up gives a smooth point.

Solution 1.5.7. (a) At P = (0, 0, 0), all partials are zero, so P is singular. If Q = (a, b, c) , P, we may assume
c , 0, and denote g(x, y) = f (x, y, 1). Since Y is nonsingular, g x , gy will not be simultaneously zero. Sense
f x (a, b, c) = cd−1 f x ( ac , bc , 1), f x , fy will not be simultaneously zero, hence nonsingular.
(b) Take x, y, z, s, t, u as the coordinates in A3 × P2 . Denote U = (s = 1). Then X̃ ∩ U is given by y = tx and

Yang Pi‑Yeh 14 Hartshorne Solutions

1 Varieties 1.5 Nonsingular Varieties

z = ux and f (x, tx, ux) = 0. Since f (x, tx, ux) = ud f (1, t, x). Thus X̃ ∩ U is given by y = tx, z = ux and f (1, t, u) = 0.
Denote g(t, u) = f (1, t, u). Since f is nonsingular, g is nonsingular too. Thus X̃ is nonsingular.
(c) Since X̃ ∩ U is isomorphic to Z( f (1, t, u)) ⊂ A3 , we know that X̃ ∩ U ∩ E is isomorphic to Z(g(t, u)) ⊂ A3 .
Similar to (t = 1) and (u = 1), we have that φ−1 (0)  X̃ ∩ E  Y.

Solution 1.5.8. Since f is homogeneous, rank (J)|P is independent of the coordinate of P. So we may assume
that x0 = 1 and denote gi (x1 , . . . , xn ) = fi (1, x1 , . . . , xn ). Then (∂xi f j )|P is (∂xi g j )|P adding (∂x0 fi )|P at the top. By
Euler’s lemma, these two matrices have the same rank. Then by some condition in affine case, the exercise
has proved.

Solution 1.5.9. If f = gh, then by 1.3.7., there exists a point P such that g(P) = h(P) = 0. Then

∂ x f (P) = g(P) · ∂ x h(P) + h(P) · ∂ x g(P) = 0

∂y f (P) = g(P) · ∂y h(P) + h(P) · ∂y g(P) = 0 ∂z f (P) = g(P) · ∂z h(P) + h(P) · ∂z g(P) = 0

which clearly makes a contradiction.

Solution 1.5.10. (a) By 5.2.A, we know dim T P (X) = dim m/m2 ≥ dim X. And the equality holds iff OP is
regular.
(b) Since we have local ring morphism Oϕ(P) → OP , we have T P (ϕ) : T P (X) → T ϕ(P) (Y).
(c) Since ϕ∗ : O(0) → O(0,0) makes m(0) 7→ m2(0,0) . And m(0) = (x), we have x 7→ (x = y2 ) ∈ m2(0,0) , i.e. a zero
map.

Solution 1.5.11 (The Elliptic Quartic Curve in P3 ). Denote f1 = x2 − xz−yw, f2 = yz−xw−zw and g = y2 z− x3 +xz2 ,
π : P3 → P2 as (x, y, z, w) 7→ (x, y, z). First, since g = (x + z) f1 + y f2 , we know f1 = f2 = 0 implies g = 0, i.e.
π(Y − P) = W. Second, if g(a, b, c) = 0 and Q = (a, b, c) , (1, 0, −1) = P0 , we have b , 0 or a + c , 0.
In the first case, define ψ(x, y, z) = (x, y, z, x −xz
2
y
). In the second case, define ψ(x, y, z) = (x, y, z, x+z
yz
). Thus we
0 0
construct a map ψ : (W − P ) → Y − P as the inverse of ϕ : Y − P → W − P . So by 1.5.9. we know that W is
irreducible, hence W − P0 is, then Y − P, hence Y is.

Solution 1.5.12 (Quadric Hypersurfaces). (a) Since every quadric form has form xT Ax for some symmetric
h i
matrix A, and k is algebraic closed, so A is similar to 0I 00 of some Ir×r .
(b) If f = gh is reducible, g and h must be linear form. If g and h are proportional, gh = x02 . And if g and h
are nor proportional, gh = x02 + x12 .
(c) Since for any P ∈ SingQ, we have that every partial derivatives are zero, i.e. x1 = . . . = xr = 0. Thus
SingQ is variety for dimension n − r − 1.
(d) Define Pr = (xr+1 = . . . = xn = 0), and Q0 = Q ∩ Pr . If P is a point on line L of R ∈ Q0 , and R0 ∈ Z = SingQ.
Then we may define P = (aR0 , . . . , aRr , bR0r+1 , . . . , bR0n ). Then f (R) = 0 implies f (P) = 0. Conversely, if P =
(Pi ) ∈ Q, we know that R = (P0 , . . . , Pr , 0, . . . , 0) has f (R) = 0, i.e. R ∈ Q0 , and (0, . . . , 0, Pr+1 , . . . , Pn ) ∈ SingQ.

Solution 1.5.13. We may assume X is affine, and denote the integral closure A(X) ∈ K(X) as A, and A =
( f1 , . . . , fm ). If fi = ghii , fi is defined exactly on D(hi ). Then OX is integrally closed⇔ fi ∈ OX ⇔ hi (X) , 0 for all
T
i. Then X is normal⇔ x ∈ D(hi ), i.e. open.

Solution 1.5.14 (Analytically Isomorphic Singularities). (a) This isomorphic induces mrP 7→ mrQ , so mrP /mr+1
P 7→
mQs /mQs+1 implies r = s, i.e. µP (Y) = µQ (Z).
(b) Similar process, we have fr+1 = ht g s+1 + g s ht+1 , and so on.
(c) If f has an ordinary 2‑fold point on P, hence we may assume f = f1 f2 . Then we have an automorphism
ϕ of k[[x, y]] such that f1 7→ x and f2 7→ y. Hence ϕ induces an isomorphism ϕ : k[[x, y]]/(xy) → k[[x, y]]/( f ).
Thus every ordinary 2‑fold point is analytically isomorphic to P = 0 on g = xy.

Yang Pi‑Yeh 15 Hartshorne Solutions

1 Varieties 1.6 Nonsingular Curves

If f has an ordinary 3‑fold point on P, hence we may assume f = f1 f2 f3 . Then we have a unique automor‑
phism ϕ of k[[x, y]] such that f1 7→ x, f2 7→ ay and f3 7→ x + y for some a ∈ k. Hence ϕ induces an isomorphism
ϕ : k[[x, y]]/(xy(x + y)) → k[[x, y]]/( f ). Thus every ordinary 2‑fold point is analytically isomorphic to P = 0 on
g = xy(x + y).
Denote ga = xy(x+y)(x+ay) for some a ∈ k. If ϕ : k[[x, y]]/(ga )  k[[x, y]]/(gb ), we know that ϕ : (x, y) 7→ (x, y)
and (ga ) 7→ (gb ), i.e. x 7→ x, y 7→ u1 y, x + y 7→ u2 (x + y) and x + ay 7→ u3 (x + by). Then clearly u1 = u2 = u3 = 1, so
a = b. So, if a , b, ga is not isomorphic to gb .
(d) If y2 − xr and y2 − x s for some r ≤ s are analytically isomorphic at P = 0, we have automorphism ϕ of
k[[x, y]] as y2 − xr 7→ u · (y2 − x s ) = y2 + (higher terms of y) − x s + (higher terms with only x), where u = 1+higher
terms. Denote l x as a linear term of ϕ(x), and ly as a linear term of ϕ(y). So ϕ(y)2 = y2 + ϕ(x)r + (terms of
degree> 2). Then we have ly2 = y2 , i.e. ly = y. Thus we may assume l x = ax + by, and clearly a , 0. So r = s, by
comparing terms of x in above equation.
Then for any double points on f (x, y), if f = f2 g for some g ∈ (x, y)3 , and f2 = l1 l2 with l1 , l2 , we may
as in above, map ( f2 ) 7→ (x2 − y2 ). If f2 = l2 , we may assume l = x under some coordinate changing. Since
g ∈ (x, y)3 , we can find r ≥ 3 and h ∈ m, such that g = −(ax + y)r + h(x2 − (ax + y)r ) for some a ∈ k. Then we have
an automorphism ϕ of k[[x, y]] as x 7→ x and y 7→ ax + y. So (1 + f ) ◦ ϕ(x2 − yr ) = (1 + f )(x2 − (ax + y)r ) = x2 + g.
Hence (x2 − yr ) 7→ (x2 + g), i.e. f is analytically isomorphic to x2 − yr .

Solution 1.5.15 (Families of Plane Curves). (a) Clearly we have a correspondence

(points of PN ) ←→ (algebraic sets in P2 )

X
(arst )r+s+t=d −→ arst xr y s zt = 0
r,s,t
X
(arst )r+s+t=d ←− algebraic set ⊂ P2 defined by f = arst xr y s zt = 0
r,s,t

∂f ∂f
(b) The 1‑1 correspondence is in (a). Then by elimination theorem, for ∂x0
, . . . ∂xn
), there are polynoials in
coefficients of f which are zero when f is singular or reducible.

1.6 Nonsingular Curves

Solution 1.6.1. (a) Since every non‑singular curve is isomorphic to some abstract nonsingular curve, hence
isomorphic to an open set of some projective curve. And since Y is not isomorphic to P1 , we know Y must be
isomorphic to some open subset of A1 .
Q
(b) We may assume Y  A1 − {a1 , . . . , an } by (a), then Y is isomorphic to y · (x − ai ) = 1 in A2 , hence affine.
(c) Since A(Y)  k[x](x−a1 ,...,x−an ) , it is clearly a UFD with all primes (x − ai ).

Solution 1.6.2 (An Elliptic Curve). (a) If P = (x, y) ∈ Y is a singular point, we have y2 = x3 − x, 2y = 0
and −3x2 + 1 = 0, which have no zeros, i.e. Y is nonsingular. Thus y2 − x3 + x is irrducible, hence A(Y) =
k[x, y]/(y2 − x3 + x) is a domain. Moreover, since Y is nonsingular, hence normal, i.e. A(Y) is integral closed.
(b) k[x] is obviously a polynomial ring. Since y2 ∈ k[x], then y is in the closure of k[x], i.e. A ⊂the closure
of k[x]. Conversely, since A is integral closed, then equals.
(c) If f (x, y) ∈ A, it must have the form y·g(x)+h(x) for some g, h ∈ k[x]. So N( f ) = h2 −y2 g2 = h2 −g2 (x3 − x) ∈
k[x]. And clearly, N(1) = 1 · 1 = 1, N(a · b) = N(a) · N(b).
(d) If a ∈ A is a unite, there exists a b ∈ A such that ab = 1. Then N(a)N(b) = 1 in k[x], i.e. N(a) ∈ k× . If
a = yg + h, then h2 − g2 (x3 − x) ∈ k× , i.e. g = 0 and h ∈ k× . So a = h ∈ k× . Since we clearly notice that no element
has norm with degree 1, x and y are clearly irreducible in A. Finally, y2 = y · y = x(x − 1)(x + 1) has two ways
of factorization, i.e. A is not a UFD.

Yang Pi‑Yeh 16 Hartshorne Solutions

1 Varieties 1.7 Intersections in Projective Space

(e) Since A is non‑trivial and not a UFD, by 6.1.(c), A is not isomorphic to A(X) for any rational curve X,
hence Y is not a UFD.

Solution 1.6.3. (a) A2 − {(0, 0)} → P1 , (x, y) 7→ (x, y).

(b) P1 − {(1, 0)} → A1 , (x, y) 7→ yx .

Solution 1.6.4. f induces a map Y → A1 as x 7→ f (x). Then by 6.8. we have ϕ : Y → P1 . If Y is irreducible, since
f is non‑constant, Imϕ = P1 and ϕ is dominant, we know that ϕ induces a ϕ∗ : k(Y) → k(P1 ). If P ∈ P1 , ϕ−1 (P) is
closed. Since ϕ is non‑constant, ϕ−1 (P) are just finite points. If Y is reducible, we may assume Y = Y1 ∪ . . . ∪ Yn .
So for every Yi , ϕ−1 (P) ∩ Yi is finite, so ϕ−1 (P) is finite.

Solution 1.6.5. Since X̄ is a curve, if x ∈ X̄ − X, by 6.8, for any X → X can induce a X ∪ {x} → X, which makes
a contradiction. Hence X = X̄.

Solution 1.6.6 (Automorphisms of P1 ). (a) If ad − bc , 0, then x 7→ ax+b

cx+d
has inverse x 7→ (ad − bc)−1 −cx+a
dx−b
, hence
an automorphism.
(b) By 6.12, clearly AutP1 = Autk(x).
(c) If ϕ ∈ Autk(x), we may assume ϕ(x) = g(x)
h(x)
for some (g, h) = 1. Then ψ = ϕ−1 is rational, i.e. if x , y,
ϕ(x) , ϕ(y). But if f or g has degree> 1, h = a has more than 1 solution, which makes a contradiction. So f
g

and g are linear transformation. So PGL(1)  Autk(x)  AutP1 .

Solution 1.6.7. Denote ϕ : A1 − {P1 , . . . , P s }  A1 − {Q1 , . . . , Qt }. Then ϕ can be extended as isomorphism

ϕ : P1 → P1 , which gives a 1‑1 correspondence between {P1 , . . . , P s } and {Q1 , . . . , Qt }.
Since any set with at most 3 points in P1 can be mapped to any other set with same size under Aut(P1 ).
so the converse if true when r ≤ 3, but cannot hold in case r ≥ 4.

1.7 Intersections in Projective Space

 x+n   (xd)n
Solution 1.7.1. (a) Clearly Pd−uple (x) = Pn (xd), and Pn (x) = n
. So Pd−uple = xd+n
n
= n!
+lower‑terms, hence
deg(d − uple) = dn .
   x+s
xr x s
(b) Since Pr×s (x) = Pr (X)×P s (x). So Pr×s (x) = x+r · s = · +lower‑terms, hence deg(Segre embedding) =
r+s r r! s!
r!·s!
(r+s)!
= r .

Solution 1.7.2. (a) Clearly PPn (0) = 1, then pa (Pn ) = 0.

   x−d+2
(b) If Y is a plane curve in degree d, then PY (x) = x+2 − 2 = d2 (2x − d + 3). Then pa (Y) = 12 (d − 1)(d − 2).
2    x−d+n  
(c) If H is any hypersurface of degree d, then PH (x) = x+n − n . So pa (H) = d−1 .
 x+3  x−a+3  x−b+3  x−a−b+3 n n
(d) Clearly PY (x) = 3 − 3 − 3 + 3
. So pa (Y) = 2 ab(a + b − 4) + 1.
1

(e) Since PY×Z (x) = PY (x)·PZ (x), then pa (Y×Z) = (−1)r+s (PY (0)·PZ (0)−1) = pa (Y)pa (Z)+(−1) s pa (Y)+(−1)r p1 (Z).

Solution 1.7.3 (The Dual Curve). Clearly if Pis nonsingular and Y is defined by f . So we may denote P =
(x0 , y0 , z0 ), and the coordinates of tangent line in ( ∂∂xf , ∂∂yf , ∂∂zf )|(x0 ,y0 ,z0 ) , which is well‑defined. Thus P 7→ T P (Y) is
a morphism.

Solution 1.7.4. If a line meets Y in d points, this line is not a tangent line, and does not pass through any
singular point. Since Y has only finite points, all lines pass through and form a closed set of P2∗ . So by 1.7.3.
all tangent lines contain in a closed set of P2∗ . Thus out of these two closed sets is an open set U.

Solution 1.7.5. (a) If Y has a point with multiplicity≥ d, we may assume it is P = (0, 0). Then if f defines Y,
f has all terms of degree d, i.e. a product of linear factors, which is contradict with that Y is irreducible of
d > 1.

Yang Pi‑Yeh 17 Hartshorne Solutions

1 Varieties 1.8 What is Algebraic Geometry?

(b) In A2 we may assume that f defining Y has the form f = g + h with g homogeneous of degree d − 1
and h homogeneous of degree d. Then denoting t = yx , we have f = 0 ⇔ y = − g(t,1)
h(t,1)
and x = yt, which is the
inverse rational map of Y → A1 . Hence Y is rational.

Solution 1.7.6 (Linear Varieties). (⇐) Trivial. (⇒) Since Y has pure dimension r, by 7.6., Y is irreducible.
And since the Hilbert polynomial of divisor variety has leading term as linear degree, if Y has degree 1, for
hyperplane H not containing Y, we have deg(Y ∩ H) = 1, i.e. linear.

Solution 1.7.7. (a) Denote the hyperplane of infinity in Pn as H. Then we may assume that H does not contain
P and Y. So we can define X → Cone(Y) as PQ−(points in infinity) 7→affine line on Cone(Y), which has Q 7→ Q
and P 7→the cone point. Thus this map is birational, hence X is a variety with dimension= dim(Cone(Y)) = r+1.
(b) If dim Y = 0, the deg Y = d means Y has d points, i.e. X has d − 1 lines. If dim Y = r > 0, H is hyperplane
containing P but not containing Y, by 7.7. and 7.6., deg X ∩ H = deg X. But deg X ∩ H ≤ deg Y ∩ H < deg Y < d.

Solution 1.7.8. 1.7.7. means Y is contained in a variety H with degree 1, dimension r. So 1.7.6. means that H
is a linear variety, which is isomorphic to Pr+1 . So Y is isomorphic to quadric hypersurface in Pr+1 .

1.8 What is Algebraic Geometry?

The reason of my death.

Yang Pi‑Yeh 18 Hartshorne Solutions

2 Schemes

2 Schemes
2.1 Sheaves
Solution 2.1.1. Denote the constant presheaf as P. Then for any x ∈ X, we just need to prove that P x =
A x . Trivially we have P x = A. Otherwise, for any open set U containing x, we may denote the connected
component of x in U as V. Then V ⊂ U is open, i.e. A (V) = A. So A x = lim A (U) = A. i.e. A x = P x for any
−−→ x∈U
x ∈ X. Thus A = P + .

Solution 2.1.2. (a) Since the sheaf of kernal is just the presheaf of kernal, we clearly have (ker ϕ)P = lim (ker ϕ)(U) =
−−→P∈U
ker(lim (ϕ)(U)) = ker(ϕP ), where the second equality is from the fact that (ker ϕ)(U) = ker(ϕ(U)) for all open
−−→P∈U
set U.
Denote P as the image of ϕ as presheaf and C as the cokernal of ϕ as presheaf, i.e. P + = Imϕ, C + =
cokerϕ. Then (Imϕ)P = PP = (ker(G → cokerϕ))P = ker(GP → (cokerϕ)P ) = ker(GP → CP ) = ker(GP →
coker(ϕP )) = Im(ϕP ).
(b) ϕ is injective⇔ ker ϕ = 0 ⇔ ker(ϕP ) = 0 for all P ∈ X ⇔ ϕP is injective. And same for the surjective case.
(c) For any i, ker ϕi = Imϕi−1 ⇔ (ker ϕi )P = (Imϕi−1 )P for all P ∈ X ⇔ ker ϕiP = ImϕiP for all P ∈ X. So the
exactness of sequence of sheaves is equivalent to the exactness of sequence of every stalks.

Solution 2.1.3. (a) (⇐) If ϕ is surjective, by 2.1.2., we know that ϕP is surjective for every P ∈ X. So for every
P ∈ U, sP have a preimage (V, t) such that (ϕ(t))P = sP . Shrinking V if necessary, we have some sets (Vi , ti )i∈I
such that ϕ(ti ) = sVi .
(⇒) We clearly have that ϕP is surjective for any P ∈ X. Then by 2.1.2., we have done.
(b) Take X = C − {0}, F = G are the sheaves of holomorphic functions on X and ϕ( f ) = exp( f ) for every
section f . Then if open set U ⊂ X is not simply connected, we know that ϕ(U) is not surjective.

Solution 2.1.4. (a) By 2.1.2., we know that (Imϕ+ )P = Im(ϕ+P ) = Im(ϕP ) = (Imϕ)P = 0 for all P ∈ X, where the
last equality if from the fact that ϕ is injective. So by 2.1.2., ϕ+ is injective.
(b) Since Imϕ is a subsheaf of G , we have a morphism g : Imϕ+ → G by sheafification. And gP : Imϕ+P 
ImϕP → GP is injective. So g is injective, i.e. Imϕ+ is a subsheaf of G by first part of 2.1.6.(b) below.

Solution 2.1.5. By proposition 1.1., the morphism of sheaves is an isomorphism iff it is an isomorphism on
every stalks. So it is equivalent to the injection and surjection on every stalks, i.e. the injection and surjection
of the morphism by 2.1.2.

Solution 2.1.6. (a) The natural map is F (U) → F (U)/F 0 (U) → (F /F 0 )(U). And the surjection and exactness
can be confirmed on stalks.
(b) Since ϕ : F 0 → F is an injection, if ϕ(s) = 0 ∈ F (U) for some s ∈ F 0 (U), we know for every p ∈ U,
ϕ(s)P = 0, and by injection on stalks we know that sP = 0, i.e. s = 0. So F 0 is a subsheaf of F .
On every P ∈ X we know 0 → FP0 → FP → FP00 → 0 is exact by 1.1.1.(c). So FP00  FP /FP0 = (F /F 0 )P ,
that is, F 00  F /F 0 , i.e. F 00 is a quotient sheaf of F .

Solution 2.1.7. (a) Since we have an exact sequence 0 → ker ϕ → F → coImϕ → 0, and the category is a
Abelian category, i.e. coImϕ  Imϕ, we have Imϕ  F / ker ϕ by 2.1.6.(b).
(b) Since we have an exact sequence 0 → Imϕ → G → cokerϕ → 0, we have cokerϕ  G /Imϕ by 2.1.6.(b).

Solution 2.1.8. Denote f : F 0 → F and g : F → F 00 . Then the injection of f is in 2.1.6.(b), so we just need
to prove the exactness at F (U).
For any P ∈ U, we have (gU ◦ fU )P = lim (g ◦ fU )V = lim (g ◦ f ) = (g ◦ f )P = 0. Moreover, if
−−→P∈V⊂U U −−→P∈V⊂U V S V
we have some s ∈ F (U) such that f (s) = 0 in FU00 , we can decompose U = i∈I Vi and s|Vi has a preimage ti in

Yang Pi‑Yeh 19 Hartshorne Solutions

2 Schemes 2.1 Sheaves

FV0 i . And on Vi ∩ V j for any i, j ∈ I, f (ti |Vi ∩V j ) = f (t j |Vi ∩V j ) ∈ F (Vi ∩ V j ) = s|Vi ∩V j . So by injection of f , we know
that ti |Vi ∩V j = t j |Vi ∩V j , i.e. we may glue all ti to get a t ∈ F 0 (U) which maps to s.

Solution 2.1.9 (Direct Sum). It is trivially a sheaf. For any sheaf morphism f : H → F and g : →G , we can
define f ⊕ g as ( f ⊕ g)(s) = f (s) ⊕ g(s) ∈ (F ⊕ G )(U) for any s ∈ H (U), which is a morphism too. So F ⊕ G
we defined is the direct product of F and G in the category of sheaves of abelian groups. Since in abelian
category, the direct product of finite objects is just the direct sum and vice versa, F ⊕ G is the direct sum of
F and G too.

Solution 2.1.10 (Direct Limit). Denote F = lim Fi as presheaf, and F + is the direct limit as sheaf. Then
−−→
for any collection of Fi → G compatible with the maps of direct system, on every open set U ⊂ X we have
Fi (U) → G (U) compatible with the maps of direct system, hence we have a morphism lim Fi (U) → G (U),
−−→
i.e. we get a presheaf morphism F → G . By sheafification, this morphism factor as F → F + → G . And we
may compose Fi → F → F + , and we have Fi → F + → G .

Solution 2.1.11. Denote this presheaf as F . Then we need to prove that is a sheaf, i.e. to cheek the condition
of glueing.
S S
If we have U = k∈I Uk , then by noetherian we may assume that I is finite and U = nk=1 Uk . And if we
have a set of sections sk ∈ F (Uk ) such that sk |Uk ∩Ul = sl |Uk ∩Ul for all k, l, for sk and sl for k, l = 1, . . . , n, we may
find some ikl such that sikkl ∈ Fikl (Uk ) and silkl ∈ Fikl (Ul ) as the preimage of sk and sl such that sikkl |Uk ∩Ul = silkl |Uk ∩Ul .
Then take a i > maxk,l {ikl }, we have sik |Uk ∩Ul = sil |Uk ∩Ul . Hence those sik can be glued up in Fi to get a si . Then
just take the image s of si in F (U). And by the previous process, the s is unique since Fi is a sheaf.
S
Solution 2.1.12 (Inverse Limit). We may denote F = lim Fi as presheaf. If we have U = k∈I Uk and a set
←−−
of sections sk ∈ F (Uk ) such that sk |Uk ∩Ul = sl |Uk ∩Ul for all k, l, we have sik |Uk ∩Ul = sil |Uk ∩Ul for all k, l, where sik is
the image of sk in Fi and same to sil . So {sik } can be glued up to a unique si ∈ Fi (U), which has a direct limit
s ∈ F (U), i.e. F is a sheaf.
If we have G → Fi satisfying the condition of inverse limit, for any open set U we have G (U) → Fi (U)
satisfying that condition, so we have a morphism G (U) → lim Fi (U) = F (U), i.e. we have defined a G → F ,
←−−
which means F satisfies the universal property we need.

Solution 2.1.13 (Espace Étalé of a Presheaf). For any s ∈ F + (U), by strongest topology we have S̄ is contin‑
uous. Then we just need to prove s is continuous. If V is open in Spé (F ) and P ∈ s−1 (V), then clearly P ∈ U.
Take a neighbourhood U 0 ⊂ U of P, such that there exists a t ∈ F (U 0 ) with s|U 0 = t. Then s|−1 −1
U 0 (V) = t (V) is an
−1 −1
open neighbourhood of P, which is contained in s (V). So every point in s V has an open neighbourhood
contained in the preimage, i.e. s is continuous.
If s : U → Spé (F ) is continuous, V ⊂ X open and t ∈ F (V), we have s(x) = t(x) for every x ∈ t−1 (s(U)),
i.e. there exists an open subset W ⊂ t−1 (s(U)) such that s|W = t|W . Since t is continuous by strongest topology,
t−1 (s(U)) is open in U for any t ∈ F (U), we know s(U) is open in Spé (F ). So if x ∈ U, s(x) = t x in F , and the
continuity of s makes s−1 (t(W)) open. So for any open W 0 ⊂ W, we have t|W 0 = s|W 0 . So s locally gives a section
of F , i.e. a section of F + .

Solution 2.1.14 (Support). If P ∈ U such that sP = 0, we know by definition there exists a subset p ∈ V ⊂ U
such that s|V = 0. So U − Supp (s) is open, Supp (s) is closed.
S
Solution 2.1.15 (Sheaf H om). It is clearly a presheaf. If we have U = k∈I Uk and a set of sections fk ∈
H om(F , G )(Uk ) such that fk |Uk ∩Ul = fl |Uk ∩Ul for all k, l, for s ∈ F (U), we have fk (s|Uk )|Uk ∩Ul = ( fk |Uk ∩Ul )(s|Uk ∩Ul ) =
( fl |Uk ∩Ul )(s|Uk ∩Ul ) = fl (s|Ul )|Uk ∩Ul . So fk (s|Uk ) can be uniquely glued up and get a t = f (s), i.e. { fk } can be uniquely
glued up. Hence H om(F , G ) is a sheaf.

Yang Pi‑Yeh 20 Hartshorne Solutions

2 Schemes 2.1 Sheaves

Solution 2.1.16 (Flasque Sheaves). (a) Since every subset of irreducible topological space is connected, we
know that for every open set U, F (U) = A, and the restriction map is always identity, hence surjective. So
constant sheaf on irreducible topological space is flasque.
S
(b) If F 0 is flasque, we take a section s00 ∈ F 00 (U). Since F → F 00 is surjective, we may take U = i∈I Ui
and si ∈ FUi whose image in F 00 is s00Ui . So on Ui ∩ U j , si j = si |Ui ∩U j − s j |Ui ∩U j has the image 0 in F 00 , hence has
a preimage in F 0 (Ui ∩ U j ), namely ti j .
Let S = {(ti )i∈J | J ⊂ I, ti ∈ F 0 (Ui ) such that ti j = ti |Ui ∩U j − t j |Ui ∩U j for all i, j ∈ J}. By Zorn’s lemma,
S must have a maximal element, namely (ti )i∈J . If J $ I, we may take i0 ∈ I − J. For any i, j ∈ J, we
S
S  
have ti |Ui ∩U j ∩Ui0 − t j |Ui ∩U j ∩Ui0 = ti j |Ui ∩U j ∩Ui0 = ti0 j |Ui ∩U j ∩Ui0 − ti0 i |Ui ∩U j ∩Ui0 . Since i∈J U i ∩ U i0 = i∈J U i ∩ U i0
and ti |Ui ∩Ui0 + ti0 i ∈ F 0 (Ui ∩ Ui0 ) satisfying the glueing condition, they can be glued as ti0 ∈ F 0 (Ui0 ). So
(ti )i∈J∪{i0 } ∈ S, which contradict with the maximality of J. So we must have I = J, i.e. we find (ti )i∈I such
that ti j = ti |Ui ∩U j − t j |Ui ∩U j . Then we have (si − ϕ(ti ))|Ui ∩U j = (s j − ϕ(t j ))|Ui ∩U j , where ϕ : F 0 → F . So we can
glue all (si − ϕ(ti )) up as s ∈ F (U), which has the image s00 ∈ F 00 (U), i.e. F (U) → F 00 (U) is surjective. So
0 → F 0 (U) → F (U) → F 00 (U) → 0 is exact.
(c) If F 0 and F are flasque, since we have the following commutative diagram for open sets U, V such
that V ⊂ U:
0 / F 0 (U) / F (U) / F 00 (U) /0

  
0 / F 0 (V) / F (V) / F 00 (V) /0
we know that F 00 (U) → F 00 (V) is surjective by the exactness of first two vertical arrow.
(d) For every open sets V ⊂ U ⊂ Y, we have f −1 (V) ⊂ f −1 (U) ⊂ X. Since F ( f −1 (U)) → F ( f −1 (V)) is
surjective, f∗ F (U) → f∗ F (V) is surjective, hence f∗ F is flasque.
(e) For every open sets V ⊂ U, and s ∈ G (V), we can define a t ∈ G (U) such that s(P) = t(P) for P ∈ V and
s(P) = 0 for P < V. So the restriction map is surjective, i.e. G is flasque.
And we can define a map ϕ : F → G as ϕ(s) = t for every s ∈ F (U), where t satisfies t(P) = sP ∈ FP . Since
if a section is zero on every stalks, this section must be zero, we know that ϕ is injective.

Solution 2.1.17 (Skyscraper Sheaves). If Q ∈ {P} and Q ∈ U for some open set U, we clearly have Q ∈ U,
so iP (A)Q = lim i (A)(U) = lim A = A. Otherwise, if Q < {P}, then there exists a very small open
−−→Q∈U P −−→Q∈U
neighbourhood V of Q such that P < V, hence iP (A)Q = lim i (A)(U) = lim i (A)(U) = 0.
−−→Q∈U P −−→Q∈U⊂V P
−1
If P ∈ U for some open set U, we have i∗ A (U) = A (i (U)) = A ({P}) = A. And if P < U for some open set
U, we have i∗ A (U) = A (i−1 (U)) = A (∅) = 0. Hence i∗ A  iP (A).

Solution 2.1.18 (Adjoint Property of f −1 ). Since f −1 f∗ F is the sheaf associated to the presheaf U 7→ lim −1 F ( f −1 (V)),
−−→U⊂ f (V)
which clearly has a morphism to F as presheaf. Then by sheafification we have a sheaf morphism f −1 f∗ F →
F . And for the other one, for any open subset U ⊂ Y, we have f ( f −1 (U)) ⊂ U. So we have G (U) →
lim −1 G (V) → f −1 G ( f −1 (U)) = f∗ f −1 G (U), i.e. we have define a morphism G → f∗ f −1 G .
−−→ f ( f (U))⊂V
f −1 φ
We can define α : Hom(G , f∗ F ) → Hom( f −1 G , F ) as (φ : G → f∗ F ) → ( f −1 G −−−→ f −1 f∗ F → F ). And
f∗ φ
similarly a morphism β : Hom( f −1 G , F ) → Hom(G , f∗ F ) as (φ : f −1 G → F ) → (g → f∗ f −1 G −−→ f∗ F ). Then
clearly α ◦ β = id and β ◦ α = id by counting all stalks for β(α(φ)) or α(β(φ)).

Solution 2.1.19 (Extending a Sheaf by Zero). (a) If P ∈ Z, we have (i∗ F )P = lim i∗ F (V) = lim F (V∩Z) =
−−→P∈V −−→P∈V∩Z
FP . If P < Z, there exists a open subset P ∈ W ⊂ X such that W ∩ Z = ∅. So (i∗ F )P = lim i∗ F (V) =
−−→P∈V
lim i F (V) = 0.
−−→P∈V⊂W ∗
(b) If P ∈ U, we have ( j! F )P = lim j F (V) = lim F (V) = FP . If P < U, we have ( j! F )P =
−−→P∈V ! −−→P∈V⊂U
lim j! F (V) = 0 since there must exist a P ∈ W ⊂ X such that W * U. Since j! F is a sheaf, it must be
−−→P∈V

Yang Pi‑Yeh 21 Hartshorne Solutions

2 Schemes 2.2 Schemes

unique to satisfy those conditions on stalks. And for any V ⊂ U, we have ( j! F )|U (V) = j! F (V)  F (V), i.e.
j! F )|U  F as sheaves on U.
(c) Since 0 → j! (F |U )P → FP → i∗ (F |Z )P → 0 is 0 → FP → FP → 0 → 0 if P ∈ U, or 0 → 0 → FP → FP →
0 if P ∈ Z by (a) and (b), they are both exact. So by 2.1.2.(c), the sequence is exact.
S
Solution 2.1.20 (Subsheaf with Supports). (a) For any open set U with a covering U = i∈I Ui , and sections
si ∈ ΓZ∩Ui (Ui , F |Ui ) satisfying the compatibility condition, we may treat si as a section in F |(Ui ) whose support
is in Z. Then obviously all si satisfy the compatibility condition, i.e. si can be glued up as a unique s ∈ F (U).
Since for any P ∈ U, sP = (si )P for some i, then the support of s is just the union of all supports of si , so is
contained in Z, i.e. s ∈ ΓZ∩U (U, F |U ). Hence HZ0 (F ) is a sheaf.
(b) It is clearly exact by counting the stalks like 2.1.19. And furthermore, since j∗ (F |U )(V) = F |U (U ∩ V) =
F (U ∩ V). Then on every open set V, F (V) → j∗ (F |U )(V) is just F (V) → F (V ∩ U), which is surjective since
F is flasque.
S
Solution 2.1.21 (Some Examples of Sheaves on Varieties). (a) For any open set U with a covering U = i∈I Ui ,
and sections fi ∈ IY (Ui ) satisfying the compatibility condition, we know that every fi is a regular function
vanishing at all points of Y ∩ Ui . Then there exists a unique f ∈ OX (U) as the glueing of all fi , hence it vanishes
on every points of Y ∩ U, i.e. IY is a sheaf.
(b) By construction, we clearly have a injective IY → OX . And for any open set U ⊂ X, coker(IY (U) →
OX (U)) = { f is a regular function on Y} = i∗ (OY )(U). So by 2.1.7.(b), we have OX /IY = ∗ (OY ).
(c) The global section of this exact sequence is 0 → 0 → k → k⊕k. So the last arrow is clearly not surjective.
(d) For any open set U we clearly have a morphism O(U) → K (U) as f 7→ f¯, where f¯ is the extension
of f on X. If f¯ = 0, f = f¯|U = 0, hence injective. Moreover, on every point P ∈ X, (K /O)P = KP /OP = IP =
P P
(iP (IP ))P = ( P∈X iP (IP ))P , hence K /O  P∈X iP (IP ).
P L
(e) Only need to show that Γ(X, K ) → Γ(X, P∈X iP (IP )) is surjective, i.e. K → P∈X P
I is surjective. So we
just need show that for every f ∈ K for some P ∈ X, then for every Q ∈ X, Q , P we have some g ∈ K such
P
that g ∈ OQ and g − f ∈ P. If f is constant, trivial. If not, we may assume f = x p · α(x) β(x)
, where α = ai xi and
P i
β = bi x for some a0 , b0 , 0. Since PGL1 is transitive, we may assume P = 0. Then if p ≥ 0, we just take
P Pi−1
g = 1. If p < 0, we define g = x p · −p i −1
i=0 ci , where c0 = b0 , and ci = x b0 (ai −
a0
j=0 c j bi− j ). Then g ∈ OQ for any
P
Q , P, and f − g = x p β(x)−1 (α(x) − β(x) −p i=0 ci ) ∈ O P .

Solution 2.1.22 (Glueing Sheaves). For any open set U ⊂ X, we just define F (U) = {(si ) ∈ F (U∩Ui ) | si |U∩Ui ∩U j =
s j |U∩Ui ∩U j }. Then F is clearly a sheaf on X.

2.2 Schemes
Solution 2.2.1. Since (A f )g = A f g , and {D( f )} form a basis of open set on Spec A, we just need to prove that
`
O(D( f )) = A f . We construct a ϕ : A f → O(D( f )), fak 7→ (D( f ) → p∈Spec A Ap , p 7→ fak , then we just need to
prove ϕ is a bijection.
If ϕ( fak = 0, for any p ∈ D( f ), we have fak = 0 in Ap , i.e. ∃t < p such that ta = 0. So Ann (a) * p for all p, i.e.
p
V(Ann (a)) ⊂ V(( f )). Then f ∈ Ann (a). So fak = 0 in A f , hence ϕ is injective.
If s ∈ O(D( f )), then we have an open covering of D( f ) as {Ui }i∈I and for any Ui there exist ai , fi ∈ A such
that fi < q and s(q) = afii for any q ∈ Ui . We may assume Ui = D(gi ) for some gi ∈ A since we have a basis of
open set {D( f )}, and this covering is finite covering since Spec A is noetherian. Then for any q ∈ D(gi ) we have
p p
fi < q, i.e. D(gi ) ⊂ D( fi ), hence (gi ) ⊂ ( fi ). So gki i = hi fi for some hi ∈ A. Then sq = hikaii . Since D(gi ) = Dgki , we
gi i

may replace gi by gki i and ai by hi ai , i.e. we have an open covering {D(gi )}i∈I for D( f ), and s(q) = ai
gi
for every
q ∈ D(gi ).

Yang Pi‑Yeh 22 Hartshorne Solutions

2 Schemes 2.2 Schemes

aj
For any q ∈ D(gi ) ∩ D(g j ), we have ai
= ∈ Aq . So for any q ∈ D(gi g j ), there exists a t < q such that
gi gj
p
t(ai g j − a j gi ) = 0, i.e. Ann (ai g j − a j gi ) * q. So V(Ann (ai g j − a j gi )) ⊂ V((gi g j )), i.e. gi g j ∈ Ann (ai g j − a j gi ).
Since I is finite, we may take a k sufficiently large, such that (gi g j )k (ai g j − a j gi ) = 0. Then we may replace our
a gk
ai
gi
by gk+1 i i
, i.e. we have an open covering {D(gi )}i∈I for D( f ), and s(q) = agii for every q ∈ D(gi ), and ai g j = a j gi for
i
every pairs i, j.
S P P
Since D( f ) = i D(gi ), we have f k = bi gi for some bi ∈ A. So for any j, we have a j f k = a j bi gi =
P P P aj
P
bi a j gi = bi ai g j = g j bi ai , i.e. g j = fbki ai . So ϕ is surjective.
S
Solution 2.2.2. Since X is a scheme, then it has an open covering X = Vi , such that (Vi , OX |Vi )  (Spec Ai , OSpec Ai )
is an affine scheme. If U is some Vi , trivial. If not, then {U ∩ Vi }i∈I form an open covering of U. And since U
S
is open in X, then U ∩ Vi is open in Vi . Then U ∩ Vi = j Spec (Ai ) fi j for some fi j ∈ Ai since {D( f )} f ∈Ai forms a
basis of open set of Spec Ai . And on every Spec (Ai ) fi j , OX |U |Spec (Ai ) fi j = OSpec Ai |Spec (Ai ) fi j  OSpec (Ai ) fi j by 2.2.1.
So (U, OX |U ) is a scheme.

Solution 2.2.3 (Reduced Schemes). (a) (⇒) If s ∈ OP is nilpotent, i.e. sn = 0 for some n, there exists a neigh‑
bourhood U of p and t ∈ O(U) such that t p = s and tn = 0. Then t = 0 because O(U) is reduced. So s = tP = 0,
hence OP is reduced.
(⇐) If s ∈ O(U) is nilpotent, i.e. sn = 0 for some n. Then snP = 0 for every P ∈ U, hence sP = 0 for every
P ∈ U, i.e. s = 0. So O(U) is reduced.
(b) If X = Spec A is affine, we just take the Spec Ared as the reduced schemes of X, since Spec A  Spec Ared
S
as topological space, and (A f )red = (Ared ) f¯, where f¯ is the image of f in Ared . For general case, if X = Spec Ai ,
then clearly OXred (Spec Ai ) = (Ai )red .
On every affine piece, the canonical morphism A → Ared induces a morphism Spec A → Spec Ared , then
we just glue them up to get a morphism X → Xred . Since on every affine piece, Spec A → Spec Ared is a
homeomorphism on topological space, hence so is X → Xred .
(c) Since X is reduced, then for any open set U ∈ Y, OY (U) → OX ( f −1 (U)) induces a morphism (OY (U))red →
OX ( f −1 (U)), hence we have a presheaf morphism (U 7→ (OY (U))red ) → f∗ OX . So by sheafification we have a
morphism OYred → f∗ OX , i.e. a scheme morphism X → Yred . By construction, X → Y obviously factors through
X → Yred .

Solution 2.2.4. We can define a β : Hom(A, OX (X)) → Hom(X, Spec A) as follows, for any φ : A → OX (X), we
S S φ
may assume X = Ui = Spec Ai is an affine covering, then the ring morphism A → − OX (X) → OX (Ui ) induces
a fi : Ui → Spec Ai . So for any Ui ∩ U j , we may cover it by some affine pieces Ui jk , where Ui jk = Spec Ai jk .
Then fi |Ui jk : Spec Ai jk → Spec A and f j |Ui jk : Spec Ai jk → Spec A are clearly the same, i.e. fi and f j are the same
on Ui ∩ U j . Hence { fi } can be glued up to get a f : X → Spec A, i.e. β is well‑defined. Clearly β is the inverse
of α, i.e. α is bijection.

Solution 2.2.5. Clearly all the prime ideals of Z are (0) and (p) for all prime numbers p, and the unique
minimal ideal of Z is (0). For every ideal (n) ⊂ Z, OZ (D(n)) = { ba | p - b for all p|n}. And on every point of
Spec Z, OZ,(0) = Q, and OZ,(p) = Z p .
Since Z is the initial object of the category of rings, then by 2.2.4., Spec Z is the final object of the category
of schemes.

Solution 2.2.6. Since zero ring has no prime ideal, the spectrum of the zero ring is ∅. Since the zero ring is
the final object of the category of rings, the spectrum of the zero ring is the initial object of the category of
schemes.

Solution 2.2.7. If we have a morphism Spec K → X, clearly the unique point of Spec K maps to a point x ∈ X.
And we have a morphism O x → K. Since K is a field, the maximal ideal m x of O x maps to zero in K, i.e. it
induces a morphism k(x) = O x /m x → K.

Yang Pi‑Yeh 23 Hartshorne Solutions

2 Schemes 2.2 Schemes

Conversely, if we fix a point x ∈ X and a morphism k(x) → K, we can define a sheaf morphism f :
Spec K → X as the point of Spec K 7→ x, and OX → f∗ OSpec K as



OX (U) → 0
 if x < U



OX (U) → O x → k(x) → K if x ∈ U

Solution 2.2.8. Since k[ε]/(ε2 ) has only one prime (ε), Spec k[ε]/(ε2 ) has only one point. If we have a morphism
Spec k[ε]/(ε2 ) → X, the point of Spec k[ε]/(ε2 ) has a image x. And it induces a local ring morphism φ : O x →
k[ε]/(ε2 ). So φ(m x ) ⊂ (ε). Since ε2 = 0, we can define a ψ : m x → k as a 7→
f racφ(a)ε, and it induces a morphism T x = m x /m2x → k. This morphism is from vector space to the constant
field, so it is induced by a element of the vector space, and the morphism is the inner product.
Conversely, if we fix an x ∈ X and an element of T x , it induces the ψ above. Then we can define φ : O x →
k[ε]/(ε2 ) as a+b → a+ψ(b)ε, where a ∈ k and b ∈ m x . So we can define a scheme morphism f : Spec k[ε]/(ε2 ) →
X as the point of Spec k[ε]/(ε2 ) → x. And OX → Ok[ε]/(ε2 ) is



OX (U) → 0
 if x < U


 φ
OX (U) → O x →
− k[ε]/(ε2 ) if x ∈ U

Solution 2.2.9. If U = Spec A is an affine piece of X and U ∩ Z , ∅, U ∩ Z must be an irreducible closed

subset of U, so has the form V(p) for some prime ideal p ⊂ A. Since V(p) = {p}, we may denote ζ as the point
corresponding to p. So the closure of Z in U is U ∩ Z, but U ∩ Z is a nonempty open subset in the irreducible
space Z, hence dense in it, i.e. Z = {ζ}. If Z = {ζ 0 } for another point ζ 0 , we have ζ 0 ∈ U ∩ Z. If p0 is the prime
ideal corresponding to ζ 0 , we have V(p0 ) = {p0 } = U ∩ Z = V(p), which means p = p0 , i.e. ζ is unique.

Solution 2.2.10. Since R[x] is a PID, its prime ideals are: i. the unique minimal prime ideal (0), whose residue
field is R(x); ii. (x + a) for some a ∈ R, whose residue field is R; iii. (x2 + bx + c) for some b, c ∈ R such that
b2 < 4c, whose residue field is C. So Space (Spec R[x])  R ∪ H ∪ {0} as a set.

Solution 2.2.11. Since F p [x] is a PID, its prime ideals are: i. the unique minimal prime ideal (0), whose residue
field is F p (x); ii. ( f ) for some irreducible polynomial in F p (x), whose residue field is F pd if deg f = d. Since
n
x p − x is the product of all prime polynomial with degree d|n, so the number of all the prime polynomial with
P P
degree d is d|n µ(d)pn/d , i.e. ]{points with residue field F pd } = d|n µ(d)pn/d .
`
Solution 2.2.12 (Glueing Lemma). We may glue the underlying space as X = Xi / ∼, where the equivalent
relation is xi ∈ Xi ∼ x j ∈ X j ⇔ ϕi j (xi ) = x j . Then we clearly have maps ψi : Xi → X. So we can define the sheaf
on X as glueing all ψi∗ OXi on X, hence (X, OX is a scheme, and the four conclusion is naturally gotten.
S
Solution 2.2.13. (a) (⇐) If X is noetherian and U is a open set in X with an open covering U = Ui . If U
does not have a finite subcovering, then we can construct a ascending chain {Vi } in U as: V1 = Ui1 for some
Ui1 , and Vn+1 = Vn ∪ Uin for some Uin * Vn . Since U does not have a finite subcovering, this construction is
well‑defined, i.e. we get a ascending chain V1 $ V2 $ . . . $ Vn $ . . ., which is contradict with the fact that X is
noetherian.
S
(⇒) If U1 ⊂ U2 ⊂ . . . ⊂ Un ⊂ . . . is an ascending chain in X, then U = i Ui has an open covering
S
i U i . Since it is quasi‑compact, U has a finite covering, which will involve U n as the maximal index, hence
Un = Un+1 = . . ..
P P
(b) If {Ui } covers X = Spec A, we may assume X − Ui = V(ai ) closed, then i ai = A, i.e. we have 1 = ni fi gi
S
for some fi ∈ A and gi ∈ ai for some ai . So X = ni=1 Ui for those Ui , hence quasi‑compact.
For example of non‑noetherian but quasi‑compact spectrum, we just take A = k[x1 , x2 , . . .], then we have
an ascending chain of closed subsets V((x1 )) $ V((x1 , x2 )) $ . . . $ V((x1 , x2 , . . . , xn )) $ . . ..

Yang Pi‑Yeh 24 Hartshorne Solutions

2 Schemes 2.2 Schemes

(c) For any descending chain Z1 ⊃ Z2 ⊃ . . . ⊃ Zn ⊃ . . . in Spec A, we have a corresponding ascending chain
of ideals in A as I1 ⊂ I2 ⊂ . . . ⊂ In ⊂ . . ., which will be stable since A is noetherian. So the descending chain
{Zi } will be stable, hence Space (Spec A) is noetherian.
(d) Take A = k[x1 , x2 , . . .]/(x12 , x22 , . . .). So Space (Spec A) is a singleton, hence noetherian. But A itself is not
noetherian.

Solution 2.2.14. (a) (⇐) If Proj S = ∅, then every graded prime ideal in S containing S + . For any prime ideal
L∞
p, we have a graded prime ideal q = d=0
p ∩ S d , then p ⊃ q ⊃ S + . Hence S + is contained in the nilpotent
radical, i.e. every element of S + is nilpotent.
(⇒) If every element of S + is nilpotent, every prime ideal in S contain S + , hence Proj S = ∅.
(b) For any p ∈ U, there exists some s ∈ S + such that ϕ(s) < p. So D(ϕ(s)) ⊂ U is a neighbourhood of p.
Hence U is open. Moreover, for every p ∈ U, we may define f (p) = ϕ−1 (p) + S + , hence well‑defined. And the
morphism on sheaves f] is given by S (ϕ−1 p) → T (p) .
(c) For any p + T + , there exists some t ∈ T k , such that t < p. So tn < p for some n such that nk ≥ d0 . Then
there exists some s ∈ S + such that s = ϕ(tn ), hence p + ϕ(S + ), i.e. U = Proj T .
p P
For any p ∈ Proj S , we denote q = ϕ(p). If a ∈ ϕ−1 q, we have ϕ(an ) ∈ ϕ(p), i.e. ϕ(an ) = bi ϕ(si ) for some
P
bi ∈ T and si ∈ p. Since ( bi ϕ(si ))m is a polynomial in ϕ(si ) with coefficients in T ≥d0 , ϕ(anm ) ∈ ϕ(p), i.e. a ∈ p.
Thus ϕ−1 (q) = p, i.e. f is surjective. Conversely, if f (p) = f (q) for some p, q ∈ Proj T , we have ϕ−1 (p) = ϕ−1 (q).
For t ∈ p ∩ T + , there exists a s ∈ S such that ϕ(s) = tk for some sufficiently large k. So tk = ϕ(s) ∈ q, i.e t ∈ q.
Hence p ⊂ q, and vice versa, which means injection.
Since all D+ (s) cover Proj S , then for sufficiently large i, f −1 D+ (si ) = D+ (t) ⊂ Proj T for some t. Then we
need to prove that S (si ) → T (t) is an isomorphism. Changing si into s, if sfn 7→ 0, we have 0 = tm ϕ( f ) = ϕ(sm f )
for some large m, i.e. sm f ∈ ker ϕ. Since S d  T d for sufficiently large d, some large powers of sm f maps to 0,
d0
i.e. sfn = 0, hence S (s) → T (t) is injective. If tgn ∈ T (t) , we know that ttn+dg0 has a preimage in S (s) , hence surjective.
(d) Since V = {maximal ideals of S which do not contain S + }, we know that t(V) = {irreducible components
of V} = {prime ideals of S which do not contain S + }, i.e. t(V) = Proj S . And clearly t(V) and Proj S have same
topology. Moreover, for every U open in Proj S , denote the preimage of U in V as U 0 , α∗ OV (U) = OV (U 0 ) =
{regular functions which have no poles in U 0 } = {regular functions which have no poles in U} = OProj S (U),
hence (t(V), α∗ OV )  (Proj S , OProj S ).

Solution 2.2.15. (a) P is not a closed point⇔the corresponding irreducible closed subset Z of P is not a
point⇔the residue field of Z is has transcendental degree≥ 1. So P is a closed points iff the residue field
of P is just k.
(b) Since we have the morphism of sheaves OY → f∗ OX , it induces a morphism of residue fields k( f (P)) ,→
k(P). Since X and Y are both fields over k, k( f (P)) and k(P) are both field extension of k. But k(P) = k, so we
have k ,→ k( f (P)) ,→ k, i.e. k( f (P)) = k.
(c) Denote this morphism by υ. For any ϕ : V → W, we can define υ(ϕ) as ϕ on closed points, and for any
nonclosed points which corresponding to irreducible closed subset Y of V, we can define υ(ϕ)(Y) = ϕ(Y). So
since υ(ϕ) is an extension of ϕ, the injection is obvious. For surjection, if ψ : t(V) → t(W) is a morphism of
schemes, we have υ−1 (ψ) = ψ|V . Since ψ maps closed points to closed points, this is well‑defined. If ψ(P) = Q
for closed points P and Q in t(V) and t(W), we may assume P = p ⊂ A in some affine piece U = Spec A and
Q = q ⊂ B in some affine piece V = Spec B ⊂ f −1 (U). Then P 7→ Q is just Spec B → Spec A on stalks, hence
υ−1 (ψ) is regular, i.e. υ−1 (ψ) ∈ HomVar (V, W).

Solution 2.2.16. (a) If x ∈ X f ∩ U, we know f x < m x , i.e. f¯x < m x , hence x ∈ D( f¯). If x ∈ D( f¯) ⊂ U, we have
f¯x < m x , i.e. f x < m, hence x ∈ X f .
S
(b) Since X is quasi‑compact, we may assume X = Ui for finite i, and Ui = Spec Ai is affine. For every
i, since a|X f = 0, we know that ai = a|Ui restricts on (Ui ) fi is zero, where fi is the restriction of f on Ui . So

Yang Pi‑Yeh 25 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

there exists a ni such that fini ai = 0. If we denote n = max{ni } since the index set is finite, we have fin ai = 0, i.e.
( f n a)|Ui = 0 for every i. Hence we have f n a = 0 on whole X.
(c) On every Ui , b|Ui ∩X f = b|D( fi ) , there exists some ni such that fini b|D( fi ) can be extended into Ui . So we
may take an n = max{ni } so that fin b|D( fi ) can be extended into Ui as some bi . On every i, j, we may assume
S
Ui ∩ U j = k∈Ii j Ui jk for some finite index set Ii j . On every Ui jk , since bi |Ui jk − b j |Ui jk = 0, there exists some mi jk
n
such that fUii jkjk (bi |Ui jk − b j |Ui jk ) = 0. Then since the index i jk is finite, we can take some m = max{mi jk } such that
f m (bi |Ui j − b j |Ui j ) = 0. So f n+m b can be extended to some global section.
(d) Since A = OX (X) → OX (X f ) maps f to zero, it induces a morphism A f → OX (X f ). But (b) means
injection and (c) means surjection. So we have A f  OX (X f ).

Solution 2.2.17 (A Criterion for Affineness). (a) Denote the inverse of f | f −1 (Ui ) as gi . Since two maps f | f −1 (Ui ) | f −1 (Ui ∩U j )
and f | f −1 (U j ) | f −1 (Ui ∩U j ) are just f | f −1 (Ui ∩U j ) , and they are obviously both isomorphism, there exists gi j are the in‑
verse of f | f −1 (Ui ∩U j ) . Clearly gi j = gi |Ui ∩U j = g j |Ui ∩U j , which means all gi can be glued together into a g : Y → X
as the inverse of f . So f is isomorphism.
(b) The identity A → O(X) induces a ϕ : X → Spec A. Since { fi } generate the unit ideal, {D( fi )} forms
an open covering of Spec A. Since clearly ϕ−1 (D( fi )) = X f , by 2.2.16.(d) we have OX (X fi )  A fi . Since ϕ is
isomorphic on ϕ−1 (D( fi )), by (a), ϕ is an isomorphism.

Solution 2.2.18. (a) f is nilpotent⇔there exists an n such that f n = 0 ⇔ D( f ) = D( f n ) = D(0) = ∅.

(b) (⇐) If ϕ is injective, for any prime q ∈ B, the fϕ]−1 (q) is just Aϕ−1 (q) → Bq , which is injective. Hence f ] is
injective.
(⇒) Just take the global section of f ] . Then we can get ϕ = f ] : A → B is injective.
T T
Furthermore, for any U ⊂ X, we have U ⊂ V( p∈U p), i.e. Ū ⊂ V( p∈U p). If Ū = V(a) for some a ⊂ A, for
S T
any p ∈ U, we have a ⊂ p, i.e. a ⊂ p∈U p. So we get Ū = V( p∈U p). Thus,
\ \ \ p
f (Y) = V( p) = V( ϕ−1 (q)) = V(ϕ−1 ( ϕ−1 q)) = V(ϕ−1 ( (0))) = V(ker ϕ) = V(0) = X
p∈ f (Y) q∈Y q∈Y

(c) If ϕ is surjective, B  A/ ker ϕ induces Spec B  Spec A/ ker ϕ = V(ker ϕ) ⊂ Spec A is closed. Furthermore,
we have fp] = Ap → B ⊗A AP is surjective, hence f ] is surjective.
S
(d) If f ] is surjective, for any b ∈ B ∈ OY (Y), there exists an open covering X = Ui such that b|Ui has
preimage f Ni in OX (Ui ). Since we may cover each Ui by affine pieces and X can be covered by such affine
ai
S
pieces, we may assume the covering X = Ui is affine covering for Ui = Spec A fi = D( fi ) for some fi ∈ A.
S
Since X is quasi‑compact, we may assume this covering is finite as X = ni=1 Ui , and N = max{Ni }. Hence we
Pn N P P
have some gi ∈ A such that i=1 fi gi = 1 since {D( fiN )} covers X. So b = ϕ( fi )N ϕ(gi )b = ϕ( gi fiN−Ni ai ).

Solution 2.2.19. (i⇒ii) If X = Spec A is disconnected, there exists two disjoint open subsets U1 , U2 . So there
exist e1 , e2 ∈ OX (X) = A such that e1 |U1 = 1, e2 |U1 = 0 in OX (U1 ) and e1 |U2 = 0, e2 |U2 = 1 in OX (U2 ) since OX is a
sheaf. So exactly e1 e2 = 0, e21 = e1 , e22 = e2 and e1 + e2 = 1.
(ii⇒iii) Just define A1 = e1 · A and A2 = e2 · A. We have A = A1 × A2 .
(iii⇒i) Since prime ideals of A1 × A2 have two kinds: p × A2 for some prime p ⊂ A1 , and A1 × q for some
`
prime q ⊂ A2 . Clearly we have Spec A = Spec A1 Spec A2 .

2.3 First Properties of Schemes

S S S
Solution 2.3.1. (⇒) If Y = Vi = Spec Bi is an affine covering of Y, and f −1 (Spec Bi ) = Spec Ai j for some
S
finitely generated Bi ‑algebra Ai j . Since V ∪Vi is open in Vi , there exists an affine covering V ∪Vi = Spec (Bi ) fik
for some fik ∈ Bi . If we denote ϕ : Bi → Ai j , then (Ai j )ϕ( fik ) is a finitely generated (Bi ) fik ‑algebra.

Yang Pi‑Yeh 26 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

S S
So we may assume that Spec B = Spec Bi with f −1 (Spec Bi ) = Spec Ai j for some finitely generated Bi
algebra Ai j . For any p ∈ Spec B, we have p ∈ Spec Bi for some i. Since Spec Bi is open in Spec B, there exists
some fp ∈ B such that p ∈ Spec B fp ⊂ Spec Bi . So we have Spec (Bi ) fp  Spec B fp . Since f −1 (Spec (Bi ) fp ) =
S
Spec (Ai j )ϕ( fp ) , we know (Ai j )ϕ( fp ) is a finitely generated B fp ‑algebra, hence a finitely generated B‑algebra.
(⇐) Obvious.
S S
Solution 2.3.2. (⇒) We may assume Y = Vi = Spec Bi is an affine open covering, and f −1 (Vi ) is quasi‑
S
compact. If V ⊂ Y is an affine piece, V ∩ Vi = D(gi j ) for some gi j ∈ Vi . Then V can be covering by these D( fi j ).
Since V is affine, i.e. quasi‑compact, we may choose finitely many D(gi j ) covering V. If for every Vi , we have
S S
f −1 (Vi ) = Spec Aik for finitely many Aik . So f −1 (D(gi j )) = Spec (Aik )ϕ(gi j ) , where ϕ : Bi → Aik , i.e. f −1 (D(gi j ))
S
is quasi‑compact. So f −1 (V) = f −1 (D(gi j )) is quasi‑compact.
(⇐) Obvious.

Solution 2.3.3. (a) (⇒) We only need to show f is quasi‑compact. If V is an affine piece in Y, then f −1 (V) can
be covered by finite affine piece in X. Since every affine piece is quasi‑compact, f −1 (V) is quasi‑compact.
(⇐) Obvious.
(b) For any affine piece V = Spec B, f is of finite type ⇔ f is locally of finite type and quasi‑compact
⇔ f −1 (V) is quasi‑compact and can be covered by Spec Ai for finitely generated B‑algebra Ai ⇔ f −1 (V) can be
covered by finitely many Spec Ai with same property.
S S
(c) Since f is of finite type, f −1 (V) = Ui = Spec Ai for finitely many finitely generated B‑algebra Ai . If
U = Spec A ⊂ f −1 (V), for any point p ∈ U ∩ Ui , there exists same f 0 ∈ A such that p ∈ Spec A f 0 ⊂ U ∩ Ui . Since
Spec A f 0 is open in Ui , there exists some fi ∈ Ai such that p ∈ Spec (Ai ) fi ⊂ Spec A f 0 . This induces a morphism
Ai → A f 0 → (Ai ) fi . So the image of fi in A f 0 has a preimage in A, namely f . So we have Spec A f  Spec (Ai ) fi .
ai,n
Clearly (Ai ) fi is a finitely generated Ai ‑algebra, hence a finitely generated B‑algebra with basis { a1i1 , . . . , 1 i , 1fi }.
S
So, we may assume that U = Spec A fi for some fi with A fi are all finitely generated B‑algebra. Since
U is quasi‑compact, we may assume this covering is finite. For any a ∈ A, we have a1 = fski for some si ∈
i
ai,n
B[ a1i1 , . . . , 1 i , 1fi ] and a fixed sufficiently large k. Thus there exists another sufficiently large l such that fil ( fik a−
S
si ). We may change k + l as k and fil si as si and have fik a = si . Since U = D( fik ), there exists some gi ∈ A such
P k P
that fi gi = 1. So a = gi si ∈ B[S ] with S = {ai j , fi for all i, j}.
S S
Solution 2.3.4. (⇒) We may assume Y = Vi = Spec Bi , f −1 (Vi ) = Ui = Spec Ai with some finitely generated
Bi ‑module Ai . If we denote ϕi : Bi → Ai , for every f ∈ Bi , clearly f −1 (D( f )) = D((ϕ( f ))) ⊂ Spec Ai . And more,
(Ai )ϕ( f ) is a finitely generated (Bi ) f ‑module.
So for any affine piece V = Spec B ⊂ Y, similar with 2.3.3., we know that V ∩ Vi can be covered by basis
open sets of both V and Vi , which preimage is basis open sets of Ui and finite over their images. So we may
S ai,n
assume V = Spec Bi with f −1 (Vi ) = Ui = Spec Ai for some finite Bi ‑module Ai with basis { afi1k , . . . , f k i }. By
S i i
2.2.17., we know f −1 (V) = U is affine. Moreover, we may assume V = Vi is finite, since V is quasi‑compact.
So if Vi = Spec B fi for P
some fi ∈ B, { f1 , . . . , fn } generates B, then the preimages of them generates A. For any
s a P
a ∈ A, we have 1a = f jk i j . So there exists an l such that fil ( fik a − s j ai j ) = 0. So we may replace k + l to k and
i P P
fil ai j to ai j . And we have fik a = s j ai j . Since D( fik ) covers U, we have some gi ∈ U such that fik gi = 1, hence
P P
a = i j gi s j ai j , i.e. A is finite over B.

Solution 2.3.5. (a) Since this question is local, we may assume Y = Spec B is affine. Hence X = f −1 (Y) = Spec A
is affine and A is finite generated B‑module. For any y ∈ Y, we have f −1 y = Spec A ⊗B κ(y), which is a finitely
generated κ(y)‑module, i.e. a vector space of finite rank. So f −1 (y) is finite.
S S
(b) We may assume Y = Vi = Spec Bi and f −1 (Vi ) = Ui = Spec Ai for some finitely generated Bi ‑module
Ai . For any closed set Z ⊂ X, we know that Z ∩ Ui is closed in Ui . If f (Z ∩ Ui ) is also closed in Vi , Vi − f (Z ∩ Ui )
is open in Y, i.e. Y − f (Z) is open in Y, hence f (Z) is closed. Hence we reduce this problem into affine case.

Yang Pi‑Yeh 27 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

So we may assume Y = Spec B and X = Spec A. Then any closed subset Z ⊂ X has the form Z = V(b) for
some ideal b ⊂ A. If we denote the morphism ϕ : A → B induced by X → Y, a = ϕ−1 (b) ⊂ B is an ideal, and
W = V(a) ⊂ Y. So we only need to prove f (Z) = W. Obviously we have f (Z) ⊂ W. Conversely, for any point
p ∈ W, i.e. a prime ideal p ⊂ B with b ⊂ p. By going‑up theorem, it can be lifted to a prime q ⊂ A such that
a ⊂ q, which means f (Z) ⊃ W, hence f (Z) = W.
(c) The ring morphism k[t] → k[t, 1t ] ⊕ k[t, t−1
1
] induces a morphism of schemes: Spec k[t, 1t ] ⊕ k[t, t−1
1
]→
Spec k[t], which is surjective, finite‑type and quasi‑finite but not finite.

Solution 2.3.6. If ξ is in some affine piece U = Spec A, it corresponds to a minimal prime ideal p ⊂ A. Since X
is an integral scheme, A must be an integral domain, hence p = (0). So Oξ = (O|U )ξ = A(0) = Frac(A) is a field.
And if X = Spec A, we have had Oξ = Frac(A).

Solution 2.3.7. Since X, Y are integral, i.e. irreducible, we can denote the generic points of X and Y are ηX and
ηY . Since f is dominant, we have f (ηX ) = ηY , because f (ηX ) = f ({ηX }) = f (X) = Y = ηY , and the generic point
of Y is unique. So we have an injective morphism on stalk fη]Y : OηY → OηX . Since X, Y are integral, OηX and
OηY are integral, i.e. fη]Y induces a morphism f 0 : k(Y) → k(X) with ker f 0 = 0. So k(X)/k(Y) is a field extension.
We may assume ηY ∈ V ⊂ Y for some affine piece V = Spec B, and ηX ∈ U ⊂ f −1 (V) for some affine piece
U = Spec A, then clearly k(X) = Frac(A) and k(Y) = Frac(B), and ηX and ηY are corresponding to (0) in A and
B. Since f is generically finite, there are only finitely many prime ideal in A lying over (0) ⊂ B. So if k(X)/k(Y)
is transcendental, i.e. A/B is transcendental extension, by going‑up theorem there exists infinite prime ideals
lying over (0) ⊂ B, which makes a contradiction. So k(X)/k(Y) is algebraic extension. And since A is a finitely
generated B‑algebra, k(X)/k(Y) is finite.
In the affine case, i.e. X = Spec A and Y = Spec B. We may denote {ai }i is a set of generators of A as a
B‑algebra. Then for any ai , it satisfies a polynomial with B‑coefficients with leading term bi ani i . So we may
Q
assume b = bi , then in Ab → Bb , the generator ai satisfies a monic polynomial, hence Ab is integral over Bb ,
hence Ab is a finitely generated Bb ‑module, i.e. on U = D(b), X is finite over Y. In the general case, since X
and Y are integral, i.e. irreducible, any affine piece is densed in them, so we’ve done.

Solution 2.3.8 (Normalization). For affine pieces U = Spec A and V = Spec B in X, we need to glue Ũ = Spec Ã
and Ṽ = Spec B̃ together. For every p ∈ U ∩ V, we can find a open set W ⊂ U ∩ V with p ∈ W and W is principal
open subset of both U and V as we do in 2.3.3. We may assume W = Spec A f = Spec Bg . Denoting ϕ : Ũ → U
and ψ : Ṽ → V as canonical morphisms, we have ϕ−1 (W) = Spec à f and ψ−1 (W) = Spec B̃g . Since à f and B̃g
are both normalized, so by uniqueness we have à f  B̃g , i.e. ϕ−1 (W)  ψ−1 (W), which means we can glue Ũ
and Ṽ, or other affine piece together with a canonical map φ : X̃ → X.
If we have a normal integral scheme Z with f : Z → X, for every affine piece U = Spec A ⊂ X, we have
f : f −1 (U) → U. Since Z is normal, we clearly know that f −1 (U) is normal. So f induces a ring morphism
A → OZ ( f −1 (U)), which can be extended to à → OZ ( f −1 (U)). So we have a morphism f˜ : f −1 (U) → Ũ. Then
glueing them together, we get a morphism f˜Z → X̃.
Finally, by construction, the morphism φ is clearly affine, so this problem is local. We may assume X =
Spec A and X̃ = Spec Ã. Then if X is of finite type over some field k, i.e. A is a finitely generated k‑algebra, we
know that à is finitely generated A‑module by Theorem 3.9A. in Chapter I. So φ is finite.

Solution 2.3.9 (The Topological Space of a Product). (a) Clearly A1k ×Spec k A1k = Spec (k[x]⊗k k[y]) = Spec k[x, y] =
A2k . And, (x − y) ∈ Spec k[x, y] is not contained in Space A1k × Space A1k .
(b) Obviously k(x) ⊗k k(y) = { f a(x,y)
(x)g(y)
| a(x, y) ∈ k[x, y], f (x) ∈ k[x], g(y) ∈ k[y]}. So the primes of k(x) ⊗k k(y) are
all height 1 primes of k[x, y] which are not in forms of just f (x) or g(y). Hence Spec k(x) ×Spec k Spec k(y) has
infinitely many points.

Solution 2.3.10 (Fibres of a Morphism). (a) Clearly the morphism Space Xy → f −1 (y) is induced by the pro‑
jection π : X ×Y Spec k(y) → X. Then we need to prove that π is a homeomorphism on the underlying

Yang Pi‑Yeh 28 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

space. Since the problem is local for Y, we may assume Y = Spec B. For any open subset U ⊂ X, we have
π−1 (U) = U ×Y Spec k(y) = Uy , so we only need to prove that Uy  f −1 (y) ∩ U. So we can also assume X = Spec A
affine. Denoting p is the prime ideal of B corresponding to y, ϕ : B → A is the ring homomorphism corre‑
sponding to the scheme homomorphism f : X → Y, we have X ×Y Spec k(y) = Spec A ⊗B Bp /pBp = Spec Ap /pAp .
Since the prime ideals of Ap /pAp are corresponding to primes ideals of Ap containing pAp , which is correspond‑
ing to prime ideals q ⊂ A such that q ∩ ϕ(B − p) = ∅ and ϕ(p) ⊂ q, which is corresponding to prime ideals q ⊂ A
with ϕ−1 (q) = p, we have X ×Y Spec k(y)  f −1 (y).
(b) Clearly we have Xy = Spec (k[s, t]/(s − t2 )) ⊗k[s] (k[s]/(s − a)) = Spec k[s, t]/(s − t2 , s − a). If a , 0,
k[s, t]/(s − t2 , s − a)  k ⊕ k, so Xy has only two points and each point has residue field k. If a = 0, k[s, t]/(s −
t2 , s − a)  k[t]/(t2 ), so Spec k[t]/(t2 ) is a non‑reduced one‑point scheme. Since η is corresponding to (0) ⊂ Y,
so Xη = Spec (k[s, t]/(s − t2 ) ⊗k[s] k(s) = Spec k(s)[t]/(s − t2 ). Since k(s)[t]/(s − t2 ), Xη is a singleton. Moreover,
k(s)[t]/(s − t2 ) is an extension of degree 2 over k(η).

Solution 2.3.11 (Closed Subschemes). (b) If ϕ : Y → X = Spec A is a closed immersion, we firstly prove that Y
is an affine scheme. For any point P ∈ Y, we can find an neighbourhood U ⊂ X of P such that U ∩ Y = Spec A0
is an affine subset of Y. Take f ∈ A such that D( f ) ⊂ U, then D( f ) ∩ Y is an affine subset of Y. Denote g is
the image of f under A → O(U) → OY (U ∩ Y) = A0 . So D( f ) ∩ Y = D(g)  Spec A0g . Since we can cover X − Y
by some principal open set, combining with above D( f ), we can covering X by {D( fi }i∈I such that D( fi ) ∩ Y is
affine subset of Y. Since Spec A is quasi‑compact, we may assume the above covering is finite for i = 1, . . . , n,
i.e. all fi generate the A. If we denote the image of fi under OX (X) → OY (Y) as f¯i , all f¯i generate the OY (Y).
Moreover, D( f¯i ) = D( fi ) ∩ Y is affine, we know Y is affine by 2.2.17.(b).
So we may assume Y = Spec B. Then the closed immersiom Y → X is induced by ring homomorphism
φ : A → B. Denote a = ker φ, we have a commutative diagram:

ϕ
Spec B / Spec A
t 9
ttt
t
tt
θ
 tt
Spec A/a

T T
where θ is induced by injection A/a → B. Since ϕ(Spec B) = ϕ(Spec B) = V( p∈ϕ(Spec B) p) = V( q∈Spec B φ−1 (q)) =
T √ √
V(φ−1 ( q)) = V(φ−1 ( (0))) = V( a) = V(a). So θ is a homeomorphism on the underlying topological space.
Moreover, for any f ∈ A, we have a homomorphism (A/a) f¯ → Bφ( f ) is injective, because A/a → B is injective,
i.e. θ] is injective. Since θ is a closed immersion, we have θ] is an isomorphism. So Y = Spec A/a.
(a) For any affine piece U = Spec A ⊂ X, then f −1 (U) → U is a closed immersion, so f −1 (U) = Spec A/I
for some ideal I ⊂ A. For any g : X 0 → X, take an affine piece U 0 = Spec A0 ⊂ g−1 (U), then f 0 −1 (U 0 ) =
Spec A0 ⊗A A/I = Spec A0 /IA0 , so f 0 −1 (U 0 ) → U 0 is a closed immersion. Then Y 0 → X 0 is just the glueing‑
together of all f 0−1 (U 0 ) → U 0 , so is a closed immersion.
(c) If X = Spec A is affine, then Y = Spec A/I for some radical ideal I. If Y 0 → X is a closed immersion such
√ √
that Y  Y 0 on the underlying space, then Y 0 = Spec A/I 0 , for some V(I 0 ) = V(I). So I = I = I 0 . So we have
a canonical ring homomorphism A → A/I 0 → A/I, which induces the scheme homomorphism Y → Y 0 → X.
S
For general case, we just consider the affine covering X = U, then U ∩ Y → U may factor through U ∩ Y 0 .
Then we may glue together all U ∩ Y → U ∩ Y 0 → U to get a Y → Y 0 → X.
(d) This question may need to add a condition that f is quasi‑compact and quasi‑separated, and the
proof will use the theory of sheaves of modules. Denote I = ker f ] . Since f is quasi‑compact and quasi‑
separated, we know that f∗ OZ is quasi‑coherent by preposition 5.8.(c), so I is a quasi‑coherent OX ‑module.
Then we can define (Y, OY ) = (Supp (OZ /I ), (OZ /I )|Supp (OZ /I ) ). So i : Y → X is a closed immersion. Clearly
f (Z) ⊂ Supp ( f∗ OZ ) ⊂ f (Z),
¯ since Y is closed, we have Y = f (Z). Then we can define g : Z → Y on the

Yang Pi‑Yeh 29 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

underlying space as induced by f . And f ] : OX → f∗ OZ induces OX /I → f∗ OZ = i∗ g∗ OZ , which induces

OY → g∗ OZ . Hence g : Z → Y is a scheme homomorphism, and clearly f = ig.
If i0 : Y 0 → X is a closed immersion, g0 : Z → Y 0 with f = i0 g0 , we have f (Z) ⊂ i0 (Y 0 ), i.e. f (X) ⊂ i0 (Y 0 ). So
we have a continuous map j : Y → Y 0 with i = i0 j on the underlying topological space. Denoting I 0 as the
ideal sheaf of Y 0 , we clearly know I 0 = ker i0] . Since f = i0 g0 , I 0 is contained in ker f ] , i.e. I 0 is a subsheaf of
I . So we have OX /I 0 → OX /I , i.e. i0∗ OY 0 → i∗ OY = i0∗ j∗ OY . Hence we have OY 0 → j∗ OY . So j : Y → Y 0 is a
morphism of schemes, i.e. Y → X will factor through Y 0 .
By our construction, if Z is a reduced scheme, Y is clearly the reduced induced structure on f (Z).

Solution 2.3.12 (Closed Subschemes of Proj S ). (a) Since ϕ preserves the degrees, we know ϕ(S + ) = T + .
Then U = {p ∈ Proj T | p + T + } = Proj T . Since T  S / ker ϕ, we have a one‑to‑one corresponding between
homogeneous prime ideals of S containing ker ϕ and homogeneous prime ideals of T , so f : Proj T → V(ker ϕ)
is a homeomorphism. For any a ∈ S , we have OProj S (D+ (a)) = S a → T ϕ(a) = OProj T ( f −1 (D+ (a)) is sujective, so
f ] is surjective.
L
(b) If I 0 = I , then (S /I)d  (S /I 0 )d for all d ≥ d0 . Then the morphism f : Proj (S /I) → Proj (S /I 0 )
d≥d0 d
0
induced by S /I → S /I is an isomorphism by 2.2.14.(c).

Solution 2.3.13 (Properties of Morphisms of Finite Type). (a) If f : Z → X is a closed immersion. For any
affine piece U = Spec A of X, f −1 (U) = Z ∩ U → U is also a closed immersion, hence Z ∩ U = Spec A/I for
some ideal I of A. Since A/I is a finitely generated A‑algebra, we know f is of finite type.
(b) If i : Y → X is an open immersion. For any affine piece U = Spec A of X, U ∩ Y is open in U, hence
S S
can be U ∩ Y = i D( fi ) = i Spec A fi for some fi ∈ A. Since A fi is a finitely generated A‑algebra, we know f
is locally of finite type. So f is of finite type since f is also quasi‑compact.
(c) If f : X → Y and g : Y → Z is of finite type. So for any affine piece U = Spec A of Z, there are finitely
S
many Vi ⊂ Y such that g−1 (U) = Vi and Vi = Spec Bi for some finitely generated A‑algebra Bi . Then there
S
are finitely many W ij ⊂ X such that f −1 (Vi ) = W ij and W ij = Spec C ij for some finitely generated Bi algebra C ij .
So all C ij are finitely generated A‑algebra, and the index set of i, j is finite. Hence g ◦ f is of finite type.
(d) If f : X → Y is of finite type, we need to show that X 0 = X ×Y Y 0 → Y 0 is of finite type.
(i) If X = Spec A, Y = Spec B, Y 0 = Spec B0 , we have A is finitely generated B‑algebra, hence A ⊗B B0 is
finitely generated B0 ‑algebra, i.e. X 0 → Y 0 is of finite type.
S
(ii) If Y = Spec B and Y 0 are affine, there exists a finite open affine covering X = Ui such that Ui =
S
Spec Ai for some finitely generated B‑algebra Ai . Hence X 0 has an finite open affine covering X 0 = Spec Ai ⊗B
B0 , where Ai ⊗B B0 is finitely generated B0 ‑algebra, i.e. X 0 → Y 0 is of finite type.
S
(iii) If Y = Spec B is affine, and Y 0 = Yi0 is an open affine covering of Y 0 such that Yi0 = Spec B0i for some
B0i , we have X ×Y Yi0 is finite type over Yi0 by (ii). Sine X ×Y Yi0 is the preimage of Yi0 , we know that X 0 ×X Y is
finite type over X 0 .
S
(iv) Denote the morphism of Y 0 → Y as g. If Y = Yi is an open affine covering of Y such that Yi = Spec Bi
for some ring Bi , f −1 (Yi ) ×Yi g−1 (Yi ) is finite type over g−1 (Ui ) by (iii). So X ×Y g−1 (Ui ) → g−1 (Ui ) is of finite type.
So X 0 → Y 0 is the glueing‑together of all above morphism, hence of finite type.
(e) Since X ×S Y → Y is the base change of X → S , hence of finite type. Since X ×S Y → S is the compound
of X ×S Y → Y and Y → S , hence of finite type.
(f) For any affine piece Spec C ⊂ Z, any affine piece Spec B ⊂ g−1 (Spec C) ⊂ Y, and any affine piece
Spec A ⊂ f −1 (Spec B) ⊂ X, we have Spec A ⊂ (g ◦ f )−1 (Spec C), and A is a finitely generated C‑algebra with the
morphism C → B → A. If {ai }ni=1 generates A as C‑algebra, we have a surjective morphism C[x1 , . . . , xn ] → A as
xi 7→ ai . Since this morphism is factored through C[x1 , . . . , xn ] → B[x1 , . . . , xn ] → A, we know B[x1 , . . . , xn ] → A
is surjective. So A is a finitely generated B‑algebra, i.e. g ◦ f is locally of finite type. Since g is of finite type,
hence quasi‑compact, and f is quasi‑compact too, we know that g ◦ f is also quasi‑compact. Hence g ◦ f is of
finite type.

Yang Pi‑Yeh 30 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

S
(g) Since Y is noetherian, we may assume Y = Yi for some Yi = Spec Bi is a finite open affine covering of
Y. So f −1 (Yi ) can be covered by finitely many Xi j = Spec Ai j with some finitely generated Bi ‑algebra Ai j . Since
each Bi is noetherian, {Spec Ai j } is a finite covering of X with Ai j noetherian, hence X is noetherian.

Solution 2.3.14. This problem is local, so we may assume that X = Spec A is affine. Since X is of finite type
over a field k, we know that A is a finitely generated k‑algebra, hence a Jacobson ring. Denote V = Spm A ⊂ X
as the set of all maximal ideals of A, i.e. the closed points set of X. If some open subset U ⊂ X satisfies
U ∩ V = ∅, as we may assume U = D( f ) as a principal open subset, we have f ∈ m for all maximal ideal of A.
Then f is nilpotent since A is Jacobson, i.e. U = D( f ) = ∅. Hence V is dense in X.
For the counterexample, we just need to take a integral local ring A, then Spec A = {(0), m}. Then taking
an f ∈ m\{0}, we know m < D( f ), hence not dense.

Solution 2.3.15. (a) (i⇒ii) and (iii⇒i) is trivial. (ii⇒iii): We only need to show that if K/k is purely inseperable,
then π : XK → X is a homeomorphism. If XK is not irreducible, we may have two affine piece V1 and V2 disjoint,
then U K defined as the direct sum of those affine pieces is also affine, and clearly U K is not integral. Hence
we only need to consider U K and U = π(U K ), so this problem is local, i.e. we may assume X = Spec A is affine.
And since Spec A is homeomorphic to Spec Ared , we may assume that A is reduced. Firstly we assume that
K/k is generated by a single element, i.e. K  k[T ]/(T q − c) for some c ∈ k and q is a power of char(k). Then for
any prime ideal p ⊂ A, and prime ideal q ⊂ AK lying above p. Then pAK ⊂ q. Conversely, if aq ∈ p for all a ∈ q,
√ √
we have q ⊂ pAK . So q = pAK , i.e. π is injective. The surjection is trivial. Moreover, for any ideal I ⊂ AK
and J = I ∩ A, we have I q ⊂ J, i.e. π(V(I)) = V(J). So π is closed, hence homeomorphism. Secondly for K/k is
a purely seperable finite extension, then K/k is made up of a series of simple extension, then trivial. Finally
for general case, we will prove that XK is homeomorphic to some XK 0 for K 0 /k is finite. For any closed subset
ZK ⊂ XK , we may assume ZK = V(IK ) for some redical ideal IK ⊂ AK . Since XK is a variety, we know that AK is
noetherian, i.e. IK is generated by f1 , . . . , fr . So there exists a finite extension K 0 of k, such that f1 , . . . , fr ∈ AK 0
and generate an ideal IK 0 . So IK 0 induces a closed subset ZK 0 of AK 0 and clearly ZK = ZK 0 ×K 0 K, i.e. the closed
subset of XK is homeomorphic to a closed subset of some XK 0 , hence homeomorphic to a closed subset of X.
So XK is homeomorphic to X.
(b) (i⇒ii) and (iii⇒i) is trivial. (ii⇒iii): We only need to prove that if K/k is seperable, and X is reduced,
we have XK is reduced. This problem is clearly local, we may assume that X = Spec A is affine. For all
minimal prime ideals pi of A (since A is finitely generated k‑algebra, there exist only finitely many pi ), we
Q Q
have A → i A/pi is injective, then AK → i (A/pi )K is injective since ×k K is flat. So we may assume that
A is integral. Since A is a subring of Frac(A), we only need to show that for all separable field extension
F/k, we have F K = F ⊗k K is reduced. Since every element F K is contained in some F K 0 for some finite
separable extension K 0 /k, so we only need to show that Fk is reduced for finite separable extension K/k. Since
K = k[T ]/ f for some separable polynomial f ∈ k[T ], we have F K = F[t]/ f , and f ∈ F[T ] is also separable. So
F K is reduced.
`
(c) Take X = Spec (R/(t2 + 1))  Spec C, then X is integral. But XC = Spec (C/(t2 + 1))  Spec C Spec C is
neither irreducible nor reduced.

Solution 2.3.16 (Noetherian Induction). Denote S = {V | V is a closed subset of X which do not hold the
property P}. If X does not hold P, S is not empty. Since X is noetherian, S has a minimal element, namely
Y. Since every proper closed subset of Y is not in S , they all hold the property P, then Y also holds the P,
which makes a contradiction.

Solution 2.3.17 (Zariski Spaces). (d) If x ∈ X is the generic point, and x < U ⊂ X for some proper open subset
U, then X = {x} ⊂ X − U, which makes a contradiction.
(a) Only need to show that if X is a notherian scheme, every irreducible subscheme Y of X has a unique
generic point. If U = Spec A is an affine piece of X with U ∩ Y , ∅, U ∩ Y ⊂ U is an irreducible closed subset,

Yang Pi‑Yeh 31 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

i.e. U ∩ Y = V(p) for some prime ideal p ⊂ A. Clearly V(p) = {p} in U. Denote y ∈ Y corresponding to p. Then
U ∩ Y ⊂ {y} in Y. Since U ∩ Y is open in Y, and Y is irreducible, we know that U ∩ Y is dense in Y. Hence
{y} ⊃ U ∩ Y = Y, i.e. {y} = Y. If y and y0 are both generic point of Y, by (d) we know that y, y0 ∈ U ∩ Y, and there
√ √
exist two corresponding prime ideal p and p0 in A. So V(p0 ) = U ∩ Y = V(p), i.e. p0 = p, hence y = y0 .
(b) If Y is a minimal nonempty closed subset, for every y ∈ Y, we know {y} ⊂ Y is closed. Since Y is
minimal, we have {y} = Y, i.e. y is the generic point of Y. By the uniqueness of generic point, Y has only one
point.
(c) For any two points x, y ∈ X. If x ∈ X/{y} or y ∈ X/{x}, we’ve done. So, if x ∈ {y} and y ∈ {x}, we know
that {x} = {y} := Z, i.e. x and y are both generic points of Z. By uniqueness we know that x = y. So X satisfies
the axiom T 0 .
(e) If x ∈ X is a minimal point, then if y ∈ {x}, we have y = x by minimality. Hence {x} is a singleton. Then
by (b) we know that x is a closed point. If x ∈ X is a maximal point, we know x ⊂ Y for some irreducible
component of X. Denote y as the generic point of Z. Clearly x ∈ {y}, then x = y by maximality of x. So x is a
generic point of some irreducible component. If Z ⊂ X is a closed subset, z ∈ Z, and x is a specialization of z,
clearly we have x ∈ {z} ⊂ Z.
(f) Clearly t(X) is noetherian since every closed subset of t(X) is induced by some closed subset of X.
For any nonempty closed irreducible subset V ⊂ t(X), we have a closed irreducible subset Z ⊂ X such that
V = t(Z). Then Z itself is a point of t(Z), and clearly Z is the generic point of t(Z), hence t(X) is a Zariski
space. Furthermore, if X itself is a Zariski space, then every closed irreducible subset of X is one‑to‑one
corresponding to its generic point, i.e. this is a bijection t(X) → X. And the continuity is obvious.

Solution 2.3.18 (Constructible Sets). (a) (⇐) Conditions (1) and (3) imply every closed subset of X belongs
to F. Condition (2) implies locally closed subset of X belongs to F. Conditions (2) and (3) imply finite union
of elements of F is in F, so any finite union of distinct locally closed subset of X belongs to F.
(⇒) Denote G as the set of all subset of X which can be written as a finite disjoint union of locally closed
`
subset, then G ⊂ F. Clearly every open subset of X belongs to G, i.e. G satisfies condition (1). If i (Zi ∩
` ` ` `
Ui ), j (Z 0j ∩ U 0j ) ∈ G, we have ( i (Zi ∩ Ui )) ∩ ( j (Z 0j ∩ U 0j ) = i, j ((Zi ∩ Z 0j ) ∩ (Ui ∩ U 0j )) ∈ G, hence G satisfies
condition (2). For condition (3), if U is an open subset of X and Z is a closed subset of Z, then X − (U ∩ Z) =
` ` T
(X − U) ∪ (X − Z) = (X − U) (U ∩ (X − Z)) ∈ G. For general case, X − ( i (Zi ∩ Ui )) = (X − (Zi ∩ Ui )). Since
`
X − Zi ∩ Ui ) ∈ G, by condition (2) we know that X − ( i (Zi ∩ Ui )) ∈ G, hence G satisfies condition (3). Then by
minimality of F, we have G = F, i.e. every element of F is a disjoint union of locally closed subset of X.
(b) (⇐) If η ∈ V with V ∈ F, obviously V̄ ⊃ {η} = X, i.e. V̄ = X, hence dence.
` S S
(⇒) If V = i (Zi ∩ Ui ) ∈ F is dense in X, we have Zi ⊃ V̄ = X, hence Zi = X. Then η ∈ Zi for some i.
Since Ui are all open, by 2.3.17.(d) we know that η ∈ Ui . So η ∈ Zi ∩ Ui ⊂ V.
Furthermore, since every nonempty open subset contains the generic point by 2.3.17.(d), the (⇐) is trivial.
S
Conversely, since Zi = X and X is irreducible, there must exist an i such that Zi = X. So Zi ∩ Ui = Ui , i.e.
Ui ⊂ V.
(c) (⇒) is trivial. (⇐) Obviously we only need to consider the cast S = Z ∩U. If x ∈ S , we have {x} ⊂ S since
S is stable under specialization. For S̄ , denote all generic points of its irreducible components as {η1 , . . . , ηn }.
Since {ηi } ⊂ S̄ , we have {ηi } ⊂ Z j ∩ U j for some j. Then ηi ∈ U j by 2.3.17.(d). So {ηi } ⊂ S , hence S̄ ⊂ S , i.e. S
is closed. Furthermore, if T is open, it is obviously constructed and stable under generalization. Conversely,
if T is constructed and stable under generalization, X − T is constructed and stable under specialization, so
X − T is closed, i.e. T is open.
`
(d) Clearly if f is continuous, f −1 (Zi ) is closed and f −1 (Ui ) is open. So f −1 (V) = f −1 ( i (Zi ∩ Ui )) =
` −1 −1
i ( f (Zi ) ∩ f (U i )) is constructible of X.
S
Solution 2.3.19. (a) Since X is noetherian, X = Spec Ai for finitely many i and Spec Ai are all irreducible.
So if every f (Spec Ai ) is constructible, f (X) is constructible, hence we reduce the question to the case that

Yang Pi‑Yeh 32 Hartshorne Solutions

2 Schemes 2.3 First Properties of Schemes

S
X = Spec A is affine and irreducible. Since Y is also noetherian, Y = Spec Bi for finitely many i and all
Spec Bi are all irreducible. So if every f (X) ∩ Spec Bi is constructible in Spec Bi , we know that f (X) ∩ Spec Bi
is constructible in Y, hence f (X) is constructible in Y. So we reduce the question to the case that Y = Spec B is
also affine and irreducible. By 2.2.3.(b), we clearly can assume that X and Y are reduced, i.e. we may assume
X and Y are integral. Finally, we may denote f (X) = Z ⊂ Y, then we have Z = Spec B/I for some reduced I by
2.3.11.(b). So f (X) = f (X) ∩ Z, the constructibility are the same for those two forms, so we may assume that
Z = Y, i.e. f is dominant.
(b) Firstly we prove this algebraic result by using induction on the number r of generators of B as A‑
algebra. If r = 1, we have B  A[x] or A[x] quotient something. If B  A[x], every b ∈ B has the form
b = bn xn +. . .+b0 , then we just define a = bn . Then for ϕ : A → K with ϕ(a) , 0, the polynomial ϕ(an )xn +. . .+ϕ(a0 )
in K has only finitely many roots in K since K is algebraic closed. Taking an α ∈ K is not a root of this
polynomial, and define ϕ0 (x) = α, we clearly have ϕ0 (b) , 0. For the second condition, we may assume x is a
generator of B as an A‑algebra, then it is algebraic over A with the minimal polynomial am xm + . . . + a0 = 0.
Since b is also algebraic over A, we may assume bn bn +. . .+b0 = 0. And we can formally write bn +. . .+b0 b−n = 0.
Then we can define a = am b0 . For any φ : A → K such that φ(a) , 0, then φ can be extended to a morphism
lim A[a−k ] → K. Since x is integral over A[a−1 ], we know B must be contained in some lim A[a−k ], hence we
←−−k ←−−k
have a morphism φ0 : B → K. And moreover, we know b−1 is also integral over A[a−1 ], it is also contained
in lim A[a−k ]. Since b ∈ B, we know that b has valuation 0 in the valuation ring lim A[a−k ], i.e. an invertible
←−−k ←−−k
element. So φ0 (b) , 0.
For general r, we know that B  A[x1 , . . . , xr ]/p for some prime ideal p. We clearly have a morphism ψ :
A[x1 , . . . , xr−1 ] ,→ A[x1 , . . . , xr ] and ψ−1 (p) ⊂ A[x1 , . . . , xr−1 ] is a prime ideal. Then we have A → A[x1 , . . . , xr−1 ]/ψ−1 (p) →
B are ring extension, so ϕ : A → K can be extended to ϕ00 : A[x1 , . . . , xr−1 ]/ψ−1 (p) → K satisfies the condition,
which can be extended to ϕ0 : B → K by induction. Since x is integral over A[a−1 ],
Back the the question, we may assume X = Spec A, Y = Spec B for noetherian integral domain A, B and f
is dominant. Then f : X → Y is induced by a ϕ : B → A. Then for 1 ∈ A we have an b ∈ B as the algebraic result
above. Then we just need to prove that D(b) ⊂ f (X). For every p ∈ D(b), i.e. b < p, then ϕ : B → Frac(B/p) is a
morphism such that ϕ(b) , 0. Then we know that ϕ can be extended to a ϕ0 with ϕ0 (1) , 0. Then q = ker ϕ0 ⊂ A
is a prime ideal. Clearly q ∩ B = p, which means f (q) = p. So D(b) ⊂ f (X).
(c) For V(b) = Spec B/(b), we have a morphism ϕ̃ : B/(b) → A/(b)A, which induces a f˜ : Spec A/(b)A →
Spec B/(b). Since ϕ is injective, ϕ̃ is injective, i.e. f˜ is dominant. Decompose (b) as (b) = ∩pi for some primary
S √ √ √
ideal pi . Then V(b) = V( pi ). Since we have fi : Spec A/ pi A → Spec B/ pi , by noetherian induction we
S
know that Im fi is constructible in V(pi ). So Im f˜ = Im fi is constructible in Spec B/(b), i.e. f (X) ∩ V(b) = Im f˜
`
is constructible in V(b). Hence f (X) = D(b) ( f (X) ∩ V(b)) is constructible.
(d) Consider X = A1C and Y = P2C , and f : X → Y is f (x) = (x : 1 : 0). Then Im f is neither closed nor open.

Solution 2.3.20 (Dimension). (a) Clearly, since X is integral scheme, for any P ∈ X and affine neighbourhood
U = Spec A of P, if P corresponds to p ⊂ A, we have {Z ⊂ X | Z closed irreducible with P ∈ Z}  {Z ⊂ U | Z
closed irreducible with P ∈ Z}  I ⊂ A | I prime with I ⊂ p}  {I ⊂ Ap = OP | I prime}. Hence (a) is obvious
since we assume more that P is a closed point.
(b) For closed P, clearly FracOP = K(X). So tr.d.K(X)/k = tr.d.FracOP /k = dim OP = dim X.
(c) By the one‑to‑one corresponding in (a), trivial.
(e) Since K(X) = K(U) and (b), trivial.
(d) For the case that X = Spec A and Y = Spec B are both affine and Y is irreducible, we have B = A/p
for some prime p ⊂ A. So dim Y + codim(Y, X) = height p + dim A/p = dim A = dim X. If X and Y are not
necessarily affine but Y is still irreducible, we may pick an affine subset X 0 ⊂ X, and Y 0 = X 0 ∩ Y is affine. Then
clearly dim X 0 = dim X, dim Y 0 = dim Y and codim(Y, X) = codim(Y 0 , X 0 ) by (e) or just counting the dimension
of stalk, so dim X = dim X 0 = dim Y 0 + codim(Y 0 , X 0 ) = dim Y + codim(Y, X). For general case, if Y is not

Yang Pi‑Yeh 33 Hartshorne Solutions

2 Schemes 2.4 Separated and Proper Morphisms

irreducible, we may pick an irreducible component with maximal dimension Z of Y, then dim Y = dim Z and
codim(Y, X) = codim(Z, X), then obvious.
(f) By (e) we only need to consider the affine case, i.e. X = Spec A. Then dim X = tr.d.Frac(A)/k =
tr.d.Frac(A) ⊗k k0 /k0 = dim X 0 .

Solution 2.3.21. (a) If mR = (u) for some u ∈ R, then I = (ut−1) ⊂ R[t] is a maximal ideal. Since R[t]/I  Frac(R),
which has dimension 1. And since X clearly has dimension 2, so dim OP < dim X.
(d) Just take Y = {P}, then dim Y + codim(Y, X) = 0 + 1 < 2 = dim X.
(e) Define U = Spec (R[t])mR [t] , then dim U = 1 , 2 = dim X.

Solution 2.3.22 (Dimension of the Fibres of a Morphism). (a) This problem is local, so we may assume that
X = Spec A and Y = Spec B, f : X → Y induces a ring homomorphism ϕ : B → A. Since Y 0 ⊂ Y closed and
irreducible, we may assume Y 0 = Spec B/p for some prime ideal p ⊂ B, then it’s generic point η0 is the point
in Y corresponding to p. Since f −1 (Y 0 ) = {q ⊂ A | q prime and ϕ−1 (q) ⊂ p}. So every closed irreducible subset
Z ⊂ X with p ∈ f (Z) must has the form Spec A/q with some q satisfying f −1 (q) = p. For every point P ∈ Y 0
corresponding to p0 , i.e. p0 ⊂ p, then there exists a point in Q ∈ Z ∩ f −1 (Y 0 ) corresponding to q0 ⊂ q is the
preimage of P. Hence this induces a surjective morphism OP,Y → OQ,X , hence dim OP,Y ≥ dim OQ,X . Then by
2.3.20.(c), codim(Y 0 , Y) = inf{dim OP,Y | P ∈ Y 0 } ≥ inf{dim OQ,X | Q ∈ Z} = codim(Z, X).
(b) Denote the irreducible component of Xy as Z. Take Y 0 = {y}, then by (a) we have codim(Z, X) ≤
codim(Y 0 , Y). By 2.3.20.(d), we have codim(Z, X) = dim X − dim Z and codim(Y 0 , Y) = dim Y − dim Y 0 , we have
dim X ≥ dim X − dim Y = e.
(c) This problem is local for Y, so we may assume Y = Spec B is affine. If X = Spec A is also affine, we know
that A is a finitely generated B‑algebra. So we can pick t1 , . . . , tn ∈ A as a transcendental base of K(X)/K(Y).
Define X1 = Spec B[t1 , . . . , te ], we clearly know that X1  AeY , and g : X → X1 is generically finite. Then by 3.7.
we know there exists an open dense U 0 ⊂ X 0 , such that g−1 (U 0 )  U 0 , then U = g−1 (U 0 ) ⊂ X is open and dense.
And obviously, h : X 0 → Y clearly have dimension e on every fiber Xy0 . Hence U ⊂ X is the open set we need.
For general integral scheme X, since every affine piece V ⊂ X is open and dense, and for V we have a U ⊂ V
open dense with the property we need, we know that U ⊂ X is also open and dense.
(d) (1) By (b) this is trivial. (2) By (c), there exists a open set U such that Eh ⊂ X − U for h > e, hence Eh
cannot be dense. (3) For h > e, we have Eh ⊂ X − U, then we may define X 0 = X − U with the reduced induced
closed subscheme structure, then clearly X 0 is an integral scheme. Then f : X → Y induces a f 0 : X 0 → Y,
and Eh is clearly the set of all point x ∈ X 0 such that dim X 0f (x) ≥ h. Then by induction of dimension of X, Eh is
closed in X 0 , hence closed in X.
(e) Clearly Ch = f (Eh − Eh−1 ) = f (Eh ) − f (Eh−1 ), so we only need to show each f (Eh ) is constructible. By
2.3.19. this is obviously true by closedness of Eh , hence we’ve done.

Solution 2.3.23. Clearly t(V) ×Spec k t(W) is a integral scheme of finite type over k, i.e. t(V) ×Spec k t(W) = t(X) for
some variety X. Since every closed point of t(X) will project to closed points on t(V) and t(W), i.e. every point
of X corresponds to a point in V and a point in W, which gives us a bijection X  V × W. And the continuity
is obvious.

2.4 Separated and Proper Morphisms

Solution 2.4.1. If f : X → Y is finite, by definition f is affine, hence seperated. Finiteness clearly means
finitely type. And since finiteness is stable under base change by 2.3.13.(d), and by 2.3.5.(b) we know that
finiteness means closed, so f is universally closed.

Solution 2.4.2. Firstly consider the morphism of underlying space. Since f and g are S ‑morphisms, we can
define a h = f × g : X → Y ×S Y, then h = ∆ ◦ f on U. Since Y is seperated over S , ∆(Y) is closed in Y ×S Y,

Yang Pi‑Yeh 34 Hartshorne Solutions

2 Schemes 2.4 Separated and Proper Morphisms

hence h−1 (∆(Y)) is closed is X. Since U ⊂ h−1 (∆(Y)) and U is dense, we know that h−1 (∆(Y)) = X, i.e. f = g on
whole X. Then we just need to consider the sheaf structure. For x ∈ X, we have a y = f (x) = g(x) ∈ Y, we
have fy] , g]y : OY,y → OX,x . We may assume that y is in an affine piece Spec B corresponding to a prime ideal
p and Spec A ⊂ f −1 (Spec B) with x ∈ Spec A. Hence we have two morphisms ϕ, ψ : B → A corresponding to
f, g : Spec A → Spec B. For any b ∈ B we can define a = ϕ(b) − ψ(b) ∈ A. Clearly we have U ∩ Spec A ⊂ V(a)
closed in Spec A and U ∩ Spec A is dense in Spec A, we know that V(a) = Spec A. Since X is reduced we have
a = 0, i.e. f = g on Spec A. Since x is arbitrary, f = g on whole X.
For counterexamples. (a) Define X = Y = Spec k[x, y]/(x2 , xy). Then X and Y are homeomorphic to A1k but
have a nilpotent at (0). Then we may define the f : X → Y be the identity, and g : X → Y by the morphism
corresponding to k[x, y]/(x2 , xy) → k[x, y]/(x2 , xy) with x 7→ 0, y 7→ y. Then f = g on X − {(0)} but not on (0). (b)
Define X and Y as the affine line over k with P doubled as in Example 2.3.6., and f : X → Y as the identity,
g : X → Y as the switching of the doubled point. Then f = g out of the doubled point, but not on whole X.

Solution 2.4.3. We may assume that S = Spec A. Since X is seperated over S , we know that ∆ : X → X ×S X is
a closed immersion. If we denote two projections from X ×S X to X as π1 and π2 , we have U ∩ V = ∆−1 (π−1 1 (U) ∩
π−1
2 (V)). So we have a closed immersion U ∩ V → π −1
1 (U) ∩ π−1
2 (V) = U × S V. Since U × S V is clearly affine,
U ∩ V is also affine.
For counterexample, define X to be the affine surface with a doubled point P, U and V are two copies of
A2 , then U ∩ V = A2 − P is not affine.

Solution 2.4.4. Since Z → S is proper and Y → S is separated, we know that Z → Y is proper, hence f (Z)
is closed. Moreover, since f (Z) → Y is a closed immersion, hence of finite type by 2.3.11.(a), and Y → Z is
of finite type, we know that f (Z) is of finite type over Z. Since ∆( f (Z)) = ∆(Y) ∩ ( f (Z) ×S f (Z)), and ∆(Y) is
closed in Y ×S Y, we know that ∆( f (Z)) is closed in f (Z) ×S f (Z), i.e. f (Z) → S is separated. Finally, since
properness and surjection is stable under base change, we only need to show f (Z) → S is closed. Denote
the morphism X → S and Y → S as p and q. Then for any W ⊂ f (Z), we know that W = f ( f −1 (W)), so
q(W) = q( f ( f −1 (W))) = p( f −1 (W)). Since p is proper, we know that p( f −1 (W)) is closed, hence f (Z) → S is
closed.

Solution 2.4.5. (a) If R a valuation ring of K has center x, then mR ∩ OX,x = m x . So we have a commutative
diagram
Spec K :/ X
u
uu
uuu
uu
 uu 
Spec R / Spec k

Since x is unique, the morphism Spec R → X is unique, hence X is seperated over k.

(b) If R is a valuation ring of K, and X is proper over k, then the following commutative diagram

Spec K /X

 
Spec R / Spec k

induces a morphism Spec R → X. Then obviously the point in X corresponding to the point mR in Spec R is
the center of R.

Yang Pi‑Yeh 35 Hartshorne Solutions

2 Schemes 2.4 Separated and Proper Morphisms

(c) If S is a valuation ring with fraction field L with a commutative diagram

Spec L /X

 
Spec S / Spec k

we may assume the image of Spec L in X is the generic point x of X, or we may take a closed subset Z ⊂ X
with generic point x and change X to Z. Then Spec L → X induces a field extension K → L. Hence valuation
on L induce a valuation ring on K with a valuation ring R = {r ∈ K | v(r) ≥ 0} ⊂ K. So we have the following
commutative ring
Spec L / Spec K /X

  
Spec S / Spec R / Spec k

In the case of (a), if we have two morphisms Spec S → X, then the different two images of mS is the different
two images of mR in the morphism Spec R → X, which makes a contradiction. Hence X is seperated over k.
In the case of (b), by assumption we have a unique morphism Spec R → X, i.e. Spec S → Spec R → X. Hence
X is proper over k.
(d) If a ∈ OX (X) with a < k, since k is algebraically closed, a is transcendental over k, so k[a−1 ](a−1 ) is
a local ring contained in K. Hence there exists a valuation ring R ⊂ K dominates it, i.e. a−1 ∈ mR . By
morphism Spec R → X, we have a morphism OX (X) → R, hence the image of a in R has valuation≥ 0. Thus
v(1) = v(a/a) = v(a) + v(a−1 ) > 0, which makes a contradiction.

Solution 2.4.6. We may assume X = Spec A and Y = Spec B for some finitely generated k‑algebra A and B.
Then we have a k‑algebra morphism ϕ : B → A. Denote K = Frac(A). And there exists a valuation ring R with
fraction field K containing Imϕ. Then we have a commutative ring

Spec K /X

 
Spec R /Y

Since X → Y is proper, we have a morphism Spec R → X, i.e. a ring morphism A → R. By 4.11.A, the integral
closure of Imϕ in K is the intersection of all valuation ring in K containing Imϕ, so A is contained in this
integral closure, i.e. every element of A is integral over B. Since f is of finite type, f is finite.

Solution 2.4.7 (Scheme over Spec R). (a) First consider the case that X = Spec A is affine. Since X has a
semilinear involution σ, we have a ring morphism τ : A → A commutating with the conjugation on C. Then
B = {a ∈ A | τ(a) = a} is clearly a subring of A over R of finite type, since A is a finitely generated C‑algebra.
Then we can just define X0 = Spec B, then X0 ×R C = X.
For general case, for every x ∈ X, we have an affine piece Spec A containing x and σ(x). Since X is
seperated over C, Spec A ∩ σ(Spec A) is affine and contains x and σ(x), hence there exists a ring A x such that
x, σ(x) ∈ Spec A x and σ(Spec A x ) = Spec A x . Then by affine case we know that there exists a finitely generated
R‑algebra Bx can be base changed to A x . Since {A x } is an affine covering of X, we only need to glue all Spec Bx
together to get a X0 . For any A x and Ay , as before Spec A x ∩ Spec Ay is affine, namely Spec A. And there exists

Yang Pi‑Yeh 36 Hartshorne Solutions

2 Schemes 2.4 Separated and Proper Morphisms

a ring B, which can be base changed to A. Hence we have a commutative diagram

Spec A x o Spec A / Spec Ay

  
Spec Bx o Spec B / Spec By

Hence all Bx can be glued together to be a X0 , which satisfies X0 ×R C = X.

(b) If X0 is affine, then X is obviously affine. Conversely, if X = Spec A is affine, then by construction
above, X0 = Spec B for the ring B = {a ∈ A | σ(a) = a}, hence affine.
(c) Clearly if we have a f0 , we clearly have a f = f ×R C such that f ◦ σX = σY ◦ f . Conversely, in the
case that X = Spec A and Y = Spec B, we have a ring homomorphism ϕ : B → A. Since f ◦ σX = σY ◦ f ,
we know ϕ induces a morphism ϕ : Bτ → Aτ , where τ is the complex conjugation on X or Y. Then by the
construction above we know a morphism f0 : Spec Aτ → Spec Bτ , which can be base changed to f . In general
case, for Y we can take affine piece Spec B stable under σ as above, and for any x ∈ f −1 (Spec B), we know that
f (σ(x)) = σ( f (x)) ∈ Spec B, i.e. f −1 (Spec B) is stable under σ, so we can take affine piece Spec A ⊂ f −1 (Spec B)
stable under σ as above. So we have the f0 as in the affine case. Hence we may glue all f0 together to get a
f0 , which can be base changed to f as we want.
(d) If X = A1C = Spec C[t], we know that X0 = Spec (C[t])τ = Spec R[t] = A1R .
(e) If X = P1C and σ has a fixed closed point x, we may define U = X − {x}, which is affine and isomorphic
to Spec C[t]. Then σ : U → U induces a morphism τ : C[t] → C[t], and τ is an involution commutating with
complex conjugation. Hence we may assume that τ(at + b) = āt + b̄, i.e. U0 = Spec R[t] by construction in (a).
Hence X0 = P1R .
If sigma does not have any fixed closed point, for any closed point x ∈ X, we may define U = X − {x, σ(x)},
which is isomorphic to Spec C[t, t−1 ]. Then σ : U → U induces an involution τ : C[t, t−1 ] → C[t, t−1 ] commutat‑
ing with complex conjugation. Since τ(τ(t)) = t, it has two cases: τ(t) = −t or τ(t) = at−1 for some a ∈ C. (1) If
τ(t) = −t, τ will fixed C, which is contradict with the assumption that σ does not have any fixed closed point.
√ √ √ √
(2) If τ(t) = at−1 , then τ(a) = τ(tτ(t)) = τ(t)t = a, hence a ∈ R. If a ≥ 0, then τ(t − a) = at−1 − a = − at−1 (t − a),
√
i.e. σ can fix the prime ideal (t − a), which makes a contradiction, hence a < 0. Under the coordinate chang‑
ing t 7→ a−1/2 t, we may assume that a = −1, i.e. τ(−t) = 1/t. Since we have an isomorphism C[ XZ , YZ ]/(1 + XY Z2
) via
−t 7→ YZ and t−1 7→ XZ , and the corresponding C[−t]  C[ YX , XZ ]/( YX + ( XZ )2 ) and C[t−1 ]  C[ YX , YZ ]/( YX + ( YZ )2 ),
we know that τ switch X and Y. Since P1C  Proj C[X, Y, Z]/(XY + Z 2 ), we may change the coordinates
X+Y
2
7→ U and i(Y−X) 2
7→ V and have P1C  Proj C[U, V, Z]/(U 2 + V 2 + Z 2 ). So by the construction above,
X0 = Proj C[U, V, Z]/(U 2 + V 2 + Z 2 ).

Solution 2.4.8. (d) If f : X → Y and g : Z → W have P, by base change, X × A → Y × A and Y × A → Y × B

have P. So the composition X × A → Y × B has P, which is just f × g.
(e) Since X → X ×Z Y is the base change of Y → Y ×Z Y by f × id, and Y is separated, i.e. Y → Y × Y is closed
immersion, X → X ×Z Y has P. Since X ×Z Y → Y is the base change of g ◦ f , so X → X ×Z Y → Y has P.
(f) Since Xred → X is a closed immersion, we know Xred → X → Y has P. Since Xred → Y can factor
through Xred → Yred by 2.2.3.(c), and Yred → Y is separated, we know Xred → Yred has P.

Solution 2.4.9. If f : X → Y and g : Y → Z are projective, we may assume they have factorizations X → PnY →
Y and Y → PmZ → Z. Since (PmZ ×Z PnZ )×PnZ Y  PmZ ×Z Y  PmY , hence g◦ f factorize as X → PmY → PmZ ×Z PnZ → PnZ → Z.
Since X → PmZ ×Z PnZ is clearly a closed immersion, we only need to prove that there exists a closed immersion
PmZ ×Z PnZ → Pmn+m+n
Z , namely the Segre immersion.
For the case Z = Spec Z, we know that PmZ = Proj Z[x0 , . . . , xm ], PnZ = Proj Z[y0 , . . . , yn ], Pmn+m+n
Z = Proj Z[z00 , . . . , zmn ].
On open piece D+ (xi ) × D+ (y j ) → D+ (zi j ) we have the closed immersion induced by Z[z00 , . . . , zmn ](zi j ) →

Yang Pi‑Yeh 37 Hartshorne Solutions

2 Schemes 2.4 Separated and Proper Morphisms

Z[x0 , . . . , xm ](xi ) ⊗Z Z[y0 , . . . , yn ](y j ) , zzkli j → xxki ⊗ yylj . So glueing them together we have a closed immersion
PmZ ×Z PnZ → PZmn+m+n . For general Z, we only need to prove that projective property is stable under base
change.
If f : X → Y is projective, it may be factorized as X → PnY → Y. Then for any Y 0 → Y and X 0 = X ×Y Y 0 , we
have f 0 : X 0 → Y 0 can be factorized as X 0 → PnY ×Y Y 0 → Y 0 . Since PnY ×Y Y 0 = PnZ ×Z Y ×Y Y 0 = PnZ ×Z Y 0 = PnY 0 , f 0
is projective, i.e. the property of projective is stable under base change.
For (a) in 2.4.8., closed immersion X → Y is clearly projective via factorizing P0Y . And we’ve done the (b)
and (c). So by 2.4.8. projective morphisms have properties (d)‑(f).

Solution 2.4.10 (Chow’s Lemma). (a) Since S is noetherian, X is also noetherian, hence X have irreducible
components X1 , . . . , Xn . For every Xi , we denote Vi = Xi − ∪ j,i X j . Then Vi is open dense in Xi , and open in X.
Since f : X → S is proper, the induced morphism fi : Xi → S is proper. If we have gi : Xi0 → Xi have the
` 0
property we need, and Ui ⊂ Xi open dense is isomorphic to g−1 i (U i ), we just only need to define X =
0
Xi and
0
we have a morphism g : X → X such that g and f g is projective. Take U = ∪(Ui ∩ Vi ). Then U is dense in X
and clearly U  g−1 (U). So we only need to consider the case that X is irreducible.
(b) S has a finite affine covering S = ∪Spec Bi , and each f −1 (Spec Bi ) has a finite affine covering f −1 (Spec Bi ) =
∪Spec Ai j for some finitely generated Bi ‑algebra Ai j . If a1 , . . . , an is a set of generators of Ai j , we have a surjective
morphism Bi [x1 , . . . , xn ] → Ai j as xk → ak , which induces a closed immersion Spec Ai j → Spec Bi [x1 , . . . , xn ].
So we have an immersion Spec Ai j → PnBi . Since we have an open immersion Spec Bi → S , which induces an
immersion PnBi → PnS , i.e. we have an immersion Spec Ai j → PnS . So it can be factorized as Spec Ai j → Pi j → PnS
for some open immersion Spec Ai j → Pi j . And we only need to change notations Spec Ai j to Ui and Pi j to Pi .
(c) Since X and P are both proper over S , X ×S P is proper over S , so X 0 closed in X ×S P is also proper
over S . Then by corollary 4.8.(e), X 0 is proper over P. So we only need to show that X 0 → P is an immersion.
Define Wi as the preimage of Ui of the projection P → Pi . Then we have h−1 (Wi ) = X 0 ∩ (X ×S Wi ). Since
X 0 ∩ (X ×S Wi ) form an open covering of X 0 , we only need to show that X 0 ∩ (X ×S Wi ) → Wi is a closed
immersion. Consider Wi → Wi → X, the graph map Wi → X ×S Wi is a closed immersion. Since the image of
U → P is contained in every Wi , we have morphisms U → Wi . Similarly we have a morphism U → X ×S Wi
induced by U → X ×S P, and this morphism factorize Wi . Since X 0 ∩ (X ×S Wi ) is the scheme‑theoretic image of
U → X ×S Wi , the morphism X 0 ∩ (X ×S Wi ) → X ×S Wi factorizes through the closed immersion Wi → X ×S Wi .
So X 0 ∩ (X ×S Wi ) → Wi are closed immersions.
(d) Clearly X 0 ∩ (U ×S P) is the scheme‑theoretic image of U → U ×S P and this is closed since it is a graph
morphism, so g−1 (U) = X 0 ∩ (U ×S P) is clearly U itself.

Solution 2.4.11. (a) If L/K has transcendental degree n, then we may find x1 , . . . , xn ∈ L such that L/K(x1 , . . . , xn )
is finite. Then O[x1 , . . . , xn ](m[x1 ,...,xn ]) is a noetherian local domain with quotient field K(x1 , . . . , xn ), hence
we just need to consider the case that L/K is finite. Consider a system of parameters {x1 , . . . , xn } of m i.e.
√ √
(x1 , . . . , xn ) = m and x1 < (x2 , . . . , xn ). Then in the ring O 0 = O[x2 /x1 , . . . , xn /x1 ], mO 0 = (x1 ) is not the unit
ideal. Hence for any minimal prime ideal p lying over (x1 ), then by Krull’s principal ideal theorem we have
height p = 1, i.e. dim Op0 = 1. Hence by Krull‑Akizuki theorem, dim Õp0 is noetherian of dimension 1. So if R is
any localization of Õp0 at one of its maximal ideals, R is a discrete valuation ring in L dominating O since R is
principal.
(b) For any valuation ring O and its fraction field K with the commutative diagram

Spec K /X

 
Spec O /Y

Yang Pi‑Yeh 38 Hartshorne Solutions

2 Schemes 2.5 Sheaves of Modules

We have a field extension L/K with a discrete valuation ring R ⊂ L dominating O satisfying the commutative
diagram
Spec L / Spec K /X

  
Spec R / Spec O /Y

Then every morphism Spec O → X corresponds to a morphism Spec R → X by composing Spec R → Spec O.
Conversely, every morphism Spec R → X corresponds to a morphism Spec O → X set‑theoretically. On the
sheaf, if we denote x ∈ X as the image of (0) ⊂ R, we have O x → R(0) = L factors through K, i.e. it induces a
morphism O x → O(0) . And if we denote x ∈ X as the image of mR ⊂ R, since R dominate O, the morphism
O x → RmR factors through Om , i.e. we have a morphism of schemes Spec O → X. Hence the morphism from
Spec R to X is one‑to‑one corresponds to the morphism from Spec O to X, so the criteria of separable or proper
is just need to consider the case of discrete valuation rings.

Solution 2.4.12 (Examples of Valuation Rings). (a) Consider the smooth projective curve C K , we have the
following commutative diagram
Spec K / CK

 
Spec R / Spec k

Then we denote the image of the maximal ideal m of R in C X as x. If x is the generic point, we have a morphism
K = OX,x → R, which makes a contradiction with that R is not a field. Hence x is a closed point, and we have
a morphism OX,x → R, which means R dominate OX,x . By theorem 6.1.A in chapter I, we know that OX,x is a
maximal element in the set of local ring in K, hence OX,x  R, i.e. R is a discrete valuation ring.
(b.1) Take an affine piece U = Spec A containing x1 . If x1 corresponds to a prime ideal p ⊂ A, we know
that K = Frac(A) and OX,x1 = Ap . Since Y has codimension 1, we know that height p = 1, so dim Ap = 1. Since X
is nonsingular, A is normal, hence Ap is normal. And since Ap is clearly noetherian, Ap is a discrete valuation
ring. Obviously Ap has center x1 .
(b.2) We may assume that X 0 is smooth or we just need to blow up it several times to make it smooth.
Then by (b.1) R is a discrete valuation ring. Since f induces an injection OX,x0 → R, we clearly have R dominate
OX,x0 , i.e. R has center x0 .
(b.3) By definition R is a valuation ring and dominate R0 . Since R0 dominate OX,x0 , clearly R dominate
OX,x0 , i.e. has center x0 .

2.5 Sheaves of Modules

Solution 2.5.1. (a) For every open subset U ⊂ X, we can define φ : E (U) → Eˇˇ (U) = Hom(H om(E , OX )|U , OX |U )
as s 7→ {ρ s,V }V , where V runs through all open subset of U, and ρ s,V : Hom(E , OX )(V) → OX (V) as t 7→ tV (s|V ).
Since E is locally free of finite rank, we have H om(E , OX )P = Hom(EP , OX,P ) for every point P ∈ X. Hence on
stalks, the φ induces EP → Hom(Hom(EP , OX,P ), OX,P ) as sP 7→ ρ s,P for some ρ s,P (tP ) = tP (sP ). Since EP is free of
finite rank, the morphism φP is clearly an isomorphism, hence φ is an isomorphism.
(b) For every open subset U ⊂ X, we can define φ : Eˇ (U) ⊗ F (U) → Hom(E |U , F |U ) as ε ⊗ s 7→ θε, f,U ,
where θε, f,U (e) = ε(e) · f |V for any e ∈ E (V) for some open subset V of U. So on every P ∈ X, we have
φP : EˇP ⊗ FP → Hom(EP , FP ) as φP (ε ⊗ f ) = θε, f,P where θε, f,P (e) = ε(e) · f . It is clearly an isomorphism, hence
φ is an isomorphism.
(c) We may define a φ : HomOX (E ⊗ F , G ) → HomOX (F , H omOX (E , F )) as on every V ⊂ U ⊂ X both
open, φ(ψ)(s)(t) = ψV (t ⊗ s|V ) for s ∈ F (U) and t ∈ E (V). Firstly, if φ(ψ) = 0, for every U ⊂ X open, s ∈ F (U)

Yang Pi‑Yeh 39 Hartshorne Solutions

2 Schemes 2.5 Sheaves of Modules

and t ∈ G (U), we have ψ(t × s) = 0, i.e. ψ is a zero map. Hence φ is injective. Conversely, for any morphism
θ : F → H omOX (E , G ), on open subset U ⊂ X, we can define a ψ : E (U) ⊗ F (U) → G (U) as ψ(t ⊗ s) = θ(s)(t)
for any s ∈ F (U) and t ∈ G (U). Hence ψ is a morphism of sheaves as the preimage of θ, so φ is also surjective.
(d) (Projection Formula) f∗ F ⊗OY E  H omOY (Eˇ , f∗ F )  f∗ H omOX ( f ∗ (Eˇ ), F )  f∗ (F ⊗OX f ∗ E ).

Solution 2.5.2. (a) For every OX ‑module F , we may define M = F (X) and L = F (U) where U = {(0)} open
in X, then we clearly have the morphism ρ. Conversely, for every M, L, ρ, we may define a OX ‑module F as
F (X) = M and F (U) = L, and the restriction map F (X) → F (U) is the map M ⊗ 1 → L.
ρ
(b) If F is quasi‑coherent, i.e. F = M̃, then L = Mm = M ⊗R K → − L. So ρ is an isomorphism. Conversely,
if ρ is an isomorphism, i.e. L = Mm . So F = F ](X) = M̃, i.e. quasi‑coherent.
φ(m)|
Solution 2.5.3. For any φ : M → F (X), we may define ψ : M̃ → F as M f → F (D( f )), fmn 7→ f nD( f ) on every
principal open subset D( f ) for some f ∈ A. Conversely, for any ψ : M̃ → F , we only need to define φ = ψ(X).

Solution 2.5.4. (⇒) For every x ∈ X, there exists an affine open neighborhood U = Spec A of x. If we denote
F (U) = M, we know F |U = M̃ since F is quasi‑coherent. Since M is an A‑module, it must be a cokernal
of a morphism between free A‑modules. Hence on U, F |U is a cokernal of a morphism between free OX |U ‑
modules. Moreover, if F is coherent, i.e. M is finitely generated A‑module, we can take two free A‑modules
above finitely generated, i.e. F is a cokernal of a morphism between free OX |U ‑modules of finite rank.
(⇐) For every x ∈ X we have a neighborhood U ⊂ X such that F |U is the cokernal of a morphism between
free OX |U ‑modules. Shrinking U to make U = Spec A is affine, we know on U, free OX |U ‑module is of the form
Ãn for any cardinality n, i.e. quasi‑coherent, so the cokernal is quasi‑coherent. So F |U = M̃ for some A‑module
M. Hence F is quasi‑coherent. Moreover, if F |U is the cokernal of a morphism between free OX |U ‑modules
of finite rank, i.e. cokernal of a morphism between two coherent sheaves, it is clearly coherent.

Solution 2.5.5. (a) Just take X = Spec k[s, t] and Y = Spec k[s]. Then f∗ OX is not coherent, since k[s, t] is not
finitely generated over k[s].
(b) If f : Z → X is a closed immersion, for some affine piece U = Spec A ⊂ X, we have U ∩ Z is closed in
U, hence homeomorphic to some Spec A/I by corollary 5.10. On U, since A/I = OX (U ∩ Z) → OZ (U ∩ Z) = B
is surjective, we have A → A/I → B is surjective. And since B is clearly a finite A‑module, f is finite.
(c) Since f : X → Y is finite, for any affine piece U = Spec A in Y, f −1 (U) = Spec B for some finite A‑module
B. Since F is coherent, we know that F | f −1 U = M̃ for some finitely generated B‑module M. So by proposition
5.2., f∗ F |U = AgM, where A M means to treat M as an A‑module, which is clearly finitely generated. So f∗ F is
coherent.

Solution 2.5.6 (Support). (a) Supp m = {x ∈ Spec A | m x , 0} = {p ⊂ A prime | mp , 0} = V(Ann m).

S
(b) Clearly Supp F = m∈M V(Ann m). Since M is finitely generated, we may take a set of generators
S S T
{m1 , . . . , mn }, hence m∈M V(Ann m) = ni=1 V(Ann mi ) = V( ni=1 Ann mi ) = V(Ann M).
S
(c) If X is a noetherian scheme, we have an affine finite covering X = Spec Ai for some ring Ai , and F
on every affine piece Spec Ai has the form M̃i for some Ai ‑module Mi . Since Supp F ∩ Spec Ai = V(Ann Mi )
S
is closed, Supp F = V(Ann Mi ) is also closed.
(d) By 2.1.20. we have an exact sequence 0 → HZ0 (F ) → F → j∗ (F |U ), where U = X − Z and j : U → X
is the open immersion. Then by proposition 5.8., j∗ (F |U ) is quasi‑coherent, so HZ0 (F ) is quasi‑coherent by
proposition 5.7. Since A is noetherian, we know ΓZ (F ) = {m ∈ M | Supp m ⊂ V(a)} = {m ∈ M | an m = 0 for some
n} = Γa (M), so HZ0 (F ) = Γ] a (M).
(e) Similarly with (d), we have the exact sequence 0 → HZ0 (F ) → F → j∗ (F |U ). By proposition 5.8.,
j∗ (F |U ) is quasi‑coherent (resp. coherent), then HZ0 (F ) is quasi‑coherent (resp. coherent).

Solution 2.5.7. (a) For any affine neighbourhood U = Spec A of x, we may assume x corresponds to prime
ideal p ⊂ A, F |U = M̃ for some finitely generated A‑module M, and Mp  A⊕n
p for some n ∈ Z. If M is generated

Yang Pi‑Yeh 40 Hartshorne Solutions

2 Schemes 2.5 Sheaves of Modules

Q
by {m1 , . . . , mk }, we may assume that m1i corresponds to ( asi1i1 , . . . , asinin ) ∈ Anp . So we may define s = i, j si j , hence
P ai j
mi |D(s) = j si j e j , where e j ∈ Mp corresponding to (0, . . . , 0, 1 j , 0, . . . , 0) ∈ A⊕n p . So we have a surjection φ s :
A⊕n
s → M s . Moreover, since A is noetherian, M is finite presentation, i.e. ker φ s = Im(θ) for some θ : A s → A s
⊕m ⊕n

and some m. Since θ can be written as a matrix (ai j )i, j , and ker φ s will vanish in Ap , there exist some ti j ∈ A − p
Q
such that ai j ti j = 0. So we just need to define t = s · i, j ti j , then D(t) ⊂ D(s), and ker φ s vanish in D(t), i.e.
F|D(t)  OX |nD(t) .
(b) (⇒) Trivial. (⇐) By (a), trivial.
(c) (⇒) Just define G = H omOX (F , OX ). So we have a morphism F ⊗ H omOX (F , OX ) → OX , and it is
clearly isomorphism on stalks since F is locally free of rank 1, i.e. this morphism is an isomorphism.
(⇐) Since F ⊗ G  OX , for every P ∈ X we have FP ⊗ GP  OX,P . Since FP and GP are both OX,P ‑modules,
and OX,P is a local ring, we know FP  OX,P . Then by (b), F is locally free of rank 1.

Solution 2.5.8. (a) For every n, and x ∈ X such that ϕ(x) = k < n, we can choose an affine neighbourhood
U = Spec A of x. So we may assume F |U = M̃ for some finitely generated A‑module M, and M is generated
by {m1 , . . . , mr }. If prime ideal p ⊂ A corresponds to x, we have F x ⊗ k(x)  Mp /pMp . So we can take a set of
elements {u1 , . . . , uk } in M such that they form a basis of Mp /pMp as k(x) vector space. Then Nakayama lemma
a P a Q
implies {u1 , . . . , uk } generate Mp as Ap ‑module, so there exist fii jj ∈ Mp such that m j = i fii jj ui . Define f = i, j fi j .
Then p ∈ D( f ), and for any q ∈ D( f ), m j ∈ Mq , then {m1 , . . . , mr } generate Mq as an Aq ‑module. Hence {u1 , . . . , uk }
generate Mq as an Aq ‑module, i.e. ϕ(q) ≤ k < n. Thus {x ∈ X | ϕ(x) < n} is open, i.e. ϕ is upper semi‑continuous.
(b) For any n ∈ N, we can define Un = ϕ−1 (n) ⊂ X. For any x ∈ Un , by 5.7.(a), there exists an open
neighbourhood U of x such that F |U  OX |nU is free. Hence for every y ∈ U, we have ϕ(y) = n. So Un is an open
S
set. Since Ui ∩ U j = ∅ for i , j, and Ui = X by definition, and since X is connected, we know there exists
some n ∈ N such that Un = X, and Ui = ∅ for all i , n, i.e. ϕ ≡ n is a constant function.
(c) For any x ∈ X and any affine neighbourhood U = Spec A of x, we may assume F |U = M̃ for some finitely
generated A‑module M. Take a set {m1 , . . . , mn } of M forming a basis of Mp /pMp as an k(x) vector space. Then
by Nakayama lemma, {m1 , . . . , mn } generate Mp as Ap ‑module, hence they generate Mq as Aq ‑module for all
P
q ⊂ p. Since ϕ is constant, the images of m1 , . . . , mn in Mq /qMq are linearly independent. So if ai mi = 0 in
Mp , we have ai = 0 in all Aq /qAq for all q ⊂ p. Since X is reduced, we know ai = 0 in Ap , hence mi are linearly
independent over Ap , i.e. Mp is a free Ap ‑module. Then by 5.7.(b), F is locally free.

Solution 2.5.9. (a) Since S 1 generates S , if S 1 = {si }i , {D+ (si )}i is an affine covering of X. For all m ∈ M, it must
be contained in some Md , hene m has degree 0 in M(d) si = M(n))(D ] + (si )) for all si , i.e. a section in D+ (si ). Since
D+ (si ) ∩ D+ (s j ) = D+ (si s j ), we know the sections defined by m above agree on all intersections, hence they
can be glued together to be a section of Γ∗ ( M̃), i.e. we have a morphism α : M → Γ∗ ( M̃). Moreover, for any
m ∈ Md , s ∈ S d0 , we have s · α(m) is the image of m × s in Γ(X, M(d) ] ⊗ OX (d0 )) = Γ(X, M(d ] + d0 )), i.e. α preserves
the grade.
(b) First consider the case M̃ = 0 and we need to prove that Md = 0 for d  0. By proposition 7.4.
in Chapter I, we have a filtration of M as 0 = M0 ⊂ . . . Mn = M, with 0 → Mi−1 → (S /pi )(li ) → 0 for all
i, where pi are homogeneous prime ideals in S and li ∈ Z. Since M̃ = 0, by induction of exact sequences
0 → M˜i−1 → M̃i → S ] /pi (li ) → 0, we know that M̃i = 0 and Sg /pi = 0 for all i. So pi ⊂ S + for all i, i.e. (S /pi )(li ) has
non‑zero elements in only one degree. Since the filtration is finite, we know that M has non‑zero elements in
finitely many some degrees, i.e. Md = 0 for all d  0. For general M, we may denote M 0 = Γ∗ ( M̃) and have
M 0 ⊂ M. Then M/M 0 satisfies the case above, i.e. (M/M 0 )d = 0 for all d  0. Then Md = Md0 for all d  0.
(c) By (b), if M is finitely generated, M is equivalent to Γ∗ ( M̃). And by proposition 5.15., we know that
if F is quasi‑coherent, Γ∗ (F˜ )  F . So we only need to show that for any quasi‑finitely generated graded
S ‑module M, M̃ is coherent, and conversely for any coherent sheaf F , Γ∗ (F ) is quasi‑finitely generated.
Firstly, for any quasi‑finitely generated graded S ‑module M, there exists a finitely generated graded S ‑
msdi
module M 0 such that for some d, M≥d  M≥d
0
. So for any si ∈ S 1 , since m
sni
= sn+d
, M(si )  M(s0 i ) , hence finitely
i

Yang Pi‑Yeh 41 Hartshorne Solutions

2 Schemes 2.5 Sheaves of Modules

generated. Since {D+ (si )}i forms an affine covering of X, and M̃|D+ (si ) are all coherent, we clearly have M̃ is
coherent.
Conversely, for every coherent sheaf F , by theorem 5.17., F (n) is generated by a finite number of global
sections for n  0. So we can define M as the submodule of Γ∗ (F ) generated by these sections. Then we
have an injection M̃ → F . Since we obviously have the isomorphism M(n) ]  F (n), M̃  F , and hence
]
Md  M(d)(X)  F (d)(X) = Γ∗ (F )d for d  0, i.e. Γ∗ (F ) is quasi‑finitely generated.

Solution 2.5.10. (a) For any s ∈ I, ¯ we have s = P sd for some homogeneous elements sd with degree d. Since
P
xin s = xin sd ∈ I, xin sd are homogeneous elements, and I is a homogeneous ideal, we have xin sd ∈ I for each i
and d, i.e. sd ∈ I¯ for all d, hence I¯ is a homogeneous ideal.
(b) Clearly for any homogeneous ideal I we have I xi  I¯xi for all xi by definition. So I1 and I2 define the
same closed subscheme ⇔ I˜1 = I˜2 ⇔ I1,xi = I˜1 (D+ (xi )) = I˜2 (D+ (xi )) = I2,xi ⇔ I¯1 = I¯2 .
(c) Denote I = Γ∗ (IY ). Since I¯ is homogeneous, for any homogeneous s ∈ I¯ with degree d, i.e. xin s ∈ I for
all xi and some n. Since xin s has degree n + d, i.e. xin s ∈ Γ(IY (n + d)). So s = xi−n xin s ∈ Γ(Ui , IY (n + d) ⊗ OX (−n)) =
Γ(Ui , IY (d)) for all i. Since {Ui } forms an open covering of X, there exists an s ∈ Γ(IY (d)), i.e. I = I. ¯
(d) By (a), (b) and (c), trivial.

Solution 2.5.11. For any homogeneous element f ∈ S and g ∈ T with same degree greater than 0, we have
m n+m s⊗ f n+m gn t
an isomorphism S f ⊗A T g  (S ×A T ) f ⊗g as ( fsn , gtm ) 7→ f g ( f ⊗g) n+m and ( fsn , gtn ) ←[ ( fs⊗t
⊗g)n
. Hence we have an
isomorphism D+ ( f ) ×A D+ (g)  D+ ( f ⊗ g). Moreover, for any f, f ∈ S , g, g0 ∈ T homogeneous for deg f = deg g
0

and deg f 0 = deg g0 , we have a cumbersome but trivial commutative diagram for restriction

D+ ( f ) ×A D+ (g)

/ D+ ( f ⊗ g)
O O

D+ ( f f 0 ) ×A D+ (gg0 )

/ D+ ( f f 0 ⊗ gg0 )

 
D+ ( f 0 ) ×A D+ (g0 )

/ D+ ( f 0 ⊗ g0 )

So, since D+ ( f ) for all f ∈ S homogeneous with degree greater than 0 form an open covering for S , and same
for T , we have Proj (S ×A T )  X ×A Y by construction of fibre product.
To prove O(1)  p∗1 (OX (1)) ⊗ p∗2 (OY (1)), we just need to check on all affine piece D+ ( f ⊗ g). But actually
O(1)(D+ ( f ⊗g)) = (S ×A T ) f ⊗g , and (p∗1 (OX (1))⊗ p∗2 (OY (1)))(D+ ( f ⊗g)) = (p∗1 (OX (1)))(D+ ( f ⊗g))⊗(p∗2 (OY (1)))(D+ ( f ⊗
g)) = OX (1)(D+ ( f )) ⊗ OY (1)(D+ (g)) = S f ⊗A T g , then trivial.

Solution 2.5.12. (a) We may assume in : X → PnY and im : X → PmY are two closed immersions and L  i∗n OPnY (1),
M  i∗m OPmY (1). Then consider i : X → PnY ×Y PmY → PYN , where N = nm + n + m, the first arrow is just in × im ,
and the second arrow is the Segre embedding. Then L ⊗ M = i∗n (OPnY (1)) ⊗ i∗m (OPmY (1)) = i∗ (OPN (1)), hence very
Y
ample.
(b) We may assume in : X → PnY and im : Y → PmZ are two closed immersions and L  i∗n OPnY (1), M 
∗
im OPmZ (1). Then consider i : X → PnY  PnZ ×Y → PnZ ×PmZ  PnZ ×PmZ ×Z  PZN ×Z  PZN , where N = nm+n+m, and the
three arrows are in , id×im and the Segre embedding×id. Then L ⊗ f ∗ M = i∗n (OPnY (1))⊗ f ∗ i∗m (OPmZ (1)) = i∗ (OPN (1)),
Z
hence very ample.

Solution 2.5.13. Clearly, S (d) is generated by S d as an S 0 ‑algebra. So {D+ ( f )} f ∈S d forms an open covering on
Proj S (d) and also X. So we just need to check on each affine piece, i.e. to prove the isomorphism ϕ : S ((d)f )  S ( f ) ,
which is just fsn 7→ fsn , so we’ve done. Moreover, on every affine piece D+ ( f ), O(1)(D+ ( f )) = S (d) (1)( f ) =
S (d)( f ) = OX (d)(D+ ( f )). So just glue together all isomorphism, we have O(1)  ϕ∗ OX (d).

Yang Pi‑Yeh 42 Hartshorne Solutions

2 Schemes 2.5 Sheaves of Modules

Solution 2.5.14. (a) X is reduced since it’s normal, so X is irreducible since it’s connected. So S is a domain.
L
Defining I = O (n) as hint, then we have Ip = { sf ∈ S p | deg s ≥ deg f } for all p ∈ X. Since S p is
n≥0 X
integrally closed, Ip is integral over S p , i.e. I is a sheaf of integrally closed domain. Then S 0 = Γ(X, I ) is
integrally closed, which clearly contains S . Moreover, S 0 is obviously the minimal one which contains S and
is integrally closed as we’ve seen in the proof of theorem 5.19., i.e. S 0 is the integral closure of S .
(b) By definition we clearly know that S̃  S˜0 , then by 2.5.9.(b) we know that S d = S d0 for d  0.
0
(c) By (b), we know for d  0, we have S nd = S nd for all n since S 0 = S 00 = k, i.e. S (d)  S 0(d) . Then for any x
integral over S (d) , i.e. x satisfies xn + sn−1 xn−1 + . . . + s0 = 0 for some si ∈ S (d) , we know x ∈ S 0 and clearly has
degree being divisible by d. So x ∈ S 0(d) = S (d) . So S (d) is integrally closed.
(d) (⇒): If X is projectively normal, S is integrally closed, i.e. S = S 0 . Then by (a), we have S n = Γ(X, OX (n)).
Since A[x0 , . . . , xr ] → A[x0 , . . . , xr ]/I = S is clearly surjective and perverse the degree, we have Γ(Pr , OPr (n)) →
Γ(X, OX (n)) is surjective.
(⇐): Just need to prove S = S 0 . If not, the projection A[x0 , . . . , xr ] → S → S 0 is clearly not surjective, i.e.
Γ(P , OPr (n)) → Γ(X, OX (n)) is not surjective, which makes a contradiction.
r

Solution 2.5.15 (Extension of Coherent Sheaves). (a) If X = Spec A is an noetherian affine scheme, F is a
quasi‑coherent sheaf on X. We may assume F = M̃ for some A‑module M. Since X is noetherian, A is also
S
noetherian, so M = Mα for all finitely generated A‑module Mα . Then on any principal open subset D( f ) for
S
some f ∈ A, clearly we have M f = Mα, f . And for any open subset U ⊂ X, we just take a direct limit from
S
principal open subset to U to get that F = Fα , where Fα = M̃α .
S
(b) By proposition 5.8., we know that i∗ F is quasi‑coherent. Then by (a), we have i∗ F = Fα for all
coherent subsheaves Fα = M̃α for i∗ F . Since A is noetherian, M̃α has maximal element, namely M, so i∗ F =
F 0 := M̃, i.e. coherent.
(c) Clearly ρ−1 (i∗ F ) is quasi‑coherent, and clearly ρ−1 (i∗ F )|U  F . Then as the same method of (b),
S
we may take ρ−1 (i∗ F ) = Fα and take a maximal element of coherent sheaf F 0 of the union, we know
0 −1
F = ρ (i∗ F ) is the extension of F .
S
(d) If X = Vi is an affine open covering, which we can assume this covering is finite since X is noetherian,
by induction we just need to prove that F on U can be extended to a coherent sheaf on U ∩ V1 . Since FU∩V1 is
a coherent sheaf on U ∩ V1 , it can be extended to be a sheaf G on V1 . Since F and G is compatible on U ∩ V1 ,
they can be glued together to be a coherent sheaf F 0 on U ∪ V1 . By induction we can get a coherent sheaf F 0
S
on U ∪ ( Vi ) = X.
S
(e) Obviously F ⊃ Fα for all coherent Fα ⊂ F . Conversely, for any open subset U ⊂ X and s ∈ F (U),
s can generate a coherent sheaf G on U. Then by (d), G can be extended to be a coherent subsheaf F 0 of F
S
such that s ∈ F 0 (U) = G (U). Hence F ⊂ Fα , i.e. equal.

Solution 2.5.16 (Tensor Operations on Sheaves). (a) Since F is locally free of rank n, for any point x ∈ X,
there exists a neighbourhood U of x such that F |U = OX |nU with basis e1 , . . . , en . Then T r (F )|U is free with
basis {ei1 ⊗ . . . ⊗ eir | 1 ≤ i1 , . . . , ir ≤ n}, S r (F )|U is free with basis {ei1 ⊗ . . . ⊗ eir | 1 ≤ i1 ≤ . . . ≤ ir ≤ n} and finally
Λr (F )|U is free with basis {ei1 ⊗ . . . ⊗ eir | 0 ≤ i1 < . . . < ir ≤ n}. So T r (F ), S r (F ) and Λr (F ) are locally free with
  
rank nr , n+r−1
n−1
and nr .
(b) For any point x ∈ X, and a neighbourhood U of x such that F |U = OX |nU with basis e1 , . . . , en . Then we
clearly have a morphism

Λr (F )|U ⊗ Λn−r (F )|U → Λn (F )|U

(ei1 ⊗ . . . ⊗ eir ) ⊗ (e j1 ⊗ . . . ⊗ e jn−r ) 7→ (−1)σ(i1 ,...,ir , j1 ,..., jn−r ) e1 ⊗ . . . ⊗ en

where the σ is the sign of permutation. It is clearly a perfect pairing.

Yang Pi‑Yeh 43 Hartshorne Solutions

2 Schemes 2.5 Sheaves of Modules

(c) Similarly, take open subset U such that F 0 |U , F |U , F 00 |U are free, then F |U = F 0 |U ⊕ F 00 |U . So we have
Lr Lr
S F |U =
r
(S p F 0 |U ⊗ S r−p F 00 |U ). So we just define F i = (S p F 0 |U ⊗ S r−p F 00 |U ) to get such filtration.
p=0 p=i Lr
(d) Similarly with (c), F |U = F |U ⊕ F |U are all free and n = n0 + n00 . Then Λr F |U =
0 00
p=0
(Λ p F 0 |U ⊗
00 0 00 n0 0 n00 00
Λ F |U ). But if r > n, we have Λ F |U = 0 and same for F and F , so Λ F |U = Λ F |U ⊗ Λ F |U because
r−p r n

the rest terms are all zero.

(e) For T r , clearly f −1 can commute with T r , then we do this by induction. If r = 1, trivial. If f ∗ (T r−1 (F )) =
T ( f (F )), we have T r ( f ∗ (F )) = f ∗ (F ) ⊗OX T r−1 ( f ∗ (F )) = f −1 (F ) ⊗ f −1 OY OX ⊗OX f ∗ (T r−1 (F )) = f −1 (F ) ⊗ f −1 OY
r−1 ∗

f −1 (T r−1 (F )) ⊗ f −1 OY OX = f −1 (T n (F )) ⊗ f −1 OY OX = f ∗ (T r (F )). For S r or Λr , we may denote I as the ideal of

x ⊗ y − y ⊗ x or x ⊗ x as in definition, we have S r ( f ∗ (F )) = T r ( f ∗ (F ))/ f ∗ I = f ∗ (T r (F ))/ f ∗ I = f ∗ (T r (F )/I ) =
f ∗ (S r (F )), and the same for Λr .
S
Solution 2.5.17 (Affine Morphisms). (a) If f : X → Y is affine, and Y = Vi is an affine open covering for
some Vi = Spec Bi , we have Ui = f −1 (Vi ) = Spec Ai are all affine. Then for any affine subset V ⊂ Y, we can take
S
an affine covering V ∩ Vi = D( fi j ) for some fi j ∈ Bi , and f −1 (D( fi j )) = Spec (Ai ) fi j are all affine. So we now
may assume that V = Spec B can be covered by some affine Vi = Spec Bi and f −1 (Vi ) = Ui = Spec Ai is affine.
Moreover, we can cover Vi by some D( fi j ) for some fi j ∈ B, and denote fi0j as the image of restriction map
OY (V) → OY (Vi ) of fi j . Then clearly D( fi j ) = D( fi0j ), and f −1 (D( fi0j ) = f −1 Spec (Bi ) fi0j = Spec (Ai ) fi0j . So we now
can assume that V = Spec B can be covered by some principal affine subset Vi = D( fi ) = Spec B fi for some
fi ∈ B, and f −1 (Vi ) = Spec A fi . Finally, since V = Spec B, we can assume this covering is finite. Then denote
U = f −1 (V), and gi as the image of fi under B → OX (U), so {gi } generates OX (U). Furthermore, we know that
all D(gi ) cover U, and Ugi = f −1 (D( fi )) are all affine. Then by 2.2.17., we know that U is affine.
S
(b) If f : X → Y is affine, we may take an affine open covering Y = Vi and f −1 (Vi ) = Ui are all affine.
Then since affine schemes are quasi‑compact, i.e. Vi and Ui are all quasi‑compact, so f is quasi‑compact.
Moreover, since f : Ui → Vi are separated by theorem 4.1., and the diagonal morphism X → X ×Y X factors
S S
through X → Ui → Ui ×Vi Ui → X ×Y X, we know that X → X ×Y X is a closed immersion, i.e. f is separated.
And finally, if f is finite, then by definition it must be affine.
(c) If Y = Spec B is affine, we just define X = Spec A for A = A (Y). Then for principal affine subset D(g) for
g ∈ B, we have f −1 (D(g)) = Spec Ag = Spec (A (D(g))). And for general open subset V ⊂ Y, V is a direct limit of
principal open subset, and similar about f −1 (V) and A (V). So X = Spec A as we want. And the uniqueness
of X is trivial by the property we want.
For general Y, we can define XV = Spec A (V) for any affine V = Spec B ⊂ Y, and ϕV : XV → V is the
corresponding morphism. Then we just need to glue all XV together. For V = Spec B and V 0 = Spec B0 affine
S
in Y, by we have V ∩ V 0 = Wi for some Wi = D( fi ) = D( fi0 ) principal in both two affine piece for some fi ∈ B
and fi0 ∈ B0 . Then by axiom of sheaves, A (V) → A (Wi ) is a morphism of rings, then we have a morphism
of affine schemes XWi → XV . And since Wi is principal in V, A (Wi ) is a localization of A (V). Therefore,
ϕ−1 −1 0
V (Wi )  XWi  ϕV 0 (Wi ), i.e. XV and XV are compatible on the part of V ∩ V . So we can glue XV all together
0

to get a X. Finally, the uniqueness of X is from the uniqueness of glueing lemma and the uniqueness of the
affine case.
(d) On any affine piece V ⊂ Y, we have f∗ OX (V) = OX ( f −1 (V)) = A (V). Since all affine subsets form a
basis of open subsets, we have f∗ OX  A . Conversely, for any affine piece U ⊂ V ⊂ Y, we have OX ( f −1 (V)) =
f∗ OX (V) = A (V), and the restriction maps are clearly compatible. Moreover, A (V) = A := OX ( f −1 (V)) is a
B‑algebra for B = OY (V), so clearly a B‑module, i.e. A is quasi‑coherent. Then by existence and uniqueness
of Spec A , we have X  Spec A .
(e) For any quasi‑coherent A ‑module M , we can define a quasi‑coherent OX ‑module M˜ like M˜( f −1 (V)) =
A (V) for all affine piece V ⊂ Y. Furthermore, for any quasi‑coherent OX ‑module F , we clearly know that
fg g
∗ F is the same with F on every affine piece, so f∗ F  F . Conversely, for any quasi‑coherent A ‑module
M , f∗ M˜ is the same with M at every affine piece too, so f∗ M˜  M . Hence f∗ and˜are inverse to each other,

Yang Pi‑Yeh 44 Hartshorne Solutions

2 Schemes 2.6 Divisors

i.e. those two categories are equivalent.

Solution 2.5.18 (Vector Bundles). (a) Just need to confirm the transition function. For any affine Ui = Spec Ai
where E is free, by definition we have f −1 (Ui ) = Spec Ai [x1 , . . . , xn ]. For any affine Ui = Spec Ai , U j = Spec A j
in Y where E is free, and some affine V = Spec B ⊂ Ui ∩ U j , we have two canonical morphisms ρi : B → Ai
and ρ j : B → A j from restriction maps. So the transition function is just ψ = ψ j ◦ ψ−1 −1 −1
i = Spec (ρ j ρ j ρi ρi ) = id,
hence linear. So (X, f, {U}, {ψ}) is a vector bundle of rank n over Y.
(b) This problem is local, so we may assume Y = Spec A, X = Spec AnA and define a OY ‑module structure
of S (X/Y). So now, a section s : Y → X induces a A‑algebra homomorphism θ : A[x1 , . . . , xn ] → A, which is
depended on a n‑tuple (e1 , . . . , en ) for θ(xi ) = ei . Hence S (X/Y)  An , it clearly has an A‑module structure. So
S (X/Y) is a locally free OY ‑module for rank n.
(c) For any open set V where E is free on. For every s ∈ Γ(V, Eˇ ) = Hom(E |V , OY |V ), it induces a morphism
ssym : S (E |V ) → OY |V . So ssym induces a morphism Spec s : Spec OY |V → Spec S (E |V ). Since V  OY |V ,
the morphism Spec s : V → Spec S (E |V ) ,→ X is a section of S (X/Y). Conversely, if we have a section
σ : V → X with image in Spec S (E |V ), it induces a morphism of global section S (E |V ) → OY |V , which is
exactly a morphism s|σ : E |V → S (E |V ) → OY |V . So we have S (X/Y)  Eˇ .
(d) For every locally free sheaf E of rank n, by (a), we have a geometric vector bundle Spec S (E ). Con‑
versely, if we have a vector bundle f : X → Y, by (b), we have a locally free sheaf of sections (S (X/Y))ˇ of
rank n. So by (c) and 2.5.1.(a), we know this corresponding is one‑to‑one.

2.6 Divisors
Solution 2.6.1. Clearly X × Pn is integral and separated. And since we can cover X × Pn by n pieces of X × An ,
hence regular. Moreover, we have a exact sequence

i j
Z→
− Cl X × Pn →
− Cl X × An → 0

where i(1) = Z, where Z = π−1 2 ([0, . . . , 0, 1]), for π2 : X × P → P . Firstly, if n ∈ ker i for some n > 0, i.e.
n n

nZ ∼ 0. Since the function field of X × P is K(t1 , . . . , tn ), where K is the function field of X, then we have a
n

f = f (t1 , . . . , tn ) ∈ K(t1 , . . . , tn ) such that vZ ( f ) = n and vY ( f ) = 0 for any other prime divisor Y. We may assume
f = t1r1 . . . tnrn · hg for some g, h ∈ K[t1 , . . . , tn ] prime to each other and have no factors of t1 , . . . , tn . If there exists
an i satisfies ni , 0 and g, h have degree 0 about ti , then clearly vY ( f ) = −ni for Y = π−1 2 ([0, . . . , 0, 1, 0, . . . , 0])
for 1 in the i‑th tuple. So g or h has non‑zero degree, hence has a non‑zero irreducible factor in K[t1 , . . . , tn ]
corresponding to a Y = π−1 2 (t) for some t ∈ P , so vY ( f ) , 0, which makes a contradiction. So ker i = 0.
n

Secondly, we may consider the morphism Cl X → Cl X × P1 , Y 7→ π−1 1 (Y), where Y is a prime divisor of X,
and π1 : X ×Pn → X. So the component Cl X → Cl X ×Pn → Cl X ×An  Cl X, we have Y 7→ Y ×Pn 7→ Y ×An 7→ Y,
hence the exact sequence above is split, so Cl X × Pn  (Cl X) × Z.

Solution 2.6.2 (Varieties in Projective Space). (a) Since V ∩X is noetherian, hence has finitely many irreducible
P P
components, i.e. ni Yi is a finite sum. And by 1.1.8., we know that Yi all have codimension 1 in X, so ni Yi is
P
a Weil divisor of X. So we have a morphism Div Pn → Div X as linearly extension of V 7→ ni Yi since Div Pn
is generated freely.
(b) We may assume D corresponds to f ∈ k(Pn ). Since we have the immersion X ,→ Pn , it induces an
endomorphism i : k(Pn ) → k(X). Denoting f¯ as the image of f by above morphism, we have D.X = ( f¯). Hence
we have a morphism Cl Pn → Cl X.
(c) Recall the definition i(X, V; Y) = µp (S /(IV + IX )), where p is the prime ideal corresponding to Y. If we
take a prime decomposition of IV + IX = pr11 . . . prnn , we have a filtration of S /(IV + IX ) as 0 = S /S ⊂ S /p1 ⊂
. . . ⊂ S /pr11 . . . prnn −1 ⊂ S /pr11 . . . prnn = S /(IV + IX ). So by the uniqueness of proposition 7.4. in chapter I, we

Yang Pi‑Yeh 45 Hartshorne Solutions

2 Schemes 2.6 Divisors

know that µpi (S /(IV + IX )) = ni for p = pi , and µp (S /(IV + IX )) = 0 for any other p. For any Ui with Y * Ui ,
we know that vY ( f¯) = vp (IV + IX ), since vp is the extension of vY to the function field of Pn , hence equals
P
to i(V, X; Y). Furthermore, by theorem 7.7. in chapter I, we have i(V, X; Y) deg Y = deg V · deg X, we have
deg(V.X) = deg V · deg X. And we may extend this equality linearly and have deg(D.X) = deg D · deg X.
(d) Since D is principal, there exists some g ∈ k(X) such that D = (g). Since we have an endomorphism
i : k(Pn ) → k(X), we may pick a f ∈ i−1 (g). Then by (b), we have ( f ).X = (g) = D. Hence deg D = 0. And the
rest is obvious.

Solution 2.6.3 (Cones). (a) Define Ui = V ∩ D+ (xi ) for i = 0, . . . , n. Then Ui ⊂ An is an affine variety. We may
assume Ui is generated by I = ( f1 , . . . , fd ) for some homogeneous f = f (x0 , . . . , 1, . . . , xn ) ∈ k[x0 , . . . , x̂i , . . . , xn ].
Then C(Ui ) is generated by I = ( f¯1 , . . . , f¯d ) for f¯ = f¯(x0 , . . . , xn ) = xideg f · f ( xx0i , . . . , 1, . . . , xxni ). So C(Ui ) is generated
by I = (F1 , . . . , Fd ) for F = F(x0 , . . . , xn , xn+1 ) = f¯(x0 , . . . , xn ), which just ignore the last coordinate. So clearly
P P
π−1 (Ui ) = C(Ui ) − P = Ui × A1 . Moreover, since the morphism π∗ is given by ni Yi 7→ ni π−1 (Yi ). Firstly, if
D ∈ Div V, and π∗ (D) = ( f ) for some f ∈ k(X̄ − P) = k(X̄) = k(V)(t). Clearly π∗ (D) only involve prime ideals in
the form π−1 (Y) for some prime divisor Y of X, we know f ∈ k(V), i.e. π∗ (D) is principal, hence π∗ is injective.
Secondly, we just need to prove that any prime divisor Z in X̄ − P in the form π(Z) = V is linearly equivalent
P
to some divisor ni π−1 (Yi ). Since Z corresponds to a prime ideal p ∈ k(X̄ − P) = k(V)[t]. Since p is principal,
P
we may find a generator f , then ( f ) − Z has the form ni π−1 (Yi ) for some prime divisors Yi in X, hence π∗ is
surjective. So we have Cl (V)  Cl (X̄ − P)  Cl (X̄).
P
(b) Define H∞ = (xn+1 = 0), then clearly V = X̄ ∩ H∞ . So if g = an+1 xn+1 + ni=0 ai xi , and H = (g), we have
P
V * H. So if V = ( f1 , . . . , fd ), we have V.H = ( f1 , . . . , fd , g), and π−1 (V.H) = ( f1 , . . . , fd , g0 ), where g0 = ni=0 ai xi . So
if we set f = xn+1g
∈ k(Pn+1 ), and f¯ is the image of f under the morphism k(Pn+1 ) → k(X̄), hence V +( f¯) = π−1 (V.H).
Furthermore, by proposition 6.5., we have Z → Cl X̄ → Cl X → 0, and the first arrow is just 1 7→ 1 · V. Since
deg V , 0, the first morphism is injective, hence we have 0 → Z → Cl X̄ → Cl X → 0. Then by (a), we
have 0 → Z → Cl X̄ → Cl X → 0, and the first morphism is just 1 7→ V.H and the H is determined by the
isomorphism Cl V  Cl X̄ as above.
(c) (⇒): (1) Clearly the unique factorization domain is integrally closed, hence V is projectively normal.
(2) By proposition 6.2., we have Cl X = 0 since S (V) is a unique factorization domain, hence by (b), we have
Z  Cl V.
(⇐): Similarly with 1.3.18., since S (V) is integrally closed, for every prime p (may not homogeneous),
S (V)p is integrally closed, hence X = Spec S (V) is normal. So by proposition 6.2., S (V) is a unique factorization
domain.
(d) Consider the maximal ideal (x0 , . . . , xn ) ⊂ k[x0 , . . . , xn ], it induces a maximal ideal m = (x0 , . . . , xn )S (V) ⊂
S (V), so OP = S (V)m . Since X = Spec S (V), and Spec OP = Spec S (V)m , we have a morphism β : Spec OP → X
P P
induced by the morphism α : S (V) → S (V)m . Then β induces a Div X → Div Spec OP as ni Yi 7→ ni β−1 (Yi ),
where β−1 (Yi ) = Z(α(I(Yi ))). So we have a morphism β∗ : Cl X → Cl Spec OP , then we need to prove this is
isomorphic. Firstly, if β−1 (Yi ) = Z(α(I(Y))) = 0 for some prime divisor Y ⊂ X, i.e. Y does not contain 0, since Y
is corresponding to a prime ideal p ⊂ S (V) for height 1. Since p * m, p = ( f ) is principal for some f ∈ S (V),
P
hence Y = ( f ), i.e. Y ∼ 0. So if β∗ ( ni Yi ) = (g) for some g ∈ S (V)m , we may assume that all Yi containing the
P P
vertex. So ni Yi = (β(g)), i.e. ni Yi ∼ 0, hence β∗ is injective. Secondly, Since every prime ideal p ⊂ S (V)m
corresponds to a prime ideal p0 ⊂ S (V) contained in m, so β∗ is clearly surjective. So Cl X  Cl Spec Op .

Solution 2.6.4. Denote K = FracA, B = k[x1 , . . . , xn ], and L = FracB. For any element gg+zh 0 +zh0 ∈ K for some
g+zh g0 −zh0 (gg0 − f hh0 )+(g0 h−gh0 )z
g, g0 , h, h0 ∈ B, then gg+zh
0 +zh0 = g0 +zh0 · g0 −zh0 = g02 − f h02
. So K = L[z]/(z 2
− f ), i.e. K/L is an extension of degree
2, hence Galois. So for any α = g + zh ∈ K for some g, h ∈ L, its minimal polynomial is just X 2 − 2gX + (g2 − f h2 ).
So α is integral over B iff 2g, g2 − f h2 ∈ B, i.e. g, h ∈ B since f is square‑free. So A is the integral closure of B in
K, hence integrally closed.

Yang Pi‑Yeh 46 Hartshorne Solutions

2 Schemes 2.6 Divisors

Solution 2.6.5 (Quadric Hypersurfaces). (a) By 1.5.12., when r ≥ 2, f = x02 + . . . + xr2 is irreducible, hence
square‑free. So by 2.6.4., A = k[x0 , . . . , xn ]/( f ) is integrally closed, so X = Spec A is normal.
(b) Just take a linear change x0 7→ x0 +x 2
1
and x1 7→ x20√−x−11 , then f will be changed to f = x22 + . . . + xn2 − x0 x1 .
(1) The case r = 2 is just example 6.5.2., so Cl X  Z/2Z. (2) If r = 3, similarly with example 6.5.2., we may
denote x̄1 ∈ A as the image of x1 from k[x0 , . . . , xn ]. Then we may define Z = V(x1 ) closed in X, hence we have
an exact sequence Z → Cl X → Cl (X − Z) → 0. Since X − Z = Spec A x̄1 for A x̄1  k[x2 , . . . , xn ](x1 ), which is a
unique factorization domain. Hence Cl (X − Z) = 0 by proposition 6.2., so we have a surjection Z → Cl X. So
Cl X only has three choices: Z, Z/nZ for some n, and 0. Moreover, if we also take a linear change x2 7→ x2 +x 2
3

x2 −x3
and x3 7→ 2 −1 , then f = x0 x1 + x2 x3 . So X is the cone of V = Proj A. By example 6.6.1., Cl V = Z ⊕ Z. And
√

by 2.6.3.(b), we have an exact sequence 0 → Z → Z ⊕ Z → Cl X → 0, so Cl X = Z. (3) Same with (2), we have

a surjection Z → Cl X as 1 7→ 1 · Z. So we just need to prove that the corresponding ideal ( x̄1 ) of Z is prime.
Since A/( x̄1 )  k[x0 , x2 , . . . , xn ]/(g) where g = x22 + . . . + xn2 , and we already have g is irreducible in k[x0 , x2 , . . . , xn ]
since r ≥ 4 by 1.5.12. So A/( x̄1 ) is a domain, hence ( x̄1 ) is a prime ideal, i.e. Cl X = 0.
(c) (1) By 2.6.3.(b), we have an exact sequence 0 → Z → Cl Q → Z/2Z, where the first arrow is 1 7→ Q.H.
Tensoring with Q, we have Q → Cl Q ⊗ Q → 0, we know that Cl Q  Z or Z ⊕ T with some torsion T . Tensoring
with Z/pZ for some prime p , 2, we have Z/pZ → Cl Q ⊗ Z/pZ → 0, so T  (Z/2Z)n if exists. Tensoring with
Z/2Z, we have Z/2Z → Cl Q ⊗ Z/2Z → Z/2Z → 0, so T  Z/2Z if exists. In example 6.5.2., we know that
Y = (x1 = x2 = 0) or Y 0 = (x0 = x2 = 0) is the generator of the group Cl X = Z/2Z. Since the preimage of Y or
Y 0 in Cl Q is R = [1 : 0 : 0] or R0 = [0 : 1 : 0]. Since Q.H = R + R0 , if T exists, the preimage of generator of Z/2Z
must be a torsion element, which makes a contradiction. So we know T does not exists, so we have Cl Q = Z
with generator R ∼ R0 . (2) The case r = 3 is just the example 6.6.1., we have Cl Q  Z ⊕ Z. (3) Since Cl X = 0,
we have Cl Q  Z by 2.6.3.(b).
(d) Similarly with (a), we have S (Q) is integrally closed. And moreover by (c) we have Cl Q  Z, we know
that S (Q) is a unique factorization domain by 2.6.3.(c). Since Y is a irreducible subvariety or codimension 1
in Q, it corresponds to a prime ideal p in S (Q) with height 1, hence principal. So p = ( f¯) for some f¯ ∈ S (Q), it
has a preimage f ∈ k[x0 , . . . , xn ]. Then we just need to define V = Z( f ), so V ∩ Q = Y is a complete intersection.
Solution 2.6.6. (a) (⇒) If P, Q, R are collinear, i.e. there exists a line L with L∩ X = {P, Q, R} by Bezout theorem.
Since L ∼ (z = 0), i.e. P + Q + R = 3P0 = 0.
(⇐) If P + Q + R = 0, we may denote L as the line of P and Q. So L ∩ X = {P, Q, T } for some T by Bezout
theorem. Since P + Q + T = 0, we have R = −P − Q = T , i.e. R is on the line L.
(b) Denote the tangent line of P as L, so L intersect X at P with multiplicity ≥ 2. Then we may assume
that L intersect X as {P, P, T } for some T ∈ X, i.e. P + P + T = 0. Since P + P = 0, we have T = P0 , which means
L passes through P0 . Conversely, if L passes through P0 , we have L ∩ X = {P, P, P0 }, i.e. P + P = P + P + P0 = 0.
(c) If P + P + P = 0, then by (a), there exists a line L0 intersect X with 3P, then L0 is the tangent line, i.e. L.
So P is an inflection point. Conversely, if L intersect X with 3P, we have P + P + P = 0 by (a).
(d) Clearly P0 ∈ X(Q). Moreover, if R = P + Q with P, Q ∈ X(Q), we have a line L intersect X with P, Q, −R.
The intersection forms a set of functions, so by Vieda’s theorem, we know that the coordinates of R are in Q
since the coordinates of P and Q are in Q. Hence X(Q) is a group. Furthermore, by Mordell’s theorem, X(Q)
is a finitely generated abelian group.
Solution 2.6.7. Similarly with example example 6.11.4., we have a one‑to‑one correspondence from closed
points in X − Z to Cartier divisor classes of degree 0, i.e. elements in CaCl 0 (X). Then Since we have an
isomorphism Gm → X − Z as t 7→ (4t − 4t2 , 4t + 4t2 , (1 − t)3 ), and y−x
y+x
←[ (x, y, z). So we have CaCl 0 (X)  Gm .
Solution 2.6.8. (a) Clearly we have a map f ∗ : Pic Y 7→ Pic X. For the group structure, if L , M ∈ Pic Y, for
every point P ∈ Y, there exists an affine neighbourhood V = Spec B such that L = M̃ and M = Ñ on V. So
for any U = Spec A ⊂ f −1 (V), we have f ∗ (L ⊗ M )|U = f ∗ ( M] N ⊗B A  (M ⊗B A)]
⊗B N)|U = M ⊗B] ⊗A (N ⊗B A) =
∗ ∗ ∗
f (L ) ⊗ f (M ). Hence we have a group homomorphism f : Pic Y → Pic X.

Yang Pi‑Yeh 47 Hartshorne Solutions

2 Schemes 2.6 Divisors

(b) For any D = (Ui , fi ) ∈ Γ(Y, K ∗ /O ∗ ), we can define f CaCl ∗ (D) = ( f −1 (Ui ), f ∗ ( fi )), hence a Cartier divisor
on X. So we have a morphism f CaCl ∗ : CaCl Y → CaCl X. On one hand, for any prime divisor V ⊂ Y,
P
if Ui ∩ V , ∅, then clearly f Cl ∗ (vV ( fi )V) = vV ( fi ) f Cl ∗ (V) = vV ( fi ) · f (U)=V vU (t)U, where t ∈ K(Y) such that
vV (t) = 1. But vV ( fi ) · vU (t) = vV ( fi ), hence the morphism f Cl ∗ corresponds to f CaCl ∗ . On the other hand, the
corresponding between f CaCl ∗ and f Pic ∗ is trivial by definition. Hence f Cl ∗ , f CaCl ∗ and f Pic ∗ is the same, we
may denote it as f ∗ .
(c) For any prime Weil divisor V ⊂ Pn , then clearly for any local parameter fi of Yi on Ui , we have f¯i = fi |Ui ∩X
P
is the f¯i in 2.6.2., so we have f Cl ∗ (V) = vYi ( f¯i )Yi = V.X. Hence the morphism f Cl ∗ corresponds to the f ∗
defined in 2.6.2., so by (b), these all f ∗ are the same.
Solution 2.6.9 (Singular Curves). (a) We already have a morphism π∗ : Pic X → Pic X̃ as in 2.6.8. Denote
the morphism θ : Γ(X, K ∗ ) → Γ(X, K ∗ /OX∗ ), and θ̃ : Γ(X̃, K ∗ ) → Γ(X̃, K ∗ /OX̃∗ ), then Pic X  CaCl X =
coker(θ) and Pic X̃  coker(θ̃), ker(θ) = Γ(X, OX∗ ) and ker(θ̃) = Γ(X, OX̃∗ ). As the hint, we consider the exact
sequence 0 → π∗ OX̃∗ → K ∗ /OX∗ → K ∗ /π∗ OX̃∗ → 0, then since H 1 (π∗ OX̃∗ ) = 0, we have an exact sequence
0 → Γ(X, π∗ OX̃∗ ) → Γ(X, K ∗ /OX∗ ) → Γ(X, K ∗ /π∗ OX̃∗ ) → 0. Then by snack lemma, we have an exact sequence
Γ(X, OX∗ ) → Γ(X, OX̃∗ ) → Γ(X, π∗ OX̃∗ /OX∗ ) → Pic X → Pic X̃ → 0. So ker π∗ = Im(Γ(X, OX̃∗ ) → Γ(X, π∗ OX̃∗ /OX∗ )) =
L L
Γ(X, π∗ OX̃∗ /OX∗ ) = P∈X
ÕP∗ /OP∗ . Hence we have the exact sequence 0 → P∈X
ÕP∗ /OP∗ → Pic X → Pic X̃ → 0.
(b) Clearly whenever X is cuspidal or nodal cubic curve, the normalization of X is just the blow up at the
L
singular point, i.e. P1 . Since clearly Pic P1 = Z by corollary 6.17., we have 0 → ÕP∗ /OP∗ → Pic X → Z → 0.
∗ ∗
L P∈X
If P ∈ X is regular point, we have OP  ÕP , so the is non‑vanish on those singular points. If X is the plane
cuspidal cubic curve y2 = x3 , the unique singular point is just Z = (0, 0), then ÕZ∗ /OZ∗  k = Ga . If X is the nodal
cubic curve y2 z = x3 + x2 z, the unique singular point is just Z = (0, 0, 1), then ÕZ∗ /OZ∗  k∗ = Gm .
Solution 2.6.10 (The Grothendieck Group K(X)). (a) If X = A1k = Spec k[x], F = M̃ for some finitely generated
module M. Since k[x] is a principal ideal domain, there exists an exact sequence 0 → (k[x])n → (k[x])m → M →
0, i.e. 0 → OXn → OXm → F → 0. Hence in K(X), we have γ(F ) = (m − n)γ(OX ). Conversely, we clearly have
γ(OXn ) = nγ(OX ). So we have an isomorphism K(X)  Z.
(b) For any exact sequence 0 → F 0 → F → F 00 → 0, we have 0 → Fξ0 → Fξ → Fξ00 → 0, hence
dimK Fξ = dimK F xi0 + dimK Fξ00 , hence the morphism K(X) → Z is well‑defined. Moreover, for any n ∈ Z, we
have dimK (OXn )ξ = n, so surjective.
(c) For any coherent sheaf F on X − Y, then F can be extended to be a coherent sheaf F 0 on X by 2.5.15,
so K(X) → K(X − Y) is surjective. For the exactness at the middle, for any F ∈ ker(K(X) → K(X − Y)), on every
affine piece U = Spec A ⊂ X, we have Y ∩ U = Spec A/I for some ideal I ⊂ A, i∗ i∗ F |U = M/I ]M, where F |U = M̃
for some A‑module M. So η : F → i∗ i∗ F is surjective. Then we may define F0 = F , and Fi = ker(Fi−1 →
i∗ i∗ Fi−1 ) for i > 0, so Fi |U = I i˜M. Since M is finitely generated, I i M = 0 for sufficiently large i, so the filtration
F0 ⊃ F1 ⊃ . . . ⊃ Fn is finite. And by definition, Fi /Fi+1 is a coherent sheaf extended from a coherent sheaf
P
on Y, so γ(F ) = γ(Fi /Fi+1 ) ∈ Im(K(Y) → K(X)), i.e. ker(K(X) → K(X −Y)) ⊂ Im(K(Y) → K(X)). The converse
is obvious, so K(Y) → K(X) → K(X − Y) → 0 is exact.
Solution 2.6.11 (The Grothendieck Group of a Nonsingular Curve). (a) Define FP as the skyscraper sheaf
of coker(ID,P → OX,P ) = OX,P /mnP at P, where n is the coefficient of P in D. Then we have an exact sequence
L L P
0 → ID → OX → P∈D
FP → 0 since all P ∈ D are closed points. So OD  FP , i.e. γ(OD ) = γ(FP ).
Since we have a natural OX,P ‑modules exact sequence 0 → mi−1 /mi → OX,P /mi−1 → OX,P /mi → 0, and since
mi−1 /mi  m/m2  k, we have γ(FP ) = nγ(k(P)). So we have γ(OD ) = ψ(D). If D ∼ D0 are both effective divisors,
we have ψ(D) = γ(OD ) = γ(OX ) − γ(ID ) = γ(OX ) − γ(L (−D)) = γ(OX ) − γ(L (−D0 )) = ψ(D0 ). For general
divisor D, we can choose two effective divisor D+ and D− such that D = D+ − D− , so ψ(D) = ψ(D+ ) − ψ(D− ) is
independent of linear equivalence, hence we have a morphism ψ : Cl X → K(X).
(b) We may embed X into Pn , and define Y as the closure of X in Pn . So F can be extended to be a coherent
sheaf F 0 on Y, so there exists a free module E with surjection E → F 0 by corollary 5.18. Restrict on X, we

Yang Pi‑Yeh 48 Hartshorne Solutions

2 Schemes 2.6 Divisors

have a free module E0 = E |X with surjection E0 → F . Denote the kernal of this surjection as E1 . Since for
every closed point P ∈ X, the module E1,P is a submodule of E0,P = OX,P n
. Then since OX,P is a reduced regular
local ring of dimension one, hence principal ideal domain, we have E0,P is a free module. So by 2.5.7., E0 is
locally free of finite rank.
If we also have another resolution 0 → E10 → E00 → 0, we may denote G = ker(E0 ⊕ E00 → F . Then by
nine lemma we have two exact sequences 0 → E1 → G → E00 → 0 and 0 → E10 → G → E0 → 0. Then
0 0
clearly rank G = r0 + r10 = r00 + r1 . So det F = (∧r0 E0 ) ⊗ (∧r1 E1 )−1 = (∧r0 E0 ) ⊗ (∧r1 E1 )−1 ⊗ (∧r0 E00 )−1 ⊗ (∧r0 E00 ) =
0 0 0 0
(∧r0 E0 ) ⊗ (∧rank G G )−1 ⊗ (∧r0 E0 ) = (∧r0 E0 ) ⊗ (∧r0 E0 )−1 ⊗ (∧ 1 E10 )−1 ⊗ (∧ 0 E00 ) = (∧ 0 E00 ) ⊗ (∧ 1 E10 )−1 . Clearly, if we
r r r r

have 0 → F 0 → F → F 00 → 0, we have det F = det F 0 ⊗ det F 00 , so we have a morphism det : K(X) → Pic X.
If D is effective, we have an exact sequence 0 → ID → OX → OD → 0. Since OX and ID is locally free of
rank 1, then det(ψ(D)) = OX ⊗ID−1 = ID−1 = L (−D)−1 = L (D). Then for general divisor D, we have D = D+ −D−
for two effective divisors D+ and D− , we have det(ψ(D)) = det(ψ(D+ )) ⊗ (det(ψ(D− )))−1 = L (D+ ) ⊗ (L (D− )) =
L (D).
(c) Take an affine open covering {Ui } of X, Ui = Spec Ai . Then we may assume F |Ui = M̃i for some Ai ‑
module Mi . Since the generic point η is contained in every Ui , and corresponds to the ideal (0) in every Ai , so
m
Fη = Frac(Ai ) ⊗Ai Mi . Denote {e1 , . . . , er } be a set of basis of Fη as K(X)‑module, then we may assume e j = aii j
T
for all Ai and e j . So we need to construct a Cartier divisor about these ai . Denote Z = V(ai ). If Z = ∅, we
define Vi = Ui −V(ai ). If not, since Z is a closed subset of X, it is just finite set of closed points since X is a curve.
`
Then we may divide Z as Z = Zi , such that Zi ⊂ Ai . Then we may define Vi = Ui − (V(ai ) − Zi ). Hence for any
i , j, V(ai ) ∩ Vi ∩ V j = ∅, so ai is invertible in OX (Vi ∩ V j ). Then we can define a Cartier define D0 = (Vi , ai ). So
m
Fη is generated by aii j ∈ Γ(Vi , L (D0 ) ⊗ F ), and this induces an injection OXr → L (D0 ) ⊗ F . Hence if we define
D = −D0 , then we have an injection L (D)r → F , so we may denote T as the cokernal of this injection. Then
we only need to prove that T is a torsion sheaf. But consider 0 → (L (D)r )η → Fη → Tη → 0, the first two
terms are both vector spaces of rank r over K(X), hence Tη = 0, then T is a torsion sheaf.
To prove γ(F ) − rγ(OX ) ∈ Imψ, since γ(F ) − rγ(OX ) = r(γ(L (D)) − γ(OX )) + γ(T ), we just need to prove
that γ(L (D)) − γ(OX ) and γ(T ) are both in the image of ψ. Firstly, if D is effective, we clearly have ψ(OD ) =
ψ(OX ) − ψ(L (D)) by (a), then we’ve done. For general D, we may denote D = D+ − D− . Then ψ(D) = γ(OD+ ) −
γ(OD− ) = γ(L (D− )) − γ(L (D+ )) = γ(L (D− ) ⊗ (L (D− ))−1 ) − γ(L (D+ ) ⊗ (L (D− ))−1 ) = γ(OX ) − γ(L (D)). Secondly,
by 2.5.6., we know Supp T is a closed subset of X, hence a finite set of closed points. So T = ⊕T P for finitely
some skyscraper sheaves T P . For any P, we may choose an affine subset U = Spec A containing P, then p is
the corresponding prime ideal of P, and T P = M̃ for some A‑module M. Then by 2.5.6.(b), we have an N such
that pN M = 0. Since we have an exact sequence 0 → pi+1 M → pi M → pi M/pi+1 M → 0, and pi M/pi+1 M  (A/p)⊕ni
P P
for some ni . So γ(T P ) = ni γ(k(P)) ∈ Imψ, hence γ(T ) = γ(T P ) ∈ Imψ. Thus, γ(F ) − rγ(OX ) ∈ Imψ.
(d) Define a morphism φ : Z → K(X) as n 7→ nγ(OX ). Since we have rank : K(X) → Z, det : K(X) → Pic X,
ψ : Pic X → K(X) and φ such that rank ◦ ψ = 0, det ◦ψ = idPic X and rank ◦ φ = idZ , we clearly have
K(X) = Pic X ⊕ Z.

Solution 2.6.12. By 2.6.11., we have det F ∈ Pic X, then we can define deg F as the degree of the Weil divisor
corresponding to det F . So the properties (1) and (3) is immediately following from (a) and (b) in 2.6.11. For
L P P
(2), by 2.6.11.(d), F = F , so deg F = deg F P , where F P is the skyscraper sheaf of FP at P. For any P,
we can pick U = Spec A containing P, p corresponds to P, and F P = M̃ on U, then deg F P = length (Mp ) =
P
length FP , hence deg F = length FP . For uniqueness, if F has rank zero, (2) insures the uniqueness. If F
has rank one, (1) implies it. For higher rank, the problem can be reduced to rank one case by the induction
from (3).

Yang Pi‑Yeh 49 Hartshorne Solutions

2 Schemes 2.7 Projective Morphisms

2.7 Projective Morphisms

Solution 2.7.1. By 1.1.2., we just need to check on stalks. Since L and M are both invertible, hence rank 1, for
every point P ∈ X we can take an affine neighbourhood U = Spec A of P such that L |U  OX |U and M |U  OX |U .
So this original question is equivalent to prove that if we have a surjection of Ap ‑module as ϕ : Ap → Ap , then
ϕ is isomorphic. Since ϕ is a morphism of Ap ‑modules, it just depends on ϕ(1), i.e. ϕ(a) = a · ϕ(1). Since ϕ is
surjective, 1 has a preimage, named ι. Then ϕ−1 (b) = b · ι ∈ Ap , hence ϕ is an isomorphism.
P
Solution 2.7.2. Since {si } and {t j } generate the same linear space, we may assume that si = ai j t j . Then on
P P P
OPm (1), we define ui = ai j x j , hence ϕ∗ (ui ) = ai j ϕ∗ (xi ) = ai j t j = si . So if we define L = Z(u0 , . . . , un ), by the
uniqueness in theorem 7.1., we know that ρ◦ϕ = ψ with the projection ρ : Pm −L → Pn , (u0 , . . . , un , v0 , . . . , vm−n−1 ) 7→
(u0 , . . . , un ).

Solution 2.7.3. (a) By theorem 7.1.(2), the morphism ϕ is equivalent to an invertible sheaf L on Pn with global
sections s0 , . . . , sm generate L . Since every invertible sheaf on Pn is isomorphic to some O(d), then we may
assume L  O(d) and si ∈ k[x0 , . . . , xn ](d) . In the case m < n, if d ≥ 1, then V+ (s0 ) ∩ . . . ∩ V+ (sm ) has dimension
> 0, hence non‑empty, which means there exists some point P such that s0 , . . . , sm cannot generate P, hence
contradict. So d ≤ 0 in this case, then there are only trivial global section, hence Imϕ is just a point. In the
case m ≥ n, if ϕ is surjective, then dim(Imϕ) = m ≥ n, but dim(Imϕ) ≤ n, hence dim(Imϕ) = n. If not, there
exists some P ∈ Pm not in the image of ϕ, so we have a morphism ϕ0 : Pn → Pm − P → Pm−1 , where the second
arrow is the projection π : Pm − P → Pm−1 . Then by induction, we have Imϕ0 is a singleton or dim(Imϕ0 ) = n. If
Imϕ0 = {Q} is a singleton, then Imϕ ⊂ π−1 (Q)  A1 . Since clearly Imϕ is closed and connected, Imϕ is a point
in A. Else if dim Imϕ0 = n, we have dim Imϕ ≥ n. Since dim Imϕ ≤ n, we have dim Imϕ = n.
e e
(b) As we discuss in (a), we know that L  Od for some d ≥ 1. Then we may assume si = x0i,0 . . . xmi,n for
P n+d
j ei, j = d. So for d‑uple embedding ρd : P → P for N = − 1, we have ρd,∗ L ⊂ OPN ,1 as a subsheaf, with
n N
n
si 7→ ui = uei,0 ,...,ei,n . Then defining L = V+ (u0 , . . . , um ), we have a projection PN − L → Pm with ui 7→ yi , hence ϕ
factors through this projection.

Solution 2.7.4. (a) If X has an invertible ample sheaf L , which means has an invertible very ample sheaf
L n , hence induces a closed immersion i : X → PmA for some m and L n  i∗ O(1). Then i is proper, hence
separated. Since PmA → Spec A is separated, X → Spec A is also separated.
(b) We may denote two copies of affine line as U0  U1  Spec k[x], and U0 ∩ U1 = V  Spec k[x, x−1 ]. Then
we already have that Cl U0 = Cl U1 = Cl V = 0. So for any invertible sheaf L on X, we have L0 = L |U0  OU0 ,
and L1 = L |U1  OU1 , and M = L |V = OV , with L0 |V  M  L1 . So the morphism L0 |V → L1 |V is
determined by an automorphism of k[x, x−1 ]. Since Aut(k[x, x−1 ]) = {axn | a ∈ k, n ∈ Z}, every L actually
corresponds to a Cartier divisor Daxn = {(U0 , 1), (U1 , axn )}. If Daxn ∼ Dbxm , we clearly have n = m, and vice
versa. So Pic X  Z.
For Cartier divisor Dn = {(U0 , 1), (U1 , xn )} with n > 0, and Ln is the corresponding invertible sheaf. Then
s ∈ Γ(X, Ln ) means, s0 = s|U0 ∈ k[x] and s1 = s|U1 ∈ x−n k[x] agree on V. Since x−n k[x] have a homogeneous
component of non‑negative degree, we know that the origin of U1 cannot be generated by global sections.
For n < 0, Dn = {(U0 , 1), (U1 , xn )} ∼ {(U0 , x−n ), (U1 , 1)}, similarly we know that U0 cannot be generated by global
sections. Clearly we have Ln ⊗ Lm  Ln+m , i.e. Lnm = Lnm , so Ln is not ample since Lnm is not very ample for
any m. For n = 0 case, we have L0  OX , hence not ample.

Solution 2.7.5. (a) Clearly M n is generated by global sections. Then for any coherent sheaf F , F ⊗ L n is
generated by global sections for large n, hence F ⊗ (L ⊗ M )n = (F ⊗ L n ) ⊗ (M n ) is generated by global
sections, hence L ⊗ M is ample.
(b) Since M is at least coherent, M ⊗ L n is generated by global sections for sufficient large n. For general
coherent sheaf F , F ⊗ L m is generated by global sections for some n, hence F ⊗ (M ⊗ L n+1 )m = (F ⊗ L m ) ⊗
(M ⊗ L n )m is generated by global sections, i.e. M ⊗ L p is ample for p ≥ n + 1.

Yang Pi‑Yeh 50 Hartshorne Solutions

2 Schemes 2.7 Projective Morphisms

(c) For any coherent sheaf F , F ⊗ L n is generated by global sections for n > n1 , and OX ⊗ L m is generated
by global sections for n > n2 . So for n > max{n1 , n2 }, F ⊗ (L ⊗ M )n = (F ⊗ L n ) ⊗ (OX ⊗ M n ) is generated by
global sections, hence L ⊗ M is ample.
(d) Since we have a closed immersion ι : X → Pn and a morphism ϕ : X → Pm such that L = ι∗ OPn and
M = ϕ∗ OPm . Since we have a closed immersion f : X → X × Pm → Pn × Pm → PN , where the last arrow is the
Segre embedding, and the second arrow is a closed immersion because it is the base change of ι, we know
L ⊗ M = f ∗ OPN (1) is very ample.
(e) Since L m is very ample for some m, and L d is generated by global sections for d > d0 , then we just
take n0 = m + d0 and by (d) we know that L n is very ample for n > n0 .

Solution 2.7.6 (The Riemann‑Roch Theorem). (a) Since ι : X → Pn is a closed embedding, we may take S =
k[x0 , . . . , xn ]/I(X) and have X = Proj S . Since L is very ample w.r.t. ι, we have L = ι∗ OPn = Sg (1). By 2.5.9. we
have Γ(X, L n ) = Γ(X, Sg (n))  S (n) for sufficiently large n. So dim |nD| = dim Γ(X, L n )−1 = dim S n −1 = PX (n)−1.
(b) If r|n, we know that nD = 0, then |nD| is trivial, hence dim |nD| = 0. In the case r - n, if nD ∼ E for some
effective divisor E, then 0 = deg r(nD) = deg rE > 0, which makes a contradiction. So dim |nD| = −1.

Solution 2.7.7 (Some Rational Surfaces). (a) By definition, the embedding ι : P2 → P5 induced by |D| satisfies
ι∗ (ti jk ) = xi y j zk for i + j + k = 2. Hence ι induces a morphism of graded rings: k[t200 , . . . , t110 ] → k[x, y, z] as
ti jk 7→ xi y j zk , which is just the definition of 2‑uple embedding, hence Imι is the Veronese surface.
(b) For any two points P = (a, b, c) and P0 = (a0 , b0 , c0 ) in P2 , if a = a0 = 0, since y2 , z2 and xy − yz can generate
O(2) on P1 = (a = 0), hence V separates P and P0 . If a = 0 and a0 = 1, x2 separated P and P0 . If a = a0 = 1
and (b0 , c0 ) , (−b, −c), we know that y2 and z2 can separates them. If a = a0 = 1, b0 = −b and c0 = −c, xy − yz
can separates them. For P = (a, b, c), if c = 1 and a , 0, the section x2 − a2 and xz − yz − a + b can separate the
tangent space of P. If c = 1 and a = b = 0, xy − yz and x2 can separate the tangent space of P. The case of P on
D(a) or D(b) are same with this, hence this corresponding morphism is a closed embedding.
(c) We may assume P = (0, 0, 1), then d =< x2 , y2 , xy, xz, yz >. Then the morphism maps U to an open
subset U 0 of V(zw − xt, yw − zt). Since U is open dense in X̃, we know that Ū 0 is the image of X̃, which is
corresponding to the divisor V(x, y, z) + V(x, z, w) + V(x, w, t) in P4 , hence degree 3. Every line thought P has
the form αa + βb = 0, so it corresponds to the line V(αx + βz, αy + βz, αt + βw). Hence if α : β , α0 : β0 , two
different line in X̃ do not meet.

Solution 2.7.8. This is just the easy version of proposition 7.12. about id : X → X.

Solution 2.7.9. (a) We can define a morphism ϕ : Pic X × Z → Pic P(E ), (L , n) 7→ π∗ L ⊗ O(n). For injectivity,
if π∗ L ⊗ O(n)  OP(E ) , we have OX = π∗ OP(E ) = π∗ (π∗ L ⊗ O(n))  L ⊗ π∗ O(n). Then by proposition 7.11.,
we know that π∗ O(n)  S n (E ), hence there we need it to be invertible and get n = 0, and L  OX . For
surjectivity, we may choose a finite affine covering {Ui } of X such that E is trivial on each Ui = Spec Ai .
Then all Vi = Pr−1 Ai  U i × P
r−1
cover P(E ). So we have Pic Vi = Pic Ui × Z. For any M ∈ Pic P(E ), we have
Mi = M |Vi ∈ Pic Vi , hence there exists Li ∈ Pic Ui and ni such that Mi  π∗i Li ⊗Oi (ni ). Denote Ui j = Ui ∩U j and
Vi j = Vi ∩V j . Clearly Mi |Vi j = M j |Vi j , hence we have a transition isomorphism π∗i Li ⊗Oi (ni )|Vi j  π∗j L j ⊗O j (n j )|Vi j ,
hence Li ⊗ π∗ Oi (ni )|Vi j  L j ⊗ π∗ O j (n j )|Vi j . By proposition 7.11. we have ni = n j = n. Since Oi |Vi j  O j |Vi j , we
have Li |Ui j  L j |Ui j . So all Li can be glued together to be an L , i.e. M  L ⊗ O(n).
(b) If we have an f : P(E )  P(E 0 ), there exists an L ∈ Pic X such that f ∗ O 0 (1)  O(1) ⊗ π∗ L . So
E 0  π0∗ (O 0 (1)) = π∗ (O(1)⊗π∗ L ) = E ⊗L . Conversely, if E 0  E ⊗L , we have a surjection π∗ E 0  π∗ E ⊗π∗ L →
O(1) ⊗ π∗ L . Then by theorem 7.12., we have a morphism P(E ) → P(E 0 ).

Solution 2.7.10 (Pn ‑Bundles Over a Scheme). (a) We may define a projective n‑space bundle over X as a
scheme P with a morphism π : P → X, such that there exists an open covering {Ui } of X such that π−1 (Ui ) 

Yang Pi‑Yeh 51 Hartshorne Solutions

2 Schemes 2.7 Projective Morphisms

Ui × Pn , and for any V = Spec A ⊂ Ui ∩ U j , the transition morphism (Ui × Pn )|V×Pn → (U j × Pn )|V×Pn is an A‑linear
automorphism. (Is this really a question?)
(b) Since E is locally free, we have an affine open covering {Ui = Spec Ai } of X such that E |Ui  OUn+1 i
. Then
−1
π Ui = PnAi . For any V = Spec B ⊂ Ui ∩ U j , then PnAi → PnB is induced by B[x0 , . . . , xn ] → Ai [x0 , . . . , xn ] →
Ai [x0 , . . . , xn 0, where the first arrow is just xk 7→ xk and morphism on coefficient B → Ai , and the second arrow
is linear. Hence the transition morphism is linear, so P(E ) is a Pn ‑bundle.
(c) If P is a projective bundle with projection π : P → X, we have an affine open cover {Ui } of X such
that π−1 (Ui ) = Ui × Pn . So we can define Li = OUi ×Pn (1). For any V = Spec B ⊂ Ui ∩ U j , since the transition
morphism (Ui × Pn )|V×Pn → (U j × Pn )|V×Pn is linear, it induces the isomorphism Li |V×Pn → L j |V×Pn , hence all
Li can be glued together to be an L on P. Since all Li is invertible, L is invertible too. So we can define a
locally free sheaf E = π∗ L . Then by definition we clearly have P(E )  P.
(d) By definition, we know that P(EP(F ) )  P(F ). And by 2.7.9.(b), EP(F )  F ⊗L for some invertible sheaf
L . So we have an equivalence between the category of projective bundles and the category of equivalent
classes locally free sheaves.
L∞ L∞
Solution 2.7.11. (a) For any affine piece U ⊂ X, by 2.5.13. we already have Proj ( n=1 I (U)nd )  Proj ( n=1 I (U)n ).
L
And clearly those isomorphism can be glued together by 2.5.13., so we have an isomorphism Proj In 
L n
Proj I .
nd
n L L L L
(b) Clearly, n (I · J )n = ( n I n ) ∗ J , hence by lemma 7.9., we have Proj n
(I · J )n  Proj n
I n.
(c) Actually, we will prove that for any blow‑up f : X̃ = BlI X → X, the open set U = X − Z(I ) is
the maximal open subset of X such that f −1 (U)  U, where Z(I ) is the closed subset corresponding to I .
For any affine piece V = Spec B with V ∩ Z(I ) , ∅, I (V) $ O(V) = B is a proper ideal, namely I. Then
L∞ n
π−1 (V) = Proj n=0
I , hence not isomorphic to V. So U is the maximal open subset of X such that f −1 (U)  U.
So in theorem 7.17., if Z = BlI (X) and U ⊂ X is the maximal open subset such that f −1 (U)  U, we must have
Supp I = X − U.

Solution 2.7.12. If P ∈ Ỹ ∩ Z̃ ⊂ X̃, we can take an affine piece U = Spec A containing π(P). Then we may
L
denote IY , IZ as the ideals corresponding to Y ∩ U and Z ∩ U. So π−1 (U) = Proj (IY + IZ )n . Since π−1 (Y ∩ U) =
L −1
L
Proj ((IY + IZ )/IY ) and π (Z ∩ U) = Proj
n
((IY + IZ )/IZ ) , and the embedding π−1 (Y ∩ U) → π−1 (U) is
n
L L
induced by quotient ϕY : (IY + IZ ) →
n
((IY + IZ )/IY )n and so does π−1 (Z ∩ U) → π−1 (U), we know that P
L L n L n
corresponds to a prime ideal in (I + I )n and containing ker ϕY = IY and ker ϕZ = IZ , which makes a
L n L n Y LZ
contradiction because IY + IZ = ((IY + IZ )/IY ) .
n

Solution 2.7.13 (A Complete Nonprojective Variety). (a) Since A1 → Spec k is clearly proper, π|C×(P1 −{0}) :
C × (P1 − {0}) → C is proper because it is the base change of A1 → Spec k. And so does π|C×(P1 −{∞}) , hence π is
proper.
L
(b) By 2.6.9.(a), we have an exact sequence 0 → P∈C×A1
ÕP /OP → Pic (C × A1 ) → Pic (P1 × A1 ) → 0. Since
we only need to consider the singular locus P = P0 × A term of the direct sum, and ÕP = k[t, z](t) and OP =
1

k[t2 , t3 , z](t2 ,t3 ) , so we have 0 → Gm → Pic (C × A1 ) → Z → 0. Since we clearly have a morphism Z → Pic (C × A1 )
as 1 7→ f∗ O(1), where f : P1 → P1 × A1 → C × A1 , this exact sequence is split, hence Pic (C × A1 )  Gm × Z.
L
For C × (A1 − {0}), we also have 0 → Õ /OP → Pic (C × (A1 − {0})) → Pic (P1 × (A1 − {0})) → 0.
P∈C×(A1 −{0}) P
And we only need to consider P = P0 × (A − {0}), and ÕP /OP  k[t, z, z−1 ](t) /k[t2 , t3 , z, z−1 ](t2 ,t3 )  Gm . Moreover,
1

Pic (P1 × (A1 − {0}))  Z × Z by using proposition 6.7., and the exact sequence is split by similar reason, so
Pic (C × (P1 × (A1 − {0}))  Gm × Z × Z.

Yang Pi‑Yeh 52 Hartshorne Solutions

2 Schemes 2.8 Differentials

(c) We clearly have a commutative diagram

0 / Gm / Pic (C × A1 ) /Z /0

i j k
  
0 / Gm / Pic (C × (A1 − {0})) / Z×Z /0

Since clearly i is the identity, and k : Z → Z × Z, n 7→ (0, n), we know that j : Pic (C × A1 ) → Pic (C × (A1 − {0}))
is just (t, n) 7→ (t, 0, n). For the other one, we have a commutative diagram

0 / Gm / Pic (C × (A1 − {0})) / Z×Z /0

i j k
  
0 / Gm / Pic (C × (A1 − {0})) / Z×Z /0

And similarly i is the identity, and k is just (d, n) 7→ (d + n, n), so j : (t, d, n) 7→ (t, d + n, n).
(d) By (b) and (c) we have the following diagram

Pic (C × (P1 − {0})) / Pic (C × (P1 − {0} − {∞}))

o7
o oooo
ooo
ooo
Pic X O
OOO
OOO
OOO
OO' 
Pic (C × (P1 − {∞})) / Pic (C × (P1 − {0} − {∞}))

Then if any element in Pic X with image (t, n) and (t0 , n0 ) in Pic (C × (P1 − {0})) or Pic (C × (P1 − {∞})), we have
(t, 0, n) = (t0 , n0 , n0 ), hence t = t0 and n = n0 = 0. So the image of Pic X in Pic (C × A1 ) is just {(t, n) | t ∈ Gm }. Since
Pic (C × A1 ) → Pic (C × {0}) is just identity, the image of the restriction map Pic X → Pic (C × {0}) consists
entirely of divisors of degree 0 on C. If X is projective over k, there exists a closed immersion X → Pn . Then
we have a morphism Pic Pn → Pic X → Pic (C × {0}). Since Pic Pn → Pic (C × {0}) is just n 7→ (1, n), hence not
in the image of Pic X in Pic (C × {0}), which makes a contradiction.

Solution 2.7.14. (a) Take X = P1 , and E = O(−1). Then if OP(E ) (1) is very ample relative to X, i.e. E is the
pullback of OPn (1) on X, which is contradict with the non‑negativity.
(b) Since L is ample, L n is very ample on X for sufficiently large n. By theorem 7.10., OP (1)⊗π∗ L m is very
ample on P relative to X for sufficiently large m. Hence by 2.5.12.(b), OP (1) ⊗ π∗ L m+n is very ample relative to
Y.

2.8 Differentials
Solution 2.8.1. (a) By proposition 8.4. we already have cokerδ = Ωk(B)/k . For injectivity of δ, it equivalent to
show that δ0 : Derk (B, k)Homk (ΩB/k ⊗ k, k) → Homk (m/m2 , k) is surjective. By theorem 8.25., we know that k(B)
is already contained in A, so for any b ∈ B, b can uniquely write as b = λ + c for λ ∈ k(B) and c ∈ m. For any
h ∈ Homk (m/m2 , k), we can define db = h(c̄), hence δ0 is surjective, i.e. δ is injective.
(b) Similarly, k(B) ⊂ B. (⇒) By (a) we have dimk(B) ΩB/k ⊗ k(B) = dim B + dim Ωk(B)/k = dim B + tr.d.k(B)/k.
Then by lemma 8.9., we only need to show that dimK ΩB/k ⊗ K = dim B + tr.d.k(B)/k. By theorem 8.2., we have
ΩB/k ⊗B K = ΩK/k . Since k is perfect, so K is separably generated, i.e. dimK ΩK/k = tr.d.K/k, then dimK ΩB/k ⊗ K =
tr.d.K/k. For any prime ideal p ⊂ A, we have FracA = FracAp , and height p = dim Ap . So tr.d.K/k = dim A =

Yang Pi‑Yeh 53 Hartshorne Solutions

2 Schemes 2.8 Differentials

height p+dim A/p = dim B+dim A/p = dim B+tr.d.k(B). So dimk(B) ΩB/k ⊗k(B) = dimK ΩB/k ⊗K = dim B+tr.d.k(B)/k,
then ΩB/k is free of rank dim B + tr.d.k(B)/k.
(⇐) By theorem 8.6, we have dim Ωk(B)/k = tr.d.k(B)/k. So by (a), we have dim m/m2 = dim B, hence regular.
(c) For any affine neighbourhood U = Spec A of x and x corresponds to a prime ideal p ⊂ A. By (b), OX,x
is regular iff ΩAP /k = ΩA/k = (ΩX/k ) x is free of rank dim Ap + tr.d.k(Ap )/k = dim A = dim X.
(d) By corollary 8.16., there exists an open dense set V such that OX,x is regular if x ∈ V, hence U ⊇ V is
dense. For openness, if x ∈ U, since ΩX/k is locally free by (c), there exists an open neighbourhood W of x such
that ΩX/k |W is free of rank n = dim X. So every for every y ∈ W, OX,y is free of rank n, hence w ∈ U, i.e. U is
open.

Solution 2.8.2. Define B = {(x, s) | s x ∈ m x E x } ⊂ X × V with projection πX : B → X and πV : B → V. For any

x ∈ X, then π−1
X (x) = ker(V ⊗k k(x) → E x ⊗O x k(x)  E x /m x E x . Since E is generated by global section, this map is
surjective. And since dim ker = dim V − rank E , we have dim B = dim X + dim V − rank E . Since r > n, we have
dim B < dim V, i.e. πV is not surjective. So there exists s ∈ V such that for any x ∈ X, we have s x < m x E x . Fixed
such an s, we have an injection OX → E as ×s. Then by 2.5.7.(b), we know that the cokernal is locally free of
rank (rank E − 1).

Solution 2.8.3 (Product Schemes). (a) By proposition 8.10., we have Ω(X×S Y)/Y  p∗1 ΩX/S and Ω(X×S Y)/X  p∗2 ΩY/S .
Then by proposition 8.11., we have Ω(X×S Y)/Y → Ω(X×S Y)/S → Ω(X×S Y)/X → 0, and Ω(X×S Y)/X → Ω(X×S Y)/S →
Ω(X×S Y)/Y → 0, i.e.
p∗1 ΩX/S → Ω(X×S Y)/S → p∗2 ΩY/S → 0

p∗2 ΩY/S → Ω(X×S Y)/S → p∗1 ΩX/S → 0

On every Spec A ⊂ X, Spec B ⊂ Y and Spec C ⊂ S with Spec A and Spec B contained in the preimage of Spec C,
we have Ω(A×C B)/A → ΩB/C ⊗B (B ⊗C A) → Ω(A×C B)/C → Ω(A×C B)/A , with d(1 ⊗ b) 7→ db ⊗ (1 ⊗ 1) 7→ d(1 ⊗ b) 7→ d(1 ⊗ b).
So the above two sequences are split on every affine pieces, hence Ω(X×S Y)/S = p∗1 ΩX/S ⊕ p∗2 ΩY/S .
(b) Denote the dimension of X and Y as n and m. Then clearly by 2.5.16., ωX×Y = Λmn Ω(X×Y)/S = Λmn (p∗1 ΩX/S ⊕
p2 ΩY/S ) = p∗1 Λn ΩX/S ⊗ p∗2 Λm ΩY/S = ωX/S ⊗ ωY/S .
∗

(c) By 1.7.2.(b), we have pa (Y) = 1, then pa (Y × Y) = −1 by 1.7.2.(e). By example 8.20.3., we have ωY  OY .

Then ωY×Y  p∗1 OY ⊗ p∗2 OY  OY×Y . Since Y is proper over k, i.e. Y × Y is proper over Y, hence proper over k.
Then by 2.4.5.(d), we have Γ(Y × Y, OY×Y )  k, i.e. pg = 1.

Solution 2.8.4 (Complete Intersections in Pn ). (a) (⇒) If IY = ( f1 , . . . , fr ), we only need to define Hi = Z( fi ) and
get Y = H1 ∩ . . . ∩ Hr .
(⇐) If Y = H1 ∩ . . . ∩ Hr , and each Hi corresponds to a prime ideal Ii = ( fi ) ⊂ S . Since fi+1 is not a zero
divisor in S /( f1 , . . . , fi ), ( f1 , . . . , fr ) is a regular sequence of and ( f1 , . . . , fr ) ⊂ IY . Since S /( f1 , . . . , fr ) has degree
P
deg Hi , there exists an ideal J such that ( f1 , . . . , fr ) = I ∩ J for codimJ > 2. By unmixedness theorem, the
primary components of ( f1 , . . . , fr ) has codimension ≤ 1, hence J = ∅. So IY = ( f1 , . . . , fr ).
(b) If Y is normal, we know that SingY has codimension ≥ 2, hence SingCone(Y) has codimension ≥ 2.
So by proposition 8.23.(b), we know that S (Cone(Y)) is integrally closed, i.e. S (Y) = S (Cone(Y)) is integrally
closed, hence Y is projectively normal.
(c) Since Y is projectively normal, we have S (Y) = S /IY , i.e. the projection S → S Y is surjective. Hence
Γ(P , OPn (l)) → Γ(Y, OY (l)) is just the l‑grade of the projection above, i.e. surjective. Taking l = 0, we have a
n

surjection k → Γ(Y, OY ), i.e. dim Γ(Y, OY ) ≤ 1, hence the number of connected component ≤ 1, i.e. connected.
(d) For any hyperplane H, there exists a nonsingular hypersurface H1 in |dH| such that H1 has degree d1 .
Since Pn |H1  Pn−1 , then we repeat this process in Pn−1 , . . . , Pn−r+1 , to get a subscheme Y = H1 ∩ . . . ∩ Hr such
that Y is nonsingular with deg Hi = di .

Yang Pi‑Yeh 54 Hartshorne Solutions

2 Schemes 2.8 Differentials

(e) Clearly, ωH1 = OPn (−n − 1) ⊗ OH1 (d1 ) = OH1 (d1 − n − 1). Then ωH1 ∩H2 = OH1 (d1 − n − 1) ⊗ OH1 ∩H2 (H1 .H2 ) =
OH1 (d1 − n − 1) ⊗ (OH1 (d2 )|H1 ∩H2 )  OH1 ∩H2 (d1 + d2 − n − 1). Repeating this method, we have ωY = ωH1 ∩...∩Hr =
P
OY ( di − n − 1).
(f) By (c), we have an exact sequence 0 → IY → OPn → OY → 0. Since deg Y > d − n − 1, which means
there are no section of degree d − n − 1 vanishing on Y, hence dimk Γ(Y, IY (d − n − 1)) = 0. So by this and (e), we
 
have pg (Y) = dimk Γ(Y, ωY ) = dimk Γ(Y, OY (d − n − 1)) = dimk Γ(Pn , OPn (d − n − 1)) = d−1 . By 1.7.2., we’ve already
d−1 n
know that pa (Y) = n .
(g) Denote Y = H ∩ H 0 , where deg H = d, deg H 0 = e and H, H 0 is generated by f and g. Then we have an
exact sequence 0 → (OP3 (−4))|Y → (OP3 (d − 4) ⊕ OP3 (e − 4))|Y → (OP3 (d + e − 4))|Y → OY (d + e − 4) → 0, where the
second arrow is s 7→ ( f s, gs), and the third arrow is (s, t) 7→ (gs − f t). Hence pg (Y) = dimk Γ(Y, OY (d + e − n − 1)) =
d+e−1 −1 d−1 e−1
3
+ 3 − 3 − 3 + 1 = 12 de(d + e − 4) + 1, hence equal to the arithmetic genus by 1.7.2.
Solution 2.8.5 (Blowing up a Nonsingular Subvariety). (a) By proposition 6.5., we have an exact sequence
Z → Pic X̃ → Pic (X̃ − Y 0 ) → 0. Since we have X̃ − Y 0  X − Y, and Y has codimension ≥ 2, we know that
Pic (X̃ − Y 0 )  Pic (X − Y)  Pic X. Moreover, the morphism Z → Pic X̃ is just 1 7→ Y 0 . And by theorem 8.24., we
know that OX̃ (nY 0 )|Y 0  OY (n). So this morphism is injective, i.e. we have an exact sequence 0 → Z → Pic X̃ →
Pic X → 0. Since we have a morphism π∗ : Pic X → Pic X̃, this sequence is split, i.e. Pic X̃  Pic X ⊕ Z.
(b) By (a) we may assume ωX̃  f ∗ M ⊗ L (qY 0 ) for some invertible sheaf M on X and q ∈ Z. Denote
U = X − Y, we have π−1 (U)  U. Since ωX̃ |U  ωU  ωX |U , and as in (a), Pic X  Pic U, we know that
ωX  ωX̃ |U = M , hence ωX̃  f ∗ ωX ⊗ L (qY 0 ). To determine q, by theorem 8.24., we have ωY 0  ωX̃ ⊗ L (Y 0 ) ⊗
OY 0 = f ∗ ωX ⊗ L ((q + 1)Y 0 ) ⊗ OY 0 . By proposition 6.18., we have L ((q + 1)Y 0 ) = IY−q−1
0  OY 0 (−q − 1), hence
ωY 0  f ∗ ωX ⊗ OY 0 (−q − 1). Fix a closed point y ∈ Y, and Z = π−1 (y) = {y} ×Y Y 0 , then by 2.8.3.(b), we have
ωZ = π∗1 ωy ⊗ π∗2 ωY 0 = OZ ⊗ π∗2 ωY 0 = ωZ (−q − 1). By theorem 8.24., Z is a projective space of dimension r − 1, i.e.
ωZ = OZ−r , hence q = r − 1, i.e. ωX̃  f ∗ ωX ⊗ L ((r − 1)Y 0 ).
Solution 2.8.6 (The Infinitesimal Lifting Property). (a) Denote the morphism I → B → B0 as ψ and ϕ. Since
ϕ ◦ θ = f − f = 0, we have Imϕ ⊂ I. Since g and g0 are both ring homomorphism, θ(1) = g(1) − g0 (1) = 1 − 1 = 0.
Hence for every a ∈ k, we have θ(a) = θ(a) · θ(1) = 0. Moreover, for any a, b ∈ A, θ(ab) = g(ab) − g0 (ab) =
g(a)g(b) − g0 (a) − g0 (b) − g(a)g(b) − g0 (a)g(b) + g0 (a)g(b) − g0 (a)g0 (b) = g0 (a)θ(b) + g(b)θ(a). So θ ∈ HomA (ΩA/k , I).
Conversely, if θ ∈ HomA (ΩA/k , I), we have a morphism ψ ◦ θ ◦ d : A → I → B. Then clearly ϕ ◦ ψ ◦ θ ◦ d = 0 by
exactness, we know that g + θ is another k‑linear homomorphism lifting f . Moreover, since I 2 = 0, we have
θ(a)θ(b) = 0 for any a, b ∈ A, then g(ab) + θ(ab) = g(a)g(b) + g(a)θ(b) + g(b)θ(a) + θ(a)θ(b) = (g(a) + θ(a))(g(b) + θ(b)),
hence g + θ is a ring homomorphism.
(b) For any xi , we fix a bi ∈ B0 as the lifting of f ( x̄i ). Then the morphism h : P → B0 is defined as xi → bi . If
a ∈ J ⊂ P, we know that the image of a in B is zero, hence h(a) ∈ I. Moreover, if a ∈ J 2 , we have h(a) ∈ I 2 = 0.
Hence h induces a morphism h̄ : J/J 2 → I.
(c) Take theorem 8.17. on Spec P and Spec A, we have an exact sequence 0 → J/J 2 → ΩP/k ⊗ A → ΩA/k → 0.
Since A is nonsingular, ΩA/k is locally free, hence ExtiA (ΩA/k , I) = 0 for all i > 0. So we have an exact sequence
0 → HomA (ΩA/k , I) → HomA (ΩP/k , I) → HomA (J/J 2 , I) → 0. Then we have θ : ΩP/k → I as the lifting of h̄. So
we can define θ0 : P → ΩP/k → I → B0 as a k‑homomorphism. Define h0 = h − θ, then for any a ∈ J we have
h0 (a) = 0, i.e. h0 induces a morphism g : A → B0 .
Solution 2.8.7. This problem is just an algebraic question as: If A0 is a ring, I ⊂ A0 is an ideal with I 2 = 0,
A0 /I  A, and A‑module M is isomorphic to I, then we need to prove that A0 = A ⊕ M as abelian groups, with
multiplication (a, m)·(a0 , m0 ) = (aa0 , am0 +a0 m). By 2.8.6., the identity morphism on A can be lifted to a morphism
A → A0 . Then the exact sequence 0 → I → A0 → A → 0 is split as an exact sequence of abelian groups, i.e.
A0 = A ⊕ M as abelian groups. Moreover, for any a ∈ A, we clearly know that (a, 0) · (a0 , m0 ) = (aa0 , am0 ) by
the A‑module structure on A. And for any m ∈ I, we have (0, m) · (a0 , m0 ) = (0, a0 m + mm0 ) = (0, a0 m) since
mm0 ∈ I 2 = 0. Hence A0 is the trivial extension.

Yang Pi‑Yeh 55 Hartshorne Solutions

2 Schemes 2.9 Formal Schemes

Solution 2.8.8. If X 0 is a nonsingular variety birational to X with morphism X → X 0 . We can take V as the
largest open subset of X representing this rational morphism, and f : V → X 0 is the representing morphism.
By proposition 8.11., we have f ∗ ΩX0 → ΩV , which induces morphisms f ∗ ωnX0 → ωnV and f ∗ ΩqX0 → ΩqV . By
corollary 4.5. in chapter 1, there exists an open subset U ⊂ V such that f −1 (U)  U, so ΩV |U  ΩX0 | f −1 (U) . Hence
we have the following two commutative diagrams

Γ(X 0 , ωnX0 ) / Γ(V, ωn ) Γ(X 0 , ΩqX0 ) / Γ(V, Ωq )

V V

   
Γ( f −1 (U), ωnf −1 (U) )

/ Γ(U, ωn ) Γ( f −1 (U), Ωqf −1 (U) )

/ Γ(U, Ωq )
U U

Since f (U) is open dense in X 0 , and global section of ωnX0 or ΩqX0 will not vanish on f (U), hence we have
injections Γ(X 0 , ωnX0 ) → Γ(V, ωnV ) and Γ(X 0 , ΩqX0 ) → Γ(V, ΩqV ). As in the proof of theorem 8.19., X − V has codi‑
mension > 1. For every point in X, take an affine neighbourhood U of that point such that ωnX or ΩqX is
free on U. By proposition 6.3., we have Γ(U, OU )  Γ(U ∩ V, OU∩V ) since U − V has codimension > 1 in U,
hence we have Γ(X, ωnX )  Γ(V, ωnV ) and Γ(X, ΩqX )  Γ(V, ΩqV ). So we have injections Γ(X 0 , ωnX0 ) → Γ(X, ωnX ) and
Γ(X 0 , ΩqX0 ) → Γ(X, ΩqX ), i.e. Pn (X 0 ) ≤ Pn (X) and hq,0 (X 0 ) ≤ hq,0 (X). Then by symmetry we have Pn (X 0 ) = Pn (X)
and hq,0 (X 0 ) = hq,0 (X).

2.9 Formal Schemes

Solution 2.9.1. (a) By theorem 8.17., we have an injection I /I 2 → ΩX/k ⊗ OY . Since X = Pn , we have an
injection ΩX/k → OX (−1)n+1 by theorem 8.13., which induces an injection ΩX/k ⊗ OY → OY (−1)n+1 since they are
both locally free. Hence we have an injection I /I 2 → OY (−1)n+1 .
(b) For r = 1, we already have Γ(Y, I /I 2 ) ,→ Γ(Y, OY (−1)n+1 ) = 0, hence Γ(Y, I /I 2 ) = 0. For r > 1, we
have S r (I /I 2 )  I r /I r+1 , then Γ(Y, I r /I r+1 ) = S r (Γ(Y, I /I 2 )) = 0.
(c) For r = 1, Γ(Y, OX /I ) = Γ(Y, OY ) = k, since Y has positive dimension. For greater, by induction we
may assume Γ(Y, OX /I r ) = k. Since we have an exact sequence 0 → I r /I r+1 → OX /I r+1 → OX /I r → 0,
hence Γ(Y, OX /I r+1 ) ,→ Γ(Y, OX /I r ) by (b). Since they are both k‑algebra, and we’ve already had a non‑zero
morphism k = Γ(Y, OY ) → Γ(Y, OX /I r+1 ), we know that Γ(Y, OX /I r+1 ) = k.
(d) Clearly Γ(X̂, OX̂ ) = lim Γ(Y, OX /I r ) = k.
←−−r
Solution 2.9.2. We may replace Z as the scheme‑theoretic image of f , so we may assume that f is dominant.
Then we have an injection OZ → f∗ OX . Hence it induces an injection OẐ → f∗ OX̂ . Then Γ(Ẑ, OẐ ) ,→ Γ(X̂, OX̂ ) = k
by 2.9.1., where Ẑ is the completion of Z along P. Since Γ(Ẑ, OẐ )  ÔZ,P is the completion of OZ,P along its
maximal ideal. So if Z has positive dimension, we have tr.d.OZ,P /k > 0, hence contradict. So Z has zero
degree. Since X is connected, we know that Z is just a single point, i.e. f (X) = P.

Solution 2.9.3. (a) We may assume X is the completion of X = Spec A along I, ˜ and F0 = Fˆ 0 , and F 0 = M̃.
Then F0 /In F0 = M/I ] n M = M̃ . For any affine open subset U ⊂ X, it is induced from U = Spec A ⊂ X. For
n f
00
any s ∈ Γ(U, F ), for any x ∈ U there exists a open neighbourhood D( f g) ∩ X of x such that s|D( f g)∩X lifts to
section t ∈ (F/In F0 )(D( f g) ∩ X). Since we can cover U with finite open sets D( f gi ) ∩ X such that for each i,
s|D( f gi )∩X lifts to ti ∈ (F/In F0 )(D( f g) ∩ X). Since D( f g) ∩ D( f gi ) = D( f ggi ), then t, ti ∈ (F/In F0 )(D( f ggi ) ∩ X) both
lift s, hence t − ti ∈ (F0 /In F0 )(D( f ggi ) ∩ X). Then there exists n > 0 such that gn (t − ti ) extends to a section
ui ∈ (F0 /In F0 )(D( f gi ) ∩ X). So we may pick an n sufficiently large such that ti0 = gn ti + ui is a lifting of gn s on
D( f gi ) for all i, and furthermore ti0 and gn t agree on D( f ggi ). Since we have ti0 and t0j on D( f gi g j ) of (F/In F0 ), and
they are both lifting of gn s, so ti0 −t0j ∈ (F0 /In F0 )(D( f gi g j )∩X). Since ti0 = t0j on D( f ggi g j )∩X, we have gm (ti0 −t0j ) = 0

Yang Pi‑Yeh 56 Hartshorne Solutions

2 Schemes 2.9 Formal Schemes

for some m. Hence we can glue all gm ti0 together for sufficient large m to get a section t00 ∈ Γ(U, (F/In F0 )), which
lifts gm s. So there exists some n > 0, gn s can lift to a section in Γ(U, (F/In F0 )).
Since we can cover U by finite many D( f fi )∩X such that s|D( f fi )∩X lifts to a section of (F/In F0 ) over D( f fi )∩X
P
for all i. Then there exists an n such that ti ∈ Γ(U, (F/In F0 )) is a lifting of fin s. Since we have f = ai fin for
P
some ai ∈ A, we may define t = ai ti ∈ Γ(U, (F/In F0 )) whose image in Γ(U, F00 ) is s, hence surjective.
(b) By proposition 9.6., we know that F0 /In F0 → F0 /Im F0 is surjective for n ≥ m, hence {F0 /In F0 } satisfies
(ML) condition and F0 = lim F0 /In F0 . Moreover, lim F/In F0 = lim F · F0 /In F0 = F · F0 = F. Then by proposition
←−− ←−− ←−−
9.1., we have an exact sequence 0 → lim Γ(U, F0 /In F0 ) → lim Γ(U, F/In F0 ) → lim Γ(U, F00 ) → 0, hence 0 →
←−− ←−− ←−−
Γ(U, F0 ) → Γ(U, F) → Γ(U, F00 ) → 0. Since X is affine, we have 0 → Γ(X, F0 ) → Γ(X, F) → Γ(X, F00 ) → 0.

Solution 2.9.4. Since this problem is local, we may assume X = X̂ is affine, where X = Spec A and X̂ is the
completion of X along I. ˜ (Here we may need to add a condition that A is I‑adic complete.) Hence we have
0 → Γ(X, F ) → Γ(X, F) → Γ(X, F00 ) → 0 since F0 is coherent. We may assume that F0 = M 0∆ and F00 = M 00∆ ,
0

hence 0 → M̂ 0 → N → Mˆ 00 → 0, where N = Γ(X, F). Since A is I‑adic complete, we know that N is an A‑

module, M̂ 0 = M 0 , M̂ 00 = M 00 , and the above exact sequence is an A‑module sequence. Since on every affine
piece D( f ) ∩ X we have the same thing, and clearly have the following diagram

0 / M0 /N / M 00 /0

  
0 / M0 / N0 / M 00 /0
f f

We must have N 0 = N f . Hence Γ(U, F) = Γ(U, N ∆ ) for every principal open set U. And principal open set forms
a topological basis on X, we have F = N ∆ , i.e. coherent.
Here we will explain why we assume A is I‑adic complete. We know that every N, the extension of M̂ 0 and
M̂ 00 is corresponding to an element of Ext1 ( M̂ 00 , M̂ 0 ). And since  is a flat A‑module, we have Ext1 ( M̂ 00 , M̂ 0 ) 
Ext1A (M 00 , M 0 )ˆ. If A is not I‑adic complete, we know that the A‑module homomorphism Ext1A (M 00 , M 0 ) →
Ext1A (M 00 , M 0 )ˆ is not surjective, hence we can pick an extension N of M̂ 0 and M̂ 00 , which is not induced by
some A‑module M, then F is not coherent.

Solution 2.9.5. We may assume F can be generated by global sections {si }i . Define Ui = {x ∈ X | si,x < m x }.
S S
Then X = Ui . Since X is noetherian, we can take a finite sub‑covering {Ui }ni=1 such that X = ni=1 Ui , i.e. F can
be generated by s1 , . . . , sn .

Solution 2.9.6. (a) Denote the morphism Γ(Ym , OYm ) → Γ(Yn , OYn ) as ϕmn , and the morphism Γ(Ym , OY∗m ) →
Γ(Yn , OY∗n ) as ϕ∗mn . Since {Γ(Yn , OYn )} satisfies (ML) condition, for every n, there exists an n0 such that for all
m ≥ n0 , we have Imϕmn = Imϕn0 n . For every s ∈ Γ(Ym , OY∗m ), there exists an t ∈ Γ(Ym , OY∗m ) such that st = 1.
Since Imϕmn = Imϕn0 n , there exists a0 and b0 in Γ(Yn0 , OYn0 ) such that ϕmn (a) = ϕn0 n (a0 ) and ϕmn (b) = ϕn0 n (b0 ).
So ϕn0 n (a0 b0 ) = 1. Since clearly ker ϕn0 n is nilpotent, we can write a0 b0 = 1 +  for some nilpotent element .
Since 1 +  is invertible, a0 is also invertible, i.e. a0 ∈ Γ(Yn0 , OY∗n ), which means Imϕ∗mn = Imϕ∗n0 n . So {Γ(Yn , OY∗n )}
0
satisfies (ML) condition.
(b) Since F is coherent, for any point in X, and for every affine neighbourhood X̂ of X containing this point,
where X = Spec A and X̂ is the completion of X along I˜ for some I ⊂ A, such that F|X̂ = lim Fn for some coherent
←−−
sheaf Fn = M̃n , where Mn are A/I n ‑module. Denote F(X̂) = lim Mn = M. Then (F/In F)(X̂) = M/IM ˜ = Mn . Since
←−−
F/In F  OYn , we know that Mn = (F/In F)(X̂) = A/I n . Hence for any point in X, there exists a neighbourhood
U such that F|U  OU . Since the point is ambiguous, F  OX . So for φ : Pic X → lim Pic Yn , and F ∈ ker φ, we
←−−
know that F/In F  OYn , hence F  OX . So φ is injective.
(c) Since Ln are invertible sheaf, we can define Mn = H omYn (Ln , OYn ), by 2.5.7. we have Ln ⊗ Mn  OYn .
Clearly, Mn+1 ⊗ On = H omYn+1 (Ln+1 , OYn+1 ) ⊗ On = H omYn (Ln , OYn ) = Mn , i.e. {Mn } forms an invertible system.

Yang Pi‑Yeh 57 Hartshorne Solutions

2 Schemes 2.9 Formal Schemes

So we may define M = lim Mn , hence M ⊗ L = lim Mn ⊗ Ln = lim OYn = OX . Clearly 2.5.7.(c) is correct for
←−− ←−− ←−−
formal schemes, then L is locally free of rank 1. So obviously, Pic X → lim Pic Yn is surjective.
←−−
(d) If X is affine, we may assume X = X̂ for some completion of X = Spec A along I. ˜ Then Γ(Yn , OYn ) = A/I n ,
hence ϕmn : A/I → A/I is surjective. So {Γ(Yn , OYn )} satisfies (ML) condition. If Yn is projective over k, we
m n

know that Γ(Yn , OYn ) = k, and ϕmn is just identity on k. So {Γ(Yn , OYn )} satisfies (ML) condition.

Yang Pi‑Yeh 58 Hartshorne Solutions

3 Cohomology

3 Cohomology
3.1 Derived Functors
Nothing.

3.2 Cohomology of Sheaves

Solution 3.2.1. (a) By 2.1.17., we have 0 → ZU → ZX → iP ZP ⊕ iQ ZQ → 0. So we have a long exact sequence
0 → Γ(X, ZU ) → Z → Z ⊕ Z → H 1 (X, ZU ) → . . .. Since Z → Z ⊕ Z is not surjective, we have H 1 (X, ZU ) , 0.
(b) We can assume this ”suitably general position” means Hi = (xi ). Then we can take the Godement
resolution of ZY as M M
0 → ZY → Z Hi → ZHi ∩H j → . . . → ZH0 ∩...∩Hn → 0
i i< j

So this is a flasque resolution. Then we can have H n−1 (Y, ZY ) = Z and H n (Y, ZY ) = 0 when n > 1. Moreover, we
also have H n−1 (X, ZX ) = H n (X, ZX ) = 0 for n > 1 in next section. Since we have an exact sequence 0 → ZU →
ZX → ZY → 0, we have 0 → Z → H n (X, ZU ) → 0, i.e. H n (X, ZU ) = Z.

Solution 3.2.2. Since every open subset of X is connected, the constant sheaf K is clearly flasque. For K /O,
L
we know this is just i (I ), i.e. the direct sum of skyscraper sheaves, hence flasque. So 0 → O → K →
P∈X P P
K /O → 0 is a flasque resolution. By 2.1.21.(e), we have H 1 (X, O) = 0. For n > 1, we have the long exact
sequence . . . → 0 → H n (X, O) → 0 → . . ., i.e. H n (X, O) = 0.

Solution 3.2.3 (Cohomology with Supports). (a) For any short exact sequence of sheaves 0 → F 0 → F →
F 00 → 0, we denote ϕ : ΓY (X, F 0 ) → ΓY (X, F ) and ψ : ΓY (X, F ) → ΓY (X, F 00 ). Since ϕ and ψ is restricted from
the morphism ϕ0 : Γ(X, F 0 ) → Γ(X, F ) and ψ0 : Γ(X, F ) → Γ(X, F 00 ), we have ψ ◦ ϕ = 0, and ϕ is injective. For
any s ∈ ker ψ, there exists t ∈ Γ(X, F 0 ) such that s = ϕ0 (t). Checking the stalk, we have tP = 0 for P ∈ X − Y,since
sP = 0. So t ∈ ΓY (X, F 0 ), i.e. we have the exact sequence 0 → ΓY (X, F 0 ) → ΓY (X, F ) → ΓY (X, F 00 ).
(b) The left exactness is from (a). For any s ∈ ΓY (X, F 00 ) ⊂ Γ(X, F 00 ), there exists a t ∈ Γ(X, F ) such that
s = ψ0 (t). Checking the stalk, we have tP 7→ sP = 0 for P ∈ X − Y. Hence from the exact sequence at stalk
0 → FP0 → FP → FP00 → 0, there exists rP ∈ FP0 such that ϕ0P (rP ) = tP . The morphism rP 7→ tP is induced from
some ri ∈ Γ(Ui , F 0 ) such that ri 7→ t|Ui for some neighbourhood Ui of P. So we have an open covering {Ui } of
X − Y. Since ri |Ui ∩U j − r j |Ui ∩U j 7→ t|Ui ∩U j − t|Ui ∩U j = 0, we have ri |Ui ∩U j = r j |Ui ∩U j because Γ(Ui ∩ U j , ·) is left exact.
So all ri can be glued together to be an r ∈ Γ(U, F 0 ). Since F 0 is flasque, there exists an r0 ∈ Γ(X, F 0 ) such that
r0 |U = r. Then ψ0 (t − ϕ0 (r0 )) = s, and (t − ϕ0 (r0 ))|P = t|P − ϕ0P (r0 | p ) = 0. So t − ϕ0 (r0 ) ∈ ΓY (X, F ) is a preimage of s,
i.e. ψ is surjective.
(c) By lemma 2.4., there exists injective sheaf I , such that F ,→ I , and we may define G = I /F . Since F
and I are both flasque, by 2.1.16.(c) we know that G is flasque. By (b), we know that HY1 (X, F ) = HY1 (X, G ) = 0.
And since HYn (X, I ) = 0 for all n, we have H n (X, F )  H n−1 (X, G ). So by induction, H n (X, F ) = 0.
(d) By 2.1.20.(b), since F is flasque, we have 0 → HY0 (F ) → F → j∗ (F |X−Y ) → 0. So we have 0 →
ΓY (X, F ) → Γ(X, F ) → Γ(X − Y, F ). Since F is flasque, the morphism Γ(X, F ) → Γ(X − Y, F ) is surjective,
then we’ve done.
(e) For any injective sheaf I , by lemma 2.4. and (d), we have 0 → ΓY (X, I ) → Γ(X, I ) → Γ(U, I |U ) → 0.
Take an injective resolution I · of F , since ·|U preserves injection, I · |U is an injective resolution of F |U . So

Yang Pi‑Yeh 59 Hartshorne Solutions

3 Cohomology 3.2 Cohomology of Sheaves

we have the following commutative diagram with exact rows

.. .. ..
. . .

  
0 / ΓY (X, I k ) / Γ(X, I k ) / Γ(U, I k+1 ) /0

  
0 / ΓY (X, I k+1 ) / Γ(X, I k ) / Γ(U, I k+1 ) /0

  
.. .. ..
. . .

So the long exact sequence is by using snack lemma of the above diagram.
(f) We just need show that we have an isomorphism of functors ΓY (X, ·) → ΓY (V, ·|V ). For any U open in
X, we can treat F (U) as a set of continuous morphisms U → Spé (F ) by 2.1.13. So any s ∈ ΓY (X, F ) means a
continuous morphism s : X → Spé (F ) with s(P) = 0 ⊂ FP for all P ∈ X − Y, hence ΓY (X, F ) → ΓY (V, F |V ) is
injective. Moreover, for any s ∈ ΓY (V, F |V ), then we can define a morphism t : X → Spé (F ) as t(P) = s(P) if
P ∈ Y, and t(P) = 0 ⊂ FP if P ∈ X − Y. We can take an open covering {Ui } of X such that for each i, we have
either Ui ⊂ V or Ui ∩ Y = ∅. So if Ui ⊂ V, we know that t(Ui ) = s(Ui ), hence continuous, if Ui ∩ Y = ∅, t(Ui ) = 0,
hence continuous. So t is a continuous morphism, hence t ∈ ΓY (X, F ). So ΓY (X, ·)  ΓY (V, ·|V ). For every sheaf
F with injective resolution I · , we have an injective resolution I · |V of F |V . Hence the isomorphism above
induces the isomorphism of cohomology group HYi (X, F )  HYi (V, F |V ).

Solution 3.2.4 (Mayer‑Vietoris Sequence). For every injective sheaf I , we have the following commutative
diagram

0 0 0

  
0 / ΓY1 ∩Y2 (X, I ) / ΓY1 (X, I ) ⊕ ΓY2 (X, I ) / ΓY1 ∪Y2 (X, I ) /0

  
0 / Γ(X, I ) / Γ(X, I ) ⊕ Γ(X, I ) / Γ(X, I ) /0

  
0 / Γ(X − Y1 ∩ Y2 , I ) / Γ(X − Y1 , I ) ⊕ Γ(X − Y2 , I ) / Γ(X − Y1 ∩ Y2 , I ) /0

  
0 0 0

The last two rows are clearly exact, and three columns are exact since I is flasque, so by nine lemma we
know the first row is exact. Hence for injective resolution I · of F , we have an exact sequence of complexes
0 → ΓY1 ∩Y2 (X, I · ) → ΓY1 (X, I · ) ⊕ ΓY2 (X, I · ) → ΓY1 ∪Y2 (X, I · ) → 0. Taking the long exact sequence of this, we
have . . . → HYi 1 ∩Y2 (X, F ) → HYi 1 (X, F ) ⊕ HYi 2 (X, F ) → HYi 1 ∪Y2 (X, F ) → HYi+1
1 ∩Y2
(X, F ) → . . ..

Solution 3.2.5. We first show that ΓP (X, F )  ΓP (XP , FP ). Since Γ(XP , FP ) = lim Γ(U, F |U ) = FP . So we
←−−P∈U
have a morphism Γ(X, F ) → FP  Γ(XP , FP ), hence it induces a morphism ΓP (X, F ) → ΓP (XP , FP ). For
injectivity, if s, t ∈ ΓP (X, F ) with sP = tP , since for every Q ∈ X with Q , P, we have sQ = tQ = 0, hence s = t.

Yang Pi‑Yeh 60 Hartshorne Solutions

3 Cohomology 3.3 Cohomology of Noetherian Affine Scheme

For surjectivity, for any sP ∈ ΓP (XP , FP )  FP , there exists a neighbourhood U of P and a s ∈ Γ(U, F ) such
that sP = s|P . We may shrink U such that sQ = 0 for any Q ∈ U such that P < {Q}, then we have a section
t ∈ ΓP (X, F ) as the extension of s with 0 outside, hence t|P = sP , i.e. ΓP (X, F ) → ΓP (XP , FP ) is surjective. So
ΓP (X, F )  ΓP (XP , FP ). For higher cohomology, if we have some sheaf F with injective resolution I · , there
exists an injective resolution of F |XP as I · |XP like we’ve done in 3.2.3.(f), hence HPi (X, F ) = HPi (XP , FP ).

Solution 3.2.6. Followed by the hint, we firstly prove that I is injective iff for any U open in X, and for any
sheaf R ⊂ ZU , the morphism R → I can be extended to be a morphism ZU → I . The only if part is trvial.
Conversely, for any sheaves F ⊂ G with morphism φ : F → I , we can define S = {H | F ⊂ H ⊂ G such
that phi can be extended to a morphism H → I }. Then S is a ordered set. By Zorn’s lemma, there exists a
maximal element of S as H . If H $ G , there exists an open set U ⊂ X, and s ∈ G (U) − H (U). Then we can
define ZU as a subsheaf of G generated by s, and R = ZU ∩ F . So the morphism R → I can be extended to
ZU → I , which contradict with the maximality of H , hence H = G , i.e. I is injective.
Secondly we prove that for any finitely generated sheaf R with morphism ϕ : R → lim Iα , this morphism
−−→
must factor through some Iα . We may assume R is generated by some si ∈ R(Ui ). Then there exists ti ∈
Iαi (Ui ) representing the image of si . Since A is a direct system, there exists a α greater than all αi , so we may
choose all ti ∈ Iα (Ui ). Hence ϕ must factor through Iα .
Finally, any subsheaf R of ZU must be finitely generated, since any morphism ϕ : R → lim Iα must factor
−−→
through Iα , then R → Iα can be extended to ZU → Iα , hence ϕ can be extended to ZU → lim Iα . So lim Iα
−−→ −−→
is injective.

Solution 3.2.7. (a) By definition, the constant sheaf Z is the sheaf Z(U) = {s : U → Z | s locally constant}. Then
we may fix a P ∈ S1 , we can define Y = S1 −{P}, and Z as Z (U) = Z(U ∩Y). Then Z is a sheaf, with ZQ  Z for
all Q , P, and ZP  Z ⊕ Z. Hence we the injection Z → Z has cokernal iP (Z), i.e. 0 → Z → Z → iP (Z) → 0.
So we have 0 → H 0 (S1 , Z) → H 0 (S1 , Z ) → H 0 (S1 , iP (Z)) → H 1 (S1 , Z) → H 1 (S1 , Z ) → . . .. Since we clearly
have H k (S1 , Z )  H k (I, ZI ), where I = (0, 1) and ZI is the constant sheaf on I of Z, and I is contractible, we
have H 1 (I, ZI ) = H 1 (pt, Zpt ) = 0, so we have 0 → Z → Z → Z → H 1 (S1 , Z) → 0. Since the second arrow is
isomorphic, we have H 1 (S1 , Z) = Z.
(b) Define D as D(U) = {s : U → R}. D is clearly flasque, hence H 1 (S1 , D) = 0, i.e. 0 → H 0 (S1 , R) →
H (S , D) → H 0 (S1 , D/R) → H 1 (S1 , R) → 0. So we just need to prove that Γ(S1 , D) → Γ(S1 , D/R) is surjective.
0 1
P S
Since every element in Γ(S1 , D/R) can be described as s = ∞ i=1 (si , U i ) for some covering S =
1
Ui with
si ∈ D(Ui ), and for any Ui ∩ U j , ∅, we have si |Ui ∩U j − s j |Ui ∩U j ∈ R(Ui ∩ U j ), where the sum is finite sum
P
because S1 is compact. For s = i (si , Ui ), we may shrink Ui and assume (Ui ∩ Ui+1 ) ∩ (Ui+1 ∩ Ui+2 ) = ∅. So
we can define ri = si+1 |Ui ∩Ui+1 − si |Ui ∩Ui+1 ∈ R(Ui ∩ Ui+1 ). Then there exists a continuous function ri0 such that
ri0 |Ui ∩Ui+1 = ri |Ui ∩Ui+1 , and ri0 |Ui−1 ∩Ui = 0. Then we can define a t ∈ Γ(S1 , D) such that t(P) = si (P) + ri0 (P) if P ∈ Ui .
So t is a preimage of s, i.e. Γ(S1 , D) → Γ(S1 , D/R) is surjective.

3.3 Cohomology of Noetherian Affine Scheme

Solution 3.3.1. (⇒) By definition, trivial. (⇐) Define N as the nilpotent sheaf of X, then N d = 0 for suffi‑
ciently large d since X is noetherian. So for any coherent sheaf F on X, we define Gd = N d · F /N d+1 · F .
Since X and Xred have the same underlying topological space, and OXred = OX /N , so Gd is a coherent OXred ‑
module when d ≥ 0. By theorem 3.5., we have H n (X, Gd ) = H n (Xred , Gd ) = 0. By definition we clearly have
0 → N d+1 F → N d F → Gd → 0, hence H 1 (X, N d+1 F ) → H 1 (X, N d F ) is surjective. Since H 1 (X, N d F ) = 0
for sufficiently large d, hence by induction, we know that H 1 (X, F ) = 0. So by theorem 3.7., X is affine.

Solution 3.3.2. (⇒) Since every irreducible component is closed in X, hence affine by 2.3.11.(b). (⇐) For any
sheaf of ideal I on X corresponding to closed subset Z, then for any irreducible component Y ⊂ X, we have an

Yang Pi‑Yeh 61 Hartshorne Solutions

3 Cohomology 3.3 Cohomology of Noetherian Affine Scheme

exact sequence 0 → IY∪Z → IZ → F → 0, where F = i∗ IY∩Z , and i : Y → X is the closed embedding. Since
H 1 (X, F ) = H 1 (Y, IY∩Z ) = 0, where the second equality is because Y is affine. So we have surjective morphism
H 1 (X, IY∪Z )  H 1 (X, I ). Then we can have a series of surjections H 1 (X, IZ∪Y1 ∪...∪Yn )  . . .  H 1 (X, IZ∪Y1 ) 
H 1 (X, I ). Since X is reduced, we have H 1 (X, IX ) = 0. But we clearly have Z ∪ Y1 ∪ . . . ∪ Yn = X, we know that
0 = H 1 (X, IX )  H 1 (X, I ), hence H 1 (X, I ) = 0. So X is affine.
f g
Solution 3.3.3. (a) Suppose we have exact sequence 0 → M 0 → − M→ − M 00 → 0. If m ∈ ker(Γa (M 0 ) → Γa (M)), m
is clearly in ker(M → M), hence m = 0. If m ∈ ker(Γa (M) → Γa (M )), similarly m ∈ ker(M → M 00 ), hence there
0 00

exists m0 ∈ M 0 such that f (m0 ) = m. Since for some large n, we have an m = 0, hence f (an m0 ) = an m = 0. Since f
is injective, we have an m0 = 0, i.e. m0 ∈ Γa (M 0 ). So we have 0 → Γa (M 0 ) → Γa (M) → Γa (M 00 ).
(b) For any M with injective resolution 0 → M → I · , we have a flasque resolution 0 → M̃ → I˜· of M̃.
So we only need to prove that Γa (M) = ΓY (X, M̃). For any m ∈ Γa (M), there exists a big n such that an m = 0.
n
Then for any p < Y, i.e. there exists some a ∈ a but a < p. Then an is invertible in Mp , hence m = aanm = 0
in Mp , i.e. m ∈ ΓY (X, M̃). Conversely, if s ∈ ΓY (X, M̃), then Supp (s) ⊂ V(a). By 2.5.6.(a), we know that
p
Supp (s) = V(Ann (s)), so a ⊂ Ann (s). We may denote a = ( f1 , . . . , fn ), there exists ki such that fiki ∈ Ann (s).
P
So we may fix an N > ki , we know that aN ⊂ Ann (s), hence aN s = 0, i.e. s ∈ Γa (M).
(c) Clearly Γa (Hai (M)) ⊂ Hai (M). Conversely, fix an injective resolution 0 → M → I · of M. Then we have
0 → Γa (M) → Γa (I · ). So Hai (M) = ker di+1 /Imdi , hence every element can be represented as some m ∈ ker di+1 .
So Hai (M) ⊂ Γa (Hai (M)).

Solution 3.3.4 (Cohomological Interpretation of Depth). (a) If deptha M ≥ 1, there exists at least an x ∈ a is
not a zero divisor of M. If m ∈ Γa (M), then an m = 0 for sufficiently large n, hence xn m = 0. So we only have
m = 0, i.e. Γa (M) = 0. Conversely, if Γa (M) = 0 and M is finitely generated, for any m ∈ M, there exists an x ∈ a
S
such that xn m , 0 for any n ≥ 0. So a * Ann (m), then a * Ann (m). Hence a * Ann (M), i.e. deptha (M) , 0.
(b) We prove this by induction. The case n = 0 is just (a). Suppose deptha (M) ≥ n is equivalent to
Hai (M) = 0 for all i < n. (i⇒ii) If M is an A‑module such that deptha (M) ≥ n + 1, there exists some x ∈ a
×x
such that deptha (M/xM) ≥ n. Since we have an exact sequence 0 → M −−→ M → M/xM → 0, we have
×x ×x
. . . → Han−1 (M/xM) → Han (M) −−→ Han (M) → . . .. By induction, we have Han (M) −−→ Han (M) is injective. Hence
by 3.3.3.(c), we have Han (M) = 0. (ii⇒i) Since Hai (M) = 0 for i < n + 1, by same exact sequence we have
Hai (M/xM) = 0 for i < n. Hence deptha (M/xM) ≥ n, i.e. deptha (M) ≥ n + 1.

Solution 3.3.5. Since this problem is local, we may assume U = Spec A is affine, and P is corresponding to
p ⊂ A. In this question, we will prove there two conditions are equivalent to (iii) depthp A ≥ 2.
(i⇒iii) Since OP = Ap , if ( yx11 , yx22 ) is a regular sequence of Ap , then (x1 , x2 ) is a regular sequence of A and
x1 , x2 < p.
(iii⇒i) If (x1 , x2 ) is a regular sequence of A with x1 , x2 < p, then ( x11 , x12 ) is a regular sequence of Ap .
(ii⇒iii) By 3.2.3.(e) we have 0 → HP0 (X, OX ) → H 0 (U, OX ) → H 0 (U − P, OX ) → HP1 (X, OX ) → H 1 (U, OX ) → . . ..
Since U is affine, we have H 1 (U, OX ) = 0. If any section of OX over U − P can be extended uniquely to a section
over U, we have H 0 (U, OX )  H 0 (U − P, OX ). So we have HP0 (X, OX ) = HP1 (X, OX ) = 0, i.e. Hp0 (A) = Hp1 (A) = 0.
Then by 3.3.4.(b), we know that depthp A ≥ 2.
(iii⇒ii) If depthp A ≥ 2, we know that HP0 (X, OX ) = HP1 (X, OX ) = 0, hence H 0 (U, OX )  H 0 (U − P, OX ) for same
reason.

Solution 3.3.6. (a) For every injective morphism F → F 0 of quasi‑coherent sheaves on X, and morphism
F → G , the morphism F → fi∗ (I˜i ) corresponds to F |Ui → I˜i . Since I˜i is injective, we have a morphism
F 0 |Ui → I˜i , which corresponds to F 0 → fi∗ (I˜i ). So the morphism F → G can be extended to a morphism
F 0 → G , i.e. G is injective.
(b) If F and G are two quasi‑coherent sheaves on U with an injection F → G , then i∗ F → i∗ G is also
injective, and those are both quasi‑coherent, where i : U → X is the embedding map. Since F → i−1 I

Yang Pi‑Yeh 62 Hartshorne Solutions

3 Cohomology 3.4 Čech Cohomology

induces a map i∗ F → I . So it induces a map i∗ G → I , which is corresponding to the morphism G → i−1 I .

Hence I |U is injective in Qco(U). For any open subset V in X, we want to prove that I (X) → I (V) is
S
surjective. Suppose X has a finie affine open covering i Ui for Ui = Spec Ai . For any s ∈ I (V), it induces
an si ∈ I (V ∩ Ui ). Since I |Ui is injective in Qco(Ui ), Ii is injective in Mod(Ai ), where I |Ui = I˜i . Then I |Ui is
injective in Ab(Ui ), so I (Ui ) → I (V ∩Ui ) is surjective, i.e. si has a preimage ti . Since ti ’s are clearly compatible,
there exists a t ∈ I (X), which is the preimage of s. Hence I is surjective.
(c) By (b), any injective resolution in Qco(X) is a flasque resolution, hence trivial.
S
Solution 3.3.7. (a) Suppose a = ( f1 , . . . , fr ), then U = D( fi ). For any element s ∈ HomA (an , M), we have
s( fin ) s( fin ) s( f jn ) Q
fin
∈ M fi = Γ(D( fi ), M̃). Since fin
and f jn
are same in M fi f j , the map HomA (an , M) → i Γ(D( fi ), M̃),
s( f n ) s( f n )
s 7→ ( f n1 , . . . , f nr )
induces a morphism HomA (an , M) → Γ(U, M̃). So we can take a direct limit and get a
1 r
morphism φ : lim HomA (an , M) → Γ(U, M̃). If φ(s) = 0 for some s = (sn ) ∈ lim HomA (an , M), for sufficiently
−−→ s ( f n ) −−→
s ( f n)
large n we have ( f n1 , . . . , n f nr ) = 0. Then fiki sn ( fin ) = 0 for some ki . Denote k = max{ki }. Then so for
n

1 r
N > r(k + n), we have sk ( fik ) = 0 for all i, hence sN (a) = 0. So φ is injective. For any t ∈ Γ(U, M̃), we may
k−ki
fi ti
denote the restriction of t in D( fi ) as ti
k . Denote k = max{ki }, we can replace fik
, hence t can be represent by
fi i
tj
( t1
f1k
,..., tr
frk
). Since clearly ti
fik
= f jk
in M fi f j , there exists li j such that ( fi f j ) li j
(ti f jk − t j fik ) = 0. Denoting l = max{li j },
fil ti
we may replace fik+l
by ti
fin
. Then t can be represented by ( tf1k , . . . , tr
frk
) such that ti f jn = t j fin . Since if N > rn, an is
1
generated by fin , then we can define s : aN → M as s( fin ) = ti , hence φ is surjective.
(b) For any open subset U in X, we can take X − U = V(a). By (a) we just need to prove lim HomA (An , I) →
−−→
lim HomA (an , I) is surjective. Since I is surjective, any morphism s : an → I can be extended to a morphism
−−→ n
t : A → I, hence the morphism HomA (An , I) → HomA (an , I) is surjective. Take a direct limit and we know
that the restriction map is surjective, hence I˜ is flasque.

Solution 3.3.8. If I → I x0 is surjective, the element x10 ∈ I x0 must have a preimage m ∈ I, hence there exists
some n such that x0n (x0 m − 1) = 0. Then 0 = xn+1 x0n (x0 m − 1) = xn+1 x0n+1 m − xn+1 x0n . Since xn+1 x0n+1 = 0, we have
xn+1 x0n = 0, which makes a contradiction.

3.4 Čech Cohomology

Solution 3.4.1. Take an affine covering {Vi } of Y. Since Y is separated, Vi0 ,...,in is also affine. And since f is affine,
{Ui = f −1 (Vi )} is an affine covering of X. So if F |Vi0 ,...,in = M̃, we have F (Vi0 ,...,in ) = M and f∗ F (Ui0 ,...,in ) = M.
Hence the Čech complex of F and f∗ F are the same, i.e. Ȟ i (X, F )  Ȟ i (Y, f∗ F ). Then by theorem 4.5.,
H i (X, F )  H i (Y, f∗ F ).

Solution 3.4.2. (a) Denote function fields of X and Y as KX and KY . Since f is finite, we know that KX is
a finite generated KY ‑module with generators e1 , . . . , er . We may assume e j ∈ OX,ηX can be represented by
L
s j ∈ Γ(U j , OX ). If i j : U j → X is the canonical embedding, we can define M = i j,∗ (s j OU j ). Hence we have a
canonical morphism α : OYr → f∗ M , and ( f∗ M )ηY = KY [e1 , . . . , er ] = KX .
(b) Just take H om(·, F ) to α to get a morphism β : H om( f∗ M , F ) → H om(OYr , F ) and it is isomorphic
at the generic point of Y. Since clearly H om(OYr , F )  F r , and by 2.5.17.(e) we know that H om( f∗ M , F ) is a
quasi‑coherent f∗ OX ‑module, hence H om( f∗ M , F )  f∗ G for some coherent sheaf G on Y, then β is actually
β : f∗ G → F r .
(c) By 3.3.1. we know that X, Y are affine iff Xred , Yred are affine. And clearly if f is finite surjective, fred is
also finite and surjective. So we may reduce the problem to the reduced case. If Y has irreducible components
Y1 , . . . , Yn , then fi : f −1 (Yi ) → Yi is also finite and surjective. Then by 3.3.2. we just need to prove every Yi is
affine, hence we may reduce the problem to the case that Y is irreducible. Since f : X → Y is surjective, there

Yang Pi‑Yeh 63 Hartshorne Solutions

3 Cohomology 3.4 Čech Cohomology

exists an irreducible component of X containing the preimage of ηY , hence we just replace X to this irreducible
component, and by 3.3.2. we may assume additively that X is irreducible. Hence we reduce the problem to
the case that X and Y are both integral.
For any coherent sheaf F on Y, by (b) there exists such a β : f∗ G → F with (ker β)ηY = 0 and (cokerβ)ηY = 0.
Since we have two exact sequences 0 → ker β → f∗ G → Imβ → 0 and 0 → Imβ → F → cokerβ → 0. By 3.4.1.
we have H i (Y, f∗ G )  H i (X, G ) = 0, hence H i (Y, Imβ)  H i+1 (Y, ker β). Moreover, since (ker β)ηY = 0, the support
of ker β is contained in a closec subset Z ⊂ X. Denote the embedding j : Z → X, we have ker β = j∗ j∗ ker β,
and j∗ ker β is coherent, hence H i (Y, ker β)  H i (X, j∗ ker β) = 0. And similarly H i (Y, cokerβ) = 0 for all i. Then
H i (Y, Imβ) = 0 for all i, and H i (Y, F ) = 0 for all i. Hence by theorem 3.7. we know that Y is affine.

Solution 3.4.3. Take U x = Spec k[x, y, x−1 ] and Uy = Spec k[x, y, y−1 ]. Then U x and Uy cover U, and U x ∩ Uy =
d
U xy = Spec k[x, y, x−1 , y−1 ]. Then we have the Čech complex 0 → k[x, y, x−1 ] ⊕ k[x, y, y−1 ] →
− k[x, y, x−1 , y−1 ] → 0,
−1 −1 P
where d( f, g) = f − g. Then H (U, OU )  k[x, y, x , y ]/Imd = { ai j x y | i, j < 0}, i.e. H (U, OU ) is isomorphic
1 i j 1

to the k‑vector space spanned by {xi y j | i, j < 0}.

Solution 3.4.4. (a) By the condition, the restriction map F (Uλ( j) ) → F (V j ) induces a map λn : C n (U, F ) →
C n (V, F ). Since all λn are commutative with d in the complexes C · (U, F ) and C · (V, F ), they induces mor‑
phisms λi : Ȟ i (U, F ) → Ȟ i (V, F ).
(b) We may define λn : C n (U, F ) → C n (V, F ) in the same way. If we have an injective resolution 0 →
F → I · , the morphism C · (U, F ) → I · and C · (U, F ) → C · (V, F ) → I · are both induced from the identity
of F , hence homotopic. So the following diagram is commutative

λ
/ Ȟ i (V, F )
i
Ȟ i (U, F )
LLL
LLL
LLL
L% 
H i (X, F )

hence compatible.
(c) Following the hint, we define D· (U) as the cokernal of C · (U , F ) → C · (U , G ). Then by proposi‑
tion 4.3. we have Ȟ i+1 (U, F )  H i (D· (U)). Since we have the natural map D· (U) → C · (U, R), it induces
the map H i (D· (U)) → Ȟ i (U, R). By induction, we have H i (X, R)  lim Ȟ i (U, R). So we only need to prove
−−→
lim H i (D· (U)) = lim Ȟ i (U, R).
−−→ −−→
Since for any s ∈ Γ(U, R) for some open subset U, there exists an open covering {Ui } of U such that s|Ui
has a preimage ti ∈ Γ(Ui , G ). So any element of C i (U, R), there exists a refinement V with a preimage in
C i (V, G ), hence lim C i (U, G ) → lim C i (U, R) is surjective, i.e. we have an exact sequence 0 → lim C · (U, F ) →
−−→ −−→ −−→
lim C · (U, G ) → lim C · (U, R) → 0. But since direct limit is an exact functor, we have an exact sequence 0 →
−−→ · −−→ ·
lim C (U, F ) → lim C (U, G ) → lim D· (U) → 0, i.e. lim D· (U)  lim C · (U, R). Then we have lim Ȟ i+1 (U, F ) 
−−→ i −−→ −−→ −−→ −−→ −−→
lim D (U)  lim Ȟ i (U, R)  H i (X, R)  H i+1 (X, F ). Then we just take i = 0.
−−→ −−→
Solution 3.4.5. For any open covering U = {Ui }i , we can define Pic (U) to be the subgroup of Pic (X) consisting
of the isomorphic classes of all L satisfying L |Ui  OUi . Then clearly we have Pic (X) = lim Pic (U). So if we
−−→
can prove Ȟ 1 (U, OX∗ )  Pic (U), by 3.4.4. we will get H 1 (X, OX∗ )  Pic (X).
Now let’s prove Ȟ 1 (U, OX∗ )  Pic (U). Denote Z 1 (U, OX∗ ) = ker(C 1 (U, OX∗ ) → C 2 (U, OX∗ )) = {(si j ) | si j |Ui jk s jk |Ui jk =
sik |Ui jk } and B1 (U, OX∗ ) = Im(C 0 (U, OX∗ ) → C 1 (U, OX∗ )) = {(si j ) | si j = ti−1 |Ui j t j |Ui j for some (ti )}. Then by definition,
we have Ȟ(U, OX∗ )  Z 1 (U, OX∗ )/B1 (U, OX∗ ). For any (si j ) ∈ Z 1 (U, OX∗ ), we can denote φi j : OUi |Ui j → OU j |Ui j as ×si j .
So if (si j ) ∈ Z 1 (U, OX∗ ), all OUi j can be glued together through {φi j } to get an L ∈ Pic (U). Hence we define a
morphism Φ : Z 1 (U, OX∗ ) → Pic (U). For any L ∈ Pic (U), we have a set of φi : L |Ui → OUi . Then φ j φ−1 i is an
∗ ∗
automorphism of OUi j , which corresponds to an si j ∈ OX (Ui j ). Then clearly (si j ) ∈ Z (U, OX ), i.e. Φ is surjective.
1

Yang Pi‑Yeh 64 Hartshorne Solutions

3 Cohomology 3.4 Čech Cohomology

If (si j ) ∈ ker Φ, i.e. we have a morphism ψ : OX  L , since we have LUi  OUi by construction, compositing
with ψ, we get an automorphism on OUi , which corresponds to an element ti ∈ OX∗ (Ui ). Then clearly we have
si j = t j |Ui j ti |−1 ∗ ∗
Ui j , i.e. ker Φ = B (U, OX ). So we have Ȟ (U, OX )  Pic (U).
1 1

∗
Solution 3.4.6. On every point P ∈ X, we can denote A = OX,P and have IP = I, OX,P = A∗ , OX∗0 ,P = (A/I)∗ . Since
I 2 = 0, we clearly have an exact sequence 0 → I → A∗ → (A/I)∗ → 0. So we have 0 → I → OX∗ → OX∗0 → 0.
Taking the long exact sequence and using 3.4.5. we have . . . → H 1 (X, I ) → Pic X → Pic X0 → H 2 (X, I ) → . . ..

Solution 3.4.7. Since X does not contain (1, 0, 0), we clearly know that U∩V = X. And clearly U = Spec k[x, y] f (x,1,y) ,
V = Spec k[z, w] f (z,w,1) and U ∩ V = Spec k[s, t, t−1 ] f (s,t,1) , the map d : Γ(U, OX ) ⊕ Γ(V, OX ) → Γ(U ∩ V, OX ) is
(g(x, y), h(z, w)) 7→ g(st−1 , t−1 ) − h(s, t).
If (g, h) ∈ ker d, there exists some e(s, t) ∈ k[s, t, t−1 ] such that g − h = f e. Since (1, 0, 0) < X, we may assume
P
f (x, y, 1) = 0≤i, j≤d ai j xi yi with ad0 = 1. Then we may split e(s, t) as e = e0 + e1 + e2 , where e0 is the sum of all
terms with i ≤ −d − j, e1 is the sum of all terms with j ≥ 0, and the rest is e2 . So e0 e ∈ Im(k[x, y]/( f (x, 1, y)) →
k[s, t, t−1 ]/( f (s, t, 1))), and e1 ∈ Im(k[z, w]/( f (z, w, 1)) → k[s, t, t−1 ]/ f (s, t, 1))). So we must have e2 = 0, g = e0 e + C,
h = e1 e + C, hence ker d = {(g, h) = (C, C) | C ∈ k}, and dim H 0 (X, OX ) = 1.
P
If e ∈ cokerd, we may assume e can be represented by i≥0, j∈Z ai j xi y j ∈ k[x, y, y−1 ]. Since every term xi yi
with j ≥ 0 has preimage (0, xi y j ), and every term xi y j with − j ≥ i has preimage (xi y− j−i , 0), we may assume e is
P
represented by 0<− j<i ai j xi y j . Since (1, 0, 0) < X, we may assume f (x, y, z) = xd + f 0 (x, y, z). Then xd = − f 0 (x, y, 1)
P P
in k[x, y]/( f (x, y, 1)), hence e can be uniquely represented by 0<− j<i<d ai j xi y j , and every 0<− j<i<d ai j xi y j is in the
cokernal. Hence dim H 1 (X, OX ) = 12 (d − 1)(d − 2).

Solution 3.4.8 (Cohomological Dimension). (a) For any quasi‑coherent sheaf G , by 2.5.15. we know that
G = lim Fα for some coherent sheaves Fα . If H n (X, F ) = 0 for all coherent sheaf F , by proposition 2.9., we
−−→
know that H n (X, G ) = lim H n (X, Fα ) = 0. Hence we can only consider coherent sheaves on X.
−−→
(b) By corollary 5.18. in chapter II, there exists a locally free sheaf E with exact sequence 0 → G → E →
F → 0. Then if when i > n we have H i (X, E ) = 0, there exists isomorphism H i (X, F )  H i+1 (X, G ). So we may
take induction from i > dim X to smaller, then we have H i (X, F ) = 0. Hence we can only consider locally free
sheaves.
(c) If X can be covered by r + 1 open affine subsets, then there are no indices in C i (U, F ) if i > r. Hence
clearly cd(X) ≤ r.
(d) If X is a quasi‑projective scheme of dimension r over a field k, there exists an embedding X → Pnk . Then
we can define Ui = X ∩ {xi , 0}, then {Ui } is an affine covering of X. Then by (c) we have cd(X) ≤ r = dim X.
(e) By definition Y is the intersection of hypersurfaces H1 , . . . , Hr . By proposition 2.5. in chapter II, we
know that X−Hi is affine. Hence {X−Hi } is an affine covering of X−Y. Then by (c) we know that cd(X−Y) ≤ r−1.

Solution 3.4.9. If Y is a set‑theoretic complete intersection in X, in the same way with 3.4.8.(e) we have
cd(X − Y) ≤ 1. Then we just need to prove that H 2 (X − Y, OX ) , 0 to get a contradiction. Since X is affine,
we have H i (X, OX ) = 0 for all i > 0. Then by 3.2.3., we have H 2 (X − Y, OX )  HY2 (X, O). By 3.2.4., we have
. . . → HY31 (X, OX ) ⊕ HY32 (X, OX ) → HY3 (X, OX ) → HP4 (X, OX ) → HY41 (X, OX ) ⊕ HY42 (X, OX ) → . . .. Since X − Y1 =
Spec k[x1 , x2 , x3 , x4 , x1−1 , x2−1 ] is affine, by 3.2.3. we have HYi 1 (X, OX )  H i (X − Y1 , OX ) = 0 for all i > 0, and same
for X − Y2 . So we have HY3 (X, OX )  HP4 (X, OX ). By 3.2.3. we also have HP4 (X, OX )  H 3 (X − P, OX ). So we need
to prove that H 3 (X − P, OX ) , 0.
S
Clearly we have X − P = 4i=1 Ui , where Ui = (xi , 0) = Spec k[x1 , x2 , x3 , x4 , xi−1 ]. By 3.4.8.(d) we
have C 4 (U, OX ) = 0, hence H 3 (X − P, OX ) = coker(C 2 (U, OX ) → C 3 (U, OX )). Since this morphism is just
k[x1 , x2 , x3 , x4 , x2−1 , x3−1 , x4−1 ] ⊕ . . . ⊕ k[x1 , x2 , x3 , x4 , x1−1 , x2−1 , x3−1 ] → k[x1 , x2 , x3 , x4 , x1−1 , x2−1 , x3−1 , x4−1 ], the cokernal is
spanned by all x1i1 x2i2 x3i3 x4i4 with all i j < 0. Hence H 3 (X − P, OX ) , 0. So H 2 (X − Y, OX )  H 3 (X − P, OX ) , 0.
If Ȳ is a set‑theoretic complete intersection, we may restrict those hypersurfaces to X. Then Y is a set‑
theoretic complete intersection, hence contradict.

Yang Pi‑Yeh 65 Hartshorne Solutions

3 Cohomology 3.5 The Cohomology of Projective Space

Solution 3.4.10. Take an affine covering U of X. For any infinitesimal extension (X 0 , I ), we have an exact
sequence 0 → I → OX0 → OX → 0. Then by 2.8.7. this sequence is split on every affine piece Ui , which
is given by a lifting αi : OX |Ui → OX0 |Ui . So on every Ui j , there exist two lifting αi |Ui j and α j |Ui j . Hence there
exists βi j ∈ HomO(Ui j ) (ΩX/k (Ui j ), I (Ui j ))  (F ⊗ T )(Ui j ) such that αi |Ui j − α j |Ui j = βi j . Since on each Ui jk , we
have βi j |Ui jk + β jk |Ui jk + βki |Ui jk = αi |Ui jk − α j |Ui jk + α j |Ui jk − αk |Ui jk + αk |Ui jk − αi |Ui jk = 0. Hence (βi j ) ∈ Z 1 (U, F ⊗ T ).
For any other lifting α0i : OX (Ui ) → OX0 (Ui ), we can get (β0i j ) ∈ Z 1 (U, F ⊗ T ). By 2.8.6.(a), there exists γi ∈
HomOX (Ui ) (ΩX/k (Ui ), I (Ui )) such that α − α0 = γ. So βi j − β0i j = γi |Ui j − γ j |Ui j , hence (βi j ) and (β0i j ) are in the same
class of H 1 (U, F ⊗T ). Thus we have a map {isomorphism classes of infinitesimal extension} → H 1 (X, F ⊗T ),
(X 0 , I ) 7→ (βi j ).
Conversely, on each Ui we have trivial lifting OX |Ui → OX |Ui ⊗ F |Ui . Then for any (βi j ) ∈ Z 1 (U, F ⊗ T ), we
may glue all OX |Ui ⊗ F |Ui together through βi j to get a sheaf OX0 . Hence we have the inverse H 1 (X, F ⊗ T ) →
{isomorphism classes of infinitesimal extension}, i.e. the map is bijective.

Solution 3.4.11. Here we need to use spectral sequence. Define a presheaf H q (F ) as H q (F )(U) = H q (U, FU ).
Consider the bicomplex (K pq ) = (C p (U, I q )), where I · is an injective resolution of F . Then HIp HIIq (K ·· ) =
Ȟ p (U, H q (F )). Since HIq (K ·,p ) = Ȟ q (U, I p ), and I p are all injective, we have



 if q ≥ 1
0
HIIp HIq (K ·· ) =


H p (X, F ) if q = 0

Hence the second spectral sequence of K ·· degenerates and H n (X, F ) = H n (K ·· ). So the spectral sequence of
K ·· is Ȟ P (U, H p (F )) ⇒ H p+q (X, F ). So if all H q (Ui0 ...in , F ) = 0, this spectral sequence is degenerated. So
H p (X, F ) = Ȟ p (U, H 0 (F )) = Ȟ p (U, F ).

3.5 The Cohomology of Projective Space

Solution 3.5.1. By the short exact sequence we have 0 → H 0 (X, F 0 ) → H 0 (X, F ) → H 0 (X, F 00 ) → H 1 (X, F 0 ) →
. . .. Since F 0 , F , F 00 are all coherent, and X has finite dimension, this long exact sequence will stop by
P
zeros. Hence i (−1)i (dimk H i (X, F 0 ) − dimk H i (X, F ) + dimk H i (X, F 00 )) is a finite sum and equals to zero, i.e.
χ(F ) = χ(F 0 ) + χ(F 00 ).
×x
Solution 3.5.2. (a) Firstly we need to find some x ∈ OX (1) such that φ x : F (−1) −−→ F is injective. If φ x is not
injective, there exists at least one point P ∈ X such that φ x,P is not injective. Assume P ∈ U = Spec A, F |U = M̃
x
and P corresponds to a prime p. Then Mp → − Mp is not injective, i.e. p is an associated prime of M, and xP ∈ p.
Since X is noetherian, the associated points on X are finite, we only need to take a hyperplane H do not pass
through all associated points, and this hyperplane corresponds to x. This is trivial in linear algebra. Then we
×x
have some x ∈ OX (1) such that F (−1) −−→ F is injective.
So, if we denote the cokernal of this morphism as G , we have χ(G (m)) = χ(F (m))−χ(F (m−1)). Moreover,
clearly Supp (G ) = Supp (F ) ∪ V(x), i.e. dim Supp (G ) = dim Supp (F ) − 1. Hence by induction, ∆(χ(F (m)))
is a polynomial, we have χ(F (m)) is a polynomial, namely the Hilbert polynomial P(m).
(b) Since H i (X, F (m)) = 0 for i > 0 and m  0, we have P(m) = χ(F (m)) = dim H 0 (X, F (n)) = dim Mn =
P M (m) for m  0.

Solution 3.5.3 (Arithmetic Genus). (a) Since X is integral, we clearly know that X is a projective variety. Then
P
by theorem 3.4.(a) in chapter I, we have H 0 (X, OX ) = k. Hence pa (X) = r−1 i=0 (−1) dimk H (X, OX ).
i r−i

(b) By 3.5.2.(b) we have χ(OX ) = PX (0), i.e. (−1)r (χ(OX ) − 1) = (−1)r (PX (0) − 1) are equivalent.
(c) If X and Y are both nonsingular projective curve over k and birational equivalent, we only need to
show X  Y, hence trivially pa (X) = pa (Y). We only need to define the morphism X → Y and Y → X. Define

Yang Pi‑Yeh 66 Hartshorne Solutions

3 Cohomology 3.5 The Cohomology of Projective Space

the function field of X and Y as K. For any point P ∈ X, since P is nonsingular, OP is a valuation ring with
fractional field K. Hence we have a commutative diagram

Spec K /: Y
t tt
∃! ttt
t
ttt
 t 
X / Spec OP / Spec k

Since Y is proper over Spec k, by valuation criterian there exists a morphism X → Spec OP → Y. Hence we
have a set‑theoretic map X → Y. By the proof of theorem 6.9. in chapter I, the above map is a morphism from
abstract curve X to Y. Hence we have a morphism X → Y. Similarly we have Y → X as the inverse of X → Y,
i.e. X  Y.

Solution 3.5.4. (a) For any sheaf F , we may define P(γ(F )) = P(F ) ∈ Q(z) by 3.5.2.(a). By 3.5.1. we know
that for any 0 → F 0 → F → F 00 → 0 we have P(γ(F )) = P(γ(F 0 )) + P(γ(F 00 )). Then by definition of
Grothendieck group, we have a additive homomorphism P : K(X) → Q[z].
(b) By hint, we firstly prove (1⇒2). If K(X) is a free abelian group generated by {γ(OLi )}. Since there exists
  P
a linear embedding j : Pik → Prk such that OLi = j∗ OPi . So P(γ(OLi )) = P(γ(OPi )) = i+z . So if α = ai γ(OLi ) such
P   k k i
that P(α) = 0, we have ai i+z i
= 0, i.e. ai = 0 for all i. Hence P is injective.
Then we prove (1) and (2) simultaneously by induction on r. The case r = 0 is trivial. We may assume
the case r − 1. We may assume L0 ⊂ L1 ⊂ . . . ⊂ Lr . By 2.6.2. we have K(Pr−1 ) → K(Pr → K(Pr − Pr−1 ) → 0.
We may take Lr−1  Pr−1 . Then the morphism Pr−1 : K(Pr−1 ) → Q[z] factors through K(Pr−1 ) → K(Pr ) → Q(z),
hence K(Pr−1 ) → k(Pr ) is injective. Moreover, we have K(Pr − Pr−1 ) = K(Ar ) = OAr · Z, i.e. we have 0 → Zr−1 →
K(Pr ) → Z → 0 by induction. Since Ext1 (Z, Zr−1 ) = 0, we know K(Pr ) must be the unique extension of Z by
Zr−1 , i.e. Zr .

Solution 3.5.5. (a and c) We will use induction on the codimension of Y. If Y has codimension 0, i.e. Y = X.
Hence trivially H 0 (X, OX (n)) → H 0 (Y, OY (n)) is surjective. For higher codimension d of Y, by 2.8.4. we know
Y = H1 ∪ . . . ∪ H s for some hypersurfaces H1 , . . . , H s . Then we may define Z = H1 ∪ . . . ∪ H s−1 . By induction
we have H 0 (X, OX (n)) → H 0 (Z, OZ (n)) is surjective and H i (Z, OZ (n)) = 0 for 0 < i < dim q. Denote i : Y → Z as
the canonical closed embedding. Then if d = deg H s , we have an exact sequence 0 → OZ (n − d) → OZ (n) →
i∗ OY (n) → 0. Since H 1 (Z, OZ (n−d)) = 0 by theorem 5.2., we have H 0 (Z, OZ (n)) → H 0 (Y, OY (n)) is surjective, hence
H 0 (X, OX (n)) → H 0 (Z, OZ (n)) → H 0 (Y, OY (n)) is surjective. Moreover, for any 0 < i < dim Y = dim Z − 1, we have
. . . → H i (Z, OZ (n − d)) → H i (Z, OZ (n)) → H i (Y, OY (n)) → H i+1 (Z, OZ (n − d)) → . . .. By induction hypothesis
H i (Z, OZ (n − d)) = H i+1 (Z, OZ (n − d)) = 0, then H i (Y, OY (n)) = H i (Z, OZ (n)) = 0.
(b) Since k = H 0 (X, OX ) → H 0 (Y, OY ) is surjective, and both cohomology groups are k‑modules, we clearly
have H 0 (Y, OY ) = k. Since k has no idempotent, Y is connected.
(d) By theorem 5.2. we have H i (Y, OY ) = 0 for any i > dim Y. Then χ(OY ) = dimk H 0 (Y, OY )+(−1)q dimk H q (Y, OY ) =
1 + (−1)q H q (Y, OY ). Then pa (Y) = (−1)q (χ(OY ) − 1) = dimk H q (Y, OY ).

Solution 3.5.6 (Curves on a Nonsingular Quadric Surface). (a) (1) We denote i : Q → X and j : P1 → Q as
the canonical embeddings. If a = b, there exists an exact sequence 0 → OX (a − 2) → OX (a) → i∗ OQ (a) → 0.
Since H 1 (X, OX (a)) = H 2 (X, OX (a − 2)) = 0, we have H 1 (Q, OQ (a)) = 0. If a = b + 1, we have an exact sequence
0 → OQ (a, a + 1) → OQ (a + 1) → j∗ OP1 (a + 1) → 0. Since we’ve already have H 1 (Q, OQ (a + 1)) = 0, and
H 0 (Q, OQ (a+1)) → H 0 (Y, OP1 (a+1)) is surjective because this is just the restriction. Hence H 1 (Q, OQ (a, a+1)) = 0.
And the case (a + 1, a) is the same.
(2) If a ≤ b, we may write a = b−n for some n ≥ 0. Then we have 0 → OQ (b−n, b) → OQ (b) → j∗ OP1 (b)n → 0.
Hence similarly, H 1 (Q, OQ (b)) = 0 and H 0 (Q, OQ (b)) → OP1 (b)n )) = 0 is surjective. Then H 1 (Q, OQ (a, b)) = 0 if
a, b < 0.

Yang Pi‑Yeh 67 Hartshorne Solutions

3 Cohomology 3.5 The Cohomology of Projective Space

(3) Denote n = −a ≥ 2. Then we have 0 → OQ (−n, 0) → OQ → j∗ OPn1 → 0. Hence the long exact sequence
is 0 → 0 → k → kn → H 1 (Q, OQ (a, 0)) → 0, i.e. H 1 (Q, OQ (a, 0)) = kn−1 , 0.
L
Actually we can use Künneth formula H n (X × Y, F  G ) = p+q=n
H p (X, F ) ⊗ H q (Y, G ) to do this exercise.
(b) (1) Clearly IY = OQ (−a, −b). Then H 0 (Q, OQ (−a, −b)) = H 1 (Q, OQ (−a, −b)) = 0. Moreover, by the exact
sequence 0 → IY → OQ → i∗ OY → 0, where i : Y → Q is the canonical embedding. Then the long exact
sequence is 0 → 0 → k → H 0 (Y, OY ) → 0. Then H 0 (Y, OY ) = k, i.e. Y is connected.
(2) Since OQ (a, b) corresponds to a closed embedding Q  P1 × P1 → Pa × Pb → PN . Then by Bertini’s
theorem and the bracket in the theorem 8.18. in chapter II, i.e. the remark 7.9.1., there exists a hyperplane
H ⊂ PN such that H * Q, and Y = H ∩ Q is irreducible smooth. Since Q has dimension 2, Y is a curve.
(3) Clearly Q is normal. Then by 2.8.4.(b) we know that Q is projective normal. Then by 2.5.14.(d),
Γ(PN , OPN (n)) → Γ(Q, OQ (n)) is surjective for every n. Since Y is normal, by 2.5.14.(d) Γ(Q, OQ (n)) → Γ(Y, OY (n))
is surjective for every n iff Y is projective normal. Since 0 → IY (n) → OQ (n) → OY (n) → 0, we know Y is pro‑
jective normal iff H 1 (Q, IY (n)) = 0. Since IY (n) = OQ (−a, −b) ⊗ OQ (n) = OQ (n − a, n − b). Then H 1 (Q, IY (n)) = 0
iff |(n − a) − (n − b)| ≤ 1, i.e. |a − b| ≤ 1.
`
(c) Since Y is in the form (a, b), we have Y = Y1 Y2 for Y1 = a copies of P1 and Y2 = b copies of P1 .
Then since IY2 is clearly flat, we have an exact sequence 0 → IY1 ⊗ IY2 → IY2 → OY1 ⊗ IY2 → 0, i.e.
0 → OQ (−a, −b) → OQ (0, −b) → OY1 ⊗OQ (0, −b) → 0. By (a) we have H 0 (Q, OQ (−a, −b)) = H 1 (Q, OQ (−a, −b)) = 0.
Then we only need to calculate H 2 (Q, OQ (−a, −b). As in (a.3), we have 0 → OQ (0, −b) → OQ → j∗ OPb1 ) → 0.
Hence 0 → 0 → k → kn → H 1 (Q, OQ (0, −b) → 0 → 0 → H 2 (Q, OQ (0, −b) → 0 → . . .. Hence H 2 (Q, OQ (0, −b)) =
0. By (a), we have H 1 (Q, OQ (0, −b)) = kb−1 , and by similar method of (a), we have H 1 (Q, OQ (0, −b)⊗OY ) = ka(b−1) .
Then for exact sequence 0 → OQ (−a, −b) → OQ (0, −b) → OY1 ⊗ OQ (0, −b) → 0, we have 0 = H 1 (Q, OQ (−a, −b) →
kb−1 → ka(b−1) → H 2 (Q, OQ (−a, −b)) → 0 → . . .. Hence pa (Y) = χ(OQ (−a, −b)) = dimk H 2 (Q, OQ (−a, −b)) =
a(b − 1) − (b − 1) = ab − a − b + 1.

Solution 3.5.7. (a) For any coherent sheaf F on Y, we have H i (Y, F ⊗ (i∗ L )n ) = H i (X, i∗ (F ⊗ (i∗ L )n )) =
H i (X, i∗ F ⊗ L n )) = 0 for i > 0 and n  0 since L is ample. Hence i∗ L is ample.
(b) Clearly Xred → X is a closed embedding, we only need to prove (⇐). For any coherent F on X, if we
denote N is the nilpotent sheaf of OX , i.e. N r = 0 for some r, we have a filtration F % N F % . . . % N r F = 0.
And clearly we have N j F /N j+1 F is simultaneously a coherent OXred ‑module and a coherent OX ‑module.
Hence H i (X, (N j F /N j+1 F ) ⊗ L n ) = H i (Xred , (N j F /N j+1 F ) ⊗ Lred n
) = 0 for n  0 and i > 0 since Lred
is ample. By the filtration we have 0 → N F → N F → N F /N j+1 F → 0, hence H i (X, F ⊗ L n ) =
j+1 j j

H i (X, N F ⊗ L n ) = . . . = H i (X, 0 ⊗ L n ) = 0. Hence L is ample.

(c) For any irreducible component Xi of X, the canonical embedding Xi → X is closed, hence by (a) we only
S
need to prove (⇐). By (b) we may assume X is reduced. Write X = mi Xi for some irreducible components
Xi . Then we prove it by induction on m. Denote the ideal sheaf of X1 by I . Then for any coherent sheaf F on
X, we have 0 → I F → F → F /I F → 0. Clearly Supp (I F ) ⊂ X2 ∪ . . . ∪ Xm and Supp (F /I F ) ⊂ X1 . By
induction hypothesis we know H i (X, I F ⊗ L n ) = 0 and H i (X, F /I F ⊗ F n ) = 0 for i > 0 and n  0. Then
clearly H i (X, F ⊗ L n ) = 0 for i > 0 and n  0.
(d) (⇒) For any coherent F on X, we have f∗ F is coherent on Y. Then by 3.4.1. we have H i (X, F ⊗
( f ∗ L )n ) = H i (Y, f∗ (F ⊗ ( f ∗ L )n )) = H i (Y, f∗ F ⊗ L n ) = 0 for i > 0 and n  0. Hence f∗ L is ample.
(⇐) By (b) and (c) we can reduced to the case that X and Y are both integral. Then we will prove this
by noetherian induction. By 3.4.2.(b), there exists a coherent sheaf G on X such that u : f∗ G → F ⊕m is an
isomorphism at the generic point of Y. Then Supp (ker(u)) $ Supp (F ) and Supp (coker(u)) $ Supp (F ).
So by induction we have H i (Y, ker(u) ⊗ L n ) = 0 and H i (Y, coker(u) ⊗ L n ) = 0 for i > 0 and n  0. Hence
H i (Y, F ⊗ L n ) = H i (Y, f∗ G ⊗ L n ) = H i (X, G ⊗ ( f ∗ L )n ) = 0 for i > 0 and n  0. Hence L is ample on Y.

Solution 3.5.8. (a) (Is this question an interrogative sentence?)

Yang Pi‑Yeh 68 Hartshorne Solutions

3 Cohomology 3.6 Ext Groups and Sheaves

(b) By theorem 6.2A. in chapter I we know that normality implies nonsingularity. And the completeness
is clear. Hence by (a), X̃ is projective. For any very ample invertible sheaf L on X̃, there exists a closed
embedding i : X̃ ,→ Pm for some m such that L = i∗ O(1). Since the preimage in X̃ of singular points in X
P
is just finite point, by Bertini’s theorem, there exists a hyperplane H ⊂ Pm such that D = i∗ H = Pi is an
P
effective divisor on X̃ and f (Pi ) are all nonsingular points on X. Then we denote D0 = f (Pi ), and L0 is the
invertible sheaf on X corresponding to D0 . Hence L = f ∗ L0 . Then by 3.5.7.(d), we know L0 is ample on X,
hence L0m is very ample on X for some m  0. So X is projective.
Sn S
(c) If X = i=1 Xi for some irreducible components Xi , we denote Y = ni=2 Xi , and X = X1 ∪ Y. Since we
have an exact sequence 0 → OX∗ → OX∗1 ⊕ OY → OX∗1 ∩Y → 0. Since X1 ∩ Y is just points, hence H 1 (X1 ∩ Y, OX∗1 ∩Y ) =
Pic (X1 ∩ Y) = 0. So Pic X → Pic X ⊕ Pic Y is surjective. By induction on the number of irreducible components
L
of X, we know Pic X → Pic Xi is surjective. By (b), Xi are all projective, hence we have invertible sheaves
Li on Xi such that Li ’s are all very ample. Then by 3.5.7.(c), we know that L = L1  . . .  Ln is ample on X.
So L m is very ample on X for some m  0, i.e. X is projective.
(d) Take the nilpotent sheaf N of OX . Then N r = 0 for some r. We take i > 2r and I = N i . By 3.4.6.
we have an exact sequence 0 → I → OX∗ → OX∗0 → 0, where X0 = (X, OX /I ). Then since X has dimension
r
1, we know H 2 (X, I ) = 0, hence Pic X → Pic X0 is surjective. Since OX0 /N [ 2 ] = OXred , hence we can use the
induction of r and know Pic X → Pic X0 → . . . → Pic Xred is surjective. Then X is projective by 3.5.7.(b).

Solution 3.5.9 (A Nonprojective Scheme). Suppose k = 0, we know that δ(O(1)) = (− xx02 ) · (− xx12 ) · (− xx22 ) =
1 2 0
−x0−1 x1−1 x2−1 ∈ H 2 (X, ω). Hence the morphism δ is injective. So The morphism H 1 (X, ω) → Pic X 0 is surjective.
Since H 1 (X, O(−3)) = 0, we have Pic X 0 = 0. So X 0 is not projective.

Solution 3.5.10. For the exact sequence F 1 → . . . F r , we have r exact sequences: 0 → keri → F i → Imi → 0.
Then by Serre’ theorem, there exists ni such that for all n > ni , we have 0 → Γ(X, keri (n)) → Γ(X, F i (n)) →
Γ(X, Imi (n)) → 0. So take an N = max{ni }. Then for any n > N, we have Γ(X, F 1 (n)) → . . . → Γ(X, F r (n)).

3.6 Ext Groups and Sheaves

Solution 3.6.1. We denote the set E(F 00 , F 0 ) = {all extension of F 00 by F 0 up to isomorphism}. Then we
need to prove the morphism Φ : E(F 00 , F 0 ) → Ext1 (F 00 , F 0 ) is a bijection set‑theoretically.
We need to define a map Ψ : Ext1 (F 00 , F 0 ) → E(F 00 , F 0 ). Fixing an injective resolution 0 → F 0 → I · ,
we know every element e ∈ Ext1 (F 00 , F 0 ) has a representing element e0 : F 00 → Z 1 (I · ) ∈ Hom(F 00 , Z 1 (I · ).
So we may define Ψ(e) = F as the pulling‑back of 0 → F 00 → I 0 → Z 1 (I · ) → 0 by e0 , i.e. F = I 0 ⊕e0 F 0
consisting of all section (s, t) of I 0 ⊕ F 0 which satisfying d0 (s) = e0 (t). The morphism F 00 → I 0 ⊕e0 F 0 is
the direct sum of F 00 → I 0 and the zero morphism. And the morphism I 0 ⊕e0 F 0 → F 0 is the projection.
Suppose e00 : F 00 → Z 1 (I · ) represents e too. There exists a morphism f : F 00 → I 0 such that e0 − e00 = d0 f .
Then we have an isomorphism I 0 ⊕e0 F 00 → I 0 ⊕e00 F 00 as (s, t) 7→ (s − f (t), t) for every section (s, t). Hence
the Ψ is well‑defined.
Easy to see that Φ and Ψ are inverse to each other. So E(F 00 , F 0 )  Ext1 (F 00 , F 0 ).

Solution 3.6.2. (a) Suppose there exists such a P. For any open subset U ⊂ X and any closed point x ∈ U,
we know OX → i x (k(x)) is surjective, where i x (k(x)) is the skyscraper sheaf of the local field k(x). Then P →
i x (k(x)) → 0 is surjective. For any V $ U open and containing x (the existence of such V is based on the fact
that k is infinite), we have another surjection j! (OX |V ) → i x (k(x)), where j : V → X is the embedding. So by the

Yang Pi‑Yeh 69 Hartshorne Solutions

3 Cohomology 3.6 Ext Groups and Sheaves

projectivity of P, we have a morphism P → j! (OX |V ), and the following commutative diagram

P / OX /0

 
j! (OX |V ) / i x (k(x)) /0

Hence the morphism P(U) → i x (k(x))(U) = k(x) must factor through zero. So every section in P(U) has the
zero stalk on x. By the ambiguity of x, P(U) = 0. Hence P = 0, which makes a contradiction.
(b) Suppose there exists a such P. In the Coh(X) case, there exists some n such that Γ(X, P(n)) → Γ(X, OX (n))
is surjective by theorem 5.2. So Γ(X, P(n)) is not empty. In the Qco(X) case, we can pick an affine piece
U = Spec A of X. Then P|U → OX |U → 0 is induced by an A‑module injection A → M. So clearly the image
M 0 of this injection is a finitely generated A‑module, which induces a surjection M̃ 0 → OX |U . By 2.5.15. there
exists a coherent subsheaf P 0 for P, such that there exists a surjection P 0 → OX and P 0 |U = M̃ 0 . Then
by theorem 5.2. for n  0 we have Γ(X, P 0 (n)) → Γ(X, OX (n)) is surjective. So Γ(X, P 0 (n)) is not empty, i.e.
Γ(X, P(n) is not empty.
Since i x (k(x)) ⊗ OX (−n − 1)  i x (k(x)), we have a surjection OX (−n − 1) → i x (k(x)) → 0. Hence we have a
morphism P → OX (−n − 1), which induces a twisting P(n) → OX (−1). But Γ(X, OX (−1)) is empty, which
makes a contradiction.

Solution 3.6.3. (b) By theorem 6.3. we may assume X = Spec A is affine. Then F = M̃ for some finitely
generated A‑module M and G = Ñ for some A‑module N. Since M is finitely generated, there exists a finite
free resolution An· → M → 0. Then Exti (M, N) = H i (Hom(An· , N)) = H i (N n· ) is an A‑module. Hence by theorem
6.7. E xti (F , G ) = (Exti (M, N))∼ is quasi‑coherent.
(a) As in (b), since N is finitely generated, Exti (M, N) = H i (N n· ) is also finitely generated. Hence E xti (F , G )
is coherent.

Solution 3.6.4. By theorem 1.3. we just need to show that (E xti (·, G ) is a coeffaceable functor. Since Coh(X)
has enough locally frees, for any F ∈ Coh(X), there exists a locally free sheaf L such that L → F → 0. So
we only need to show for any locally free L and G ∈ Mod(X), we have E xti (L , G ) = 0 for i > 0.
For every affine piece U = Spec A such that L |U = OUn , we have E xti (L , G )|U = E xti (OUn , G |U ) by theorem
6.7. Take an injective resolution 0 → G → I · . We have E xti (L , G )|U = H i (H om(OUn , I · |U ) = H i (I · |U )n = 0.
Hence E xti (L , G ) = 0 in the global.

Solution 3.6.5. (a) (⇒) We’ve done in 3.6.4.

(⇐) If E xt1 (F , G ) = 0, we have Ext1 (F x , G x ) = E xt1 (F , G ) x = 0 for every x ∈ X. Then by ambiguity of G ,
we know that F x is projective. Since it is also finitely generated, F x is free. Then by 2.5.7., F is locally free.
(b) (⇒) By theorem 6.5. and (a), trivial.
(⇐) Since we may take a locally free resolution 0 → L n . . . → L 0 → F → 0, denoting H = ker(L 0 → F ),
we have a locally free resolution 0 → L n . . . → L 1 → H → 0 for H , hence hdH ≤ n − 1. Since we
have 0 → H → L0 → F → 0, we know that E xti (F , G ) = E xti−1 (H , G ). Hence by induction we know
E xti (F , G ) = 0 for all i > n.
(c) Take the locally free resolution 0 → L n . . . → L 0 → F → 0 of F . Then we have 0 → L x· → F x →
0. Hence hdF ≥ sup x pdOx F x . Conversely, if hdF > sup x pdOx F x strictly, by theorem 6.10. we know
Exti (F x , N) = 0 for every x ∈ X, i ≥ hdF and A‑module N. So for every G , we know E xti (F , G ) = 0 for
i ≥ hdF , which makes a contradiction with (b).

Solution 3.6.6. (a) (⇒) Trivial by definition.

(⇐) Firstly, we know Exti (M, An ) = 0 for all n by induction on n via the exact sequence 0 → An−1 →
A → A → 0. Then for any finitely generated A‑module N, there exists 0 → K → An → N → 0 for some
n

Yang Pi‑Yeh 70 Hartshorne Solutions

3 Cohomology 3.6 Ext Groups and Sheaves

n and finitely generated K. Hence we have Exti (M, N) = Exti+1 (M, K). Then by decreasing induction and
proposition 6.11., we know Exti (M, N) = 0 for all i > 0. Since M is finitely generated, there exists L and m such
that 0 → L → Am → M → 0. Then Exti (M, L) = 0 for all i. Hence Hom(M, Am ) → Hom(M, M) is surjective. So
the identity M → M factors through M → Am → M, hence M is a summand of Am , i.e. projective.
(b) (⇒) By proposition 6.10., trivial.
(⇐) We will prove it by induction on n. If n = 0, this is just (a). For generous n, since M is finitely generated,
there exists an L and m such that 0 → L → Am → M → 0. Hence Exti−1 (N, A) = Exti (M, A) = 0 for all i > n. By
induction pdN ≤ n − 1. Hence pdM ≤ pdN + 1 = n.

Solution 3.6.7. Take a free resolution of M as An· → M → 0. Then we have a locally free resolution of M̃ as
OXn· → M̃ → 0. Then ExtiX ( M̃, Ñ) = H i (HomX (OXn· , Ñ)) = H i (HomA (An· , N)) = ExtiA (M, N), and E xtXi ( M̃, Ñ)) =
H i (H omX (OXn· , Ñ)) = H i (HomX (An· , N)∼ ) = (H i (HomX (An· , N))∼ = ExtiA (M, N)∼ .

Solution 3.6.8. (a) For every open subset U ⊂ X, we denote Z = X − U. Firstly we consider the case that Z is
irreducible. In this case, Z is a Weil divisor of X, which corresponds to a Cartier divisor D = {(Ui , fi )}. Then
consider the invertible sheaf L = L (D), which satisfies L |Ui  OUi , fi−1 ←[ 1. Then we consider the global
section s ∈ Γ(X, L ) glued by all fi−1 on Ui . So X s = X − Z = U. For general case, if Z = Z1 ∪ . . . ∪ Zn , there exists
N N
Li and global sections si such that X si = X − Zi . Then consider L = Li and the global section s = si . So
S
we have X s = X si = U.
(b) Since F is coherent, we may cover X by some Ui = Spec Ai such that F |Ui = M̃i for some finitely
generated Ai ‑modules, i.e. F |Ui is generated by finitely many section mi j ∈ Mi . By (a), there exist si j ∈
Γ(Ui , Li j ) for some Li j such that Ui is covered by X si j and mi j ∈ Γ(X si j , F ). Then by theorem 5.14. in chapter
n n n n
II, si ji j mi j ∈ Γ(X, Li j i j ⊗ F ), which determine a morphism OX → Li j i j ⊗ F . Tensoring with Li j i j and direct
L ni j
summing up all i, j, we have a morphism i, j
Li j → F . Since mi j generates F locally, this morphism is
surjective.

Solution 3.6.9. (a) By 3.6.8. the existence of locally free resolution of F is obvious. By theorem 6.11A. and the
regularity of X, we know pdF x ≤ dim OX,x ≤ dim X for all x ∈ X. And by 3.6.5. we know hdF ≤ sup x pdF x ≤
dim X < ∞, hence there exists a finite locally free resolution of F .
(b) STEP 1. We need to show if there exists two surjections F 0 → F and F 00 → F , there exists a locally
free sheaf E and a commutative diagram
E / F0

 
F 00 /F
with every arrows surjective.
Denote the kernal of F 0 ⊕ F 00 → F as G . Since F 0 → F and F 00 → F are both surjective, we know the
morphism G → F 0 ⊕ F 00 → F 0 and G → F 0 ⊕ F 00 → F 00 are both surjective. Then by 3.6.8., there exists a
locally free sheaf E with a surjection E → G , hence the E is the one we need.
STEP 2. We need to show that if we have exact sequences 0 → G 0 → E 0 → F 0 → 0 and 0 → G 00 → E 00 →
F → 0, and two surjections F 0 → F and F 00 → F , there exists an exact sequence 0 → G → E → F → 0
00

Yang Pi‑Yeh 71 Hartshorne Solutions

3 Cohomology 3.6 Ext Groups and Sheaves

with a commutative diagram

0O 0O 0O

/ G0 / E0 / F0 /0
0 O O O

0 /G /E /F /0

  
0 / G 00 / E 00 / F 00 /0

  
0 0 0

with every rows and strips exact.

By Step 1, there exists a locally free sheaf E1 satisfying the diagram

E1 /F

 
E0 / F0

with all arrows surjective. Then there exists a locally free sheaf E2 satisfying the diagram

E2 / E1 /F

 
E 00 / F 00

with all arrows surjective. So this E2 satisfies the following diagram

0O 0O

EO 0 / F0 /0
O
a

E2
c
/F /0

b
 
E 00 / F 00 /0

 
0 0

d e
− G 0 → 0 and E4 →
By 3.6.8. we have two locally free sheaves E3 and E4 such that E3 → − G 00 → 0. So we may

Yang Pi‑Yeh 72 Hartshorne Solutions

3 Cohomology 3.6 Ext Groups and Sheaves

define E = E2 ⊕ E3 ⊕ E4 and G = ker(c) ⊕ E3 ⊕ E4 and

0O 0O 0O

/ G0 g
/ E0 / F0 /0
0 O O O
a|ker(c) ⊕0⊕e a⊕0⊕ge

0 / ker(c) ⊕ E3 ⊕ E4 / E2 ⊕ E3 ⊕ E4 c⊕0⊕0
/F /0
b|ker(c) ⊕d⊕0 b⊕hd⊕0
  
0 / G 00 h
/ E 00 / F 00 /0

  
0 0 0

Hence, if we have two locally free resolutions E 0· → F → 0 and E 00· → F → 0, there exists a locally free
resolution E · → F → 0 with surjections E · → E 0· and E · → E 00· with all things commutative.
STEP 3. We need to show the independence of the choice of the resolution of F for δ.
By Step 2 we only need to consider the case that two choices E · → F → 0 and E 0· → F → 0 have the
P P
surjections E · → E 0· . We denote the kernal of this surjections as G · . Then i (−1)i [E i ] = i (−1)i ([E 0i ] + [G i ]) =
P 0i P P 0i
i (−1) [E ] + i (−1) [G ] = i (−1) [E ]. Hence the definition of δ is well‑defined.
i i i i

STEP 4. We need to show that δ is actually a group morphism, i.e. for any exact sequence 0 → F 00 →
F → F 0 → 0, we have δ([F ]) = δ([F 0 ]) + δ([F 00 ]).
c
Since we can find two locally free sheaf E 00 and E 000 with surjections E 00 → F 0 → 0 and E 000 → − F 00 → 0.
By Step 1, there exists a locally free sheaf E and diagram

E
a
/ E 00

b
 
F / F0

with all arrows surjective. Then we have a diagram

0 / ker(a) ⊕ E 000 / E ⊕ E 000 a⊕0

/ E0 /0
b|ker(a) ⊕c b⊕dc
  
0 / F 00 d
/F / F0 /0

  
0 0 0

So by this progress, we construct three locally free resolution E · → F → 0, E 0· → F 0 → 0 and E 00· → F 00 → 0

Yang Pi‑Yeh 73 Hartshorne Solutions

3 Cohomology 3.6 Ext Groups and Sheaves

with commutative diagram

0 0 0

  
··· / E 001 / E 000 / F 00 /0

  
··· / E1 / E0 /F /0

  
··· / E 01 / E 00 / F0 /0

  
0 0 0

So clearly δ([F ]) = δ([F 0 ]) + δ([F 00 ]), i.e. δ is a group morphism.

STEP 5. δ and ε are clearly inverse to each other. Hence K(X)  K1 (X).

Solution 3.6.10 (Duality for a Finite Flat Morphism). (a) For any affine piece U = Spec A of Y, we may assume
G |U = Ñ, and f −1 (U) = Spec B ⊂ X. Then H omY ( f∗ OX , G )|U = (HomA (B, N))∼ . Since B is a finite A‑module, we
know that HomA (B, N) is a B‑module. Hence H omY ( f∗ OX , G ) is a quasi‑coherent f∗ OX ‑module.
(b) Firstly we define a morphism α : f∗ H omX (F , f ! G ) → H omY ( f∗ F , f∗ f ! G ) as for any φ ∈ Hom f −1 (U) (F | f −1 (U) , f ! G | f −1 (U) )
maps to ψ ∈ HomU ( f∗ F |U , f∗ f ! G |U ) as ψW = φ f −1 (W) for any open subset W ⊂ U. Then by (a) we have an isomor‑
phism f∗ f ! G  H omY ( f∗ OX , G ), which induces a morphism β : H omY ( f∗ F , f∗ f ! G ) → H omY ( f∗ F , H omY ( f∗ OX , G )).
Finally, since f∗ OX is an OY ‑algebra, we have a morphism H omY ( f∗ OX , G ) → H omY (OY , G )  G , hence
we have a γ : H omY ( f∗ F , H omY ( f∗ OX , G )) → H omY ( f∗ F , G ). So we just need to define δ = γ ◦ β ◦
α : f∗ H omX (F , f ! G ) → H omY ( f∗ F , G ). On every affine piece U = Spec A ⊂ Y, we assume f −1 (U) =
Spec B ⊂ Y, F = M̃ and G = Ñ for some A‑algebra B, B‑module M and A‑module N. The δ on U is just
HomB (M, HomA (B, N)) → HomA (M ⊗A B, N) as φ 7→ (m ⊗ 1 7→ φ(m)(1)), which is an isomorphism trivially.
Hence δ is isomorphic.
(c) As we’ve done in (b), we have a natural translation H omX (F , ·) → H omY ( f∗ F , f∗ ·). Composing with
Γ(·), we have HomX (F , ·) → HomY ( f∗ F , f∗ ·). Since ExtiX (F , ·) and ExtiY ( f∗ F , f∗ ·) are both δ‑functors, this
makes a morphism of δ‑functors, ExtiX (F , ·) → ExtiY ( f∗ F , f∗ ·). Hence we have a morphism ExtiX (F , f ! G ) →
ExtiY ( f∗ F , f∗ f ! G ). Moreover, we have a morphism f∗ f ! G → G as in (b), by the functorality of Ext we have a
morphism φi : ExtiX (F , f ! G ) → ExtiY ( f∗ F , f∗ f ! G ) → ExtiY ( f∗ F , G ).
(d) We’ll prove this by induction on the homological dimension of F . If hdF = 0, i.e. F is locally
free, by proposition 6.2. we only need to consider the case that F = OX . Then ExtiX (F , f ! G ) = H i (X, f ! G ).
Since f is affine, 3.4.1. implies H i (X, f ! G )  H i (Y, f∗ f ! G )  H i (Y, ( f∗ OX )∨ ⊗ G ). And by theorem 6.3., we know
H i (Y, ( f∗ OX )∨ ⊗ G )  ExtiY ( f∗ OX , G ). For general case, for any F , there exists a locally free sheaf E and a
coherent sheaf H such that hdH = hdF − 1 satisfying 0 → H → E → F → 0. Since f is affine, we have
0 → f∗ H → f∗ E → f∗ F → 0. Hence for every i > 0, we have

Exti (H , f ! G ) / Exti (E , f ! G ) / Exti (F , f ! G ) / Exti+1 (H , f ! G ) / Exti+1 (E , f ! G )

  φi  
    
Exti ( f∗ H , G ) / Exti ( f∗ E , G ) / Exti ( f∗ F , G ) / Exti+1 ( f∗ H , G ) / Exti+1 ( f∗ E , G )

So by five‑lemma, we know φi : Exti (F , f ! G )  Exti ( f∗ F , G ).

Yang Pi‑Yeh 74 Hartshorne Solutions

3 Cohomology 3.7 The Serre Duality Theorem

3.7 The Serre Duality Theorem

Solution 3.7.1. Since X is projective and L is ample, we know there exists an n such that L n is very ample.
Hence if i : X → PkN is the canonical embedding, we have dimk Γ(X, L n ) = dimk Γ(X, i∗ O(1)) = dim X + 1. Hence
×s
if Γ(X, L −1 ) is not empty, there exists some nonzero s ∈ Γ(X, L −n ). So the morphism Γ(X, L n ) −→ Γ(X, OX ) has
image for more than dimension 1 over k. But dimk Γ(X, OX ) = 1, which is contradict.

Solution 3.7.2. (a) For every coherent sheaf F on Y, clearly we have a morphism HomY (F , ω◦Y )  ExtnY (OY , F )0 .
Then for every coherent sheaf G on X, we know f∗ G is coherent since f is finite. Then by 3.6.10., HomX (G , f ! ω◦Y ) =
HomY ( f∗ G , ω◦Y ) = ExtnY (OY , f∗ G )0 = ExtnX (OX , G )0 . Hence f ! ω◦Y is a dualizing sheaf of X. So by proposition 7.2.,
we have ω◦X  f ! ω◦Y .
(b) By corollary 7.12., we know that ωX  ω◦X and ωY  ω◦Y . Hence ωX  f ! ωY . So we have a natural
morphism t : f∗ ωX  f∗ f ! ωY → ωY .

Solution 3.7.3. By 2.5.16.(c), we have a filtration ∧r (OX (−1)n+1 ) = F 0 ⊃ . . . F r ⊃ F r+1 = 0, which satisfies
F p /F p+1  ΩXp ⊗ ∧r−p OX . Clearly for r , 0, 1, we have ∧r OX  0, and for r = 0, 1, we have ∧r OX  O.
So F 0 = . . . = F r−1 , hence the filtration is actually is just ∧r (OX (−1)n+1 ) ⊃ F r ⊃ F r+1 = 0 such that F r 
F r /F r+1  ΩrX ⊗ ∧0 OX  ΩrX , and ∧r (OX (−1)n+1 )/Ωr = F 0 /F r  Ωr−1 ⊗ ∧1 OX  Ωr−1 X . Moreover, since
 r 
X
⊕N
OX (−1) is a line bundle, we have ∧ (OX (−1) )  OX (−r) for N = n+1 . So we have an exact sequence
r n+1

0 → ΩrX → OX (−r)⊕N → Ωr−1 X → 0.

Since H i (X, OX (−r)) = 0 for all i < n or r < n + 1, we have H i (X, ΩrX )  H i−1 (X, Ωr−1
X ) for 1 ≤ i when r < n + 1,
or 1 ≤ i < n when r ≥ n + 1. Since H 0 (X, Ω0X )  H 0 (X, OX )  k, we have H i (X, ΩiX )  k for all 0 ≤ i ≤ n.
Moreover, for i < n, we have H i (X, ΩnX )  H i (X, OX (−n − 1))  0, so when i < r and 0 ≤ r ≤ n we already have
0
H i (X, ΩrX ) = 0. Then by corollary 7.13., we have H i (X, ΩrX ) = H n−i (X, Ωn−r
X ) = 0 for all i > r and 0 ≤ r ≤ n. So we
have H i (X, ΩrX ) = 0 for all i , r.

Solution 3.7.4 (The Cohomology Class of a Subvariety). (a) In this case, we know that p = n, and H 0 (X, OX ) 
H 0 (P, OP )  k. The morphism H 0 (X, OX ) → H 0 (P, OP ) is just the restriction s 7→ sP , which is just the identity
on k. Hence it corresponds to the element 1 in H n (X, ωX )  k, i.e. tX (η(P)) = 1.
(b) By Bertini’s theorem, there exists a hyperplane H of dimension n − p, such that Z = H ∩ Y completely.
Then we restrict the morphism H n−p (X, Ωn−p X ) → H
n−p
(Y, Ωn−pY ) on the H as H
n−p
(H, Ωn−p
H ) → H
n−p
(Z, Ωn−p
Z ). Since
Z is just deg(Y) points. Then by (a) we know this morphism is just s 7→ (sP1 , . . . , sPdeg(Y) ), i.e. the deg(Y)‑copies
of identities of k. Since the morphism H n−p (Z, Ωn−p Z )  k
deg(Y)
→ k is just a sum, we know the composition
k  H (H, ΩH ) → k is × deg(Y). Hence it corresponds to the element deg(Y) · 1 in H p (H, OHp )  H p (X, OX ),
n−p n−p

where we need the fact that the morphism k  H p (H, OHp )  H p (X, OX )  k is just identity from the proof in
3.7.3.
(c) We just need to prove that d log is a morphism of sheaves of abelian groups. Since on ΩX , we know
d( f g) = f dg + g d f . So d log( f g) = f −1 g−1 d( f g) = f −1 g−1 ( f dg + g d f ) = f −1 d f + g−1 dg = d log( f ) + d log(g).
(d) Denote deg Y = d. Then we may denote the Cartier divisor corresponding to Y as {(Ui , fi )} for some
affine covering Ui and fi ∈ Γ(Ui , OX ) with degree d. Hence the cocycle on H 1 (X, OX∗ ) corresponding to LY is
fi / f j ∈ Ui j = Ui ∩ U j . Then c(LY ) = ( fi−1 d fi − f j−1 d f j , Ui j ). So via the morphism H 1 (X, ΩX ) → H 0 (X, OX ) in 3.7.3.,
P ∂ fi
the element c(LY ) maps to fi xk ∂x k
= d. Hence c(LY ) = d = η(Y), where the second equality is from (b).

3.8 Higher Direct Images of Sheaves

Solution 3.8.1. Fix an injective resolution 0 → F → I · in Mod(X). Then f∗ I i are all injective. Moreover,
since Ri f∗ (F ) = 0, we know 0 → f∗ F → f∗ I · is exact, hence an injective resolution of f∗ F in Mod(Y). Since
Γ(X, I i )  Γ(Y, f∗ I i ), we know H i (X, F )  H i (Y, f∗ F ).

Yang Pi‑Yeh 75 Hartshorne Solutions

3 Cohomology 3.9 Flat Morphisms

Solution 3.8.2. For any affine V in Y, we know f −1 (V) is also affine, hence H i ( f −1 (V), F ) = 0 for all i > 0. Since
Ri f∗ F is the sheaf associated to the presheaf V 7→ H i ( f −1 (V), F ), which is the zero sheaf. So Ri f∗ F = 0 for all
i > 0.

Solution 3.8.3. Fix an injective resolution 0 → F → I · of F . Then since f ∗ E is locally free, by proposition
6.7. we know 0 → F ⊗ f ∗ E → I · ⊗ f ∗ E is an injective resolution of F ⊗ f ∗ E . By 2.5.1.(d), we know
f∗ (I · ⊗ f ∗ E )  f∗ I · ⊗E . Since tensoring a locally free sheaf is exact, we have Ri f∗ (F ⊗ f ∗ E ) = H i ( f∗ (I · ⊗ f ∗ E )) =
H i ( f∗ I · ⊗ E ) = H i ( f∗ I · ) ⊗ E = Ri f∗ F ⊗ E .
S S
Solution 3.8.4. (a) Since E is locally free, we may take an affine covering of X = U j = Spec A j such that
E |U j are free. Then H i (π−1 (U j ), O(l)) = H i (PnAi , O(l)) = 0 in the case 0 < i < n, l ∈ Z or the case i = n, l > −n − 1.
Then by theorem 8.1., we know Ri f∗ O(l) = 0 in these cases.
(b) By theorem 7.11.(b) in chapter II, we have a surjection π∗ E → O(1). So tensoring by O(−1), we have a
surjection π∗ E (−1) → OX , whose kernal we denote as F . So for any affine piece U = Spec A on Y such that
E is free on U, we have 0 → F |PnA → π∗ E (−1)|PnA → O|PnA → 0. Then we know F |PnA  ΩX/Y |PnA . So we have an
exact sequence 0 → ΩX/Y → π∗ E (−1) → OX → 0.
By 2.5.16.(e), we know ωX/Y = ∧n ΩX/Y  (π∗ ∧n+1 E )(−n − 1). Moreover, since we have ωX/Y |PnA  OPnA (−n − 1),
then (Rn π∗ ωX/Y )|U = Rn π∗ (ωX/Y |U ) = H n (PnA , ωPnA /A )∼ = Ã = Spec A. Hence Rn π∗ ωX/Y  OY .
(c) When l > −n − 1, by (a) and the fact that π∗ O(−l − n − 1) = 0, trivial. When l = −n − 1, we may
consider the Koszul complex 0 → π∗ (∧n+1 E )(−n − 1) → . . . → π∗ (∧2 E )(−2) → π∗ (E )(−1) → OX → 0. Denote
Li,q = Rq π∗ (π∗ ∧−i E )(i)  ∧−i E ⊗ Rq π∗ (OX (i)). Then we have a spectral sequence E2p,q = H p (L·,q ). So clearly,
E20,0 = π∗ OX  OY , E −n−1,n = ∧n+1 E ⊗ Rn π∗ (OX (−n − 1)), and E2p,q = 0 in other cases. Hence, dn+1 0,0
: OY → ∧n+1 E ⊗
n+1 ∨
R π∗ (OX (−n − 1)) is an isomorphism, i.e. R π∗ (O(−n − 1))  ∧ E . When l < −n − 1, we may consider the map
n n

π∗ (S −l−n−1 E ) ⊗ O(l)  O(−n − 1). Then by projection formula, S −l−n−1 (E ) ⊗ Rn π∗ O(l)  Rn π∗ O(−n − 1)  ∧n+1 E ∨ .
So Rn π∗ O(l)  S −l−n−1 (E ) ⊗ ∧n+1 E ∨  π∗ (O(−l − n − 1)) ⊗ ∧n+1 E ∨ .
P P2n−1 P
(d) Firstly, pa (Y) = n−1i=0 (−1) h (OY ). So pa (X) =
i n−i i 2n−i
i=0 (−1) h (OX ) = n−1
i=0 (−1) h (OX ) = (−1) pa (X).
i+n n−i n

Secondly, if ωX has global section s, on each projective piece U of X, we know s|U ∈ Γ(U, ωX |U ) = Γ(U, ωU ).
Since Γ(U, ωU ) = 0, this makes a contradiction. Hence Γ(X, ωX ) = 0, i.e. pg (X) = dim Γ(X, ωX ) = 0.
(e) Again. Is this a question?

3.9 Flat Morphisms

Solution 3.9.1. Since the morphism f |U is also of finite type, we just need to prove that f (X) is open. By
2.3.18. we’ve already know that f (X) is a constructible set. For any y ∈ f (X) with preimage x ∈ X, and any
y0 such that y ∈ {y0 }, we know that y0 corresponds to a prime ideal in OY,y . So since the morphism OY,y → OX,x
is flat, by going‑down theorem of flatness version, there exists a prime ideal in OX,x corresponding to y0 , i.e.
there exists a x0 ∈ X such that x ∈ {x0 } and f (x0 ) = y0 . Hence y0 ∈ f (X), i.e. f (X) is open.

Solution 3.9.2. In the w = 1 plane, the curve is (x, y, z) = (t3 , t2 , t). Consider the curve family Xa as (x, y, z) =
(t3 , t2 , at). Then Xa = Spec k[a, x, y, z]/I, where I = (y3 − x2 , z2 − a2 y, z3 − a3 x, zy − ax, zx − ay2 ). Hence if a = 0, we
have I = (y3 − x2 , z2 , zy, zx), i.e. the fibre at a = 0 supports on the cubic curve y2 = x3 in A2 with nilpotent point
z. The embedding point is just (0, 0), and at the prime ideal (x, y), z is a nilpotent.

Solution 3.9.3. (a) Since finite morphism is quasi‑finite, the fibre is just finite points and hence zero‑dimension.
Then by 3.10.9. we know that the morphism is flat since all fibre are in the same dimension.
(b) Take the point P = (x, y) as the intersecting point of X. Then by theorem 9.1A. we know that OX,P is a free
OY, f (P) ‑module. Since OX,P /mY, f (P) OX,P  k, we know OX,P is a free OY, f (P) ‑module of rank 1, i.e. φ : OY, f (P)  OX,P
as an OY, f (P) ‑module. So there exists a decomposition of z as z = f · φ(1) for some f ∈ OY, f (P) , which makes a
contradiction.

Yang Pi‑Yeh 76 Hartshorne Solutions

3 Cohomology 3.9 Flat Morphisms

p
(c) Clearly, (k[x, y, z, w]/(z2 , zw, w2 , xz − yw))red = k[x, y, z, w]/ (z2 , zw, w2 , xz − yw)) = k[x, y, z, w]/(z, w) =
k[x, y], i.e. Xred  Y. Since the Jacobian matrix is
 
 0 0 0 z 
 
0 0 0 −w
J =  
2z w 0 x 
 
0 z 2w −y

When z, w = 0, J has rank 1 or 0, but there does not exists nilpotent in local ring, hence no embedding points at
the locus z, w = 0. When z and w are not simultaneously zero, J has rank 3, so non‑singular, i.e. no embedding
points. Moreover, the morphism f : X → Xred has changing fibre dimension, that is, f −1 (0, 0) has dimension
0, f −1 (x, 0) and f −1 (0, y) has dimension 1, and f −1 (x, y) has dimension 2 with x, y , 0. Hence by 3.10.9., f is not
flat.

Solution 3.9.4 (Open Nature of Flatness). We may assume X = Spec B and Y = Spec A, and denote U as the
subset of U of all x at which f is flat. For any x ∈ U, i.e. Bq is flat over Ap , where q corresponds to x and p
is the image of q, there exists a g ∈ B − q such that for all prime q0 ⊂ B such that q ⊂ q0 and g < q0 , we have
Bq0 is flat over Ap by commutative algebra. Hence U ∩ {x} contains a non‑zero open subset of {x}, hence U
is constructible. Moreover, for any q such that Bq is flat over Ap , for any q0 ⊃ q, Bq0 is flat over Ap0 . Hence U
contains all generalizations of its points. Then by 2.3.18.(c), U is open.

Solution 3.9.5 (Very Flat Families). (a) Consider the flat family {Xt } that Xt = {(1 : 0 : 0), (0 : 1 : 0), (1 : 1 : t)} ⊂
P2 . Notice that only when t = 0 the three points in Xt are colinear. When t , 0, I(Xt ) = (xz−txy, yz−txy, z2 −t2 xy).
And denote the closure of Z((xz−txy, yz−txy, z2 −t2 xy)) ⊂ A4 , we know that I(Y) = (xz−txy, yz−txy, z2 −t2 xy, x2 y−
xy2 ). So Y is a closure of a flat family over a smooth curve, hence Y is flat too, and when t , 0, Yt = C(Xt ).
When t = 0, X0 supports on z = 0, hence C(X0 ) , Y0 . So C(Xt ) is not a flat family.
(b) (Seems Hartshorne’s book has so many exercises which are not exactly exercises?)
(c) Since for all d ≥ 0, the dimk(t) (S t /It )d is a constant, i.e. the Hilbert polynomial Pt are the same. Then by
theorem 9.9., {Xt } is a flat family. For {C(Xt )}, if I = I(X) ⊂ k[x0 , . . . , xn ], then I(C(X)) = I ⊂ k[x0 , . . . , xn , xn+1 ]. So
for any t we have dimk(t) (k(t)[x0 , . . . , xn+1 ]/It )d = dimk(t) (S t /It )d + . . . + dimk(t) (S t /It )0 , i.e. the Hilbert polynomial
of C(Xt ) are all the same. Hence by the same reason we know C(X) → Pn+1 is flat, so C(X) → Pn+1 is flat.
(d) Since Xt are all projectively normal varieties in Pnk , we know that dim(S t /It )d = H(d) for d ≥the Hilbert
regularity, where H(t) is the common Hilbert polynomial since {Xt } is a flat family by theorem 9.11. Consider
the integrally closed ring S t /It , it is generated by the degree 1 elements x0 , . . . , xn , hence the Hilbert regularity
is less than deg Xt − n + 1 ≤ 0. So dim(S t /It )d = H(d) for all d, i.e. {Xt } is a very flat family.

Solution 3.9.6. (⇒) Obviously. (⇐) Treating Y 0 as a closed subscheme of Pn , then I(Y 0 ) = hI(Y), Li, where
L is the linear function who generates H. So if Y 0 is a complete intersection in Pn−1 , in the ring k[x0 , . . . , xn ],
we have length I(Y 0 )/(L) = codimPn−1 (Y 0 ) = codimPn (Y). And since length I(Y 0 )/(L) = length I(Y), we have
length I(Y) = codimPn (Y), i.e. Y is a complete intersection in Pn .

Solution 3.9.7. If X = Spec A is affine, we may assume Y = Z(I) for some ideal I. Since X ⊗k D = Spec A[t]/(t2 ),
the infinitesimal deformation Y 0 corresponds to an ideal I 0 ⊂ A[t]/(t2 ). Since the generic fibre is Y, we know
that the projection of I 0 in A is just I. And since Y 0 is flat over D, by theorem 9.1.(a) we know that the kernel
×t
of the morphism A → (A[t]/(t2 ))/I 0 −→ (A[t]/(t2 ))/I 0 is contained in I 0 . Conversely, if an ideal I 0 ⊂ A[t]/(t2 )
satisfies the above two conditions, it corresponds to an infinitesimal deformation Y 0 .
For any φ ∈ HomA (I/I 2 , A/I), we define I 0 = I ∪ {a + bt|b = φ(a)} ⊂ A[t]/(t2 ). Then by the above two
conditions, I 0 is an ideal in A[t]/(t2 ). Conversely, if we have such an I 0 , since I 0 → I is surjective, for any a ∈ I,
there exists a + bt ∈ I 0 for some b, then we may define b = φ(a). Then by the second condition above, this is

Yang Pi‑Yeh 77 Hartshorne Solutions

3 Cohomology 3.9 Flat Morphisms

well‑defined. Then φ is clearly a morphism from I/I 2 to A/I. Hence we have a corresponding between the
infinitesimal extension of Y and the element of HomA (I/I 2 , A/I).
For any ring homomorphism ψ : A → B with two ideal I ⊂ A, J ⊂ B such that ψ−1 J ⊂ I, we clearly have
the following commutative diagram

HomA (I/I 2 , A/I)


/ the infinitesimal extension of Spec A/I

 
HomB (J/J 2 , B/J)

/ the infinitesimal extension of Spec B/J

So the general case is just glueing.

×t
Solution 3.9.8. Clearly we have an exact sequence 0 → k −→ D → k → 0. Since A0 is flat over D, and
×t
A0 ⊗D k  A, we have an exact sequence 0 → A −→ A0 → A → 0. Moreover, defining P0 = P ⊗k D, we have an
exact sequence 0 → P → P0 → P → 0. Then we can define a morphism f : P0 → A0 with kernel J 0 . And we
have the following exact sequence

0 0 0

  
0 /J / J0 /J /0

  
0 /P / P0 /P /0
f
  
0 /A / A0 /A /0

  
0 0 0

Hence the equivalent classes of A0 is just the choice of the ideal J 0 in P0 quotient to the choice of f .
Fix a J 0 . For any x ∈ J for the right J, we can lift it to x + yt ∈ J 0 for some y ∈ P, where we may lift
it to another x + y0 t ∈ J 0 for some y0 ∈ P but y − y0 ∈ J. So we may define a morphism φ : x → ȳ ∈ A,
where the lifting of x is x + yt ∈ J 0 . And φ ∈ HomP (J, A). Conversely, for any φ ∈ HomP (J, A), we may
define J 0 = {x + yt | x ∈ J, y ∈ P such that ȳ = φ(x) ∈ A} as an ideal in P0 . And K ∩ J 0 = J, hence we have
0 → J → J 0 → J → 0 and J 0 defines an A0 . So {choices of K}  HomP (J, A).
Moreover, by 2.8.6.(a) we know that {choices of f }  HomP (ΩP/k , A)  HomA (ΩP/k ⊗ A, A). And since
HomA (J/J 2 , A)  HomP (J, A), we have T 1 (A)  {choices of A0 }.

Solution 3.9.9. By 3.9.8. we only need to show the surjectivity of the morphism Θ : HomA (ΩP/k ⊗ A, A) →
HomA (J/J 2 , A), where P = k[x, y, z, w], J = (x, y) ∩ (z, w) = (xz, xw, yz, yw). Since J/J 2 has generators xz, xw, yz, yw
as an A‑module, any morphism ψ : J/J 2 → A is determined by the quaternity (a, b, c, d) = (ψ( xz), ¯ ψ( xw),
¯ ψ(yz),
¯ ψ(yw)).
¯
Since ΩP/k has basis dx, dy, dz, dw, we may consider the morphism φ x : ΩP/k ⊗ A → A such that φ x (dx) = 1
and φ x (dy) = φ x (dz) = φ x (dw) = 0. Then the morphism µ : J/J 2 → ΩP/k ⊗ A, where f 7→ d f ⊗ 1, maps to
(z, w, 0, 0) ∈ HomA (J/J 2 , A) via Θ. Similarly, Θ(φy ) = (0, 0, z, w), Θ(φz ) = (x, 0, y, 0) and Θ(φw ) = (0, x, 0, y).
For any (a, b, c, d) ∈ HomA (J/J 2 , A), we have ay = ψ(xyz) = xψ(yz) = cx. Similarly, aw = bz, by = dx, cw = dz.
So there exists constant α, β, γ, δ such that (a, b, c, d) = α(z, w, 0, 0) + β(0, 0, z, w) + γ(x, 0, y, 0) + δ(0, x, 0, y) =
Θ(αφ x + βφy + γφz + δφw ) by noticing that xz = xw = yz = yw = 0 in A. Hence Θ is surjective.

Yang Pi‑Yeh 78 Hartshorne Solutions

3 Cohomology 3.10 Smooth Morphisms

Solution 3.9.10. (a) By Example 9.13.2., any infinitesimal deformation corresponds to an element in H 1 (P1 , TP1 ).
Since TP1 = O(2), and H 1 (P1 , O(2)) = 0, P1 is rigid.
(b) Consider Y = Z((a − 1)x2 + (b − 1)y2 + z2 ) ⊂ A2 ⊗ P2 . The projection π : Y → A2 is smooth on S =
D(a − 1, b − 1) = Spec k[a, b, (a − 1)−1 , (b − 1)−1 ]. Hence we may define X = π−1 (S ) and f = πX . Firstly,
X0 = f −1 (0, 0) = Z(x2 +y2 −z2 )  P1 . Secondly, the generic fibre Xη is the quadratic curve (a−1)x2 +(b−1)y2 +z2 = 0
in P2k(a,b) , which has no rational points on k(a, b). So this flat family is not locally trivial everywhere.
(c) Since f is flat, every fibre has the same Hilbert polynomial, i.e. every fibre is P1 . For an affine U ⊂ T
containing t, we consider an effective divisor associated to ω−1 X/T which is supported on two distinct points in
each fibre. Then D → U is clearly flat. So we may consider T 0 = D and the family f 0 : X 0 → T 0 . Moreover, for
any x ∈ T 0 , clearly x ∈ X, hence there exists a x0 ∈ X ×T T 0 = X 0 , which define a section s : T 0 → X 0 , s(x) = x0 .
Then we will prove that if a flat morphism f 0 : X 0 → T 0 has fibres P1 and a section s : T 0 → X 0 , this family is
trivial.
Denote Y 0 ⊂ X 0 as the schema‑theoretic image of s, then Y 0 is flat over T 0 . And Y 0 intersect every fibre
on just one point, it is a Cartier divisor on X 0 . Denote L as the associated invertible sheaf on X 0 . For any
t0 ∈ T 0 , H 0 (Xt00 , Lt0 ) is a vector space of dimension 2, and H 1 (Xt00 , Lt0 ) = 0. Then consider the morphism
φi (s) : Ri f∗0 L ⊗ k(t0 ) → H i (Xt00 , Lt0 ), by theorem 12.11., H 1 (Xt00 , Lt0 ) = 0, φ1 is surjective, hence isomorphism, so
Ri f∗0 L = 0. Moreover, the zero sheaf is free, hence φ0 is surjective, hence isomorphism. Since f∗0 L is locally
free of rank 2 on T 0 , we may denote it as E . Then the nature morphism f 0∗ E → L induces the morphism
X 0 → P(E ), which is isomorphic on each fibre, hence isomorphic globally.

Solution 3.9.11. By corollary 3.11. in chapter IV, any curve is birational to a plane curve. So we just need to
P
consider the case of plane curve. Than pa (C) = 12 (d − 1)(d − 2) − P δP by 4.1.8. Hence pa (C) ≤ 12 (d − 1)(d − 2).
Moreover, pa (C) = pg (C), hence non‑negative.

3.10 Smooth Morphisms

h i
Solution 3.10.1. The Jacobian matrix of this curve is 2y0 , which is rank 1 everywhere since y , 0. So every
local ring of X is regular local ring.
If X → Spec (k) is smooth, we may consider the smooth base change Xk̄ → k̄, which is defined by y2 =
x − t = (x − t1/p ) p over k̄. Hence Xk̄ is not regular at (t1/p , 0), which makes a contradict. So X is not smooth.
p

Solution 3.10.2. Denote the relative dimension of the morphism Xy → Spec (k(y)) as n. Then ΩXy /k(y) is a
locally free sheaf of rank n on Xy . So, dimk(x) (ΩX/Y ⊗ k(x)) = dimk(x) (ΩXy /k(y) ⊗ k(x)) = n. Since ΩX/Y ⊗ k(x) =
(ΩX/Y ) x /m x (ΩX/Y ) x , by Nakayama’s lemma, there exists s1 , . . . , sn ∈ ΩX/Y (U x ) for some open neighbourhood of
x such that the images of si ’s in ΩX/Y ⊗ k(x) = (ΩX/Y ) x /m x (ΩX/Y ) x form a k(x)‑basis and they generate ΩX/Y (U x ).
So for every x0 ∈ U x , we have dimk(x0 ) (ΩX/Y ⊗ k(x0 )) ≤ n. Conversely, by theorem 8.6 in chapter II we have
S
dimk(x0 ) (ΩX/Y ⊗ k(x0 )) ≥ n, hence equals, i.e. ΩX/Y is a locally free sheaf of rank n on U x . Denoting U = x∈Xy U x ,
we know that ΩX/Y is locally free of rank n on U.
Moreover, since f is proper, the base change Xy → Spec (k(y)) is proper. So Xy is quasi‑compact, hence
S
there exists finitely many x1 , . . . , xm ∈ Xy , such that Xy ⊂ mi=1 U xi . Since f is flat, we have f (U xi ) is open and
Sm
containing y. Define V = i=1 f (U xi ), then this is a open neighbourhood of y, which preimage is smooth over
itself.

Solution 3.10.3. (ii)⇔(iii) is a consequence of theorem 8.6. in chapter II. And (i)⇔(ii) is just the definition.

Solution 3.10.4. (⇒) By 3.10.3., étale implies unramify and flatness, hence k(x)/k(y) is separable, and my O x =
Oy . Since Oy → O x is flat, we have an exact sequence 0 → mr+1 ⊗Oy O x → mr ⊗Oy O x → mr /mr+1 ⊗Oy O x → 0.
Since mr /mr+1 ⊗Oy O x = mr O x /mr+1 O x , we have Gr O x  Gr Oy ⊗Oy O x . Hence Ô x  Ôy ⊗Oy O x . So Ôy ⊗k(y) k(x) =
Ôy ⊗k k ⊗Oy O x = Ô x  Ôy ⊗Oy O x .

Yang Pi‑Yeh 79 Hartshorne Solutions

3 Cohomology 3.11 The Theorem on Formal Functions

(⇐) Unramify is clear. For flatness, if M → N is an injective morphism of Oy ‑modules, we want to show
that M ⊗Oy O x → N ⊗Oy O x is injective. Since Oy → Ôy is flat, Oy → O x ⊗Oy Ôy is also flat. So M ⊗Oy O x ⊗Oy Ôy →
N ⊗Oy O x ⊗Oy Ôy is injective. Since Ôy is faithfully flat, M ⊗Oy O x → N ⊗Oy O x is injective.

Solution 3.10.5. For any x ∈ X, there exists an étale morphism f : U → X such that f ∗ F is free OU ‑module.
So ( f ∗ F ) x0 is a free O x0 ‑module. Denote F x as an O x ‑module M. Then M ⊗Ox O x0 = ( f ∗ F ) x0 = O xn0 = O xn ⊗Ox O x0
for some n. Since O x and O x0 are local, then flatness implies faithful flatness. So M  O xn . Then by 2.5.7., F is
locally free.

Solution 3.10.6. Consider X = Spec k[t, s]/(t2 −(s2 −1)2 ). Then the morphism k[x, y]/(y2 − x2 (x+1)) → k[t, s]/(t2 −
1, t(t2 − 1)), where x 7→ s2 − 1 and y 7→ st, induces the morphism f : X → Y. For unramify, since f ∗ ΩY/k = ΩX/k ,
we have ΩX/Y = 0. For flatness, the fibre at node is X0 = Spec k[x, y]/(x, y) ⊗k[x,y]/(y2 −x2 (x+1)) k[t, s]/(t2 − 1, t(t2 − 1)) =
Spec k[s, t]/(t2 , s2 − 1, st) = Spec k[s, t]/(t, s2 − 1) = Spec k[s]/(s2 − 1). The fibre at any other point is clearly
isomorphic to X0 , hence f is flat.
P
Solution 3.10.7. (a) Suppose f (x, y, z) = i+ j+k=3,i, j,k≥0 ai jk xi y j zk ∈ d. Since (1, 0, 0), (0, 1, 0), (0, 0, 1) ∈ V( f ), we
have a300 = a030 = a003 = 0. Then because (1, 1, 0) ∈ V( f ), we have a210 = a120 , and similarly a021 = a012 and
a102 = a201 . Moreover, (1, 1, 1) ∈ V( f ), so a111 = 0. Hence f has the form f = axy(x + y) + byz(y + z) + czx(z + x).
So d is generated by 3 cubic terms, hence dimension 2. And there exists a morphism P2 − {Pi } → P2 as
(x, y, z) 7→ (xy(x + y), yz(y + z), zx(z + x)).
On affine piece D+ (x), the morphism above is (x, y) 7→ ( x(x+y) y+1
, x(x+1)
y(y+1)
. So in algebraic language, we have a
morphism g : k(s, t) → k(x, y), s 7→ y+1 , t 7→ y(y+1) . Since y · g (t) + g0 (s) = x, we have 0 = g0 (s) + g0 (s) =
0 x(x+y) x(x+1) 0
0 2 0 0 2 0 0 0 0
= g (s)(y+1)+y g (t) +yg (ts)+y g (t)+yg (ts)+g (s) +yg (s)
2 2
g0 (s) + x(x+y)
y+1 y+1
, i.e. y2 g0 (t)(g0 (t) + 1) + g0 (s)(g0 (s) + 1) = 0. So we have a
0 0
minimal polynomial y = − g0 (t)(g0 (t)+1) , hence inseparable of degree 2 on this piece. Similarly we can do the
2 g (s)(g (s)+1)

same thing on the piece D+ (y) and D+ (z).

(b) For f = axy(x + y) + byz(y + z) + czx(z + x) ∈ d, we have 0 = ∂∂xf = ay2 + bz2 , 0 = ∂∂yf = ax2 + cz2 and
√ √ √
0 = ∂∂zf = bx2 + cy2 . So f corresponds to a singular point ( c, b, a), this is a 1‑1 correspondence between d
and P2 .

Solution 3.10.8 (A Linear System with Moving Singularities Contained in the Base Locus (Any Characteris‑
tic)). Simple calculate, the cone of C in A3 is (x − 1 + zt )2 + y2 = (1 − zt )2 . Yt is (tx − tw + z)2 + (ty)2 − (tw − z)2 = 0
in P3 . Hence {Yt } forms a linear system of dimension 1 with a moving singularity at P. And the base locus of
{Yt } is {(x − w)2 + y2 = w2 } ∪ {x = y = 0}, i.e. the conic C plus the z‑axis.

Solution 3.10.9. This problem is local, so we may assume X = Spec B and Y = Spec A for two regular
ring A and B with a ring homomorphism A → B. For any y ∈ Y, it may correspond to a prime p ⊂ A, the
fibre Xy = Spec B ⊗ k(y) = Spec Bp /pBp . So if X is Cohen‑Macaulay, Bp is also CM. And we have dim Bp =
dim Ap + dim Bp /pBp . So depth(pBp , Bp ) = dim Ap . Since Ap is regular, we may take a regular sequence x1 , . . . , xd
of p, where d = dim Ap . Take the Koszul complex K = K(x1 , . . . , xd ). Then the Koszul complex of the generators
A
of pBp is K ⊗ Bp , which has no homology except for H d . Hence Tor1 p (Ap /p, Bp ) = H d−1 (K ⊗ Bp ) = 0, i.e. Bp is
flat over Ap . So B is flat over A.

3.11 The Theorem on Formal Functions

Solution 3.11.1. The sheaf Rn−1 f∗ OU is the one associated to the presheaf V 7→ H n−1 ( f −1 (V), OU | f −1 (V) ). Hence
we may take the Cech complex of U for the open covering Ui = Spec k[x1 , . . . , xn , xi−1 ]:

M
n
... → k[x1 , . . . , xn , x1−1 , . . . , xˆi−1 , . . . , xn−1 ] → k[x1 , . . . , xn , x1−1 , . . . , xn−1 ] → 0 → . . .
i=1

Yang Pi‑Yeh 80 Hartshorne Solutions

3 Cohomology 3.11 The Theorem on Formal Functions

So the n − 1‑cohomology group is the linear combinations of monomials of negative degree, which is non‑
empty. Hence Rn−1 f∗ OU , 0.
f0 g
Solution 3.11.2. Consider the Stein factorization of f : X −→ Y 0 → − Y. For any y0 ∈ Y, f −1 (g(y0 )) is discrete and
hence f (y ) ⊂ f (g(y )) is discrete. But f (y ) is non‑empty and connected. So f 0−1 (y0 ) is just one point, i.e.
0−1 0 −1 0 0−1 0

f 0 is 1‑1 and onto. Moreover, since f 0 is projective, i.e. proper, there exists a homeomorphism between the
underlying spaces. Combining with f 0] : OY 0 → f∗0 OX is an isomorphism, we know that f 0 is an isomorphism
of schemes. So f = g is finite.
f0 g
Solution 3.11.3. Consider the Stein factorization of f as X −→ X 0 → − Pnk . Then since the fibres of f 0 are all
closed and connected, we only need to prove that for all D ∈ d and the associated closed subscheme Z, f 0 (Z)
is connected. By definition of f , any D ∈ d is of the form D = f ∗ H for some Cartier divisor H in the complete
linear system of the invertible sheaf O(1) on Pn , where H is clearly ample. Since f ∗ H = f 0∗ g∗ H, and g is finite,
so g∗ H is ample. Since X 0 is normal and of dimension ≥ 2, so by 3.7.9., g∗ H is connected, i.e. f 0 (Z) is connected.

Solution 3.11.4 (Principle of Connectedness). The problem is stable under base change, so we may assume
that T is smooth. Since f is flat and projective, we have f∗ OX is torsion free. Since T smooth, we know f∗ OX
is locally free. Then since f has connected fibre over U, f∗ OX |U has rank 1, hence f∗ OX has rank 1. So f∗ OX is
the sheaf of sections of a line bundle on T . Moreover, since f∗ OX (T ) = OX (X), f∗ OX has a nowhere vanishing
section corresponding to the section 1 on OX , so f∗ OX is the trivial line bundle, i.e. f∗ OX  OT . Then by
corollary 11.3., the fibre of f is connected for every closed point of T .

Solution 3.11.5. By 2.9.6., we have Pic X = lim Pic Xn . Then we only need to show that lim Pic Xn  Pic Y.
←−− ←−−
Denote I as the ideal sheaf of Y, hence generated by some polynomial f ∈ k[x0 , . . . , xN ]. Then by 3.4.6.,
considering Xn with the sheaf I n−1 , we have an exact sequence . . . → H 1 (Xn , I n−1 ) → Pic Xn → Pic Y →
H 2 (Xn , I n−1 ) → . . .. Since Xn has the same underlying topological space with Xn−1 and Y, which is of dimension
≥ 3, and H i (Xn , I n−1 )  H i (Xn−1 , I n−1 ). Since Xn−1 = Proj k[x0 , . . . , xN ]/( f n−1 ), I n−1 = OXn−1 , by 3.5.5. we have
H i (Xn−1 , O n−1 ) = 0 for i = 1, 2. So we have Pic Xn  Pic Y for all n. Then lim Pic Xn  Pic Y.
←−−
Solution 3.11.6. (a) We have an exact sequence . . . → H N−1 (X − Y, Fˇ ) → HYN (X, Fˇ ) → H N (X, Fˇ ) → H N (X −
Y, Fˇ ) →. Since Y is closed in X, we have H N−1 (X − Y, Fˇ ) = H N (X − Y, Fˇ ) = 0, i.e. we have HYN (X, Fˇ )  H N (X, Fˇ ).
Since Fˇ is locally free, by Serre duality we have H 0 (X, F )  H N (X, Fˇ )0 . Then we only need to prove the formal
duality H 0 (X̂, Fˆ ) = HYN (X, Fˇ )0 .
We’ve already had H 0 (X̂, Fˆ ) = lim H 0 (Xm , Fm ). But H 0 (Xm , Fm ) = H 0 (X, Fm )  ExtN (Fm , ωX/k )0  ExtN (OX /IYm , Fˆ )0 ,
←−−
we have H 0 (X̂, Fˆ ) = lim ExtN (OX /IYm , Fˆ )0 = (lim ExtN (OX /IYm , Fˆ ))0 = HYN (X, Fˇ )0 .
←−− −−→
(b) (i⇒ii) For any n, we have an exact sequence 0 → I k F(n)/I k+1 F(n) → F(n)/I k+1 F(n) → F(n)/I k F(n) →
0, hence we have H 0 (Xk+1 , Fk+1 (n)) → H 0 (Xk , Fk (n)) → H 1 (X1 , (I k F/I k+1 F)(n)). Since X1 = Y is projective over
k, there exists an n0 such that for any n ≥ n0 , we have H 1 (Y, (I k F/I k+1 F)(n)) = 0. So when n ≥ n0 , the mor‑
phism H 0 (Xk+1 , Fk+1 (n)) → H 0 (Xk , Fk (n)) is surjective. Hence if we take a projective limit, we have a surjective
morphism H 0 (X̂, F(n)) → H 0 (X1 , F1 (n)). Adding n0 if necessary, we assume that F1 (n) is generated by global
section, then we may pick s1 , . . . , sm ∈ H 0 (X̂, F(n)) such that their images in H 0 (X1 , F1 (n)) can generate the
sheaf. Then for any y ∈ Y, we have a morphism OX̂,y m
→ F(n)y as ei 7→ si,y . Since the morphism OY,y m
→ F1 (n)y
induced by this morphism is clearly surjective, then by Nakayama lemma we know this morphism is surjec‑
tive. Hence F(n) is generated by global sections s1 , . . . , sm .
L
(ii⇒i) Since n is generated by global section, there exist finitely some qi and a surjective morphism OX̂ (qi ) →
F(n). And clearly the kernel of this morphism K also satisfies the Serre theorem, i.e. there exists m such that
K(m) is generated by finitely many global sections. Then there also exist finitely some pi and an exact se‑
L L
quence OX̂ (pi ) → OX̂ (qi + m) → F(n + m) → 0. Tensoring with OX̂ (−n − m) we have an exact sequence

Yang Pi‑Yeh 81 Hartshorne Solutions

3 Cohomology 3.12 The Semicontinuity Theorem

L
Ê1 → Ê0 → F → 0, where Ê1 and Ê0 have the form OX (qi ). Then by (a) we have H 0 (X, H omX (E1 , E0 ) 
H (X̂, H omX̂ (E1 , E0 ), hence F correspond to some F , i.e. F = Fˆ .
0 ˆ ˆ
(c) By 3.11.5. we have Pic Y  Pic X̂. And by 2.6.2.(d) we have an injective morphism Pic X → Pic X̂. By
(b), any F ∈ Pic X̂ is algebraizable, i.e. F = Fˆ for some F ∈ Pic X, hence we have a morphism Pic X̂ → Pic X.
Then we have isomorphism Pic X  Pic X̂.

Solution 3.11.7. (a) Similarly with 3.11.5., we have Pic X̂ = lim Pic Xn . For any Xn , we have H 1 (Xn , I n−1 ) →
←−−
Pic Xn → Pic X1 → H 2 (Xn , I n−1 ). Since Xn has dimension 1, we have H 2 (Xn , I n−1 ) = 0. Hence in limit, we have
Pic X̂ = lim Pic Xn → Pic Y is surjective. Moreover, since H 1 (Xn , I n−1 ) = H 1 (Xn−1 , I n−1 ) = k[x0 , x1 , x2 ]/( f n−1 ).
←−−
Take projective limit, we have ker(lim Pic Xn → Pic Y) = lim k[x0 , x1 , x2 ]/( f n−1 ) =the completion of k[x0 , x1 , x2 ]
←−− −−→
of ( f ), hence infinite dimensional.
(b) Denote A = k[x0 , x1 , x2 ] and  the completion of A of ( f ). Then take 1 ∈ A, it maps to 1 in H 1 (Xn , I n−1 ),
via the morphism H 1 (Xn , I n−1 ) → H 1 (Xn , OX∗n ) = Pic Xn , 1 maps to an invertible sheaf Ln . Then take a projec‑
tive limit of {Ln } and get a sheaf L on X̂. Since all Ln have inverse Mn , we have L⊗M = OX̂ , where M = lim Mn .
←−−
Hence L is invertible.
(c) Just take F = L as in (b). If F(n) can be generated by global section for all n ≥ n0 for some n0 , by 3.11.6.
we have F is algebraizable, which makes a contradiction.

Solution 3.11.8. Since H i (Xy , F ⊗ Oy /my ) = 0, and F is flat, we have 0 → Fy ⊗ mky /mk+1y → Fy ⊗ Oy /mk+1
y →
Fy ⊗ Oy /mky → 0. So by induction, we have H i (Xy , F ⊗ Oy /mky ) = 0 since mky /mk+1
y is a direct sum of copies of
Oy /my . So by formal function theorem we have (R f∗ (F ))yˆ = 0, hence there exists a neighbour of y such that
i

Ri f∗ (F ) is 0 on it.

3.12 The Semicontinuity Theorem

Solution 3.12.1. We may assume that Y is affine. Since Y is of finite type over k, we may assume Y =
Spec k[x1 , . . . , xn ]/I, where I is an ideal of k[x1 , . . . , xn ] generated by f1 , . . . , fm . Then clearly φ(y) = dimk (my /m2y ) =
∂ fi ∂ fi
n + 1 − rank ( ∂x j
(y)). Then if φ(y) = n + 1 − t, we may assume det( ∂x j
(y))1≤i, j≤t , 0. Then the open subset
∂ fi
S t = {y0 | det( ∂x j
(y0 ))1≤i, j≤t , 0} is a neighbourhood of y and for any y0 ∈ S t , we have φ(y0 ) ≤ n + 1 − t, hence φ is
upper semicontinuous.

Solution 3.12.2. For any hypersurface Y of degree d, we may assume Y = Proj k[x0 , . . . , xn ]/( f ) for some
  n+r−d
polynomial f of degree d. Then hr (Y, OY ) = dimk S (Y)r = dimk (k[x0 , . . . , xn ])r −dimk (k[x0 , . . . , xn ])r−d = n+r
r
− r−d
is a constant.

Solution 3.12.3. According to (https://mathoverflow.net/questions/90260/trouble‑with‑semicontinuity) the

book has a typo here.
Clearly Xa is parametrized with (t4 , t3 u, at2 u2 , tu3 , u4 ), then I ⊂ OP4 ⊗A1 is generated by (a2 x0 x4 − x22 , a2 x1 x3 −
x22 , x0 x2 − ax12 , x2 x4 − ax32 , ax0 x3 − x1 x2 , ax1 x4 − x2 x3 ). For all t , 0, we easily know that h0 (Xt , OXt ) = 1 and
h1 (Xt , OXt ) = h1 (P 4 , It ) = h2 (P 4 , It ) = 0. When t = 0, we have I0 is generated by (x0 x2 , x1 x2 , x22 , x2 x3 , x2 x4 ).
Hence H 0 (X0 , OX0 ) = k[x2 ]/(x22 ), i.e. h0 (X0 , OX0 ) = 2. Since H 1 (P4 , OP4 ) = 0, we have h1 (P 4 , I0 ) = 1. Moreover,
by flatness, the arithmetic genus is constant, hence h1 (X0 , OX0 ) = h0 (X0 , OX0 )−χ(OX0 ) = 2−1 = 1. And similarly
h2 (P4 , I0 ) = 1.

Solution 3.12.4. Consider F = L ⊗ M −1 . We have Fy = OXy for each y ∈ Y. Then all we need to do is define
N = f∗ F , and prove that f ∗ N = F . On every fibre Xy , we have Fy = OXy , hence h0 (Xy , Fy ) = h0 (Xy , OXy ) = 1,
a constant. Then by corollary 12.9., the sheaf N = f∗ F is locally free of rank 1. Since we’ve already had a
morphism f ∗ N = f ∗ f∗ F → F , it’s a surjective morphism between two invertible sheaves, hence isomorphic.

Yang Pi‑Yeh 82 Hartshorne Solutions

3 Cohomology 3.12 The Semicontinuity Theorem

Solution 3.12.5. Define a morphism Φ : Pic Y × Z → Pic X as (F , m) 7→ π∗ F ⊗ OX (m). First we will prove Φ
is injective. By proposition 7.11. in chapter II, we have π∗ OX  OY . Then if Φ(F , m) = OX for some F and
m, we have OY  π∗ (π∗ F ⊗ OX (m))  (π∗ OX (m)) ⊗ F . Since F is invertible, we have π∗ OX (m)  F −1 . So by
 
proposition 7.11. in chapter II, we have m > 0. But π∗ OX (m)  S m (E ), which has the rank m+r−1
r−1
> 1, where r
is the rank of E . Hence it makes a contradiction.
For surjection, for any M ∈ Pic X, we restrict it to Xy for any y ∈ Y and know that My is an invertible sheaf
on Pn , hence My  OPrk (my ) for some my . Since X is flat over Y, all My have the same Hilbert polynomial, i.e.
my are all the same. We may define m = my for some y. Then we define F = π∗ (M ⊗ OX (−m)). Since for any y,
we have (M ⊗ OX (−m))y  OXy . So by 3.12.4., we have M ⊗ OX (−m))y  OXy  π∗ F for some invertible sheaf
F on Y, i.e. M has a preimage (F , m).

Solution 3.12.6. (a) We denote the projections as p : X × T → X and q : X × Y → Y. Then q is projective.

Define M = (p∗ Lt ) ⊗ L −1 . So Mt  OX×{t}  O. Since H 1 (X, OX ) = 0, we have q∗ M ⊗ k(t) → H 1 ((X × T )t , Mt ) =
H 1 (X, OX ) = 0 is clearly surjective. Then by theorem 12.11., q∗ M is locally free of rank 1 on a neighbourhood
U of t. Shrinking U and we may assume that q∗ M is free of rank 1 on U, i.e. q∗ M |U  OU . So we can pick
some s ∈ M (X × U) corresponding to 1 ∈ OU (U). Shrinking U again as we may assume that q∗ s does not
vanish on any point of U. Hence for any t0 ∈ U, we have M |X×{t0 }  OX×{t0 }  OX . Hence every Lt for t ∈ U is
isomorphic. Since T is connected and noetherian, any two Lt and Lt0 are isomorphic.
(b) We define Φ : Pic X × Pic T → Pic X × T as (F , G ) 7→ F  G , and Ψ : Pic X × T → Pic X × Pic T
as L 7→ (Lt , q∗ L ) for some closed point t. By (a) we know that the Lt is well‑defined, and q∗ L is locally
free of rank 1. So Φ and Ψ are both well‑defined, and clearly they are invertible with each other. Hence
Pic X × Pic T  Pic X × T .

Yang Pi‑Yeh 83 Hartshorne Solutions

4 Curves

4 Curves
4.1 Riemann‑Roch Theorem
Solution 4.1.1. Take another point Q , P, and define D = 2P − Q. Then deg D = 1. We can pick an n such
that deg(nD) = n > max{2g − 2, g, 1}. By Riemann‑Roch we have l(nD) = n + 1 − g. So nD − ( f ) ∼ D0 for some
f ∈ K(X) and effective divisor D0 . So we have ( f ) ∼ D0 − 2nP + nQ, so f has a pole at P and regular everywhere
else since D0 is effective and of degree n.

Solution 4.1.2. Take another point Q , P1 , . . . , Pr and define D = 2P1 + . . . + 2Pr − (2r − 1)Q. Then deg D = 1.
We can pick an n such that deg(nD) = n > max{2g − 2, g, 1} like in 4.1.1. By Riemann‑Roch we have l(nD) =
n+1−g ≥ 1. So nD−( f ) ∼ D0 for some f ∈ K(X) and effective divisor D0 . So ( f ) ∼ D0 −2n(P1 +. . .+Pr )+n(2r−1)Q.
So f has poles at each P since D0 is effective and of degree n.

Solution 4.1.3. By remark 4.10.2.(e) in chapter II, we may embed X in a complete variety X̄ as an open subset.
Then X̄\X = {P1 , . . . , Pr }. Then by 4.1.2., there exists some f ∈ K(X̄) such that f has poles at all Pi and no poles
everywhere else. Since f gives a finite morphism from X̄ to P1 , hence affine, we have X = f −1 (A1 ) is affine.

Solution 4.1.4. By 3.3.1. and 3.3.2., we may assume X is integral. Then since X is not proper, the normalization
X̃ is not proper either. So by 4.1.3. we know that X̃ is affine. Then by 3.4.2., since X̃ → X is finite, we know X
is affine.

Solution 4.1.5. By Riemann‑Roch, we have dim |D| = l(D) − 1 = deg D − g + l(K − D). Since D is effective, we
have l(K − D) ≤ l(K). Moreover, we have l(K) − l(K − K) = 2g − 2 + 1 − g = g. So dim |D| ≤ deg D − g + g = deg D.
If the equality holds, we have l(K − D) = l(K), i.e. D = 0 or g = 0.

Solution 4.1.6. For any P ∈ X, we may define D = gP. Then deg D = g, and l(D) ≥ deg D + 1 − g = 1. Then
there exists some f ∈ K(X) which has a pole at P with order g and regular everywhere else. So f : X → P1
with D 7→ ∞, and deg f = deg D = g.

Solution 4.1.7. (a) If g = 2, we have deg K = 2g − 2 = 2 and dim |K| = l(K) − 1 = g − 1 = 1. For any P ∈ X, we
have dim |P| − dim |K − P| = 1 + 1 − g = 0. So if dim |P| = 1, there exists a rational morphism f : X → P1 , which
is contradict with the fact that g > 1. Then dim |P| = 0, hence dim |K − P| = 0. Hence dim |K − P| = dim |K| − 1,
i.e. P is not a base point. Moreover, since deg K = 2, we have a morphism of degree 2 from X to P1 , i.e. X is
hyperelliptic.
(b) If X is a curve on quadric surface corresponding to the divisor of degree (g + 1, 2), we consider the
morphism f : X → P1 , where f = π2 |X is the second projection Q = P1 × P1 → P1 . Then by theorem 6.8. in
chapter II, this morphism is finite and non‑constant. So by theorem 6.9. in chapter II, deg f ∗ (P) = deg f · deg P,
i.e. deg f = 2.

Solution 4.1.8. (a) Since X and X̃ are both projective, we have H 0 (X, OX ) = k and H 0 (X, f∗ OX̃ ) = H 0 (X̃, OX̃ ) = k.
P P
And since ÕP /OP is a direct sum of flasque sheaf, we have H 1 (X, ÕP /OP ) = 0. So we have an exact
P P
sequence 0 → H 0 (X, ÕP /OP ) → H 1 (X, OX ) → H 1 (X̃, OX̃ ) → 0. So by 3.5.3., pa (X) = pa (X̃) + P∈X dimk ÕP /OP =
P
pa (X̃) + δP .
(b) Since pa and δP are all non‑negative, if pa (X) = 0, we have δP = 0 for all P. So every local rings are
normal, then the curve X is nonsingular. Since pa (X) = 0, by example 1.3.5., X is just isomorphic to P1 .
(c) For arbitrary curve X, we may embed it into P3 . Consider the projection π : P3 → P2 , we denote
π : X̃ → π(X). So δP = length ÕP /OP = length π0 (Õ)π(P) /OπP = δπ(P) . So we may assume that X is a plane curve.
0

So we may assume we have an affine neighbourhood of the node or the original cusp P as the zero of φ(x, y)
in A2 , where φ has the form φ = g2 (x, y)+higher terms, and g2 (x, y) the the degree 2 term. So we may formally

Yang Pi‑Yeh 84 Hartshorne Solutions

4 Curves 4.2 Hurwitz’s Theorem

decompose φ as two terms ψ1 and ψ2 in degree 1. Then ÕP = OP (ψ1 ) = OP (ψ2 ). So there does not exists other
k‑module A such that OP ⊂ A ⊂ ÕP , i.e. length ÕP /OP = 1.

Solution 4.1.9 (Riemann‑Roch for Singular Curves). (a) Denote f : Xreg → X as the canonical morphism.
P
Since we have an exact sequence 0 → L (D) → f∗ Lreg (D) → P∈X ( f ∗ L (D))P /L (D)P → 0. And since L (D) is
locally free, in the local ring we have L (D)P = OX,P and ( f∗ Lreg (D))P = ÕX,P . So H 0 (X, ( f∗ Lreg (D))P /L (D)P ) =
P P
δP as in 4.1.8. Hence χ(L (D)) = χ( f∗ Lreg (D)) − P∈X δP = deg D + 1 − pa (Xreg ) − P∈X δP = deg D + 1 − pa (X).
(b) For any Cartier divisor D and very ample divisor L, there exists some n > 0 such that D+nL is generated
by global section. Then by 2.7.5.(d), M = D + nL + L = D + (n + 1)L is also very ample. Then D = M − (m + 1)L
is the difference of two very ample divisors.
(c) By (b) we only need to consider the case that L is very ample. Then there exists some closed embed‑
ding f : X → Pn such that L = f ∗ O(1). By Bertini’s theorem, we may choose a hyperplane H ⊂ Pn such that
H ∩ X ⊂ Xreg does not contain any singular point of X, hence we may define D = H ∩ X, and L = L (D).
(d) By proposition 8.23. in chapter II, we know that X is CM. Hence by theorem 7.6. in chapter III, we
have H 1 (X, L (D))  Ext0X (L (D), ω0X )0 = Ext1X (OX , L (−D) ⊗ ωX )0  H 0 (X, ωX ⊗ L (−D)). So l(D) − l(K − D) =
χ(LD ) = deg D + 1 − pa (X).

Solution 4.1.10. By 4.1.9.(d), we have deg K = pa − 1 = 0. Then if D is a divisor of degree 0, we have D + P0

corresponding to an invertible sheaf L (D0 ) for some divisor D0 supported on Xreg . And on Xreg , the divisor
D0 − P0 has degree 0 corresponding to some P − P0 on Xreg by example 1.3.7. Hence D ∼ P − P0 for some
P ∈ Xreg . Conversely, for any P ∈ Xreg , there exists some invertible sheaf L corresponding to the divisor
P − P0 clearly. Hence we have a bijection Xreg → Pic 0X .

4.2 Hurwitz’s Theorem

Solution 4.2.1. We use induction on n. If we’ve proved that for any i < n, Pi is simply connected, if f : X → Pn
is an étale covering, for any hyperplane H ⊂ Pn , f ∗ H is ample. By corollary 7.9. in chapter III, f ∗ H is connected.
Then since H is simply connected, f : f ∗ H → H is an étale covering, hence trivial. Then f |H is an isomorphism,
i.e. deg f = 1. So f is an isomorphism globally.

Solution 4.2.2 (Classification of Curves of Genus 2). (a) Since gP1 = 0, by Hurwitz’s formula, we have 2gX −2 =
2 × (0 − 2) + deg R. So deg R = 6. For any ramified point P, since eP ≤ deg f = 2 and p > 2, we have eP = 2. So R
is the sum of 6 points and each ramified point has ramification index 2.
(b) Denote the integral ring of K as OK . For any prime (x − α) ∈ k[x], it will be decomposed into two
primes in OK if and only if α = αi for some i = 1, . . . , 6. And the infinity valuation of k[x] does not split in OK .
So the morphism f : X → P1 has 6 ramification points α1 , . . . , α6 , and the ramification index of each point is
P
2. Hence the ramification divisor R = αi , deg R = 2. So by Hurwitz’s formula, we have gX = 2. Moreover,
suppose f is given by a divisor D of degree 2. We have |D| − |KX − D| = deg D + 1 − gX = 1. So |KX − D| = 0.
And since |KX − D| − |D| = deg(KX − D) + 1 − gX , we have deg(KX − D) = 0, i.e. KX − D is trivial because Pic 0 P1
is trivial. So KX = D, f is given by just KX .
(c) If one of three points is the infinity, we may assume P1 = x1 and P2 = x2 in A1 , and P3 = ∞, then the
linear transformation f (x) = xx−x 1
2 −x1
: A1 → A1 maps P1 , P2 , P3 to 0, 1, ∞. If no one of three points is the infinity,
we may assume P1 = x1 , P2 = x2 , P3 = x3 ∈ A1 . Then the fractional transformation f (x) = x−x x−x1
3
· xx22 −x3
−x1
maps
P1 , P2 , P3 to 0, 1, ∞.
(d) Every chapter has some exercises which are not exactly questions.
(e) By (a), (b) and (d), trivial.

Solution 4.2.3 (Plane Curves). (a) (⇐) (1) If P ∈ L, we may assume P = (0, 0) ∈ A2 , and L = (y = 0), T P = (x = 0).
If X ∩ A2 is generated by a polynomial f , then for every Q = (a, b) ∈ X, T Q = ( f x0 (Q)(x − a) + fy0 (Q)(y − b) = 0),

Yang Pi‑Yeh 85 Hartshorne Solutions

4 Curves 4.2 Hurwitz’s Theorem

f 0 (Q) f0
which intersects L at (a + fy0 (Q) b, 0). So if t is a local parameter at 0 ∈ A1 , we have φ∗ (t) = fy0 y + x. On the y‑axis,
x x
we have fy0 (0) = 0 and φ(0) = 0. So x vanishes at 0 of order ≥ 2, i.e. φ is ramified at 0.
(⇐) (2) If P ∈ L, all things are the same with (1). So we may assume P = (0, 0) ∈ A2 , T P = (x = 0), and L is
f 0 (Q)
the line at infinity. Then similarly, the φ : X → A1 is just Q 7→ − fy0 (Q) . Since P is an inflection point, we may
x
assume X is generated by f in A2 . Then if f restricted to x = 0 has degree ≥ 3 in y, we have fy0 ∈ m2P . Hence
for any local parameter t at P, φ∗ (t) ∈ m2P , i.e. φ is ramified at P.
(⇒) If P ∈ L, we’ve done. If else, we similarly may assume P = (0, 0) ∈ A2 , T P = (x = 0), and L is the line
f 0 (Q)
at infinity. And φ : X → A1 is Q 7→ − fy0 (Q) . For any local parameter t at P, φ∗ (t) ∈ m2P implies fy0 ∈ m2P locally.
x
Hence P is an inflection point of X.
Finally, by Hurwitz’s formula, we know that ramified points are finite, hence inflection points are also
finite.
(b) We may assume L = (z = 0) on P2 is a multiple tangent as we want. If for some P such that L is
the tangent at P, we may assume P = [a : b : 0]. Then t = ax+by bx−ay
is a local coordinate on X near p. Since
L is not an inflection line near P, near P, x = t2 f at OX,P for some invertible f ∈ OX,P . Since ÔX,P  k[[t]],
z

we may assume f (t) = f0 + f1 t + f2 t2 + . . .. So near P, there exists an open neighbourhood such that t is a
coordinate and f is invertible. So the tangent line at any t0 is (z − t02 f (t0 )x) − (2t0 f (t0 ) + t02 f 0 (t0 ))(y − t0 x) = 0, i.e.
t02 ( f (t0 ) − t0 f 0 (t0 ))x + t0 (−2 f (t0 ) − t0 f 0 (t0 ))y + z = 0. So the tangent of L in the dual curve corresponding to P is
[0 : −2 f1 : 0]. Hence L is an ordinary r‑fold point on X ∗ .
(c) We may assume that O = (0, 0, 1), P = (0, 1, 1) and L is a line at infinity which does not contain O. Then
ψ : (x, y) 7→ [x : y]. So on D(y), we may assume ψ : U → D(y) as (x, y) 7→ yx for some neighbourhood U of P. So
ψ(P) = 0. If ψ is ramified at P, we have ψ∗ (t) = yx ∈ m2P , where t is a local parameter at 0. Since y , 0, we have
x ∈ m2P , i.e. the line x = 0 is tangent to X at P. By Hurwitz’s formula, we have (d − 1)(d − 2) − 2 = −2d + deg R,
i.e. deg R = d(d − 1). Since 0 is not an inflection point or on the tangent line, R is reduced. So the number of
tangent lines is deg R = d(d − 1). And the ramification index in each P is just 2.
(d) For any point O ∈ X not on any inflections or multiple tangents, we define ψ : X → P1 as the projection
from P. So deg ψ = d − 1. By Hurwitz’s theorem (d − 1)(d − 2) − 2 = −2d + 2 + deg(R), i.e. deg R = (d + 1)(d − 2).
So the number of tangent lines is deg R = (d + 1)(d − 2).
(e) Since φ−1 (P) = {Q ∈ X | P ∈ T Q (X)}, if P is not an inflection or on a multiple tangent, by (c) we have
−1
]φ (P) = d(d − 1), i.e. deg φ = d(d − 1). So by Hurwitz’s formula, we have deg R = 3d2 − 5d. So by ignoring the
ramification of type (1) in part (a), we know that the number of inflection points is 3d2 − 6d. And the last is
obvious.
(f) Since the map φ : X → X ∗ is finite and birational, and X is normal, we have X is the normalization of X ∗ .
So pa (X ∗ ) = 21 (d(d − 1) − 1)(d(d − 1) − 2). Since pa (X ∗ ) = pa (X) + ]inflections+]bitangents, pa (X) = 12 (d − 1)(d − 2),
and ]inflections= 3d2 − 6d by (e), we have ]bitangents= 12 d(d − 2)(d − 3)(d + 3).
(g) By (e), ]inflections= 3 × 3 × (3 − 2) = 9. And since degree is just 3, they are all ordinary. Choose a
coordinate system such that P = (0, 0, 1) and Q = (0, 1, 0) are inflection points and the tangent lines are (y = 0)
and (z = 0). So the cubic curve is just yz(ax + by + cz) + dx3 = 0 for some a, b, c, d. So the line passing through
P, Q intersect X at (0, −c, b), which is an inflection point.
(h) By (f), ]bitangents= 21 × 4 × 2 × 1 × 7 = 28.

Solution 4.2.4 (A Funny Curve in Characteristic p). Since f x = z3 , fy = x3 , fz = y3 , and the common zero
(0, 0, 0) < P2 , X is smooth. Moreover, since the Hessian matrix is just 0 at all points, every point is an inflection
point. For any P = (a, b, c), the tangent line at P is c3 (x − a) + a3 (y − b) + b3 (z − c) = 0, i.e. c3 x + a3 y + b3 z = 0. So
the map X → X ∗ is just the Frobenious morphism, hence isomorphic and purely inseparable.

Solution 4.2.5 (Automorphisms of a Curve of Genus ≥ 2). (a) For any ramified point P ∈ X with eP = r, we
may denote f −1 ( f (P)) = {P1 , . . . , P s }. They form an orbit of G on X. So Pi ’s have conjugate stabilizers. So

Yang Pi‑Yeh 86 Hartshorne Solutions

4 Curves 4.3 Embeddings in Projective Space

Ps n
s =the index of the stabilizer= |G| = nr . By Hurwitz’s formula, we have 2g(X) − 2 = n(2g(Y) − 2) + i=1 (r − 1),
ri i
P r
i.e. 2g−2 = 2g(Y) − 2 + s
i=1 (1 − 1
).
n
Ps ri
Ps
(b) If g(Y) ≥ 1 and i=1 (1− r1i ) = 0, we have g(Y) ≥ 2, hence by (a), n ≥ g(X)−1. If g(Y) ≥ 1 and i=1 (1− r1i ) > 0,
Ps P
we have i=1 (1 − r1i ) ≥ 12 , so 2g(Y) − 2 + i=1 (1 − r1i ) ≥ 21 , i.e. n ≥ 4(g(X) − 1). If g(Y) = 0, then by (a) we have
s
Ps Ps
i=1 (1 − ri ) = + 2 ≥ 2. So if s ≥ 5, we have i=1 (1 − r1i ) ≥ s · (1 − 21 ) ≥ 2 + 21 . If s = 4, the minimal case
1 2g(X)−2
n
Ps
is (r1 , r2 , r3 , r4 ) = (2, 2, 2, 3), hence i=1 (1 − ri ) ≥ 2 + 6 . If s = 3, the minimal case is (r1 , r2 , r3 ) = (2, 3, 7), hence
1 1
Ps Ps
i=1 (1 − ri ) ≥ 2 + 42 . So i=1 (1 − ri ) ≥ 2 + 42 at all, and n ≥ 84(g(X) − 1).
1 1 1 1

Solution 4.2.6 ( f∗ for Divisors). (a) We may assume X and Y are both affine. For any effective divisor D,
L (−D) is quasi‑coherent. So consider the exact sequence 0 → L (−D) → OX → OD → 0. By theorem 8.1.
in chapter III, we have R1 f∗ L (−D) = 0, so 0 → f∗ L (−D) → f∗ OX → f∗ OD → 0. Then by theorem 6.11.(b)
Ln
in chapter II, we have det( f∗ L (−D))  det( f∗ OX ) ⊗ (det( f∗ OD ))−1 . Since f∗ OD  O f∗ D , i.e. det( f∗ OD ) 
N −1
i=1
det(O f∗ D ) = L ( f∗ D). So det( f∗ OD )  L (− f∗ D), and det( f∗ L (−D))  det( f∗ OX ) ⊗ L (− f∗ D). For arbitrary
D, we may write D = D1 − D2 for two effective divisors D1 and D2 and consider the exact sequence 0 →
L (D) → L (−D2 ) → OD1 → 0. And similarly we have det( f∗ L (D)) = det( f∗ L (−D2 )) ⊗ L ( f∗ D1 ) = det( f∗ OX ) ⊗
L (− f∗ D2 )) ⊗ L ( f∗ D1 ) = det( f∗ OX ) ⊗ L ( f∗ D).
(b) Since L (D) only depends on its linear equivalent class, by (a), so does f∗ D. If deg f = n, the pullback
of a point is a divisor of degree n, so f∗ f ∗ = n.
(c) By 3.7.2., we have f ! ΩY  ΩX . By 3.6.10., we have f∗ ΩX = f∗ HomX (OX , f ! ΩY ) = HomY ( f∗ OX , ΩY ) =
( f∗ OX )−1 ⊗ΩY . Since both side are locally free of rank n, we have det f∗ ΩX  det(( f∗ OX )−1 ⊗ΩY ) = (det f∗ OX )−1 ⊗Ω⊗n Y .
(d) Since KX ∼ f ∗ KY + R, we have f∗ KX ∼ f∗ f ∗ KY + f∗ R = nKY + B. So L (−B)  Ω⊗n Y ⊗ L ( f ∗ K X )−1
. By (a) we
have L ( f∗ KX )−1  det f∗ OX ⊗ (det f∗ ΩX )−1 , we have L (−B)  (det f∗ OX )2 .

Solution 4.2.7 (Étale Covers of Degree 2). (a) For every P ∈ Y, ( f∗ OX )P is a rank 2 free module over OY,P . So
LP is a rank 1 free module, hence L is invertible. So by the exact sequence 0 → OY → f∗ OX → L → 0, we
have L  det L  det f∗ OX ⊗ (det OY )−1  det f∗ OX . Then L 2 = (det f∗ OX ) = L (−B). Since f is étale, hence
unramified, i.e. B = 0, so L 2  OY .
(b) Clearly the canonical morphism f : X → Y is finite, so x is integral, separated, of finite type over k,
and of dimension 1. So X is a curve. Moreover, since X is obviously normal, it is smooth. The function field
is clearly an extension of degree 2, so by 3.10.3., f is étale.
(c) For exact sequence 0 → OY → f∗ OX → L → 0, we have a section f∗ OX → OY as σ 7→ σ+τσ 2
. So the exact
sequence is split, and f∗ OX  OY ⊕ L . So by 2.5.17., X  Spec (OY ⊕ L ).

4.3 Embeddings in Projective Space

Solution 4.3.1. (⇐) By corollary 3.2., trivial. (⇒) If D is very ample, then l(D) = l(D − P − Q) ≥ 2 for any
two points P, Q. Since g = 2, we have dim |D| , 1 so l(D) , 2, i.e. l(D) > 2. So, if deg D ≤ 1, then by 4.1.5.,
l(D) ≤ degD + 1 ≤ 2, which contradicts. If deg D = 2, l(D) = l(K − D) + 1 < l(K) + 1 = 2, which contradicts. If
deg D = 3, we have l(K − D) = 0, so l(D) = 2, which contradicts. If deg D = 4, by corollary 3.2., D has no base
point. So |D| gives a morphism to P2 and X is a plane curve. So g = 21 (4 − 1)(4 − 2) = 3 , 2, which contradicts.
Hence by all above, deg D ≥ 5.

Solution 4.3.2. (a) Define D = X.L for some line L. By g(X) = 3, we have l(K) = 3, deg K = 4. By Bézout
theorem, deg D = 4. Since dim |L| = 2, we have l(D) = 3. So l(K − D) = l(D) + g(X) − deg D − 1 = 1, i.e.
deg K − D) = 0. So by 4.1.5., K = D.
(b) Suppose D = P + Q for two points P, Q. Since K is very ample and gives an embedding to P2 , we have
dim |K| = 2. So define L as the line PQ. We have K = X.L by (a). Thus we may assume K = P + Q + R + S for
some two points R, S . So dim |D| = dim |K − P − Q| = dim |K| − 2 = 0.

Yang Pi‑Yeh 87 Hartshorne Solutions

4 Curves 4.3 Embeddings in Projective Space

(c) If deg D = 2, by (b), dim |D| cannot be 1. So there cannot exist a close morphism X → P1 of degree 2, i.e.
not hyperelliptic.
S
Solution 4.3.3. We may assume X = Hi for some hypersurfaces. By 2.8.4.(e), we have L (K)  OX (n) for
some n > 0. So |K| induces an n‑tuple embedding, hence K is very ample.
If g(X) = 2, we have deg K = 2g − 2 = 2, so K is not very ample by 4.3.1. Hence X is not a complete
intersection.

Solution 4.3.4. (a) Since P1 is projectively normal, and by 2.5.14., the d‑uple embedding is projectively normal,
the image, i.e. the rational normal curve of degree d in Pd is projectively normal. Moreover, clearly, the kernel
is generated by xi x j = xk xl for all 4‑tuple (i, j, k, l) such that i + j = k + l.
(b) For any hyperplane H ⊂ Pn , we define D = X.H. Then dim |D| = n and deg D = d. So l(D) = n + 1 ≤
deg D + 1 = d + 1 ≤ n + 1. So d = n, and by 4.1.5., g(X) = 0. So X  P1 . Since X * Pn−1 , the morphism
Γ(Pn , O(1)) → Γ(X, L (D)) is injective. So D corresponds to a (n + 1)‑dimension subspace V ∈ Γ(P1 , O(n)). Since
dimk Γ(P1 , O(n)) = n + 1, they are equal. Hence X ⊂ Pn is induced by |D|, hence it is a rational normal curve.
(c) Take n small enough such that X ⊂ Pn and X * Pn−1 for any Pn−1 . As in (b), we have n = 2 and X is a
rational normal curve of degree 2, i.e. a conic in some P2 .
(d) If X is not a plane cubic curve, by (b), X ⊂ P3 and it is a rational normal curve of degre 3, hence a
twisted cubic.

Solution 4.3.5. (a) If φ(X) is nonsingular, by 2.5.5. we have dim H 0 (φ(X), Oφ(X) (1)) ≤ dim H 0 (P2 , OP2 (1)) = 3.
By since X is not a plane curve, we have H 0 (P3 , IX (1)) = 0. So H 0 (P3 , OP3 (1)) → H 0 (X, OX (1)) is injective, i.e.
dim H 0 (X, OX (1)) ≥ 4, which contradicts.
(b) Since X is the normalization of φ(X), by 4.1.8. we have g(X) ≤ g(φ(X)) = 12 (d − 1)(d − 2). Since φ(X) is
not nonsingular, the inequality above is strict.
(c) If X0 has no nilpotent element, it is a nodal plane curve, which has more bigger genus than Xt with
t , 0, which contradicts that flat family has same genus.

Solution 4.3.6. (a) Suppose X ⊂ Pn for smallest n. If n ≥ 4, by 4.3.4.(b), we have X is a rational normal curve.
If n = 2, then g(X) = 21 (4 − 1)(4 − 2) = 3. If n = 3, by 4.3.5.(b) we have g < 3. If g = 2, by 4.3.1., any divisor of
degree 3 is not very ample, so X cannot be embedded in P3 , which contradicts. If g = 0, X is clearly a rational
quartic curve.
(b) If g = 1, we may consider the exact sequence 0 → IX (2) → OP3 (2) → OX (2) → 0. Since dim H 0 (P3 , OP3 (2)) =
1, and dim H 0 (X, OX (2)) = l(2K) = 8 + 1 − 1 = 8 by Riemann‑Roch. So dim H 0 (P3 , IX (2)) ≥ 2. So X is a complete
intersection of more than 2 irreducible surface. Since the intersection of 2 quadric curves has degree 4 by
Bézout theorem, this intersection is all of X.

Solution 4.3.7. Clearly xy + x4 + y4 = 0 has only one node (0, 0). If it can by represented as a projection of a
nonsingular curve X in P3 , X must have degree 4 and genus 2. But by 4.3.6., it is impossible.

Solution 4.3.8. (a) Obviously, (1, 0, 0) is on every tangent line.

(b) We may assume X ⊂ Pn is a closed embedding. Denote P as the strange point of X. If O < X is a point,
φ is the projection about O, φ(P) is clearly on every tangent line of φ(X). So we may assume P is in P3 . Then
by theorem 3.9. X is P1 .

Solution 4.3.9. Since the tangent variety has dimension ≤ 2, and the dimension of multisecant lines ≤ 1,
the hyperplanes factor through any tangent line or multisecant line has dimension ≤ 2 in (P3 )0 , hence there
exists an open subset U of (P3 )0 and every hyperplane H in U does not factor through any tangent line and
multisecant lines, i.e. X.H consists of d distinct points and no three of them are collinear.

Yang Pi‑Yeh 88 Hartshorne Solutions

4 Curves 4.4 Elliptic Curves

Solution 4.3.10. Consider (Pn )×n , and a hyperplane H ⊂ (Pn )×n consisting of all points (x1 , . . . , xn ) such that
x1 , . . . , xn are collinear. Hence H is closed. Consider X ×n ⊂ (Pn )×n . Since X is not contained in any Pn−1 , so
X ×n * H. Since (Pn )×n − H is open in (Pn )×n , we have X ×n ∩ ((Pn )×n − H) is open in X ×n , and it is non‑empty. So
for all {P1 , . . . , Pn } ⊂ X such that (P1 , . . . , Pn ) ∈ X ×n ∩ ((Pn )×n − H), P1 , . . . , Pn are not collinear.

Solution 4.3.11. (a) Similarly with proposition 3.4., O induces a closed immersion X → Pn−1 iff O is not on
any secant line or tangent space of X. Moreover, since Sec is locally like (X × X − ∆) × P1 , which has dimension
2r + 1, and the tangent bundle has dimension 2r, there exists point O ⊂ Pn such that O is not on any secant
line or tangent space of X since n > 2r + 1.
(b) Denote the Veronese embedding as ρ. For any point R ∈ P5 on the secant line ρ(P)ρ(Q) for some two
points P, Q ∈ P2 . Consider Y = ρ(PQ), the image of secant line in P2 . R is on the plane spanned by Y. So every
line on this plane passing through R is also a secant line to Y, i.e. to X. Moreover, any point on a secant line of
X lies on a family of secant lines of dimension 1 of X. So Sec has dimension ≤ 2 + 2 + 1 − 1 = 4. But conversely,
the secant line of the four point ρ(1 : 0 : 0), ρ(0 : 1 : 0), ρ(0 : 0 : 1), ρ(1 : 1 : 0) are linearly independent, i.e.
dim Sec ≥ 4. Hence dim Sec = 4.

Solution 4.3.12. 1. Case r = 0 and d = 2, 3, 4, 5. The curve xd + yd + zd = 0 over C is smooth.

2. Case r = 1 and d = 3, 4, 5. The curve xyzd−2 + xd + yd = 0 over C has only one node (0 : 0 : 1).
3. Case d = 4, r = 3 or d = 5, r = 6. Consider rational normal curve X of degree d in Pd . We may project it
into P2 as Y. Then deg Y = d. And g(Y) = 21 (d − 1)(d − 2). Since X is isomorphic to P1 , i.e. g(X) = 0, by 4.1.8. Y
has 12 (d − 1)(d − 2) nodes.
4. Case d = 4, r = 2. The curve x2 yz + x2 z2 + xy3 + xyz2 − y2 z2 = 0 over C has only two nodes (0 : 0 : 1) and
(1 : 0 : 0).
5. Case d = 5, r = 2. The curve −x4 y + x4 z − x3 y2 − x3 yz + x3 z2 − x2 y3 − x2 y2 z + x2 yz2 + x2 z3 − xy4 − xyz3 − y5 −
y z + y3 z2 + y2 z3 = 0 over C has only two nodes (0 : 0 : 1) and (0, 1, −1).
4

6. Case d = 5, r = 3. The curve x5 +x4 z+x√

y +x3 yz−x3 z2 −x2 y2√z−x2 yz2 +x2 z3 −xy4 +xy2 z2 −xyz3 −y5 −y3 z2 +y2 z3 = 0
3 2

over C has only three nodes (0 : 0 : 1), ( 1− 2 −3 : 1 : 1) and (− 1− 2 −3 : 1 : 1).

7. Case d = 5, r = 4. The curve x2 y2 z − x2 yz2 + x2 z3 + xy2 z2 − xyz3 + y4 z + y3 z2 + y2 z3 = 0 over C has only four
√ √
nodes (0 : 0 : 1), (1 : 0 : 0), (1 : −1 : 0) and (−1 : −1 : 0).
8. Case d = 5, r = 5. The curve x4 y − x4 z +qx3 y2 + x3 yz + x2 y2 z − x2 yz
q + x z √+ xy + xy z − xyz
2 2 3 4 2 2
q + y z√
3 4
+ y2 z3 = 0
√
over C has only five nodes (0 : 0 : 1), (1 : − −1 − −1 : −1), (1 : −1 − −1 : −1), (1 : − −1 + −1 : −1)
q √
and (1 : −1 + −1 : −1).
Mathematica is Power, France is Bacon.

4.4 Elliptic Curves

Solution 4.4.1. We may assume X ⊂ P2 has the equation y2 = x(x − 1)(x − λ) and P is the infinity point by
proposition 4.6. So we may define a morphism φ : k[x, y, t] → R as x 7→ x, y 7→ y and t 7→ 1 ∈ H 0 (X, OX (P)).
For any s ∈ H 0 (X, OX (nP)) for some n, s has only one pole at P, i.e. there exists an f ∈ k(A2 ) such that f |X = s.
So φ is surjective. Conversely, (y2 − x(x − t2 )(x − λt2 )) ⊂ ker(φ). And by Riemann‑Roch, dim H 0 (X, OX (nP)) = n,
which is equal to dim R(n) . So φ is an isomorphism.

Solution 4.4.2. We need to prove a lemma from Castelnuovo, for any coherent sheaf F on X such that
H 1 (X, F ) = H 1 (X, F ⊗ L (−D)) = 0, we have H 0 (X, F ⊗ L (nD)) ⊗ H 0 (X, L (D)) → H 0 (X, F ⊗ L ((n + 1)D))
is surjective. If Supp F has dimension 0, there exists an s ∈ L (D) such that s(x) , 0 for all x ∈ Supp F .
So the morphism H 0 (X, F ) ⊗k (s · k) → H 0 (X, F ⊗ L (D)) is an isomorphism. So H 0 (X, F ) ⊗ H 0 (X, L (D)) →

Yang Pi‑Yeh 89 Hartshorne Solutions

4 Curves 4.4 Elliptic Curves

H 0 (X, F ⊗ L (D)) is surjective. If Supp F has dimension 1, we may consider the commutative diagram

0 / H 0 (X, F ⊗ L (−D)) ⊗ H 0 (X, L (D)) / H 0 (X, F ) ⊗ H 0 (X, L (D)) / H 0 (X, G ) ⊗ H 0 (X, L (D)) /0
3
ψ hh h
h hhhh
hh hh
hhhh
α β γ
 hhhhh  
0 / H 0 (X, F ) / H 0 (X, F ⊗ L (D)) / H 0 (X, G ⊗ L (D)) /0

where G is the cokernel of F ⊗ L (−D) → F as a 7→ a ⊗ s for some s ∈ H 0 (X, L (D)) such that s| x is not
a zero divisor of F x for every x ∈ X. So we have a surjective morphism cokerβ → cokerγ. Since Supp G
has dimension 1, we’ve already have cokerγ = 0. Moreover, we can define a morphism ψ : H 0 (X, F ) →
H 0 (X, F ) ⊗ H 0 (X, L (D)) as a 7→ a ⊗ s. This morphism is commutative with the diagram, hence the morphism
cokerα → cokerβ is zero morphism. So, cokerβ = 0, i.e. H 0 (X, F ) ⊗ H 0 (X, L (D)) → H 0 (X, F ⊗ L (D)) is
surjective. For higher n, we just need to replace F to F ⊗ L ((n − 1)D) and prove by induction.
Denote the embedding X → Pn by φ. Take an effective divisor E of degree d − 2 supported on Supp i∗ OD ,
and consider the exact sequence 0 → L (−E) → OX → i∗ OE → 0. So we have 0 → L (D − E) → L (D) →
i∗ OE → 0. Since deg D − E = 2, by Serre duality we have H 1 (L (D − E)) = 0. Hence we have a commutative
diagram

H 0 (X, L (D − E)) ⊗ H 0 (X, L (nD)) / H 0 (X, L (D)) ⊗ H 0 (X, L (nD)) / H 0 (X, i∗ OE ) ⊗ H 0 (X, L (nD)) /0

f g h
  
H 0 (X, L (D − E + nD)) / H 0 (X, L (D + nD)) / H 0 (X, i∗ OE ⊗ L (nD))

So by five lemma, we have an exact sequence coker f → cokerg → cokerh. Since nD has no base point for n ≥ 1.
By Castelnuovo’s lemma we have cokerh = 0. Moreover, since D − E has degree 2, D − E has no base point.
So again by Castelnuovo’s lemma, we have coker f = 0. Hence cokerg = 0, i.e. H 0 (X, L (D)) ⊗ H 0 (X, L (nD)) →
H 0 (X, L ((n + 1)D) is surjective.
Since |D| is complete, we know H 0 (Pn , OPn (1)) → H 0 (X, OX (1)) is surjective. For n > 0, since we have

H 0 (Pn , OPn (1)) ⊗ H 0 (Pn , OPn (k)) / H 0 (Pn , OPn (k + 1))

 
H 0 (X, OX (1)) ⊗ H 0 (X, OX (k))
g
/ H 0 (X, OX (k + 1))

Since g is surjective as above, and the left arrow is surjective by induction, we have H 0 (Pn , OPn (k)) → H 0 (X, OX (k))
is surjective for all k. So by 2.5.14.(d), X is projectively normal.

Solution 4.4.3. Denote f = y2 −x(x−1)(x−λ). So we may write R = k[x, y]/( f ) as the ring of functions on X which
are regular out of P0 . Then K(X) = Frac(R). So for every g ∈ K(X), it must has the form g = a(x)+b(x)y for some
a(x), b(x) ∈ k(x). So if φ ∈ Aut(X, P0 ), we may assume φ(x, y) = (a(x) + b(x)y, c(x) + d(x)y). For any P = (x, y) ∈ X,
we have φ(P) + φ(−P) = 0, i.e. 0 = φ(x, y) + φ(x, −y) = (a(x) + b(x)y, c(x) + d(x)y) + (a(x) − b(x)y, c(x) − d(x)y). So
(a(x) + b(x)y, c(x) + d(x)y) = (a(x) − b(x)y, −c(x) + d(x)y), hence b(x) = c(x) = 0 for all x, i.e. φ(x, y) = (a(x), d(x)y).
Since φ(x, ∞) = (a(x), d(x)∞) = (x, ∞) = P0 for all x, we have d(x) is a constant d, i.e. φ(x, y) = (a(x), dy). Since
(a(x), dy) ∈ X, we clearly know that a(x) has degree 1, i.e. a(x) has the form ax + b. So φ ∈ Aut(X, P0 ) has the
form φ(x, y) = (ax + b, dy).

Solution 4.4.4. Take the automorphism y 7→ y − a1 x+a3

2
, the equation f = y2 + a1 xy + a3 y − x3 − a2 x2 − a4 x − a6
a 2 a2
transform to f = y2 − x3 −(a2 + 41 )x2 −(a4 + 12 3 )x−(a6 + 43 ) = y2 −(x−a)(x−b)(x−c) for some a, b, c. Then take the
a a
2 −λ+1)3 (a2 +b2 +c2 −ab−bc−ca)3
automorphism x 7→ x−ax−b
, f = (b − a)3 y2 − x(x − 1)(x − λ) for some λ = a−ba−c
. Then j = 28 (λλ2 (λ−1)2 = ((a−b)(b−c)(c−a))2
.

Yang Pi‑Yeh 90 Hartshorne Solutions

4 Curves 4.4 Elliptic Curves

So by Veda’s theorem and Mathematica, we have

 2 3
1728 a21 + 4a2 − 24(a1 a3 + 2a4 )
j =  2 3     3  2
a21 + 4a2 − 24(a1 a3 + 2a4 ) − −36 a21 + 4a2 (a1 a3 + 2a4 ) + a21 + 4a2 + 216 a23 + 4a6

hence a rational function with coefficient in Q. For the existence of j, we consider the following case
1. Case char(k) = 2 and j = 0. y2 + y = x3 .
2. Case char(k) = 2 and j , 0. y2 + xy = x3 + x2 + j−1 .
3. Case char(k) = 3 and j = 0. y2 = x3 + x.
4. Case char(k) = 3 and j , 0. y2 = x3 + x2 − j−1 .
5. Case char(k) , 2, 3 and j = 0. y2 = x3 + 1.
6. Case char(k) , 2, 3 and j = 1728. y2 = x3 + x.
7. Case char(k) , 2, 3 and j , 0, 1728. y2 = x3 + 2kx + 2k, where k = 1728−
j
j
.

Solution 4.4.5. (a) We may define π and π0 by the linear system |2P0 | and |2 f (P0 )|. So clearly there exists a
morphism g : P1 → P1 of degree 2 satisfying π ◦ f = g ◦ π0 .
(b) We may assume X has the function y2 = x3 + ax + bx + c. Since f (P0 ) = P0 , we just define π and π0 as
the projection of the x‑coordinate in A2 . So g is (x : z) 7→ (x2 : z2 ).
(c) Suppose X has the form y2 = x(x − 1)(x − λ). Clearly g has two branch point 0 and 1. And by Hurwitz’s
formula, π and π0 both have 4 branch points 0, 1, λ and ∞. So g is branched over 0 and ∞. Define D =
(0, 0) + (1, 0) + (λ, 0) + P0 as a divisor of X. Then Rπ◦ f = f ∗ D. Since Rg◦π0 = D + π0∗ ((0) + (∞)) = D + (2 · (0, 0) + 2 · P0 ),
and g ◦ π0 = π ◦ f , we know f ∗ D = 3 · (0, 0) + (1, 0) + (λ, 0) + 3 · P0 . So g−1 ({1, λ}) = {0, 1, λ, ∞} consists of the four
√ √ 2 −λ+1)3 2 −10λ+1)3
branch points of π0 . Easily calculate, we have (0, 1, λ, ∞) ∼ (−1, 1, λ, − λ). So 256(λ λ2 (λ−1)2
= 16(λλ(λ−1) 4 .
(d) By (c), we have 256λ − 1808λ + 5504λ − 21696λ
11 10 9 8
− 5760λ + 47008λ − 5760λ − 21696λ + 5504λ3 −
7 6 5 4
√ √ 1−3 √−7 1+3 √−7 1−3 √−7 1+3 √−7
1808λ + 256λ = 0, i.e. λ = −1, 0, 1, 3 − 2 2, 3 + 2 2, 32 , 32 , 2 , 2 . So for nonsingular curve,
2

the corresponding j = 1728, 8000, −3375.

Solution 4.4.6. (a) Similarly, we may take a line L which is not a tangent line of X. Then we can define
φ : X → L as P 7→ T p (X) ∪ L. Then consider the morphism ψ : X → L as the projection from some O < L, by
Hurwitz’s formula, 2g − 2 = −2d + deg Rψ , i.e. deg Rψ = 2g − 2 + 2d = d2 − d − 2r. So deg X ∗ = ]φ−1 (P) = d2 − d − 2r.
Then by Hurwitz’s formula, deg Rφ = 2g − 2 + 2(d2 − d − 2r) = 3d2 − 5d − 6r. Ignoring d‑ramified points, we
have ]inflection points= 3d2 − 6d − 6r.
(b) If n = 3, we can take some O ∈ P3 not in any hyperosculating hyperplane, and ψ : X → P2 is the
projection from O. Denote the image plane curve as X̄. Then X̄ has degree d, r nodes and genus g = 12 (d −
1)(d − 2) − r. In P3∗ , the hyperplane corresponding to O intersects X ∗ at deg X ∗ points. And every point
corresponding to a hyperplane in P3 which is an osculating hyperplane for some point on X. So deg X ∗ =
]hyperosculating points of X̄ = ]inflection= 6(g − 1) + 3d. So by Hurwitz’s formula, ]hyperosculating points
of X = 2g − 2 + 2 deg X ∗ − (2g − 2 + 2d) = 12(g − 1) + 4d.
If n > 3, we may prove it by induction. We may take some O ∈ Pn not in and hyperosculating hyperplane,
and the image projection from O of X, X̄, is smooth. And similarly, deg X ∗ = ]hyperosculating points of
X̄ = n(n − 1)(g − 1) + nd. So by Hurwitz’s formula, ]hyperosculating points of X = 2g − 2 + 2 deg X ∗ − ((n − 2)(n −
1)(g − 1) + (n − 1)d) = n(n + 1)(g − 1) + (n − 1)d.
(c) Clearly, the infinity point P0 is a hyperosculating point. So for any hyperosculating hyperplane X in
Pd−1 intersecting X with P, we have H.X = dP. Since all hyperplanes in Pd−1 are linearly equivalent, we have
d(P − P0 ) = 0, i.e. P is a d‑torsion point. Conversely, for any d‑torsion point, dP is linearly equivalent to dP0 ,
i.e. it is a divisor obtained by intersection X with some hyperplane in Pd−1 . Hence P is a hyperosculating
point.

Yang Pi‑Yeh 91 Hartshorne Solutions

4 Curves 4.4 Elliptic Curves

Solution 4.4.7 (The Dual of a Morphism). (a) By theorem 4.11., Pic is just Jacobian. And clearly the induced
homomorphism f ∗ is the morphism between Jacobians JacX 0 → JacX, which is just (X 0 , P00 ) → (X, P0 ) since X
and X 0 are elliptic curves.
(b) Since f ∗ is just fˆ, by (g f )∗ = f ∗ · g∗ , trivial.
(c) Separable Case. If f is separable, since f has degree n, we have ] ker f = n, i.e. every element in ker f
has order dividing n. So ker f ⊂ ker nX . By Galois theory, we have n∗X K(E) ⊂ f ∗ K(E) ⊂ K(E). Then we may
fine a map g satisfying f ∗ g∗ K(E) = n∗X K(E), i.e. g ◦ f = nX , So clearly g = fˆ.
Purely Inseparable Case. We may assume f is just the Frob p , where p = char(k). Then deg f = p. Since pX
is clearly not separable, by Galois theory, pX can be decomposed into separable and purely inseparable part,
i.e. pX = g ◦ Frobep . Hence fˆ = g ◦ Frobe−1 p .
(d) Fix an invertible sheaf L on X 0 . Define Pic 0σ as in hint. Then clearly Pic 0σ  Pic (X 0 /X)0 as f 7→ L f =
f 0∗ L , where f 0 = f × id : X × X 0 → X 0 × X 0 , L = L (∆X0 ) ⊗ π∗2 L (−P00 ) ∈ Pic 0 (X 0 /X 0 ) is the sheaf in the definition
of Jacobian, and π2 : X 0 × X 0 → X 0 is the second projection. So we have L f +g  L f ⊗ Lg . And for any x ∈ X,
we have Γ∗g L f = L f,(x,g(x)) = L f (x),g(x) . If f (x) = g(x), we have L f (x),g(x) = OX0 , f (x) . And if f (x) , g(x), we have
L f (x),g(x) = 0. So Γ∗g (L f ) = Γ∗f Lg ). Since for any F ∈ Pic X 0 , F 0 = p∗2 F ∈ Pic 0σ , there exists some h : X → X 0 such
that Lh = F 0 . And we have Γ∗f +g Lh = Γ∗h L f +g = Γ∗h L f ⊗ Γ∗h Lg = Γ∗f Lh ⊗ Γ∗g Lh . Hence ( f + g)∗ F  f ∗ F ⊗ g∗ F ,
i.e. ( f + g)ˆ  fˆ + ĝ.
(e) Since nX = 1X + . . . + 1X . And 1X = idX has the dual idX . So by (d), in the case X 0 = X, n̂X = 1̂X + . . . + 1̂X =
1X + . . . + 1X = nX . So nX = n̂X . By (c), since nX ◦ nX = n2X , we have deg nX = n2 .
(f) By (c), we have (deg f )2 = deg fˆ ◦ deg f = deg f ◦ deg fˆ = (deg fˆ)2 . So deg f = deg fˆ.
Solution 4.4.8. For any étale covering X → E, since étale means unramified, we have g(X) = g(E), i.e. X is an
elliptic curve. For any étale covering f : E 0 → E, by 4.4.7., there exists a dual morphism g : E → E 0 such that
g ◦ f = nX , where n = deg f . So π1 (E) = lim Gal(nE ). Then consider the following three cases:
←−−n
1. Case char(k) = 0. Clearly nE is an étale morphism for all n. For any P ∈ ker nE , we define τP : E → E
as Q 7→ Q + P. Then nE ◦ τP = nE , hence τP ∈ Gal(nE ). Conversely, for any f ∈ Gal(nE ), we have deg f =
deg nX / deg nX = 1. So f has the form τP for some P ∈ X. Then by nE ◦ τP = nE we have P ∈ ker nE . Hence
Q
Gal(nE ) = ker(nE ) = (Z/nZ)2 . So π1 (E) = lim Gal(nE ) = lim (Z/nZ)2 = l prime Zl × Zl .
←−−n ←−−n
2. Case char(K) = p and the Hasse invariant is 0. In this case, if (n, p) = 1, nX is étale. And if p|n,
nX is ramified. So similarly with case 1, we have Gal(nE ) = ker(nE ) = (Z/nZ)2 for (n, p) = 1. So π1 (E) =
Q
lim Gal(nE ) = lim (Z/nZ)2 = l,p Zl × Zl .
←−− p-n ←−− p-n
3. Case char(K) = p and the Hasse invariant is 1. For any étale covering f : E 0 → E, we have a decompo‑
ˆ r for some g such that p - n = deg g. So we may consider nE ◦ Frob
sition f = g ◦ Frob ˆ r = ĝ ◦ Ê, and similarly
r
with case 1, the we can only take the term (nE ◦ Frob ˆ ) in the inverse limit. And since Gal(nE ◦ Frob ˆ r) =
Q
(Z/nZ)2 × (Z/pr Z) by 4.4.15., we have π1 (E) = lim (Z/nZ)2 × (Z/pr Z) = Z p × l,p (Zl × Zl ).
←−−n,r
Solution 4.4.9. (a) If f : X → X 0 and g : X 0 → X 00 are finite, clearly g ◦ f is finite. Moreover, we have a finite
dual morphism fˆ : X 0 → X. Hence isogeny is an equivalent relation.
(b) Fix an X. If f : X → X 0 is an isogeny, clearly ker f is a finite subgroup of X. Conversely, if g : X → X 00
is another isogeny with kernel G, we have K(X 0 ) = K(X)G = K(X 00 ). Moreover, X 0 and X 00 are both smooth,
they are isomorphic, i.e. the isogeny class of X is a subset of the set of finite subgroup of X, hence countable.
Solution 4.4.10. For any M ∈ Pic (X × X), we define L1 = M |X×{P0 } , L2 = M |{P0 }×X , and F = M ⊗ (p∗1 L1 ⊗
p∗2 L2 )−1 . Then clearly F |X×{P0 } and F |{P0 }×X is trivial. Since X is smooth over k, we have X × X → X is a flat
morphism. Then F |X×{P} has the same degree with F |X×{P0 } for all P ∈ X, i.e. has degree 0. Hence we may
define a morphism Φ : Pic (X × X) → R as φ(M )(P) = ψ(F |X×{P} ) for any P ∈ X, where ψ : Pic 0 X  X. Clearly
Φ is surjective. For any M ∈ ker Φ, we know F |X×{P} is trivial for all P ∈ X, so F = p∗2 L for some L ∈ Pic X.
Hence M = p∗1 L1 ⊗ p∗2 (L2 ⊗ L ), which means ker Φ = p∗1 Pic X ⊕ p∗2 Pic X. So we have an exact sequence
0 → p∗1 Pic X ⊕ p∗2 Pic X → Pic (X × X) → R → 0.

Yang Pi‑Yeh 92 Hartshorne Solutions

4 Curves 4.4 Elliptic Curves

Solution 4.4.11. (a) Denote the area of period parallelogram of lattice L as AL . Then clearly deg f = AAαLL = |α|2 .
(b) Since f is induced by the morphism C → C as ×α, and NX is induced by ×N, where N = deg f = |α|2 =
αᾱ, we know the morphism ×ᾱ induces the dual morphism fˆ : X → X.
√
(c) Since τ ∈ Q( −d) and is integral over Z, τ2 is a linear combination of τ and 1 with integral coefficients,
so Z[τ] = Z ⊕ Zτ = Lτ , and Z[τ] ⊂ R. Conversely, for any f ∈ R, there exists some α ∈ C corresponding to f
such that αLτ ⊂ Lτ . Then α · 1 ∈ Lτ , i.e. α ∈ Z[τ]. So R = Z[τ].

Solution 4.4.12. (a) If X has automorphisms leaving P0 fixed other than ±1, End(X, P0 ) must bigger than Z.
√
Hence X has complex multiplication, so τ ∈ Q( −d) for some square‑free d > 0. If f is an automorphism of
X leaving P0 fixed corresponding to some α ∈ C, we have deg f = 1. By 4.4.11.(a), we have |α| = 1. So α = ζn
for some n, where ζn , 1 satisfies ζnn = 1. Moreover, since Q(α) ⊂ Q(τ) has extension ≤ 2 over Q, and α , ±1,
we have [Q(α) : Q] = 2. So n = 4 or 6, i.e. α = i or ω, in which case, τ = i or ω.
√
(b) In this case, X has complex multiplication too, since 2 is not a square. We may assume τ ∈ Q( d) for
some square‑free d < 0. Then we have two cases:
√ √ √
1. d ≡ 1 mod 4. Then the integral ring O of Q( d) is Z[ 1+2 d ]. So α = (a + b2 ) + b2 d for some integers a, b.
2
Then (a + 2b )2√− b4d = 2. We only have an integer solution d = −7, a = 0 and b = ±2. So in this case, we clearly
have τ = −1+2 −7 .
√ √
2. d ≡ 2, 3 mod 4. Then the integral ring is Z[ d]. So α = a + b d for some integers a, b, and a2 − b2 d = 2.
√
So we have (a, b, d) = (±1, ±1, −1), (0, ±1, −2). In each case, we have τ = i or τ = −2.

Solution 4.4.13. If E is a supersingular elliptic curve over F13 , we may assume it has the form Eλ : y2 =
P  2
x(x − 1)(x − λ) for some λ ∈ F̄13 . Then by corollary 4.22., E is supersingular iff h13 (λ) = 6i=0 6i λi = λ6 + 10λ5 +
λ
4λ4 +10λ3 +4λ2 +10λ+1 = 0. So h13 (λ) has at most 6 roots. If λ is a root, λ0 = λ, λ1 , 1−λ, 1−λ
1
, λ−1 or λ−1
λ
, λ0 must be
a root of h13 since jλ = jλ0 and by theorem 4.1., Eλ  Eλ0 . Since if jλ = 0, we have h13 (λ) = λ4 +11λ3 +1λ2 +3)(λ2 −
λ+1)−2 , 0, and if jλ = 1728, we have h13 (λ) = (7λ3 +9λ2 +5)(2λ3 −3λ2 −3λ+2)+(λ2 +12λ+4) = λ2 +12λ+4 , 0.
Hence every root of h13 is not corresponding to j = 0 or 1728. So the 6 roots of h13 correspond to one j, i.e.
we only have one supersingular elliptic curve over F13 . By Mathematica, if h13 (λ) = 0, we have j = 5 mod 13.
And in this case, E has the form y2 = x3 + 4x + 7.

Solution 4.4.14. By proposition 4.21., X(p) is supersingular iff (xyz) p−1 has coefficient 0 in (x3 + y3 − z3 ) p−1 . So
if 3 - p − 1, there cannot exist the term (xyz) p−1 . If 3|p − 1, clearly the coefficient is (((p−1)/3)!)
(p−1)!
3 , which cannot be

zero since (p − 1)! and ((p − 1)/3)!) don’t have the prime factor p. So X(p) is supersingular iff p ≡ 2 mod 3. By
Dirichlet’s density theorem, the set P has density 12 .
∗
Solution 4.4.15. By proposition 2.1. and example 2.1.5, F̂ 0 : X → X p is separable ⇔ we have 0 → F̂ 0 ΩX p → ΩX
⇔ 0 → H 0 (X p , ΩX p ) → H 0 (X, ΩX ) ⇔ by Serre’s duality, the morphism F 0∗ : H 1 (X, OX ) → H 1 (X p , OX p ) is surjective
⇔ the Hasse invariant of X is 1.
Since F 0 is purely inseparable, we have deg s (F̂ 0 ) = deg s (pX ). So ] ker pX = deg s (F̂ 0 ). Since deg F 0 = p and
deg F̂ 0 = p, we have deg s (F̂ 0 ) = 1 or p because p is a prime. So by above, if the Hasse invariant is 1, F̂ 0 is
separable, deg s (F̂ 0 ) = p, i.e. ] ker pX = p, hence ker pX = Z/p. Or if the Hasse invariant is 0, F̂ 0 is inseparable,
deg s (F̂ 0 ) = 1, i.e. ] ker pX = 1, hence ker pX = 0.

Solution 4.4.16. (a) Clearly the Frobenious morphism on Spec Fq is the isomorphism, since Fq → Fq , x 7→ xq
is an isomorphism. So the base change X → Fq is an isomorphism Xq → X. If X has homogeneous equation
f (x, y, z) = 0, we have X = Proj A for A = k[x, y, z]/( f ). Then X p = Proj A p for A p = A ⊗F 0 k = k[x, y, z]/( f p ), where
f p is the polynomial which change every coordinate of f to the qth‑power. So f p (x p , y p , z p ) = ( f (x, y, z)) p .
Hence the morphism X → X is the qth‑power map on the coordinates of points of X embedding in P2 .
(b) Suppose X has the form y2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 . Consider the invariant differential
q q−1
ω = 2y+adx 1 x+a3
of X. We have (1 − F 0 )∗ ω = ω + F 0∗ ω. Since F 0∗ ω = 2yq +ad(x1 xq +a3 = 2yqqx+a1 xdx 0 ∗
q +a = 0. So (1 − F ) ω = ω.
3

Yang Pi‑Yeh 93 Hartshorne Solutions

4 Curves 4.4 Elliptic Curves

Since ΩX is defined by ω, we have 0 → (1 − F 0 )∗ ΩX → ΩX . Hence (1 − F 0 ) is separable by proposition 2.1.

Clearly ker(1 − F 0 ) = {(x, y, z) ∈ X | (x, y, z) = (x p , y p , z p } = X(Fq ).
(c) Define a = 1 + deg F 0 − deg(1 − F 0 ). Then clearly aX = 1X + (deg F 0 )X − (deg(1 − F 0 ))X = 1X + F 0 F̂ 0 − (1X −
F 0 )(1X − F̂ 0 ) = 1X + F 0 F̂ 0 − 1X − F 0 F̂ 0 + (F 0 + F̂ 0 ) = F 0 + F̂ 0 . Moreover, by (b), ]X(Fq ) = deg s (1X − F 0 ) = deg(1X − F 0 ) =
−a + 1 + deg F 0 = q − a + 1.
2 +n2 q
(d) By (c), (m + nF 0 )(m + nF̂ 0 ) = m2 + n2 F 0 F̂ 0 + mn(F 0 + F̂ 0 ) = m2 + n2 q + amn ≥ 0. We have |a| ≤ m |mn| ≤
√
2 m2 n2 q √
|mn|
= 2 q.
(e) If q = p, in the same method with (b), we know (m + nF 0 ) is separable iff p - m. So the Hasse invariant
√
is 0 ⇔ F̂ 0 = a − F 0 is inseparable ⇔ p|a. When p ≥ 5, we have p|a and |a| ≤ 2 q. So a = 0, i.e. N = q + 1.

Solution 4.4.17. (a) If X has the Weierstrass equation y2 + a1 xy + a3 y = x3 + a2 x2 + a4 x + a6 , P1 = (x1 , y1 ) and

P2 = (x2 , y2 ) ∈ X, we have: (1) if x1 = x2 and y1 + y2 + a1 x2 + a3 = 0, then P1 + P2 = O; (2) if x1 = x2 and
  3x2 +2a2 x1 +a4 −a1 y1
y1 + y2 + a1 x2 + a3 , 0, then (x3 , y3 ) = λ2 + a1 λ − a2 − x1 − x2 , −(λ + a1 )x3 − ν − a3 , where λ = 1 2y1 +a 1 x1 +a3
and
−x13 +a4 x1 +2a6 −a3 y1   y2 −y1
ν= 2y1 +a1 x1 +a3
; (3) if x1 , x2 , then (x3 , y3 ) = λ + a1 λ − a2 − x1 − x2 , −(λ + a1 )x3 − ν − a3 , where λ = x2 −x1
2

−y2 x1
and ν = y1 xx22 −x 1
.
So in this case, P + Q = ( −a +ab+b
3 2 3
a2
, − ba3 + ba − a + b − 1). And P = (0, 0), 2P = (1, 0), 3P = (−1, −1), 4P = (2, −3),
5P = ( 4 , − 8 ), 6P = (6, 14), 7P = (− 9 , 27 ), 8P = ( 21
1 5 5 8
25
, − 125
69
), 9P = (− 20
49
, − 435
343
), 10P = ( 161
16
, − 2065
64
).
(b) For this curve, we have ∆ = 37. So for any p , 37, ∆ , 0, so the reduction X p is smooth, i.e. good
reduction.

Solution 4.4.18. By Mathematica, we at least has these points: (−3, ±2), (−2, ±4), (−1, ±4), (1, ±2), (2, ±2),
(3, ±4), (5, ±10), (9, ±26), (13, ±46), (31, ±172), (41, ±262), (67, ±548), (302, ±5248). They are all generated by P
and Q.

Solution 4.4.19. Suppose X is defined as a Weierstrass equation f (x, y, z) = 0. Then we can define X̄ as a
subscheme of P2Z as the equation f (x, y, z) = 0. So the generic fibre is X/Q, and for any closed point (p) ∈ Spec Z,
the fibre is X p . Define T = D(2∆) ⊂ Spec Z and X̄0 = π−1 (T ) ⊂ X̄. Then the generic fibre of X̄0 is also X, and X̄0
is smooth over T . Moreover, the addition morphism µ : X × X → X and the negation morphism ι : X → X
can be extended to the morphisms µ : ∆(X̄0 ) → X̄0 and ι : X̄0 → X̄0 . So nX : X̄0 → X̄0 and defined over T for
all n ∈ Z. For any (p) ∈ T , consider (nX ) p : X(p) → X(p) . And for any prime `, by theorem 6.8. in chapter II,
(`X ) p : X(p) → X(p) is either constant or flat. Since it is not constant for all p, we have `X is flat. And since for
any n = `1 . . . `r , we have nX = `1,X · . . . `r,X is flat. And by theorem 4.17., nX is also finite. So ker nX is flat over T .
Finally, for all (p) ∈ T , X(p) is nonsingular. So we have an exact sequence 0 → ker π p → X(Q) → X(p) (F p ) → 0,
where π p is the reduction at prime p. So if (n, p) = 1, we have an injection X(Q)[n] → X(p) (F p ).

Solution 4.4.20. (a) If X is defined by y2 = x(x − 1)(x − λ) = x3 − (λ + 1)x2 + λx for some λ, we have jX =
!
1 3
2 −λ+1)3 256λ−2/p (−λ−1)2/p −3λ p
256 (λλ2 (λ−1)2 . So by 4.4.16., X p is defined by y = x − (λ + 1)
2 3
x +λ
1/p 2 1/p
x, we have jX p 1 . Hence
(−λ−1)2/p −4λ p
(λ2 −λ+1)3
jXp p ≡ 256 λ2 (λ−1)2 = jX mod p. Hence j(X p ) = j(X)1/p in F̄ p .
(b) (⇐) Just take π = F 0 , then trivial. (⇒) We have pX is inseparable, and deg pX = p2 . If pX = ππ̂, so
deg π = deg π̂ = p. And at least one of π and π̂ is purely inseparable. We may assume π is purely inseparable.
Then by theorem 2.5. π is equivalent to a morphism π : Y → X with K(Y) = K(X)1/p . Since π ∈ R, we have
K(X) = K(Y) = K(X)1/p , hence X  X p since K(X p ) = K(X)1/p .
Since π : X → X is purely inseparable, there must exist some unit φ ∈ R such that π = φF 0 . The Hasse
invariant is 0 ⇔ F̂ 0 is purely inseparable ⇔ π̂ = F̂ 0 φ̂ is purely inseparable of degree p ⇔ there exists some unit
ψ ∈ R such that π̂ = ψF 0 , i.e. π̂ = ψφ−1 π for some unit ψφ−1 ∈ R.
(c) If Hasse(X) = 0, we know F̂ 0 is purely inseparable. Then pX = F 0 F̂ 0 : X → X is purely inseparable of
2 2
degree p2 . So X p2  X. Hence jX = jXp 2 = jXp , i.e. jX ∈ F p2 .
p

Yang Pi‑Yeh 94 Hartshorne Solutions

4 Curves 4.5 The Canonical Embedding

(d) For any f ∈ R, we may assume f (x, y, z) = ( f1 (x, y, z), f2 (x, y, z), f3 (x, y, z)) for three homogeneous fi with
coefficient in k. So ( fi (x, y, z)) p = fi (x p , y p , z p ) since they are all in k with char(k) = p. Hence f ◦ F = F ◦ f . So
if Hasse , 0, we have λF , 0 by definition. Then for any f ∈ R, if λ f < F p , we have λ f · F ∗ (x) , F ∗ (λ f )F ∗ (x) =
F ∗ (λx), which contradicts that they actually commutate with each other, hence λ f ∈ F p . So clearly φ : R → F p
is surjective. Define p = ker φ, we have R/p  F p .
√
Solution 4.4.21. Denote d as the discriminant of O. Then (1, ω) is a base of O as a Z‑module, where ω = d+2 d .
So if we define f as the smallest positive integer such that f ω ∈ R. For any a + bω ∈ R, we have f |b, hence
R = Z + Z( f ω) = Z + f · O.

Solution 4.4.22. Denote φ : X → A1C the family and s0 : A1C → X the section. Then for any closed t ∈ A1C , we
S
define s0 (t) as the zero of group structure of Xt . So consider the set S = t∈A1 ker 2Xt and ψ = φ|S : S → A1C .
C
Then for any t ∈ A1C , S t has only four closed points. So by theorem 9.9. in chapter II, ψ : S → A1C is a flat
family. And ψ is obviously unramified, hence ψ is étale. But since A1C is simply connected, S must be four
copies of A1C , one of the copy is just s0 . And we may denote the rest three as s1 , s2 , s3 . So for any t ∈ A1C , we
may choose a coordinate of Xt such that s0 (t) = ∞, s1 (t) = 0, s2 (t) = 1, and s3 (t) = λ. Then t 7→ s3 (t) 7→ λ is a
morphism A1C → A1C − {0, 1}. Since the morphism like this must be constant morphism, i.e. λ are the same for
all Xt . Hence all Xt are isomorphic to each other, i.e. the family is trivial.

4.5 The Canonical Embedding

Solution 4.5.1. By 4.3.3. and proposition 5.2., trivial.

Solution 4.5.2. Hyperelliptic Case. If X is hyperelliptic, there exists a unique morphism f : X → P1 of degree
2. By Hurwitz’s formula, it has 2g + 2 ramified points. For any φ ∈ Aut(X), we have f ◦ φ : X → P1 of degree 2,
hence f ◦φ and f differs by some automorphism of P1 . So φ maps ramified points to ramified points. If φ fixed
all 2g + 2 ramified points, by Lefschetz’s theorem, it is just identity. So |Aut(X)| ≤ |the group of permutations
of ramified points| < ∞.
Non‑Hyperelliptic Case. If X is not hyperelliptic, we may consider the canonical embedding f : X → Pg−1 .
Then by 4.4.6., there are g(g − 1)2 + dg hyperosculation points of X in Pg−1 . In this case, we have g ≥ 3, hence
g(g − 1)2 + dg ≥ 2g + 2. And obviously, any φ ∈ Aut(X) maps hyperosculation points to hyperosculation
points by definition. And if φ fixes all hyperosculation points, by Lefschetz’s theorem we know φ = idX . So
|Aut(X)| ≤ |the group of permutations of hyperosculation points| < ∞.

Solution 4.5.3 (Moduli of Curves of Genus 4). Hyperelliptic Case. By example 5.5.5., the hyperelliptic curves
of genus 4 form an irreducible family of dimension 2 × 4 − 1 = 7.
Non‑Hyperelliptic Case. By example 5.2.2., any nonhyperelliptic curve of genus 4 is the complete in‑
tersection of a quadric hypersurface and a cubic hypersurface in P3 , and vice versa. Since the moduli of
quadric hypersurface in P3 has dimension 9, and for any quadric hypersurface Q defined by f , we have
×f
0 → H 0 (P3 , OP3 (1)) −−→ H 0 (P3 , OP3 (3)) → H 0 (Q, OQ (3)) → 0. So dim H 0 (Q, OQ (3)) = dim H 0 (P3 , OP3 (3)) −
dim H 0 (P3 , OP3 (1)) − 1 = 20 − 4 − 1 = 15. So the moduli of nonhyperelliptic curve of genus 4 has dimension
15 + 9 quotient by PGL(3), which has dimension 15, hence has dimension 9.
Unique Trigonal Case. By example 5.5.2., any curve of genus with only one g13 is the complete intersection
of a quadric cone and a cubic hypersurface in P3 , and vice versa. Since any quadric cone in P3 corresponds
to a symmetric 4 × 4 matrix of rank 3 up to scalar, we know the dimension of moduli of quadric cone has
dimension 8. Then similarly with the non‑hyperelliptic case, the moduli of curves of genus 4 with unique
trigonal has dimension 15 + 8 − 15 = 8.

Yang Pi‑Yeh 95 Hartshorne Solutions

4 Curves 4.5 The Canonical Embedding

Solution 4.5.4. (a) (⇒) We may assume X is in P3 . If X is a nonhyperelliptic curve with two g13 , they gives
two morphism φ1 and φ2 from X to P1 of degree 3. Then we define φ = φ1 × φ2 , and Y = Imφ. So Y ⊂ P1 × P1
has form (a, b) for some a, b by example 6.6.1. in chapter II. Denote deg φ = d, we have da = 3 and db = 3
since they are from g13 ’s. If a = b = 1, clearly two g13 are the same, which makes a contradiction. So a = b = 3
and d = 1, hence X  Y. Fix a point P0 ∈ X, denote π as the projection from P0 to P2 . If π(P) = π(Q) for two
P, Q, the trisecant PQP0 must lie in X. But on X, there exist only two lines through P0 , so π(X) has only two
singularities. If there exists some quatersecant T of X, since X is the complete intersection of a quadric surface
and a cubic surface, T intersect the quadric surface at 4 points, which contradicts with Bézout’s theorem. So
X does not have any quatersecant. So π(X) has only nodes. In this case π(X) is a plane curve of genus 4 with
two nodes, clearly it has degree 5.
(⇐) If X is a plane quintic curve with two nodes, its normalization X̃ must have genus 4. Since X has
degree 5, any line passing through one of nodes of X meets X in 4 points by Bézout’s theorem. So we get two
g13 ’s corresponding to two nodes of X. If X̃ is hyperelliptic, it has a g12 and also a g13 . So we have two morphism
φ2 and φ3 from X to P1 with degree 2 and 3. Then if φ = φ2 × φ3 has degree d and image Y corresponding to
form (a, b) in P1 × P1 . So da = 2 and db = 3, i.e. d = 1. So X̃ is isomorphic to the curve (2, 3) in P1 × P1 , which
has genus 6 and makes a contradiction. Hence X is nonhyperelliptic of genus 4 with two g13 ’s.
(b) By example 5.5.2., X is the complete intersection of an irreducible quadric cone C and a cubic surface
F. Fix a point P0 ∈ X and denote π as the projection from P0 . If P0 lies in a trisecant T of X, by Bézout’s
theorem, T ⊂ C. Since for P0 , there exists only one line passing through P0 and lying in C. So π(X) has only
one singularity. If there exists a quatersecant T 0 of X, similarly with (a) it makes a contradiction. So X has
only trisecants. Moreover, since 4 = 21 (d − 1)(d − 2) − δ with δ ≤ 2, we have d = 5 and r = 2, i.e. π(X) is a quintic
plane curve with only one tacnode.
If we have a plane curve X with degree d < 6 and only r nodes of genus 4, we have 4 = 12 (d − 1)(d − 2) − r,
so the only solution is d = 5 and r = 2. By (a), the normalization of X is a curve of genus 4 with two g13 ’s,
which makes a contradiction.

Solution 4.5.5 (Curves of Genus 5). (a) Clearly the moduli of quadric hypersurface in P4 has dimension 14.
If we fix a quadric hypersurface H1 and H2 which intersect completely at a surface S , if a third hypersurfaces

H3 intersect completely with S , we have S * H3 . So, the moduli of this kinds of H3 has dimension 14 − 42 = 8.
Hence the dimension of the moduli of this kinds of curves of genus 5 is 14 + 14 + 8 up to the action of PGL(4),
which has dimension 24, i.e. the dimension of the moduli of this kinds of curves of genus 5 is 14 + 14 + 8 − 24 =
12.
(b) If X has g13 , we may take a D ∈ g13 and have l(D) = 2. Then deg(K − D) = 8 − 3 = 5. By Riemann‑Roch,
we have l(K − D) = l(D) + g − deg(D) − 1 = 3. So K − D ∈ g25 , which maps X to P2 . Since 5 = 21 (5 − 1)(5 − 2) − 1,
clearly X is represented as a plane quintic curve with one node. Conversely, if f : X → P2 with the image
as a plane quintic curve with one node, we have OX (E) = f ∗ OP2 (1) for some divisor E. So deg(E) = 5, and
deg(K−E) = 8−5 = 3. Then by Riemann‑Roch, we have l(E)−l(K−E) = 1. And h0 ( f ∗ OP2 (1)) ≥ h0 (OP2 (1)) = 3, i.e.
l(K − E) ≥ 2. So E is special, and by Clifford’s theorem, dim |E| ≤ 21 deg(E) = 2.5. So 3 ≤ l(E) = dim |E| + 1 ≤ 3.5,
i.e. l(E) = 3, and l(K − E) = 2. Hence K − E is a g13 .
Since the moduli of quintic in P2 has dimension 20, any quintic with a node satisfies a equation more, i.e.
has dimension 20 − 1 = 19. So up to the action of PGL(2), which has dimension 8, the moduli of quintic with
one node has dimension 19 − 8 = 11.
(c) Define V as the blow‑up of P2 at the node. Then by 2.7.7. V is a cubic surface in P4 . So trivially X ⊂ V
and V is the union of all trisecant of X. Hence V and g13 are both unique.

Solution 4.5.6. Suppose X has g13 , and D ∈ g13 . Then by Riemann‑Roch we have dim |K − D| = dim |D| − deg D −
1+g = 3. But |K − D| is the linear system of conics in P2 , which gives a divisor > D on C. Since the linear system

Yang Pi‑Yeh 96 Hartshorne Solutions

4 Curves 4.6 Classification of Curves in P3

of conics in P2 has dimension 5, and as the condition we impose yields 3 linearly independent conditions,
|K − D| has dimension at most 2, which makes a contradiction.
Consider a plane sextic curve Y with four nodes. Then Y has genus 6. And X is the normalization of Y.
By definition, we have a closed embedding Y → P2 → P3 . Hence X has a D ∈ g37 . So deg(K − D) = 10 − 7 = 3,
and dim |K − D| = dim |D| − deg D − 1 + g = 1, i.e. K − D ∈ g13 . So by above, X is a nonhyperelliptic curve of
genus 6 which cannot represent as a smooth plane quintic curve.

Solution 4.5.7. (a) Consider the canonical divisor K of the curve X, it induces the canonical embedding
φ : X → P2 . Then for any σ ∈ Aut(X), we have σ∗ (K) ∼ K. Hence K and σ∗ (K) induce the same embedding,
which differs by an automorphism of P2 , i.e. σ is induced by an automorphism of P2 .
(b) By Klein’s paper as in the hint, AutX  PGL(2, Z/7Z), which is a simple group of order 168.
(c) For any n > 1, if X is a curve with an automorphism σ or order n, we  may assume σ is induced by
ζ1 0 0
τ ∈ AutP2 . By linear algebra, τ is similar to a diagonal matrix in the form 0 ζ2 0 0 for some roots ζ1 , ζ2 of
0 0 1
unity of order n. So changing the coordinates, we may assume σ can be written as x 7→ ζ1 x, y 7→ ζ2 y, z 7→ z. So
if X has the form f (x, y, z) in P2 for some homogeneous polynomial of degree 4, the term of xi y j zk must have
P
coordinate a ni=1 (ζ1i ζ2j )n for some a ∈ k. Since σ is not identity, ζ1 and ζ2 cannot be 1 simultaneously. So at
least one of the coordinates of xz3 and yz3 must be zero. So curves with such automorphism induced by τ
form a family of dimension≤ 8. And changing coordinates there is a 4‑dimensional family of such X. Hence
curves having such automorphism induced by τ form a family of dimension≤ 12 inside the 14‑dimensional
family of all plane curves of degree 4.

4.6 Classification of Curves in P3

Solution 4.6.1. Denote X as the curve we need. Consider an exact sequence 0 → IX (2) → OP3 (2) → OX (2) →
0. Denote D as the hyperplane section of X. Then dim |2D| = dim |K − 2D| + 9 by Riemann‑Roch. Since

deg(K − 2D) < 0, we have dim |K − 2D| = 0. So h0 (OX (2)) = 9. Since h0 (OP3 (2)) = 53 = 10, we have h0 (IX (2)) ≥ 1,
i.e. X lies on a quadric surface Q. If X is contained in two quadric surfaces, by 2.8.4. we have g(X) = 1 , 0,
which contradicts, i.e. the quadric surface is unique. Finally, since X is a rational curve, it has n + 1 linearly
independent points in Pn . So Q is nondegenerate, hence nonsingular.

Solution 4.6.2. Denote X as the curve we need. Consider an exact sequence 0 → IX (3) → OP3 (3) → OX (3) →
0. Denote D as the hyperplane section of X. Then dim |3D| = dim |K − 3D| + 16 by Riemann‑Roch. Since

deg(K −3D) < 0, we have dim |K −3D| = 0. So h0 (OX (3)) = 16. Since h0 (OP3 (3)) = 63 = 20, we have h0 (IX (3)) ≥ 1,
i.e. X lies on a cubic surface S .
Consider the curve X = {(s5 : s4 t : st4 + s3 t2 : t5 ) | (s : t) ∈ P1 }. If quadric surface Q = (u11 x12 + u22 x22 + u33 x32 +
u44 x42 + u12 x1 x2 + u13 x1 x3 + u14 x1 x4 + u23 x2 x3 + u24 x2 x4 + u34 x3 x4 contains X, we have

u11 s10 +u12 s9 t+(u13 +u22 )s8 t2 +u23 s7 t3 +(u13 +u33 )s6 t4 +(u14 +u23 )s5 t5 +(u24 +2u33 )s4 t6 +u34 s3 t7 +u33 s2 t8 +u34 st9 +u44 t10 = 0

So if X ⊂ Q, we have u11 = u12 = u13 + u22 = u23 = u13 + u33 = u14 + u23 = u24 + 2u33 = u34 = u33 = u34 = u44 = 0, i.e.
all ui j = 0, which contradicts. So X is not contained in any quadric surface.

Solution 4.6.3. Denote X as the curve we need. Consider an exact sequence 0 → IX (2) → OP3 (2) → OX (2) →
0. Denote D as the hyperplane section of X. Then dim |2D| = dim |K − 2D| + 9 by Riemann‑Roch. Since

deg(K − 2D) < 0, we have dim |K − 2D| = 0. So h0 (OX (2)) = 9. Since h0 (OP3 (2)) = 53 = 10, we have h0 (IX (2)) ≥ 1,
i.e. X lies on a quadric surface Q. If this Q is nonsingular, then by remark 6.4.1.(d), we know X is a curve on
xy = zw of the type (2, 3). So X is projectively normal. Then by 2.5.14.(d), Q is unique. If Q is singular, then
clearly X is passing through the vertex of the cone. So if X is lying on another quadric Q0 , it must be a cone.
Since X is smooth, Q and Q0 must coincide, i.e. Q = Q0 .

Yang Pi‑Yeh 97 Hartshorne Solutions

4 Curves 4.6 Classification of Curves in P3

If X is an abstract curve of genus 2, we have deg K = 2. So consider the divisor D corresponding to

OX (1). Then by 4.3.1., |D| induces an embedding X → Pr iff deg D ≤ 5. So for any D with degree 5, we have
deg(D − 2K) = 1. If D − 2K is not effective, we have dim |D − 2K| = 0, hence E = D − K has degree 2, and l(E) = 2.
So divisors in |E| are contained in two planes, hence spans a line which is the intersection of two planes. So
this line is contained in Q since it meets X in 3 points by Bézout’s theorem. Since |D − 2E| = |2K − D| is empty,
this two different lines will not meet. So Q is a smooth quadric. If D − 2K is effective, we must have |D − 2K|
is just a point, this is the vertex of the Q.

Solution 4.6.4. If X is a curve of degree 9 and has genus 11 in P3 . Consider the exact sequence 0 → IX (2) →
OP3 (2) → OX (2) → 0. By Riemann‑Roch, we have h0 (2D) − h0 (K − 2D) = 8, where D is the hyperplane section.
If 2D is nonspecial, then X is clearly contained in a quadric surface. If 2D is special, we have h0 (2D) > 8.
And by Clifford’s theorem, dim |2D| ≤ 21 deg 2D = 9. So h0 (2D) < 10, and X is lying on a quadric surface
Q. So if Q is nonsingular, by remark 6.4.1., the equations 9 = a + b, 11 = ab − a − b + 1 have no integer
solution. If Q is the product of two hyperplanes, then X must be lying on one hyperplane, i.e. a plane curve.
So g(X) = 12 (9 − 1)(9 − 2) = 28 , 11, which makes a contradiction. If Q is a quadric cone, by remark 6.4.1.(d),
X must have the form (4, 5), which has genus 12 , 11 and makes a contradiction.

Solution 4.6.5. Since X is smooth, it is normal, hence projectively normal. So by 2.5.14., H 0 (P3 , OP3 (l)) →
  l+3−a l+3−b l+3−a−b
H 0 (X, OX (l)) is surjective for all l ≥ 0, and H 1 (X, OX (l)) = 0. So by 2.8.4., h0 (OX (l)) = l+3 − 3 − 3 + 3 .
m+3−a m+3−b m+3−a−b m+3 3
Since m < min{a, b}, we have 3 = 3 = 3
= 0, i.e. h0
(O X (l)) = 3
= h 0
(OP3 (l)). Hence we have
h0 (IX (m)) = 0.

Solution 4.6.6. Case d = 6: By Castelnuovo’s bound, we have g ≤ 4. If g = 0, X is a rational curve, so in

a plane. If g = 1, X is clearly a plane cubic curve. If g = 2, by proposition 6.3., the hyperplane section D

is nonspecial. So h0 (OX (1)) = deg D + 1 − g = 5. But h0 (OP3 (1)) = 41 = 4, which contradicts to the fact that
OP3 (1) → OX (1) is surjective. Hence by all above, g = 3, 4.
Case d = 7: By Castelnuovo’s bound, we have g ≤ 6. If g = 0, X is a rational curve, so in a plane. If
g = 1, 2, 3, by proposition 6.3., the hyperplane section D is nonspecial. So h0 (OX (1)) = deg D + 1 − g = 8 − g.

But h0 (OP3 (1)) = 41 = 4, which contradicts to the fact that OP3 (1) → OX (1) is surjective. If g = 4, we may
consider the divisor 2D, which has degree 12, hence nonspecial. So h0 (OX (2)) = deg D + 1 − g = 9. But

h0 (OP3 (2)) = 52 = 10. Hence X is lying on a quadric surface Q. So it has type (a, b) on Q. Since a + b = d = 7
and ab − a − b + 1 = g = 4, we have (a, b) = (2, 5). But by 3.5.6.(b), X is not projectively normal, which
contradicts. Hence by all above, g = 5, 6.

Solution 4.6.7. If X is a line or a conic, or a twisted cubic curve in P3 , it is clearly have no multisecants. If X
is a elliptic quartic curve in P3 , we pick a point P ∈ X and denote the projection from P to P2 as π. Consider
the π(X). It is a plane curve of degree 3 and genus 1, so clearly smooth. Hence X have no multisecants.
For any other smooth curve X in P3 , if X is a plane curve, it must have degree≥ 3 since it is not a line or
a quadric curve. So by Bézout’s theorem, any line in P2 no a tangent line is a multisecant. If X is not a plane
curve has have no multisecants, for any point P ∈ X and the corresponding projection π, we know π(X) is a
smooth plane curve of degree d − 1, hence has genus g = 21 (d − 2)(d − 3). So X has the same genus. Then by
Castelnuovo’s bound, we have 21 (d − 2)(d − 3) ≤ 14 d2 − d + 1, i.e. 2 ≤ d ≤ 4. If d = 2 or 3, we have g = 0, i.e. a
conic or a twisted cubic curve, then we have a contradiction. If d = 4, we have g = 0 or 1. If g = 1, we know
X is a elliptic quartic curve, then contradicts. If g = 0, it is a rational quartic as the type of (1, 3) on a smooth
quadric surface, hence clearly has infinitely trisecants as lines.

Solution 4.6.8. (⇒) If X has a nonspecial divisor D of degree d with no base points, we have dim |D| = d − g
by Riemann‑Roch. And since D has no base points, we have dim |D| > 0, hence d ≥ g + 1.
(⇐) If E is a effective special divisor of degree d − 1, since dim |K| = g − 1, and E is a subset of K, we
know the choices of E form a subset of X d−1 of dimension≤ g − 1. Hence {E + P} for any point of P ∈ X form

Yang Pi‑Yeh 98 Hartshorne Solutions

4 Curves 4.6 Classification of Curves in P3

a subset of X d of dimension ≤ g. Since all effective divisors of degree d ≥ g + 1 form the whole set of X d ,
which has dimension d ≥ g + 1, we can pick an effective divisor D of degree d such that for all P ∈ X, we have
D − P is an effective nonspecial divisor of degree d − 1. So clearly D is nonspecial, and by Riemann‑Roch,
dim |D − P| = dim |D| − 1 for all P, i.e. D have no base points.

Solution 4.6.9. Consider the blow‑up π : P̃ → P3 of P3 at X. Denote Y = π−1 (X). Consider the sheaf IY ⊗
π∗ OP3 (m) on P̃. We have π∗ (IY ⊗ π∗ OP3 (m))  π∗ (IY )(m). So by Serre’s theorem, it has global section for m  0,
hence IY ⊗ π∗ OP3 (m) has global section for m  0. For any s ∈ Γ(P̃, IY ⊗ π∗ OP3 (m)), we have Y ⊂ Supp s. Hence
S = π(Supp s) is a surface of degree m in P3 , and X = π(Y) ⊂ S .

Yang Pi‑Yeh 99 Hartshorne Solutions

5 Surfaces

5 Surfaces
5.1 Geometry on a Surface
Solution 5.1.1. By Riemann‑Roch, we have χ(L −1 ) = 12 C.(C − K) + χ(OX ), χ(M −1 ) = 12 D.(D − K) + χ(OX ) and
χ(L −1 ⊗ M −1 ) = 12 (C + D).(C + D − K) + χ(OX ). So χ(OX ) − χ(L −1 ) − χ(M −1 ) + χ(L −1 ⊗ M −1 ) = 21 ((C + D).(C +
D − K) − C.(C − K) − D.(D − K)) = C.D.

Solution 5.1.2. By Riemann‑Roch, we have χ(nH) = 12 (nH).(nH − K) + 1 + pa . So by definition of Hilbert

polynomial, we have P(n) = 21 n2 H 2 − 12 nH.K + 1 + pa for n >> 0, hence a = H 2 , c = 1 + pa , i.e. the degree
of X in PN is H 2 . For b, by adjunction formula, we have H.(H + K) = 2π − 2, i.e. H.K = 2π − 2 − H 2 . So
b = 21 H 2 + 1 − π. For any curve C in X, by Bertini’s theorem, there exists a hyperplane S satisfying S intersect
C and X transversally, hence C.H = ](C ∩ H) = ](C ∩ (X ∩ S )) = ](C ∩ S ) = deg C by Bézout’s theorem.

Solution 5.1.3. (a) By Riemann‑Roch, χ(OD−1 ) = 12 (−D).(−D − K) + 1 + pa . Hence χ(OD ) = χ(OX ) − χ(OD−1 ) =
− 21 D.(D + K), i.e. 2pa (D) − 2 = D.(D + K).
(b) Since χ(OD ) depends only on thel inear equivalent class of D on X, so does pa (D).
(c) For the first one, pa (−D) = 1 + 12 (−D).(−D + K) = 1 + D2 − 12 D.(D + K) = D2 − pa (D) + 2. For the second
one, pa (C + D) = 1 + 12 (C + D).(C + D + K) = 1 + 12 (C.(C + K) + D.(D + K)) + C.D = pa (C) + pa (D) + C.D − 1.

Solution 5.1.4. (a) Since we have an exact sequence 0 → I /I 2 → ΩP3 → ΩX → 0, where I is the ideal sheaf
of X in P3 , we have ωX = ωP3 ⊗ det(I /I 2 )0 , i.e. KX = (KP3 + X)|X . Since X has degree d, we have X ∼ dH, and
obviously KP3 = −4H, we have KX = (d − 4)H, where H is a hyperplane in P3 . So by adjunction formula, we
have −2 = 2g(C) − 2 = C.(C + KX ). And by 5.1.1., C.H = deg C = 1. So C 2 = 2 − d.
(b) Consider the surface X defined by f (x, y, z, w) = xzd−1 + xd−1 z−ywd−1 −yd−1 w in P3 . Then X is nonsingular,
and contains the line x = y = 0.

Solution 5.1.5. (a) As we’ve done in 5.1.3., K ∼ (d − 4)H. So K 2 = (d − 4)2 H 2 . And by 5.1.2., H 2 = d. Hence
K 2 = d(d − 4)2 .
(b) Denote p1 and p2 are the projections from X = C ×C 0 to C or C 0 . Then by 2.8.3., KX = p∗1 KC + p∗2 KC 0 . Since
C has genus g, we have deg KC = 2g − 2, i.e. KC is the sum of 2g − 2 points on C. So p∗1 KC ∼ (2g − 2)(C 0 × {pt}) =
(2g − 2)C 0 , and similarly p∗2 KC 0 = (2g0 − 2)C. Clearly by definition, we have C 2 = C 02 = 0, and C.C 0 = 1. Hence
KX2 = (p∗1 KC )2 + 2p∗1 KC .p∗2 KC 0 + (p∗2 KC 0 )2 = (2g − 2)2 (C 2 ) + 2(2g − 2)(2g0 − 2)(C.C 0 ) + (2g0 − 2)2 (C 02 ) = 8(g − 1)(g0 − 1).

Solution 5.1.6. (a) Clearly ∆.(C×{pt}) = ∆.({pt}×C) = 1. So as we’ve done in 5.1.5.(b), ∆.KX = ∆.(p∗1 KC + p∗2 KC ) =
4g − 4. Hence by adjunction formula, 2g − 2 = ∆.(∆ + KX ), i.e. ∆2 = 2g − 2 − (4g − 2) = 2 − 2g.
(b) By 5.1.5. and (a), l2 = m2 = 0, l.m = l.∆ = m.∆ = 1, ∆2 = 2−2g. So if al+bm+c∆ ∼Num 0 for some constants
a, b, c, we have l.(al + bm + c∆) = m.(al + bm + c∆) = ∆.(al + bm + c∆) = 0, i.e. b + c = a + c = a + b + c(2 − 2g) = 0.
Hence a = b = c = 0, which means l, m, ∆ are linearly independent.

Solution 5.1.7 (Algebraic Equivalence of Divisors). (a) If D ∼Alg 0 and E ∼Alg 0, we have two sequence
0 = D0 , D1 , . . . , Dn = D and 0 = E0 , E1 , . . . , Em = E. Then we have a sequence 0 = D0 , . . . , Dn = Dn + 0 =
Dn + E0 , Dn + E1 , . . . , Dn + Em = D + E, i.e. D + E ∼Alg 0. Moreover, since Di ∼PreAlg Di+1 , by definition we have
Di+1 ∼PreAlg Di , we have a sequence 0 = −D0 , −D1 , . . . , −Dn = −D. So all divisors∼Alg 0 form a subgroup of
Div X.
(b) We just need to prove that for all f ∈ K(X)∗ , ( f ) ∼Alg 0. Consider the divisor (t f − u) on X × P1 , where
t, u are the homogeneous coordinates on P1 . Then if we write ( f ) = D+f − D−f and (t f − u) = D+(t f −u) − D−(t f −u) , by
example 9.8.5. on chapter II, D+(t f −u) and D−(t f −u) are flat over T . Since D+(t f −u) and D−(t f −u) restricts to D+f and D−f
over (1, 0) and 0 over (0, 1), we have ( f ) ∼Alg 0.

Yang Pi‑Yeh 100 Hartshorne Solutions

5 Surfaces 5.1 Geometry on a Surface

(c) As the proof of theorem 1.1., every divisor is the difference of two very ample divisors. So we just
need to prove that if D ∼Alg D0 , we have D.H = D0 .H. And by definition, we only need to consider the
case D ∼PreAlg D0 . For any very ample divisor H of X, it induces an embedding X → Pn . Then we have an
embedding X × T → PnT . For any divisor E ⊂ X × T with fibres D = E0 and D0 = E1 , since E is flat over T , by
theorem 9.9. in chapter III, deg D = deg D0 , i.e. D.H = D0 .H.
Solution 5.1.8 (Cohomology Class of a Divisor). (a) Clearly, any element in Pic X is an equivalent class, so
we may assume D and E are meeting transversally. And similarly we may assume D is very ample since
every effective divisor is the difference of two very ample divisor. Fix a very ample divisor D. Then we have
a morphism f : H 1 (X, ΩX ) → H 1 (D, ΩD )  k as f (s) = i∗ (s), where i : D → X is the embedding. So by Serre
duality and the definition of c, we have f (F ) = hc(D), F i. Hence we have a commutative diagram

Pic X
cX
/ H 1 (X, ΩX )

res f
 
Pic D
cD
/ H 1 (D, ΩD )

Since res(L (E)) = L (E) ⊗ OD , and by 3.7.4., we have cD (F ) = (deg F ) · 1, we have hc(D), c(E)i = f (cX (E)) =
cD (g(E)) = cD (LE ⊗ OD ) = degD (LE ⊗ OD ) = (D.E) · 1 by lemma 1.3.
(b) Suppose D1 , . . . , Dn ∈ Pic X are numerically independent. If there exists i1 , . . . , in such that c(D1 )i1 ⊗
. . . ⊗ c(Dn )in = 1, i.e. hc(D1 )i1 ⊗ . . . ⊗ c(Dn )in , F i = 0 for all F . Take F = c(E) for any divisor E of X. Then we
P
have 0 = hc(D1 )i1 ⊗ . . . ⊗ c(Dn )in , c(E)i = ( ik Dk ).E. Since D1 , . . . , Dn ∈ Pic X are numerically independent, we
have ik = 0 for all k. So c(D1 ), . . . , c(Dn ) ∈ H 1 (X, ΩX ) are linearly independent. Since dimk H 1 (X, ΩX ) < ∞, we
have NumX < ∞.
Solution 5.1.9. (a) Define E = (H 2 )D − (H.D)H. Then we have E.H = (H 2 )(D.H) − (H.D)(H 2 ) = 0. So by Hodge
index theorem, we have 0 ≥ E 2 = (H 2 )2 (D2 ) − (H 2 )(H.D)2 . Since H 2 > 0, we have D2 H 2 ≤ (H.D)2 . And the
equality holds iff D ∼Num 0.
(b) Define H = l + m and E = l − m. Then we have D.H = a + b, D.E = a − b, H 2 = 2, E 2 = −2 and H.E = 0.
For any curve C 00 ⊂ X, if C 00 do not intersect any C × {pt} for any pt ∈ C 0 , clearly C 00 ⊂ C × {pt} for some pt ∈ C 0 .
Then C 00 must intersect {pt} × C 0 for some pt ∈ C. Since H 2 = 2, by Nakar Moishezon’s theorem we have
H is ample. So if we define D0 = (H 2 )(E 2 )D − (E 2 )(D.H)H − (H 2 )(D.E)E = −4D + 2(a + b)H − 2(a − b)E, then
D0 .H = −4(a + b) + 4(a + b) − 0 = 0. Hence by (a) we have (D02 )(H 2 ) ≤ (D0 .H)2 = 0. So 0 ≥ D02 = 16D2 − 32ab,
i.e. D2 ≤ 2ab. And moreover D2 = 2ab iff we have D0 ∼Num 0, i.e. D ∼Num bl + am.
Solution 5.1.10 (Weil’s Proof (2) of the Analogue of the Riemann Hypothesis for Curves). Since Γ is the
preimage of ∆ by morphism Frob × id, we have Γ2 = ∆2 · deg Frob · deg id = q(2 − 2g). And by definition we
clearly have Γ.∆ = N. Moreover, define l = C × {pt} and m = {pt} × C. Then Γ intersect l at f −1 (pt) × pt, and
intersect m at pt × f (pt). So Γ.l = q and Γ.m = 1. And clearly ∆.l = ∆.m = 1. Define D = rΓ + s∆, then D.l = rq + s
and D.m = r + s. So by 5.1.9., 2(rq + s)(r + s) ≥ D2 = rq(2 − 2g) + s2 (2 − 2g) + 2rsN, i.e. |N − q − 1| ≤ rs gq + rs g for
p √
all r, s. So |N − q − 1| ≤ rs gq + rs g ≤ 2 rs gq · rs g = 2g q.
Solution 5.1.11. (a) If d ≤ 0, D = C1 + . . . + Cn such that D.H = d ≤ 0, by Nakai‑Morshezon theorem, we have
Ci .H > 0, hence D.H > 0, which makes a contradiction. So we may assume d > 0. Since NumX has finite rank,
and so does H ⊥ . Then take an orthogonal basis of H ⊥ as R1 , . . . , Rk . Fix an effective divisor D with D.H = d.
P
Then for any ai ∈ Z, we have (D + ai Ri ).H = d. Denote n as the number of curves involved in the sum of
P P P
D and Ri . In order to make D + ai Ri to be an effective divisor, we may assume D + ai Ri = Ci for finite
P P P P
sum of≤ n ai = N terms. Then (D + ai Ri ).(D + ai Ri + K) ≥ Ci .(Ci + K) ≥ −2N. Hence i, j ai a j Ri .R j +lower
terms> 0. Since Ri .R j = 0 if i , j, and R2i < 0 because H is ample, the inequality above has only finite solution
of ai . So the number of D such that D.H is finite.

Yang Pi‑Yeh 101 Hartshorne Solutions

5 Surfaces 5.2 Ruled Surfaces

(b) For any σ, σ0 ∈ AutC, we denote Γ, Γ0 are their graphs. Since σ , σ0 , we have Γ , Γ0 , i.e. Γ.Γ0 ≥ 0.
Since σ and σ0 are both automorphism, we have ∆ → Γ as id × σ or ∆ → Γ0 as id × σ0 has degree 1, hence
Γ2 = Γ02 = ∆2 = 2 − 2g < 0. So Γ2 , Γ.Γ0 , i.e. Γ is not numerical equivalent to Γ. Define H = l + m from 5.1.9.,
we have Γ.H = 2 for all graph of automorphism Γ. Then the number of choice of Γ is finite by (a), hence AutC
is finite.

Solution 5.1.12. The first proposition is trivial by Nakai‑Moishezon criterion. For second, consider a curve
C with genus> 2. A divisor D with deg D = 2g is very ample iff for any P, Q ∈ C we have l(D − P − Q) = l(D) − 2.
Then by Riemann‑Roch, l(D) − l(K − D) = g + 1, l(D − P − Q) − l(K − D + P + Q) = g − 1. So D is very ample iff
D is not of the form K + P + Q. So there exists D is very ample and D0 is just ample with deg D = deg D0 = 2g.
Then in X = C × P1 , the divisor D × P1 and D0 × P1 are numerically equivalent, and the first is very ample but
second is just ample.

5.2 Ruled Surfaces

Solution 5.2.1. Since X is a birationally ruled surface, there exists a function field L such that K(X) = L⊗k k(x) =
L(x). Then by corollary 6.12. in chapter I, there exists a unique smooth projective curve C such that K(C) = L,
i.e. C is unique one such that X is birationally equivalent to C × P1 .

Solution 5.2.2. (⇒) If E is decomposable, i.e. E  L ⊕ L 0 , clearly the sections corresponding to L and L 0
has no intersection.
(⇐) If X has two sections C and C 0 such that C ∩ C 0 = ∅, we have two exact sequences E → L → 0
and E → L 0 → 0 corresponding to C and C 0 with kernels K and K 0 . Then since C ∩ C 0 = ∅, we have
K ⊕ K 0 → E is injective. Since the rank is the same, we have E  K ⊕ K 0 .

Solution 5.2.3. (a) Consider X = P(E ) and π : X → C. The generic fibre Xη  Pr−1 K , hence has a rational
point. Let ξ be a rational point in the generic fibre, and define S = {ξ} ⊂ X. Then π|S : S → C is clearly a
proper birational morphism. Since C is nonsingular, we have S  C. Consider the ideal sheaf I of S , we
have 0 → I (1) → OX (1) → OS (1) → 0. So 0 → π∗ (I (1)) → E → OS (1) on C. Denote ψ : E → OS (1). For
any open subset U ⊂ C such that E |U is free, clearly ψ|U is surjective. Hence ψ is surjective. Then we define
Er−1 = π∗ (I (1)) and Lr = OS (1), we have 0 → Er−1 → Er → Lr → 0. Clearly Er−1 and Lr are locally free of
rank r − 1 and 1. So by induction we have a sequence 0 = E0 ⊂ E1 ⊂ . . . ⊂ Er = E .
(b) If Ω is an extension of invertible sheaves, we may assume 0 → OX (a) → Ω → OX (b) → 0. Since
H 1 (X, OX (a)) = H 1 (X, OX (b)) = 0, we have h0 (Ω) = h0 (OX (a)) + h0 (OX (b)), h1 (Ω) = 0, and h2 (Ω) = h2 (OX (a)) +
h2 (OX (b)). By theorem 8.13. in chapter II, we have 0 → Ω → OX (−1)3 → OX → 0, hence similarly, h0 (Ω) +
h1 (Ω) = h0 (OX (−1)3 ) − h0 (OX ) = 0 − 1 = −1. So h0 (Ω) = −1, which is impossible.

Solution 5.2.4. (a) Since X = C × P1 , by example 2.11.1., we have e = 0 and C0 = C, hence by corollary 2.11.,
we have K ∼Num −2C + (2g − 2) f . For any section D, we may assume D = C + r f . Then by adjunction formula,
D2 = 2g − 2 − D.K = 2g − 2 − (2g − 2 − 2r) = 2r. So D2 is always an even integer. If r = 0, we just take D = C. If
d ≥ g + 1, by 4.6.8., there exists a nonspecial divisor E on C such that |E| has no base points. Then D = C + E f
is a section with D2 = 2r.
(b) (Here I think if C is hyperelliptic, r = 2 or 3 are both possible.) If X has a section D ∼Num C + b f , then
the composition C → D → P1 gives a morphism gD : C → P1 . And by definition deg gD = C.D = b. Conversely,
if we have a morphism g : C → P1 with deg g = d, the graph Dg = Γg ⊂ X is a section, and Dg .C = d, hence
Dg = C +d f . Since g(C) = 1, there are no morphism g : C → P1 with deg g = 1, hence r , 1. If C is hyperelliptic,
i.e. C has a g12 , hence r = 2 is possible. And C has a g13 by adding a point in g12 , and r = 3 is also possible. And
if C is nonhyperelliptic, by example 5.5.2., C has a g13 but no g12 , hence r = 3 is possible, but 2 is not.

Yang Pi‑Yeh 102 Hartshorne Solutions

5 Surfaces 5.2 Ruled Surfaces

Solution 5.2.5 (Values of e). (a) As we’ve done in theorem 2.12., we only need to prove Ext1 (L , OC ) is
nontrivial for deg L = −e when 0 ≤ e ≤ 2g − 2. Since Ext1 (L , OC )  H 1 (C, Lˇ )  H 0 (X, ωC ⊗ L ), and
deg ωC ⊗ L = 2g − 2 − e ≥ 0, hence Ext1 (L , OC ) is nontrivial. So we may pick a nontrivial extension E , which
is indecomposable.
(b) (⇒) If E is normalized, the morphism γ : H 0 (L (D + K − E)) → H 1 (L (−E)) is injective. So for the
ξ ∈ H 1 (L (−D)), we have
H 0 (C, OC )
δ
/ H 1 (C, L (−D))

α β
 γ

H 0 (C, L (D + K − E)) / H 1 (C, L (−E))
where δ(1) = ξ, α(1) = t and t is the section defining the divisor D + K − E, β is induced from α. So β is dual
to β0 : H 0 (C, L (E)) → H 0 (C, L (D + K)) induced by t. So for any nonzero element e ∈ H 0 (C, L (E)), β0 (e) is a
section of H 0 (C, L (D + K)) corresponding to D + K = E + (D + K − E). So varying E and (D + K − E), we can take
all divisors in |D + K|. So β0 is surjective, hence LE ⊂ |D + K|. If LE ⊂ H, we have β(ξ) = 0, which contradicts to
the injectivity of γ. Hence LE * H.
(⇐) If E is not normalized, there exists some E such that H 0 (E ⊗ L (−E)) , 0. Hence the map γ is not
injective, i.e. LE ⊂ ker(ξ) = H.
(c) By (b), we only need to find a ξ such that H = ker(ξ) does not contain any |D + K − E| + E. By Riemann‑
Roch, we have h0 (D + K) = d + g − 1, and h0 (D + K − E) = g. Then consider B ⊂ C d−1 × H 1 (C, L (−D)) consisting
all pair (E, ξ) such that LE ⊂ ker(ξ). Then B → C d−1 is surjective, and for any effective divisor E of degree d − 1,
dim BE = (d + g − 1) − g = d − 1. So B has dimension (d − 1) + (d − 1) = 2d − 2 < d + g − 1, hence there must exist
some ξ ∈ H 0 (C, L (D + K))0 such that ξ < BE for any effective divisor E.
(d) If e < −g = −2, by definition, we know if E , E 0 , we have LE , LE 0 since dim LE = dim LE 0 = 1 ≤ d − 2.
So all LE ’s form the Grassmannian G2d+1 , which has dimension 2(d − 1) since d > 2. Since all ker(ξ) form the
Grassmiannian Gdd+1 , which has dimension d. By theorem 2.12., we have e ≥ −2g = −4, i.e. 3 ≤ d ≤ 4. So we
have d ≤ 2(d − 1). So for any ξ, there exists at least one E such that LE ⊂ ker(ξ). Hence no one E is normalized,
which is impossible.

Solution 5.2.6. For some locally free sheaf E of rank r on P1 , by theorem 8.8.(c) in chapter III we have
H 1 (P1 , E (n)) = 0 for n  0. Then by Serre duality, we know H 0 (P1 , E (−n)) = 0 for n  0. So we may as‑
sume i is the largest integer such that E (−i) has a global section, which induces a morphism 0 → O → E (−i).
Hence 0 → O(i) → E → F → 0 for some cokernel F locally free of rank r − 1. By induction, we may assume
Lr−1
F = O(n j ). If n j > i, there exists a surjection E (−n j ) → O → 0, which means E (−n j ) has a global section
j=1 L
and contradicts to the definition of i. Hence all n j ≤ i. Since Ext1 (F , O(i)) = Ext1 (O(n j ), O(i)) = 0, the
L
extension is trivial, i.e. E = O(i) ⊕ ( O(n j ))).

Solution 5.2.7. By the proof in theorem 2.12., E is indecomposable with e = −1. And by theorem 2.15., this
kind of ruled surface X is unique. Fix the normalized E and a point P0 ∈ C with section E → L (P0 ) → 0.
Then any section corresponds to a surjection E → L (d) → 0. Then by theorem 2.15., d has degree 1, i.e. d = P
for some P ∈ C. Conversely, by corollary 2.16., there is a 1‑1 correspondence between all E ’s and points in C.
Then any E 0 → L (P) → 0 corresponding to a section E → L ( P+P 2
0
) → 0, since by the proof of theorem 2.15.
0
we have E  E ⊗ L ( 2 ). And L (P + Q) = L (R) for some R ∈ C since C is an elliptic curve. Moreover, if
P−P0

two section C0 and C00 are linearly equivalent, we have C0 ∼ C00 ∼ C0 + (P − Q) f , i.e. P f ∼ Q f , so L (P) = L (Q)
on C. Since C is an elliptic curve, we have P = Q.

Solution 5.2.8. Here we denote µ(E ) = deg E /rank E for any locally free sheaf E of finite rank. Then clearly,
if rank E > 1, we have µ(E ) = µ(E ⊗ L ) or any invertible sheaf L . And if we have 0 → E 0 → E → E 00 → 0,

Yang Pi‑Yeh 103 Hartshorne Solutions

5 Surfaces 5.2 Ruled Surfaces

E 0 00
E
we have deg E = deg E 0 + deg E 00 . So µ(E ) = rank rank
E 0 +rank E 00
µ(E 0 ) + rank rank
E 0 +rank E 00
µ(E 00 ). So the definition of
(semi)stability is equal to for any 0 → F → E , we have µ(F ) < (≤)µ(E ).
(a) If E = F ⊕ G is stable, we may assume µ(F ) ≤ µ(G ). Then consider the morphism E → F → 0, we
have µ(E ) < µ(F ). Since µ(F ) ≤ µ(E ) ≤ µ(G ), we have µ(E ) < µ(F ) ≤ µ(E ), which is impossible.
(b) (⇒) We may assume E is normalized. Then since we have 0 → OC → E → L → 0. Then µ(E ) > (≥
)µ(OC ) = 0, i.e. deg E > (≥)0.
(⇐) Again, we may assume E is normalized. If deg E > (≥)0 and decomposable, for any F ⊂ E , by
theorem 2.12.(a), we have deg F ≤ 0. Hence µ(F ) < (≤)µ(E ), i.e. (semi)stable. If deg E ) > (≥)0 and indecom‑
posable, then the case g(C) = 0 is clear by 5.2.6. So if E is not (semi)stable, there must exist F ⊂ E such that
deg F ≥ (>) 12 deg E > (≥)0, i.e. deg F > 0. So E ⊗ Fˇ has a section. Since deg Fˇ < 0, but since E is normalized,
we have h0 (E ⊗ Fˇ ) = 0, which makes a contradiction.
(c) If E is not semistable, by (b) we have deg E < 0. By the proof of theorem 2.12., we have 0 → OC → E →
L → 0. Then deg L = deg E < 0. So by theorem 2.12., e > 0, i.e. 0 < e ≤ 2g − 2, and deg L = −e. Moreover,
E corresponds to a non‑trivial element in Ext1 (L , OC ) = H 1 (C, Lˇ ) up to scalar. Hence all indecomposable
locally free sheaves E of rank 2 that are not semistable are classified, up to isomorphism, by these three
conditions.

Solution 5.2.9. If Y is the complete intersection of X0 with a surface of degree a ≥ 1, by 2.8.4., we have
deg Y = 2a and g(Y) = 12 × 2 × a(a + 2 − 4) + 1 = (a − 1)2 . For general case, we may treat Y0 as a divisor on X0 .
Blow up X0 at the vertex, we get a ruled surface X over P1 with projection π : X → X0 . Moreover, by example
2.11.4., the e of X is −2. And Y0 has a preimage Y. Clearly, the preimage of the ruling of X0 in X is the section
C0 , so we may assume Y ∼ aC0 + b f . Clearly, b = 2a + 1 since Y is not the complete intersection. Then for
any hyperplane H in P3 , we may denote H̃ = π−1 (H ∩ X0 ) then H̃.C0 = 1 and H̃. f = 2, i.e. H̃ ∼Num C0 + 2 f . So
deg Y0 = (H ∩ X0 ).Y0 = −2a + 2a + b = 2a + 1. And KX ∼Num −2C0 − 4 f , C02 = −2. So by adjunction formula,
2g − 2 = (aC0 + (2a + 1) f )((a − 2)C0 + (2a − 3) f ) = −2a(a − 2) + a(2a − 3) + (2a + 1)(a − 2) = 2(a2 − a − 1), i.e.
g = a2 − a.

Solution 5.2.10. By the proof of corollary 2.19., the embedding is induced by the very ample divisor D =
C0 + n f . Consider the divisor 3C0 + a f for some a. Then it can be written as a curve iff C.C0 ≥ 0, i.e. a ≥ 3e. So
if a = n + e + 2, we have a ≥ 3e, then there is a curve C ∼Num 3C0 + (n + e + 2) f . Since K ∼Num −2C0 − (e + 2) f , by
adjunction formula, 2g − 2 = C.(C + K) = 4n − 2e + 2, i.e. g = 2n − e + 2. And clearly, C has a g13 since C. f = 3.

Solution 5.2.11. (a) Consider 0 → LX (−C0 ) → OX → OC → 0. Since b is nonspecial, we have h1 (LX (b f )) =

h2 (LX (b f )) = 0. Since hi (LX (b f )) = hi (LC (b)) and hi (LX (C0 + b f )) = hi (LC (b + e)). So we have hi (LX (C0 + b f )) =
hi (LC (b)) + hi (LC (b + e)). Denote D = C0 + b f . For any P ∈ C, we have h0 (LX (D − P f )) = h0 (LC (b − P)) +
h0 (LC (b + e − P)) = h0 (LC (b)) − 1 + h0 (LC (b + e)) − 1 = h0 (D) − 2. But as the same proof of theorem 3.1. in
chapter IV, we know D is base points free on P f for some P ∈ C iff h0 (LX (D − P f )) = h0 (LX (D)) − 2, i.e. D is
base points free.
(b) For any P, Q ∈ C, h0 (LX (D − P f − Q f )) = h0 (LC (b − P − Q)) + h0 (LC (b + e − P − Q)) = h0 (LX (D)) − 4. Then
similarly as the same proof of theorem 3.1. in chapter IV, D is very ample.

Solution 5.2.12. If e ≥ 0, then by corollary 3.2. in chapter IV and 5.2.11., this problem is trivial. Hence we
only need to consider the case e = −1.
(a) Here b = deg b ≥ 1. If b ≥ 2, then by corollary 3.2. in chapter IV and 5.2.11., trivial. So we only need to
consider the case b = 1. In this case, we may assume e = E, and b = B. Then h0 (LX (C0 + B f )) = h0 (LC (E + B)) =
2. And for any P ∈ X, h0 (LX (C0 + B f − P f )) = h0 (LC (E + B − P)) − h1 (LX (B f − P f )) = 1 − 1 = 0. Hence by the
same reason with 5.2.11.(a), C0 + b f has no base points.
(b) Here b = deg b ≥ 2. If b ≥ 3, then by corollary 3.2. in chapter IV and 5.2.11., trivial. So we only
need to consider the case b = 2. In this case, we may assume b = A + B. Then h0 (LX (C0 + (A + B) f )) =

Yang Pi‑Yeh 104 Hartshorne Solutions

5 Surfaces 5.2 Ruled Surfaces

h0 (LX (C0 + B f )) + 2 = 4 by (a). And for any P, Q ∈ X, h0 (LX (C0 + (A + B − P − Q) f )) = 0. Hence by the same
reason with 5.2.11.(a), C0 + b f is very ample.

Solution 5.2.13. By 5.2.12.(b), if n ≥ e + 3, any divisor D ∼Num C0 + n f is very ample. Since D. f = 1 and
D2 = 2n − e, we know the embedding corresponding to D is an elliptic scroll of degree d = 2n − e. Moreover,
h0 (LX (D)) = h0 (π∗ LX (D)) = h0 (OC (n) + OC (n − e)) = 2n − e. So the embedding is in Pd−1 . Moreover, if we take
e = −1 and n = 2, we have d = 2n − e = 5, i.e. there is an elliptic scroll of degree 5 in P4 .

Solution 5.2.14. (a) Denote Ỹ as the normalization of Y with natural map $ : Ỹ → Y, and the projection
π : Y → C. If π ◦ $ has degree d, we have 2g(Ỹ) − 2 ≥ d(2g − 2) + deg R. By 4.1.8., we have g(Y) ≥ g(Ỹ), i.e.
2g(Y)−2 ≥ d(2g−2). Moreover, by adjunction formula, 2g(Y)−2 = Y.(Y + K) = (aC0 +b f ).((a−2)C0 +(b+2g−2−
e) f ) = −ae(a−2)+a(b+2g−2−e)+b(a−2), i.e. ae(1−a)+2b(a−2) ≥ (d −a)(2g−2). If 2 ≤ a ≤ p−1, we have d = a,
i.e. b(a − 1) ≥ 12 ae(a − 1), i.e. b ≥ 21 ae. If a ≥ p, we have d = a − np for some n, then (d − a)(2g − 2) = 2np(1 − g).
So b(a − 1) ≥ 12 ae(a − 1) + np(1 − g). Since a − d ≤ a − 1, we have b ≥ 12 ae + 1 − g. Finally, if a = 1, Y is a section
corresponding to E → L → 0, we have deg L ≥ deg E since E is normalized. Since deg L = C0 .Y, we have
b − e ≥ −e, i.e. b ≥ 0.
(b) If D is ample, by Nakai‑Noishezon criterion we have D. f > 0m i.e. a > 0, and D2 > 0, so −a2 e + 2ab > 0,
i.e. b > 12 ae. Conversely, suppose a > 0 and b > 12 ae + ap (g − 1). Since g > 2, a > 0 and b > 12 ae, we have
D. f > 0 and D2 > 0. And D.C0 = −ae + b > − 12 ae > 0. So for any irreducible curve Y , C0 , f , we may denote
Y ∼Num αC0 +β f , then D.Y = −aαe+aβ+αb. If α = 1, by (a) we have β ≥ 0. So D.Y = −ae+aβ+b > −ae+b > 0. If
2 ≤ α ≤ p − 1, by (a) we have β ≥ 21 αe, so D.Y ≤ −aαe + 12 aαe + 21 aeα = 0. If α ≥ p, by (a) we have β ≥ 21 αe + 1 − g,
so similarly D.Y = −aαe + aβ + αb ≥ −aαe + 12 aαe + a − ga + 21 aαe + ap α(g − 1) = a(1 − g)(1 − αp ) ≥ 0. Hence by
Nakai‑Moishezon criterion, D is ample.

Solution 5.2.15 (Funny Behavior in Characteristic p). (a) Denote f = x3 y + y3 z + z3 x. Since C is smooth, we
have g(C) = 12 (d − 1)(d − 2) = 3. Consider the exact sequence 0 → OP2 (−4) → OP2 → OC → 0, we have 0 →

H 1 (C, OC ) → H 2 (P2 , OP2 ) → 0, i.e. H 1 (C, OC )  H 2 (P2 , OP2 ). Hence h1 (OC ) = h2 (OP2 (−4)) = h0 (OP2 (1)) = 31 = 3.
Denote Frob as the Frobenius morphism on P2 . Then we have a diagram

0 / OP2 (−4) / OP2 / OC /0

Frob∗ Frob∗ Frob∗

  
0 / OP2 (−4p) / OP2 / OC p /0

where C p is defined as x9 y3 + y9 z3 + z9 x3 = f 3 = 0. And we have

0 / OP2 (−4p) / OP2 / OC p /0

×f2
  
0 / OP2 (−4) / OP2 / OC /0

So combining these two diagram, we have

H 1 (C, OC )

/ H 2 (P2 , OP2 (−4))

Frob∗ Frob∗
 
H 1 (C p , OCp )

/ H 2 (P2 , OP2 (−4))

 
H 1 (C, OC )

/ H 2 (P2 , OP2 (−4))

Yang Pi‑Yeh 105 Hartshorne Solutions

5 Surfaces 5.2 Ruled Surfaces

Since H 2 (P2 , OP2 (−4)) has a basis {x−2 y−1 z−1 , x−1 y−2 z−1 , x−1 y−1 z−2 } as a free F3 [x−1 , y−1 , z−1 ]‑module. So Frob∗ (x−2 y−1 z−1 ) =
x−6 y−3 z−3 and same for two others, and the image in H 2 (OP2 (−4)) at the bottom row is f 2 · Frob∗ (x−2 y−1 z−1 ) =
y3 3
x6 z
+ x2z5 + x4z y3 + x2y
3 z2 + x2 y2 + yz3 and same for two others. Then any monomial with non‑negative exponent on
2 1

x, y, z is 0, so each of the above three expression is 0, i.e. Frob∗ is identically 0.

(b) Denote H as the hyperplane section. Then we have

0 / OC (−1) / OC / OH /0

Frob∗
  
0 / OC (−3) / OC / OH 3 /0

×f2
  
0 / OC (−1) / OC / OH /0

Hence for any ξ ∈ H 1 (L (−P)), Frob∗ (ξ) = ξ3 , and its image in H 1 (C, OC (−1)) is f 2 · ξ3 . Since every element in
P
H 1 (C, OC ) has the form ai j xi y j for 1 ≤ i < 4 and −i < j < 0 on the affine open subset Uz = (z , 0). So ξ has
P P
the form (ax + by + c)−1 ( ai j xi y j ) on Uz . Hence (ax + by + c)−1 ( ai j xi y j ) has non‑negative exponent for some
suitable (ai j ), i.e. Frob∗ ξ = 0.
(c) Clearly deg π = Y. f = 3. So π is not separable with inseparable degree 3, i.e. separable degree 1,
hence purely inseparable. Then π is an isomorphism on the underlying topological space, i.e. Y is a set‑
theoretic section Ỹ of C lying on the surface. Hence Y = Ỹ p is a curve contained in X. Moreover, e = −1 since
deg L (P) = 1. Then D = 2C0 satisfies a > 0 and b > 21 ae, and C0 .D = (3C0 − 3 f ).(2C0 ) = 6 − 6 = 0, hence D is
not ample by Nakai‑Moishezon criterion.

Solution 5.2.16. Denote R = OC (C). Since C is a nonsingular affine curve, R is a PID. So for any locally free
sheaf E , it must have the form M̃ for some finitely generated projective R‑module M. Since R is a PID, M
is free. So every two locally free sheaves E and E 0 are the same, hence clearly in the same classes in the
Grothendieck group.

Solution 5.2.17. (a) Clearly the 3‑uple embedding is φ : P1 → P3 , (t, u) 7→ (x, y, z, w) = (t3 , t2 u, tu2 , u3 ), and we
have a commutative diagram

0 0

 
OC
(2)
/ OC

(t,u) (x,y,z,w)
 J

OC (1)2 / OC (4)3

(−tu ) κ
 φ 
0 / TC / TP3 |C / NC /0

 
0 0

∂(x,y,z,w) P
where J = ∂(t,u)
. Consider the functor F = n H 0 (C, OC (n)). Denote R(n) as the free k[t, u]‑module whose

Yang Pi‑Yeh 106 Hartshorne Solutions

5 Surfaces 5.3 Monoidal Transformations

generator has degree n. Then F(κ) is just R(−4) ⊕ R(−4) ⊕ R(−4) → R(−3)4 given by matrix
 
 u 0 0 
 
−t 0 2u 
 
 0 u −2t
 
0 −t 0

And F(φ) is R(−4) ⊕ R(−4) ⊕ R(−4) → R(−2) given by (t2 , u2 , 2tu). So ker F(φ) = R(−5) ⊕ R(−5), i.e. ϕ∗ (NC ) =
OP1 (5) ⊕ OP1 (5). So ϕ∗ (I /I 2 ) = OP1 (−5) ⊕ OP1 (−5).
φ
(b) If char(k) , 2, 3, all things are like (a). So 0 → TC →
− TP3 |C → NC → 0, and F(φ) : R(−5)⊕R(−5)⊕R(−6) →
R(−2) given by matrix (t3 , u3 , 6t2 u2 ). So here ker F(φ) = R(−7) ⊕ R(−7), i.e. ϕ∗ (I /I 2 ) = OP1 (−7) ⊕ OP1 (−7). If
char(k) = 2, 3, F(φ) is given by (t3 , u3 , 0). So ker F(φ) = R(−8) ⊕ R(−6), i.e. ϕ∗ (I /I 2 ) = OP1 (−8) ⊕ OP1 (−6).

5.3 Monoidal Transformations

Solution 5.3.1. Denote π : X̃ → X as the blow‑up of X at Y. Then if I is the ideal sheaf of Y in X, we have
E := π−1 (Y) = Proj I /I 2 by theorem 8.24. in chapter II. As the proof of theorem 3.4., we only need to prove
that H i (E, OE (n)) = 0 for all i > 0, n > 0.
Since codim(Y, X) = 1, I /I 2 is an invertible sheaf on Y. So we can take an affine open covering U = {Ui }
of Y such that I /I 2 is free on each Ui . If Ui = Spec Ai , we have π−1 (Ui ) = Proj Ai [x] = P1Ai . Since Y is
separated, we know any U = Ui1 ∩ . . . ∩ Uik is affine as some Spec A, hence π−1 (U) = P1A also. Since V = {Vi },
where Vi = π−1 (Ui ), is an open covering of E. And for any V = Vi0 ∩ . . . ∩ Vik , if U = Ui0 ∩ . . . ∩ Uik = Spec A,
we have V = P1A . So for any V = Vi0 ∩ . . . ∩ Vik , we have H i (V, OE (n)|V ) = H i (P1A , OP1 (n)) = 0. So by 3.4.11.,
A
H i (E, OE (n)) = Ȟ i (V, OE (n)). Clearly, by calculating the Čech complex, we have Ȟ i (V, OE (n)) = 0, we have
H i (E, OE (n)) = 0 for all i > 0, n > 0. Then by the process of the proof of theorem 3.4., hi (X̃, OX̃ ) = hi (X, OX ), i.e.
pa (X̃) = pa (X).

Solution 5.3.2. Since C̃ = π∗ C+µP (C)E and D̃ = π∗ D+µP (D)E, we have C̃.D̃ = C.D+µ p (C)·µ p (D)E 2 = C.D−µ p (C)·
µ p (D). By definition of blow‑up and intersection number, we clearly have C̃.D̃ = 0. So C.D = µ p (C) · µ p (D).

Solution 5.3.3. Since D is ample, we have D2 > 0, i.e. D2 ≥ 1. So (2π∗ D − E)2 = 4D2 − 1 ≥ 3 > 0. Moreover,
since D is very ample, for any curve C on X, we have D.C = deg C. For any curve C̃ on X̃ with π(C̃) = C, we
have C̃ = π∗ C − rE with r = µP (C) ≤ deg C. So (2π∗ D − E).C̃ = (2π∗ D − E).(π∗ C − rE) = 2 deg C − r ≥ deg C > 0.
So by Nakai‑Noishezon criterion, D is ample.
L
Solution 5.3.4 (Multiplicity of a Local Ring). (a) Consider the graded ring M = Gr m A = d≥0
md /md+1 . By
theorem 7.5 in chapter I, there exists a Hilbert polynomial φ M of M such that φ M (l) = dimk m /ml for l  0.
l−1

Since A is a noetherian local ring, we have length A/ml = dim ml−1 /ml . So we may just take PA (z) = φ M (l), then
we have PA (l) = φ M (l) = dimk ml−1 /ml = length A/ml = ψ(l) for l  0.
(b) Take the system of parameters of m as {x1 , . . . , xn }. Then Gr m A  (A/m)[x1 , . . . , xn ]. Then deg PA =
dim Z(Ann (A/m[x1 , . . . , xn ]). Since A/m is a field, we have Ann (A/m[x1 , . . . , xn ]) = 0, i.e .deg PA = dim Z(0) =
dim Pn = n. Moreover, since dim A equals to the length of system of parameters, i.e. dim A = n. So deg PA =
dim A.
(c) Every Chapter.
(d) Take a small neighbourhood U or P on X. Denote A = OX,P and m = mP . Then we may assume C has
local equation f on U. If C has multiplicity r on P, we have f ∈ mr but f < mr+1 . Since OC,P = A/( f ), if P
has local parameter x, y on U such that f has a term xr , for any l > r, we have a basis of (A/( f ))(l) as xi y j for
2 2
0 ≤ i < r and 0 ≤ j ≤ l − i, i.e. dim(A/( f ))(l) = lr − r2 + 3r2 . Then PA (l) = lr − r2 + 3r2 . Hence µP (C) = µ(OC,P ) = r.

Yang Pi‑Yeh 107 Hartshorne Solutions

5 Surfaces 5.3 Monoidal Transformations

(e) Denote X as the cone of Y with vertex P. Then we may take a Ln−r passing through P meets X only at
P
P. Then deg X = x∈X∩L i(y, X ∩ L) = i(P, X ∩ L) = µP (X). But deg X = deg Y = d, i.e. µP (X) = d.

Solution 5.3.5. Since C has equation zr−2 y2 = xr +b1 xr−1 z+. . . +br zr for some coefficients bi . Then at the infinity
point (0 : 1 : 0), the Jacobian is (rxr−1 +. . .+br−1 zr−1 , −2yzr−2 , −(r−2)zr−3 y2 +b1 xr−1 +. . .+br zr−1 )|∞ = (0, 0, 0). Hence
∞ is a singular point. Moreover, on the affine plane (y , 0), we the equation is zr−2 = xr + b1 xr−1 z + . . . + br zr .
Then clearly r∞ =the lowest term= r − 2, so δ∞ = 21 r∞ (r∞ − 1) = 21 (r − 2)(r − 3). And consider the morphism
π : C̃ → C → P1 , (x : y : z) 7→ (x : z). Then clearly deg π = 2, and all ai ’s are branch points. If r is odd, the point
(0 : 1 : 0) has just one preimage in C̃, hence a branch point. But if r is ever, the point (0 : 1 : 0) has two, so not
a branch point. Then by Hurwitz’s formula, 2g(C̃) − 2 = 2(2g(P1 ) − 2) + deg R, where deg R = r if r is even, and
deg R = r + 1 if r is odd. Then g(C̃) = 21 r − 1 if r is even, and g(C̃) = 12 (r − 1) if r is odd.

Solution 5.3.6. For any singular point P on curve C, we may denote C̃ = BlP C, and P has preimage Q1 , . . . , Qr .
Then by definition of blow‑up, we may take sufficiently small affine neighbourhood U = Spec A of P on C,
`
such that the preimage of U, i.e. BlP U, = ri=1 Vi , such that Vi = Spec Bi is a neighbourhood of Qi . Hence we
L L
have Blp A = Bi . So we may take a inverse limit for such A, and have BlmP OP,C = OQi ,C̃ . Since completion
L
is flat, we have Blm̂P ÔP,C = ÔQi ,C̃ . So the resolution is determined by the completion of local ring, i.e.
analytically isomorphic implies equivalent.
Here is a counterexample of inverse direction. Consider two curves C = (x3 +y7 = 0) and C 0 = (x3 +xy5 +y7 =
0) over C. They are equivalent but not analytically isomorphic.

Solution 5.3.7. (a) x3 + y5 = 0. Take x = x1 y1 and y = y1 , we have x13 + y21 = 0 with exceptional curve y1 = 0, and
the singular point has multiplicity 2. Then take x1 = x2 and y1 = x2 y2 , we have x2 + y22 = 0 with exceptional
curve x2 = 0, and this is a smooth curve, intersecting (x2 = 0) at a double point. Hence we may take x2 = x3 y3
and y2 = y3 , we have x3 + y3 = 0 with exceptional curve x3 = 0, which intersect x2 = 0 and the curve at the
same point. So we may take one more blow‑up to get the normal crossing.
(b) x3 + x4 + y5 = 0. Take x = x1 y1 and y = y1 , we have x13 + x14 y1 + y21 = 0 with exceptional curve y1 = 0,
and the singular point has multiplicity 2. Then take x1 = x2 and y1 = x2 y2 , we have x2 + x23 y2 + y22 = 0 with
exceptional curve x2 = 0, and this is a smooth curve, intersecting (x2 = 0) at a double point. Hence we may
take x2 = x3 y3 and y2 = y3 , we have x3 + x33 y33 + y3 = 0 with exceptional curve x3 = 0, which intersect x2 = 0 and
the curve at the same point. So we may take one more blow‑up to get the normal crossing.
(c) x3 + y4 + y5 = 0. Take x = x1 y1 and y = y1 , we have x13 + y1 + y21 = 0 with exceptional curve y1 = 0, and
this is a smooth curve, which intersect (y1 = 0 at two distinct points. Hence we get the normal crossing.
(d) x3 + y5 + y6 = 0. Take x = x1 y1 and y = y1 , we have x13 + y21 + y31 = 0 with exceptional curve y1 = 0, and the
singular point has multiplicity 2. Then take x1 = x2 and y1 = x2 y2 , we have x2 + y22 + x2 y32 = 0 with exceptional
curve x2 = 0, and this is a smooth curve, intersecting (x2 = 0) at a double point. Hence we may take x2 = x3 y3
and y2 = y3 , we have x3 + y3 + x3 y33 = 0 with exceptional curve x3 = 0, which intersect x2 = 0 and the curve at
the same point. So we may take one more blow‑up to get the normal crossing.
(e) x3 + xy3 + y5 = 0. Take x = x1 y1 and y = y1 , we have x13 + x1 y1 + y21 = 0 with exceptional curve y1 = 0, and
the singular point has multiplicity 2. Then take x1 = x2 and y1 = x2 y2 , we have x2 + y2 + y22 = 0 with exceptional
curve x2 = 0, and this is a smooth curve, intersecting (x2 = 0) at two distinct points. Hence we get the normal
crossing.
So (a), (b), (d) are equivalent to each other. And (c), (e) are not equivalent to any others.

Solution 5.3.8. (a) x4 − xy4 = 0. Take x = x1 y1 and y = y1 , we have x14 − x1 y1 = 0, with exceptional curve
y1 = 0, and the singular point has multiplicity 2. Then take x1 = x2 and y1 = x2 y2 , we have x22 − y2 = 0, with
exceptional curve x2 = 0, and this is a smooth curve, intersecting (y1 = 0) and (x2 = 0) at one point. Hence
one more blow‑up, we can get the normal crossing.

Yang Pi‑Yeh 108 Hartshorne Solutions

5 Surfaces 5.4 The Cubic Surface in P3

(b) x4 − x2 y3 − x2 y5 +y8 = 0. Take x = x1 y1 and y = y1 , we have x14 − x12 y1 − x12 y31 +y41 = 0, which exceptional curve
y1 = 0, and the singular point has multiplicity 3. Then take x1 = x2 and y1 = x2 y2 , we have x2 −y2 −x22 y32 +x2 y42 = 0,
with exceptional curve x2 = 0, and this curve is nonsingular at (0, 0).
So on these two curves, δ(0,0) are the same. But two curves are not equivalent.

5.4 The Cubic Surface in P3

Solution 5.4.1. By theorem 4.2., dim d = 3. Then by theorem 4.1., d0 has no base points, so it determines a
morphism ψ : X 0 → P3 . Changing the coordinates, we may assume P1 = (1, 0, 0) and P2 = (0, 1, 0). Then
V ⊂ H 0 (P2 , OP2 (2)) corresponding to d is spanned by x0 x1 , x1 x2 , x2 x0 , x22 , so this is the space of all conics passing
through P1 and P2 . Then we may assume ψ(x22 ) = y0 , ψ(x0 x1 ) = y1 , ψ(x1 x2 ) = y2 and ψ(x2 x0 ) = y3 . So the
Im(ψ) satisfies the equation y0 y1 = y2 y3 , which is a quadric surface Y = (xy = zw). Conversely, we clearly have
Y ⊂ Im(ψ), hence equal.
Moreover, define π : Y → P2 to be the projection (x, y, z, w) 7→ (x, z, w). So π ◦ ψ = idP2 , and ψ ◦ π = idY .
Then we denote Γ as the graph of π. By definition, Γ = Ỹ is the blow‑up of Y at Q. Since (x0 x1 , x1 x2 , x2 x0 , x22 )
generate an ideal with saturation the homogeneous ideal of I(P1 , P2 ), so Γ is also the blow‑up of P2 at P1 and
P2 , i.e. X 0  Ỹ.

Solution 5.4.2. Changing coordinates, we may assume P1 = (1, 0, 0), P2 = (0, 1, 0) and P3 = (0, 0, 1). And
we may assume C is defined by f (x, y, z) = 0 for some homogeneous polynomial f with deg f = d. Since
µP1 (C) = r1 , we have f (x, y, z) = fr1 (y, z)xd−r1 + . . . + fd (y, z) for some homogeneous polynomial fi (y, z) of degree
 
i. So f (yz, zx, xy) = fr1 (zx, xy)(yz)d−r1 + . . . + fd (zx, xy) = xr1 (yz)d−r1 fr1 (z, y) + . . . + xd−r1 fd (z, y) . Hence clearly C 0
has equation f 0 (x, y, z) = fx(yz,zx,xy) 0 0
r1 yr2 zr3 . Clearly deg C = deg f = 2d − r1 − r2 − r3 . And for Q1 = (1, 0, 0), we have

yr2 zr3 f 0 (x, y, z) = (yz)d−r1 fr1 (z, y) + . . . + xd−r1 fd (z, y). Hence µC 0 (Q1 ) = d − r2 − r3 . And similarly for Q2 and Q3 .

Solution 5.4.3. For any singular point P1 on C, we may pick any two points P2 , P3 on P2 such that P1 Pi
intersects C at P1 transitively. Then we blow up P2 at P1 , P2 , P3 to get a X̃. And as the notation of example 4.2.3.,
we may assume C̃ ∩ E1 = {R1 , . . . , Rn }. By choosing of P2 and P3 , Ri is not on L̃12 nor L̃13 . Since µ(Ri ) < µ(P1 ),
the image of quadratic transformation of the curve C 0  ψ(C̃) splits P1 into some point Ri , some of which are
P
smooth, and some of which has multiplicity less than P1 . Since P∈C µ(P) − 1 < deg C is finite, this process
will stop when C 0 has only ordinary singularities.

Solution 5.4.4. (a) Since (PP0 +QQ0 +RR0 ).C = P+Q+R+P0 +Q0 +R0 +P00 +Q00 +R00 , and (L+L0 +P00 Q00 ) intersects
C with P, Q, R, P0 , Q0 , R0 , P00 , Q00 , by corollary 4.5., (L + L0 + P00 Q00 ).C = (PP0 + QQ0 + RR0 ).C, i.e. R00 ∈ P00 Q00 ).
(b) If PQ meet C at R, P0 R meet C at T , S T meet C at U, P0 U meet C at V, by definition we have P + Q = T
and T +S = V. If S Q meet C at B, PU meet C at A, consider the line L = QPR and L0 = S UT , then by (a), B, A, P0
are collinear. Hence by definition S +Q = A, and P+A = V. So we have the associativity (P+Q)+S = P+(Q+S ).

Solution 5.4.5. Denote the conic as S . Since (AB0 +BC 0 +CA0 ).(A0 B+B0 C +C 0 A) = A+B+C +A0 +B0 +C 0 +P+Q+R,
and (S + PQ) meets (AB0 + BC 0 + CA0 ) at A, B, C, A0 , B0 , C 0 , P, Q, by corollary 4.5., (S + PQ).(AB0 + BC 0 + CA0 ) =
A + B + C + A0 + B0 + C 0 + P + Q + R, i.e. P, Q, R are collinear.

Solution 5.4.6. Since H 0 (P2 , OP2 (2)) = 15, the complete linear system |D| of all quartic curves has dimension 14.
For any Pi = (xi , yi , zi ), we define a vector vi = (xi4 , y4i , z4i , xi3 yi , xi y3i , y3i zi , yi z3i , z3i xi , zi xi3 , xi2 y2i , y2i z2i , z2i xi2 , xi2 yi zi , y2i zi xi , z2i xi yi ).
Then if the matrix (v1 , . . . , v13 ) has rank 13, clearly dim |D − P1 − . . . − P13 | = 1. So if C, C 0 ∈ |D − P1 − . . . − P13 |,
C meets C 0 at P1 , . . . , P13 and three more points P14 , P15 , P16 . For any C 00 ∈ |D − P1 − . . . − P13 |, we must have
C 00 = λC + µC 0 , hence C 00 pass through P14 , P15 , P16 .
P P P P
Solution 5.4.7. If D = al − bi ai , then d = 3a − bi . Then pa (D) = 21 (a − 1)(a − 2) − 12 b2i + 21 bi =
P P P P P P P
1
18
(d − 3 + bi )(d − 6 + bi ) − 21 ( b2i ) + 12 bi = 18
1
(d − 3)(d − 6) + 19 d · bi + 18
1
( bi )2 − 12 ( b2i ) ≤ 18
1 2
d − 12 d +

Yang Pi‑Yeh 109 Hartshorne Solutions

5 Surfaces 5.4 The Cubic Surface in P3

 P P 
1 − ( 16 bi )2 − 19 d · bi + ( 13 d)2 + 19 d2 ≤ 16 d2 − 12 d + 1 = 16 (d − 1)(d − 2) + 23 , where the first inequality is the
P P P
Schwartz’s inequality, and the second is ( 61 bi )2 − 19 d · bi + ( 13 d)2 = ( 16 bi − 13 d)2 ≥ 0. So if d ≡ 1, 2 mod 3,
since pa (D) is an integer, we have pa (D) ≤ 61 (d − 1)(d − 2).
P
For achievement, when d ≡ 0 mod 3, we just take D = dl + d3 ( bi ), then d = 3d − 6 × d3 = d, and
2
pa (D) = 21 (d − 1)(d − 2) − 3 × d9 + 3 × d3 = 16 (d − 1)(d − 2) + 32 . When d ≡ 1 mod 3, we may assume d = 3k + 1,
then clearly the divisor D = dl − k(e1 + e2 + e3 + e4 ) − (k + 1)(e5 + e6 ) achieve the bound. When d ≡ 2 mod 3, we
may assume d = 3k + 2, then clearly the divisor D = dl − k(e1 + e2 ) − (k + 1)(e5 + e6 + e3 + e4 ) achieve the bound.

Solution 5.4.8. We firstly consider the surface Xr forming by blow‑up r points P1 , . . . , Pr on P2 with 2 ≤ r ≤ 5.
Then similarly to proposition 4.8., we have Pic Xr = Zr+1 generated by l, e1 , . . . , er with l2 = 1, e2i = −1, l.ei = 0
for any i, and ei .e j = 0 for any i , j, where l is the pullback of a line on P2 to Xr . Then also, similarly to theorem
4.9., if r < 5, then Xr contains exactly lines Ei and Fi j , where Ei is the exception line of Pi , and Fi j is the strict
line of Pi P j on Xr . Moreover, Ei = ei , Fi j = l − ei − e j . And if r = 5, Xr contains exactly lines Ei , Fi j and G, where
G is the strict curve of the conic passing through P1 , . . . , P5 on P2 . And G = 2l − e1 − . . . − e5 . Similar to lemma
4.12., we may define D0 = l, D1 = l − e1 , D2 = 2l − e1 − e2 , D3 = 2l − e1 − e2 − e3 if r ≥ 3, D4 = 2l − e1 − e2 − e3 − e4
if r ≥ 4, and D5 = 3l − e1 − e2 − e3 − e4 − e5 if r ≥ 5. Then |D0 |, |D1 | correspond to the linear systems of lines
in P2 passing through or not passing through P1 , hence very ample. And by proposition 4.1., D2 , D3 , D4 are
P
very ample on Xr if r is big enough, by proposition 4.3., D5 is very ample on X5 . So on Xr , ri=0 ni Di is very
P
ample, i.e. all divisor D = al − bi ei satisfying b1 ≥ . . . ≥ b5 > 0, and a > b1 + b2 + b5 if r = 5, or a > b1 + b2
P
if r < 5, is very ample on Xr . Conversely, as the proof of theorem 4.13., if D = al + bi ei is very ample, we
have a > b1 + b2 when r < 5 or a > b1 + b2 + b5 when r = 5, where b1 ≥ . . . ≥ b5 is a sorting. In the language of
intersecting, D is very ample iff D.L > 0 for all line L on Xr . Moreover, as Bertini’s theorem, if D.L > 0 for all
line L, there exists an irreducible nonsingular curve C ⊂ Xr such that C ∼ D. (*)
Back to the question.
(i)⇒(iii). If D contains the line Ei for some i, this is the condition (a) and we’ve done. If D does not contain
and Ei , and contains a curve C, then for any Xr for 0 ≤ r ≤ 5 (here we assume X0 = P2 ) and πr : X → Xr , πr (C)
P
is a curve on Xr . If D = al + bi ei , we may sort i as b1 ≥ . . . ≥ br > 0 = br+1 = . . . = b6 . In the case r = 0, D = al,
then clearly D.L > 0. In the case r = 1, we have D = al −b1 e1 . Then we may consider the curve D0 = π1 (D) ∈ X1 .
By example 2.11.5., X1 is a ruled surface over P1 with e = 1, and clearly D0 ∼Num (a − b1 )C0 + a f . Then by
corollary 2.18., we have a − b1 = 1 and a = 0, or a = 1 and a − b1 = 0, or a − b1 > 0 and a ≥ a − b1 . Hence
a = 0, b1 = −1, which is just E1 , or a = 1, b1 = 1, which is a conic, or a > b1 ≥ 0, which is the condition (c). In
P
the case r ≥ 2, we may consider πr : X → Xr , π∗r (D) = al + ri=1 bi ei . Then D.L > 0 for all line L ⊂ Xr . Hence
clearly D.L ≥ 0 for all line L ⊂ X, which is the condition (c).
(ii)⇒(i). Trivial.
P
(iii)⇒(ii). The condition (a) and (b) are trivial. In the condition (c), we may assume D = al − bi ei . If
D.L ≥ 0 for all line L, we have D.Ei = bi ≥ 0. Then rearranging Pi , we may assume b1 ≥ . . . ≥ br > 0 = br+1 ≥ . . ..
In the case r = 0, D = al, then by Bertini’s theorem, there exists nonsingular irreducible curve C of degree a in
P2 , hence π∗ (C) is an irreducible nonsingular curve on X. In the case r = 1, consider D0 = π∗1 (D) as a divisor on
X1 . Then by corollary 2.18., there exists an irreducible nonsingular curve C on X1 . Then π∗1 (C) is an irreducible
nonsingular curve on X. In the case r ≥ 2, consider D0 = π∗r (D) similarly, then by (*) or corollary 4.11., there
exists an irreducible nonsingular curve C on Xr , then π∗r (C) is an irreducible nonsingular curve on X.

Solution 5.4.9. We may assume b1 ≥ . . . ≥ b6 . Then by 5.4.8., we have b6 ≥ 0, a−b1 −b2 ≥ 0 and 2a−b1 −. . .−b5 ≥
P P P
0. So adding these conditions and a2 − b2i ≥ 3a − bi and d = 3a − bi ≥ 8 if d is even, or d =≥ 13 if d is
P
odd, the question is equivalent to the inequality a2 − 6a + 14 ≥ (bi − 1)2 for such a, b1 , . . . , b6 , which likes a
crazy IMO question.
P
So we may fix a ≥ 0 and b = bi ≥ 0, i.e. fix the value of d, and find the minimal value of g. Then
P 2 P
bi = b2 − 2b(b2 + . . . + b6 ) + (b22 + . . . + b26 ) + (b2 + . . . + b6 )2 . So for the minimal of b2i , we must take

Yang Pi‑Yeh 110 Hartshorne Solutions

5 Surfaces 5.4 The Cubic Surface in P3

(b22 + . . . + b26 ) + (b2 + . . . + b6 )2 to be small enough. If we make it be zero, we have g = 21 (a(a − 3) − b(b − 1)). So for
the minimal value of g, we must take b to be big enough. So we may assume b = a by the condition, and we
have g = 1 − a < 1, which makes a contradiction. So we cannot make the value (b22 + . . . + b26 ) + (b2 + . . . + b6 )2
to be zero. Then if we make b2 = 1 and b1 = b − 1, we have g = 12 (a(a − 3) − b(b − 1)). For b big enough, we may
assume b = a, and g = 0, i.e. we cannot take b = a. If b = a − 1, we have d = 2a + 1, and g = a − 2 = 21 (d − 5). This
is the minimal value of g in the case d is an odd number. We try to amplify the value (b22 +. . .+b26 )+(b2+. . .+b6 )2
a little, i.e. we may assume b2 = 2, and b1 = b − 2, then g = 21 (a(a − 3) − b(b − 5) − 6). So we make b = a to be
big enough, and we have d = 2a, g = a − 3 = 21 (d − 6), which is the limit of the bound in the case d is an even
number. Clearly calculating, this two value is the minimal bound of g when g > 1.

Solution 5.4.10. We may assume b1 ≥ b2 ≥ . . . ≥ b6 > 0. Then the condition is reduced to a − b1 − b2 > 0 and
2a − b1 − . . . − b5 > 0. Then we split this question in two cases very freshly:
Case I. b1 ≥ b2 + b5 . In this case, we have 2b1 b2 ≥ 2b22 + 2b2 b5 ≥ b23 + b24 + b25 + b26 . Hence by a ≥ b1 + b2 , we
P
have a2 > b21 + b22 + 2b1 b2 ≥ b2i .
Case II. b1 < b2 + b5 . In this case, a > b1 + b2 + b3 + b4 + b5 > 2b1 + b3 + b4 . Hence a2 > 4b21 + b23 + b24 + 4b1 b3 +
P
4b1 b4 + 2b3 b4 . Since b21 ≥ b2i for i ≥ 2, we have a2 > b2i .

Solution 5.4.11 (The Weyl Groups). (a) Define the morphism φ : An → Σn as xi 7→ (i, i+1). Then clearly φ(xi )2 =
(i, i+1)2 = 1. If i < j−1, we have φ(xi x j )2 = ((i, i+1)( j, j+1))2 = 1. And φ(xi xi+1 )2 = ((i, i+1)(i+1, i+2))3 = (i, i+2, i+
1)3 = 1. Hence φ is a group morphism. And since Σn is generated by (12), (23), . . . , (n − 1, n), this morphism is
surjective. Moreover, we may define ` : An → Z as following: for any element a ∈ An , a clearly can be written
as the form xi1 xi2 . . . xik for some finite k, then we define `(a) as the minimal value of such k. Consider An−1 as
a subgroup of An generated by x1 , . . . , xn−2 . Then for any x ∈ An An−1 , we may define p = min{`(a) : a ∈ An−2 x}.
By definition of An , we know xn−1 . . . xn−p ∈ An x. So An = An−1 ∪ An−1 xn−1 ∪ An−1 xn−1 xn−2 ∪ . . . ∪ An−1 xn−1 . . . x1 .
Hence |An | = n · |An−1 |. Since A1  Z/2Z, i.e. |A2 | = 2. So by induction, |An | = n!. Since |Σn | = n!, we know φ is
an isomorphism, i.e. An  Σn .
(b) Define the morphism φ : E6 → G as in the theorem. Then since the the square of quadratic trans‑
formation is the identity, we have φ(y)2 = 1. And since quadratic transformation is symmetric for E1 , E2 , E3
and E4 , E5 , E6 , we have φ(xi )φ(y) = φ(y)φ(xi ), i.e. (φ(xi )φ(y))2 = 1. And by calculation, y(E1 ) = F23 , y(E2 ) = F13 ,
y(E3 ) = F12 , y(Ei ) = Ei for 4 ≤ i ≤ 6, y(Fi j ) = Fi j if 1 ≤ i ≤ 3 and 4 ≤ j ≤ 6, y(Fi j ) = Gk , if {i, j, k} = {4, 5, 6}, and
y(Gi ) = Gi for 1 ≤ i ≤ 3. So we clearly have (φ(x3 )φ(y))3 = 1. Hence φ is a morphism. Moreover, as in the proof
of proposition 4.10., φ is a surjection.
(c) As in remark 4.10.1., |G| = 51, 840. Consider the subgroup E5 generated by x1 , x2 , x3 , x4 , y, and the
subgroup A5 generated by x1 , x2 , x3 , y. By definition and calculate by Mathematica, we have |E6 /E5 | = 27, and
|E5 /A5 | = 16, and by (a), we have |A5 | = 5! = 120. Hence we have |E6 | = 120 × 16 × 27 = 51, 840. Hence φ is an
isomorphism.

Solution 5.4.12. Clearly we have H 0 (X, OX ) = k, H 1 (X, OX ) = 0. Then for any ample divisor D, it is very
ample by theorem 4.11., we have H 0 (X, L (D)) = k. So we may consider the exact sequence 0 → L (−D) →
OX → L (D) → 0, i.e. 0 → H 0 (X, L (−D)) → k → k → H 1 (X, L (−D)) → 0. So we only need to prove
H 0 (X, L (−D)) = 0. If not, there must exist s ∈ H 0 (X, L (−D)) such that C = Supp s, and there exists at least a
line E such that C.E ≥ 0, hence D.E ≤ 0, which contradicts with theorem 4.11.

Solution 5.4.13. (a) As we’ve done in 5.4.8., the 16 lines contained in X are Ei , 1 ≤ i ≤ 5, which are the
exceptional curve, and Fi j , 1 ≤ i < j ≤ 5, which are the preimage of line Pi P j , and G, which are the preimage
of the conic passing through all Pi ’s.
(b) Similarly to theorem 4.7., we have OX (1) = ω−1 X , and X is embedded in P by a linear system of cubics
4

passing through all Pi . So L (−2K)  OX (2). Then we may consider the exact sequence 0 → IX (2) → OP4 (2) →
OX (2) → 0, so we have 0 → H 0 (P4 , IX (2)) → H 0 (P4 , OP4 (2)) → H 0 (X, OX (2)) → 0. Since h0 (OP4 (2)) = 15. And

Yang Pi‑Yeh 111 Hartshorne Solutions

5 Surfaces 5.4 The Cubic Surface in P3

by Riemann‑Roch on X, h0 (−2K) − h1 (−2K) = 21 (−2K − K).(−2K) + 1 = 3K 2 + 1 = 3 × (32 − 5) + 1 = 13. Then

since L (−2K)  OX (2), we have H 1 (X, L (−2K)) = 0. So h0 (IX (2)) = 2, i.e. X is a complete intersection of two
quadric hypersurfaces in P4 .

Solution 5.4.14. As following.

a b1 b2 b3 b4 b5 b6 d g
8 3 3 3 3 2 2 8 7
8 4 3 3 2 2 2 8 6
9 4 3 3 3 3 2 9 9
9 4 4 3 3 2 2 9 8
9 4 4 4 2 2 2 9 7
10 5 3 3 3 3 3 10 11
11 5 5 4 3 3 3 10 10
12 5 5 5 5 3 3 10 9
10 6 3 3 3 3 2 10 8

Solution 5.4.15. (a) (⇒) If P1 , . . . , P6 are in general position, clearly any three of them are not collinear. And
we may assume P1 = (1, 0, 0), P2 = (0, 1, 0) and P3 = (0, 0, 1). If P1 , . . . , P6 all lie on a conic, we may assume
this conic is axy + byz + czx = 0. Then if Pi = (xi , yi , zi ) for i = 4, 5, 6, the image of Pi is (yi zi , zi xi , xi yi ) for i = 4, 5, 6.
They are lying on the line bx + cy + ax = 0, which contradicts to the definition of general position.
(⇐) If P1 , . . . , P6 are not collinear and not lying on a conic, we may assume P1 = (1, 0, 0), P2 = (0, 1, 0) and
P3 = (0, 0, 1) also. Then just de a quadratic transformation for P1 , P2 , P3 , if the image of P4 , P5 , P6 are lying on
a line ax + by + cz = 0, similarly P1 , . . . , P6 are lying on a conic cxy + ayz + bzx = 0, which contradicts.
(b) Trivial, because if new P01 , . . . , P0r are no in general position, three of points P0 , Q0 , R0 will be collinear
in furthermore a finite sequence of admissible transformation, i.e. P, Q, R will be collinear in a finite sequence
of admissible transformation.
(c) Consider a variety T as the union of all line Pi P j and the conic Pi1 Pi2 Pi3 Pi4 Pi5 . We have dim T = 1. Since
k is uncountable, V = P2 − T is a open dense subset of P2 . Then by (a), any point Pr+1 ⊂ V and P1 , . . . , Pr+1 are
in general position.
P
(d) Similar to 5.4.8., we may assume C = al − ri=1 bi ei . In the case of r = 7, by Schwartz’s inequality,
P P
g = 21 (a − 1)(a − 2) − bi (bi − 1) = 0 and C 2 = a2 − b2i = −1 has solution (a, b1 , . . . , b7 ) = (0, 1, 0, 0, 0, 0, 0, 0)
and rest 6, or (2, 1, 1, 0, 0, 0, 0, 0) and rest 20, or (2, 1, 1, 1, 1, 1, 0, 0) and rest 20, or (3, 2, 1, 1, 1, 1, 1, 1) and rest
6. The total number= 7 + 21 + 21 + 7 = 56. In the case of r = 8, the solution are like (0, 1, 0, 0, 0, 0, 0, 0, 0)
for 8, and (2, 1, 1, 0, 0, 0, 0, 0, 0) for 28, and (2, 1, 1, 1, 1, 1, 0, 0, 0) for 56, and (3, 2, 1, 1, 1, 1, 1, 1, 0) for 56, and
(4, 2, 2, 2, 1, 1, 1, 1, 1) for 56, and (5, 2, 2, 2, 2, 2, 2, 1, 1) for 28, and (6, 3, 2, 2, 2, 2, 2, 2, 2) for 8. And the total number=
8 + 28 + 56 + 56 + 56 + 28 + 8 = 240.
P
(e) In the case r > 9, clearly C = e1 is a line on X. Moreover, for any line C = al− bi ei , we have d = 1, g = 0,
P
C 2 = −1. Then for admissible transformation σ about P1 , P2 , P3 , σ(C) = (a + c)l − c(e1 + e2 + e3 ) − bi ei , where
P
c = a−b1 −b2 −b3 . Suppose b1 ≤ . . . ≤ br , or we just need to rearrange the index. Then since d = 3a− bi ≥ 1 > 0,
P
we have 3a > bi ≥ 3(b1 + b2 + b3 ), i.e. a > b1 + b2 + b3 , then σ(C).l = a + (a − b1 − b2 − b3 ) > a = C.l. So there
exists infinitely many line C.

Solution 5.4.16. And line L in P3 has the form a0 x2 + b0 x3 − x0 = a1 x2 + b1 x3 − x1 = 0. So if L ⊂ X the Fermat

cubic curve, we have (a0 x2 + b0 x3 )3 + (a1 x2 + b1 x3 )3 + x23 + x33 = 0, i.e. a30 + a31 = −1, b30 + b31 = −1, a20 b0 + a21 b1 = 0,
a0 b20 + a1 b21 = 0. Suppose a0 = 0, we have b1 = 0, b30 = a31 = −1. Denote ω = exp( 2πi 3
), then the 27 lines on X are
x0 + x3 ω = x1 + x2 ω = 0 for 0 ≤ j, k ≤ 2, and x0 + x2 ω = x3 + x1 ω = 0 for 0 ≤ j, k ≤ 2, and x0 + x1 ωk = x3 + x2 ω j = 0
k j k j

for 0 ≤ j, k ≤ 2.

Yang Pi‑Yeh 112 Hartshorne Solutions

5 Surfaces 5.5 Birational Transformations

5.5 Birational Transformations

P S
Solution 5.5.1. Denote ( f ) = ni Ci . And consider the curve Y = Ci , and the embedded resolution as
P
theorem 3.9. π : X 0 → X such that f −1 (Y) is normal crossing. So for any Ci , C j such that ni n j < 0, if Ci ∩C j = Pk
on X 0 for some finitely many points Pk , we may blow‑up all Pk on X 0 to separate Ci and C j . Since ( f ) is a finite
sum, we just need a finite sequence of blow‑up on X 0 to separate zeros and poles at X 00 . Hence any point on
X 00 , it cannot be a zero and a pole of f at same time, i.e. we resolve the singularities of f .

Solution 5.5.2. Denote m = −Y 2 . By theorem 5.2. in chapter III, there exists a very ample divisor H on X such
that H 1 (X, L (H)) = 0. Denote k = H.Y, and we may assume k ≥ 2 or just change H to 2H. Then we may prove
H 1 (X, L (mH + iY)) = 0 for i = 0, . . . , k. For i = 0, trivial. Then for i ≥ 1, we may consider the exact sequence
0 → L (mH + (i − 1)Y) → L (mH + iY) → OY ⊗ L (mH + iY) → 0. Since Y  P1 , we have (mH + iY).Y = m(k − i). So
OY ⊗ L (mH + iY)  OP1 (m(k − i)). So H 1 (X, L (mH + (i − 1)Y)) → H 1 (X, L (mH + iY)) → H 1 (P1 , OP1 (m(k − i)) → 0.
So by induction, we have H 1 (X, L (H + iY)) = 0.
Denote M = L (mH + kY). Since H is very ample, |mH + kY| has no base points out of Y. And on Y, we
have M ⊗ IY  L (mH + (k − 1)Y). Since H 1 (X, L (mH + (k − 1)Y) = 0, and (mH + kY).Y = mk − km = 0, we
have M ⊗ OY  OP1 , which is generated by global section i. So lift the 1 to H 0 (X, M ) and use the Nakayama
lemma, we know M is generated by global section on all X.
So M defines a morphism f1 : X → PN with image X1 . Since f1∗ O(1)  M by theorem 11.7. in chapter III,
and deg(M ⊗ OY ) = deg OP1 = 0, we know f1 (Y) = P1 for some point P1 . And the rest are all same with the
proof of Castelnuovo’s theorem.

Solution 5.5.3. By 3.8.1. and 3.8.3., we have H i (X̃, π∗ ΩX )  H i (X, π∗ π∗ ΩX )  H i (X, ΩX ). Consider the exact
sequence 0 → π∗ ΩX → ΩX̃ → ΩX̃/X → 0. Sonce ΩX̃/X is supported on ramified points, we have ΩX̃/X  ΩE . Then
0 → H 0 (X, ΩX ) → H 0 (X̃, ΩX̃ ) → H 0 (E, ΩE ) → H 1 (X, ΩX ) → H 1 (X̃, ΩX̃ ) → H 1 (E, ΩE ) → H 2 (X, ΩX ) → H 2 (X̃, ΩX̃ ) →
0 since E has dimension 1. Since H 0 (E, ΩE )  H 0 (P1 , OP1 (−2)) = 0, we have H 1 (E, ΩE )  H 1 (P1 , OP1 (2) = k.
Moreover, by Serre’s duality, we have h2 (ΩX ) = h1 (Ω2X ) = h1 (OX ). Since h1 (OX ) is a birational invariant, we
know H 2 (X, ΩX ) → H 2 (X̃, ΩX̃ ) is an isomorphism. Hence the exact sequence is 0 → H 1 (X, ΩX ) → H 1 (X̃, ΩX̃ ) →
k → 0. Since k is a field, we have H 1 (X̃, ΩX̃ )  H 1 (X, ΩX ) ⊕ k.

Solution 5.5.4. (a) Take a very ample divisor H on X 0 . Then f ∗ H is ample on X. Since f ∗ H.Y = H. f∗ Y = H.0 = 0.
Then by Hodge index theorem, we have Y 2 < 0.
(b) By (a), Yi2 < 0. Moreover, we may take further quadratic transformation if necessary and assume all
Yi are nonsingular, and if i , j, Yi ∩ Y j , ∅, then Yi ∩ Y j is just one point. Take H10 as a hyperplane section in
X 0 passing through P and H20 a hyperplane section in X 0 not passing through P. And suppose ( f ) = H10 − H20 .
P
Take Hi as the strict transform of Hi0 on X. Then we have H2 ∼ H1 + mi Yi for some mi = ordYi ( f ) > 0.
So if S = (Yi .Y j )i, j , and S 0 = (mi Yi .m j Y j )i, j , we have S 0 = MS M for diagonal matrix M = diag{m1 , . . . , mr }.
P P
Hence S i j ≥ 0 if i , j. And for any j we have i S i0 j = i (mi Yi .m j Y j ) = −H1 .m j Y j ≤ 0. So S is negative
P
indefiniteness. Moreover, since H1 passes through some Y j , we have i S i0 j < 0 for this j. So if for some
P P P P P P
real vector v = (v1 , . . . , vr )T , we have 0 = i, j vi v j S i0 j = v2i S ii0 + 2 i< j vi v j S i0 j = j ( i S i0 j )v2j − i< j S i0 j (vi − v j )2 .
S
Since v is real, we have v j = 0 for some j. Since Yi is connected, clearly we cannot split (1, . . . , r) into
(i1 , . . . , ik ) ∩ ( j1 , . . . , jr−k ) such that S i0a jb = 0 for any a, b. So the vector v must have vi = v j for all i, i.e. v = 0,
hence S 0 is negative definite, and so is S .

Solution 5.5.5. For any P ∈ X, consider the ruled surface elmP X. For any two curves C ∼ aC0 + b f and
D ∼ cC0 + d f , we have C 0 .D0 = C.D + ac − a · multP D − c · multP C. So if a = c = 1 and P ∈ C ∩ D, we have
C 0 .D0 = C.D − 1, and if P < C ∪ D, we have C 0 .D0 = C.D + 1. So for the ruled surface X with e, we pick a
point P ∈ C0 , then by above we know the new ruled surface elmP X has ẽ = e + 1. Then we blow‑up so many
times, we may assume X and X 0 has e, e0 > 2g − 2. Hence by theorem 2.12., we may assume E and E 0 are

Yang Pi‑Yeh 113 Hartshorne Solutions

5 Surfaces 5.5 Birational Transformations

decomposable with E = OC ⊕ L and E 0 = OC ⊕ L 0 . If L = L (e), for any P ∈ C0 , by definition, elmP X has

ẽ = e + P. So we may blow‑up X and X 0 more and assume e = e0 . Hence X  X 0 . So there is a finite sequence
of elementary transformations which transform X into X 0 .

Solution 5.5.6. If R is a valuation ring with valuation ν, since X is projective, hence proper, the valuation
criterion give us a morphism φ : Spec R → X. For any closed point m of Spec R, φ(m) is the center of ν. If
P ∈ X is the center, then R dominates OP . So if ν is nontrivial, the dimension of ν, d, corresponds to tr.d.R/mR ,
which corresponds to the dimension of the center. Since X is a surface, d = 0, 1. If d = 1, R and OP are discrete
valuation rings. Since R dominates OP , R = OP , this is the type (1). If d = 0, then we may blow‑up P and get
an X 0 . So the center of ν in X 0 is either the exceptional divisor or a point contained in the exceptional divisor.
For the first case, R is the type of (2). For the second case, we just repeat this process. If in finite repeats,
we get a sequence of birational morphism Xr → Xr−1 → . . . → X1 → X, and a sequence of point Pi ∈ Xi
for i < r such that Pi is in the exceptional divisor blown‑up from Pi−1 , and R corresponds to the exceptional
divisor in Xr , then R is the type of (2). If we cannot get the result in finite process, then we have an infinite
sequence X ← X1 ← . . . and (P, P1 , . . .), such that Pi lies on the exceptional divisor blown‑up from Pi−1 . Then
S
R = lim OPi ,Xi = OPi ,Xi , which is the type of (3).
←−−
Solution 5.5.7. Take the very ample divisor class |H| on X0 . Let H0 ∈ |H| be a divisor containing P, and H1 ∈ |H|
be a divisor not containing P. Then if H̃0 and H̃1 are strict transform of H0 and H1 on X, then H̃1 = H̃0 + mY
for some m. Then mY.mY = (mY − H̃1 ).mY = −H̃0 .mY < 0.

Solution 5.5.8 (A Surface Singularity). (a) Since A/(z)  k[x, y]/(x2 + y3 ), which is an integral domain, we
know z is irreducible in A. Then consider the ring A[t]. We have A[t] = k[x, y, z][t] = k[u, v][t], which is clearly
a UFD since u, v are algebraically independent over k. Then by Nagata’s criterion for factoriality, A is a UFD.
(b) First blow‑up: (x2 + y3 + z5 , xz1 = zx1 , yz1 = zy1 , xy1 = yx1 ). In the piece x1 = 1, it is (x2 + x3 y31 + x5 z51 , xz1 =
z, xy1 = y) = (x2 , xz1 = z, xy1 = y) ∪ (1 + xy31 + x3 z51 , xz1 = z, xy1 = y). The jacobian is (3x2 z5 + y3 , 3xy2 , 5x3 z4 ),
hence this piece is nonsingular. In the piece y1 = 1, it is (x12 + y + y3 z51 ). The jacobian is (2x, 3y2 z5 + 1, 5y3 z4 ),
hence this piece is nonsingular. In the piece z1 = 1, it is (z2 x12 + z3 y31 + z5 , x = zx1 , y = zy1 ) = (z2 , x = zx1 , y =
zy1 ) ∪ (x12 + zy31 + z3 , x = zx1 , y = zy1 ). The jacobian is (2x, 3y2 z, y3 + 3z2 ), hence there exists a singularity at (0, 0, 0).
Second blow‑up: (x12 + zy31 + z3 , y2 x1 = y1 x2 , x2 z = z2 x1 , y1 z2 = y2 z). In the piece x2 = 1, it is (1 + z2 y32 x12 + z32 x1 ).
The jacobian is (3y22 z2 x12 , x1 (y32 x1 + 3z22 ), z2 (2y32 x1 + z22 )), hence this piece is nonsingular. In the piece y2 = 1, it is
(x22 + y21 z2 + y1 z32 ). The jacobian is (2x2 , 2y1 z2 + z32 , y21 + 3z22 y1 ), hence there exists a singularity at (0, 0, 0). In the
piece z2 = 1, it is (x22 + z + y32 z2 ). The jacobian is (2x2 , 3y22 z2 , 1 + 2zy32 ), hence this piece is nonsingular.
Third blow‑up: (x22 + y21 z2 + y1 z32 , x2 y3 = z2 x3 , x2 z3 = y1 x3 , z2 z3 = y1 y3 ). In the piece x3 = 1, it is (1 + x22 y33 z3 +
x2 y3 z23 ), hence clearly nonsingular. In the piece y3 = 1, it is (x32 + z22 z3 + z2 z23 ). The jacobian is (2x3 , 2z2 z3 + z32 , z22 +
2z2 z3 ), hence there exists a singularity at (0, 0, 0). In the piece z3 = 1, it is (z32 + y1 y23 + y1 z2 ). The jacobian is
(y23 + z2 , 2y1 y3 , y1 + 3z22 ), hence there exists a singularity at (0, 0, 0).
Fourth blow‑up: (z32 + y1 y23 + y1 z2 , y1 y4 = y3 x4 , y3 z4 = z2 y4 , y1 z4 = z2 x4 ). In the piece x4 = 1, it is (z4 + y1 (y24 + z34 )).
The jacobian is (y24 +z34 , 2y1y4, 1+3y1z42 ), hence this piece is nonsingular. In the piece y4 = 1, it is (z32 +x4 y3 (y23 +z2 )).
The jacobian is (y3(y32 + z2), 2x4y32 + x4(y32 + z2), x4y3 + 3z22 ), which is nonsingular at y3 = z2 = 0. This is the
exceptional curve of another singular point in the third blow‑up. In the piece z4 = 1, it is (x4 + z2 + x4 y24 z2 ).
The jacobian is (1 + y24 z2 , 2x4 y4 z2 , 1 + x4 y24 ), hence this piece is nonsingular.
Fifth blow‑up: (x32 +z22 z3 +z2 z23 , x3 z5 = x5 z3 , z2 z5 = z3 y5 , x3 y5 = x5 z2 ). In the piece x5 = 1, it is (1+ x3 y25 z5 + x3 y5 z25 ),
hence clearly nonsingular. In the piece y5 = 1, it is (x52 + z2 z5 + z2 z25 ). The jacobian is (2x5 , z5 + z25 , z2 + 2z2 z5 ), hence
there exist two singularities (0, 0, 0) and (0, 0, −1). The point x5 = z2 = z5 = 0 is impossible, hence there only
exists one singularities (0, 0, −1). In the piece z5 = 1, it is (x52 + y5 (1 + y5 )z3 ), hence there exists two singularities
(0, −1, 0) and (0, 0, 0). Similarly, (0, 0, 0) is impossible, and the singularity (x5 , y5 , z3 , z5 ) = (0, −1, 0, 1) is the same
with (x5 , y5 , z3 , z5 ) = (0, 1, 0, −1).

Yang Pi‑Yeh 114 Hartshorne Solutions

5 Surfaces 5.6 Classification of Surfaces

Sixth blow‑up: (x52 + z2 z5 + z2 z25 , x5 y6 − z2 x6 , z2 z6 − y6 (z5 + 1), x5 z6 − x6 (z5 + 1)). In the piece (x6 = 1), it is
(1 + y6 z6 (−1 + x5 z6 )). The jacobian is (y6 z26 , z6 (−1 + x5 z6 ), x5 y6 z6 + y6 (−1 + x5 z6 )), hence this piece is nonsingular.
In the piece (y6 = 1), it is (x62 + z6 (−1 + z2 z6 )). The jacobian is (2x6 , z26 , −1 + 2z2 z6 ), hence this piece is nonsingular.
In the piece (z6 = 1), it is (x62 + y6 z5 ). This jacobian is (2x6 , z5 , y6 ), hence there exists a singularity (0, 0, 0).
Seventh blow‑up: (x62 + y6 z5 , x6 y7 − x7 y6 , y6 z7 − z5 y7 , z5 x7 − x6 z7 ). In the piece (x7 = 1), it is (1 + y7 z7 ), hence
clearly nonsingular. In the piece (y7 = 1), it is (x72 + z7 ), hence clearly nonsingular. In the piece (z7 = 1), it is
(x72 + y7 ), hence clearly nonsingular.
And one more blow‑up to separate all exceptional curves. Hence this Du Val singularity is of the type E8 .

5.6 Classification of Surfaces

P P
Solution 5.6.1. By 2.8.4., ω  OX ( di − n − 1). So if κ(X) = −1, we have di − n − 1 < 0. Since di ≥ 2, we
P
have n + 1 > di ≥ 2(n − 2), i.e. n < 5. And the choices of (n; d1 , . . . , dn−2 ) in the case κ = −1 are (3; 2), (3; 3),
P
and (4; 2, 2). If κ(X) = 0, we have di − n − 1 = 0. Similarly we have n ≤ 5. And the choices of (n; d1 , . . . , dn−2 )
in the case κ = 0 are (3; 4), (4; 2, 3) and (5; 2, 2, 2). Moreover, these three cases are all K3 surfaces. If κ(X) = 1,
P
then X is an elliptic surface. Then ω2 = 0, i.e. di − n − 1 = 0. But these solutions are all K3 surfaces, which
is impossible. So almost all choices of n and di are of general type.

Solution 5.6.2. If H is the hyperplane section, then deg H = d and dim |H| = n. So if h2 (OX ) = h1 (OH (d)) > 0,
by Clifford’s theorem, we have d ≥ 2n. So if d < 2n, pg (X) = 0. Then if pg (X) , 0 and d = 2n, by adjunction
formula, 2g − 2 = d + H.K ≥ d. By Clifford’s theorem again, we have n + 1 ≤ h0 (OH (d)) = 1 − g + d. Then
2n + 2 ≤ 2 − 2g + 2d ≤ d, so here we must have d = 2g − 2, i.e. d = 2n and g = n + 1, and H.K = 0. Since
X is not contained in any hyperplane, it is not a ruled surface. Then by theorem 6.2., we have |12K| , 0.
So since 12K = 0, we have κ = 0 by theorem 6.3. By Nakai Moishezon criterion, we have H − K is ample,
so h1 (K − H) = 0 by Kodaira vanishing theorem, then h1 (H) = 0 by Serre duality. So by 5.1.1., we have
deg(K − H) < 0, i.e. h0 (K − H) = 0. Then by 5.1.6., pa (X) = 1, hence X is a K3 surface by theorem 6.3.

Yang Pi‑Yeh 115 Hartshorne Solutions

Appendix

Appendix
A Intersection Theory
Solution Appx.A.6.1. We will call the definition of rational equivalence in section 1 as definition (a), and the
new definition as (b). Moreover, we can define a rational equivalence (c) as in Fulton’s Intersection Theory as:
Y ∼ Z iff there exist a cycle W of codimension r − 1 and a rational function f ∈ K(W)∗ such that Y and Z are
the zeros and the poles of f , which is similar to the definition of principal divisor as we used to do.
(b⇒c) Consider the projection πX : W → X. Then W 0 = ImπX is a subvariety of X with codimention r − 1.
So there exists a rational function f (x) = (πA1 (W ∩ π−1 −1 0 −1
X (x)) − 1) + 1 ∈ K(W ) . And the zero of f is W ∩ (X × {0}),
and the pole of f is W ∩ (X × {1}).
(c⇒b) If we have a subvariety W of codimension r − 1 and an f ∈ K(W)∗ , we may consider the subvariety
W of X × A1 as {(x, t) ∈ W × A1 | ( f (x) − 1)−1 + 1 = t}. So the zero of f is {(x, t) ∈ W × A1 | ( f (x) − 1)−1 + 1 = t, f (x) =
0

0} = W 0 ∩ (X × {0}), and the pole is W 0 ∩ (X × {1}).

(a⇔c) If V is a subvariety of X, and Ṽ is the normalization of V, we have K(V)∗ = K(Ṽ)∗ . And for any
P
f ∈ K(V)∗ , we have ordY ( f ) = ordỸ ( f ) · [K(Ỹ) : K(Y)], where the sum if over all subvarieties Ỹ of Ṽ which
map into Y, and [K(Ỹ) : K(Y)] is the degree of the field extension. So clearly (a) and (c) are equivalent.

Solution Appx.A.6.2. If f is generically finite, we may shrink X and X 0 by deleting a closed subset of codimention≥
2 and assume f is finite flat and surjective by 2.3.7. and 3.9.3.(a). For any effective divisor D, we have an exact
sequence 0 → L (−D) → OX → OD → 0. Then since f is flat, we have 0 → f∗ L (−D) → f∗ OX → f∗ OD → 0.
So the rest is same as 4.2.6.(a) and (b), if D is a principal divisor, we have f∗ (D) is also principal. Hence f∗ is
well‑defined modulo rational equivalence.

Solution Appx.A.6.3. If subvariety X is a hypersurface in Pn , then trivial. If X has codimention−r > 1, as in

example 9.8.3. in chapter III, we can find a point P ∈ Pn not in X and consider the projection π from P to Pn−1 .
Then X 0 = π(X) is a subvariety in Pn−1 of codimention r − 1. Clearly π|X is finite on an open dense subset of X,
and deg X 0 = deg X/ deg(π|X ). So after finite step of projection, we have a hypersurface X 0 ⊂ Pn−r+1 of degree
deg X/ deg(π|X ), where π is the component of all projection. And X 0 ∼ (deg X 0 ) · H for hyperplane H in Pn−r+1 .
So by the Axiom of π∗ : A(Pn−r+1 ) → A(Pn ), we have X ∼ deg(π|X ) · π∗ ((deg X 0 ) · H) = deg X · π∗ (H). Clearly π∗ (H)
is a linear subspace of dimension equals to dim X.

Solution Appx.A.6.4. By Axiom A8, we only need to prove A2 (X)  A1 (C). Since we have the projection
π : X → C and the section σ : C → X, we can just define π∗ : Z 2 (X) → Z 1 (C) as P 7→ π(P) and σ∗ : Z 1 (C) → Z 2 (X)
as P 7→ σ(P), where Z i is the group of i‑cycles. Then for any point P ∈ C, we have π∗ σ∗ (P) = P. For any point
P ∈ X, we have σ∗ π∗ (P) and P are lying on the same ruling P1 , hence rationally equivalent. And for any point
P, Q ∈ C with P ∼ Q, if P ∼ Q, there exists f ∈ K(C) such that ( f ) = P − Q. Since f can by extended to a
morphism X → C → P1 , we have σ∗ (P) ∼ σ∗ (Q). Conversely, if P, Q ∈ X with P ∼ Q, then since P ∼ σ∗ π∗ (P),
and same for Q, we have σ∗ π∗ (P) ∼ σ∗ π∗ (Q), hence clearly πP ∼ πQ . Then so clearly A2 (X)  A1 (C).

Solution Appx.A.6.5. By Axiom A3 and proposition 3.2. in chapter V, we have A1 (X̃)  A1 (X) ⊕ Z. For A2 ,
we can define Z 2 (X̃) → Z 2 (X) as P 7→ π(P). So this morphism is isomorphic out of P. If P0 , P00 ∈ X̃ such that
π(P0 ) = π(P00 ) = P, since E  P1 , we clearly have P0 ∼ P00 . So this morphism induces A2 (X̃)  A2 (X). Hence we
have a group isomorphism A(X̃)  π∗ A(X) ⊕ Z. For the ring structure, if C is a curve in X passing through P,
by proposition 3.6. in chapter V, we have π∗ (C).E = (C̃ + rE).E = Q1 + . . . + Qr − r · π∗ (P), where C̃ is the strict
transformation of C in X̃, and Qi are r intersection point of C̃ and E. Since E  P1 , we have π∗ (C).E ∼ 0. So
this induces the multiplication structure of the ring morphism A(X̃) → π∗ A(X) ⊕ Z, hence an isomorphism.

Solution Appx.A.6.6. By Axiom C7, we have cn (N∆ ) = ∆2 . Since by theorem 8.17 in chapter II, we have an

Yang Pi‑Yeh 116 Hartshorne Solutions

Appendix B Transcendental Methods

exact sequence 0 → T∆ → TX×X ⊗ OY → N∆ → 0, and clearly ∆  X, we have ct (N∆ ) = ct (TX×X ⊗ OY )/ct (T∆ ) =
ct (TX )2 /ct (TX ) = ct (TX ), we have cn (TX ) = cn (N∆ ), hence cn (TX ) = ∆2 .

Solution Appx.A.6.7. For any divisor D of X, we may consider the case E = L (D). By Axiom C1 we have
ct (L (D)) = 1 + Dt, i.e. c1 (L (D)) = D, c2 (L (D)) = 0 and c3 (L (D)) = 0. So ch(L (D)) = 1 + D + 12 D2 + 16 D3 . Since
td(L (D)) = 1 + 21 c1 + 121 (c21 + c2 ) + 241 c1 c2 , by Hirzebruch‑Riemann‑Roch, we have χ(L (D)) = deg((1 + D + 21 D2 +
D ).(1 + 21 c1 + 12
1 3
6
1
(c21 + c2 ) + 241 c1 c2 ))3 = 61 (D3 ) + 14 D2 .c1 + 12
1
D.(c21 + c2 ) + 24
1
(c1 c2 ). Since TX is the dual of ΩX , and
by Axiom C5, we have c1 (T̃ X ) = −K. So we have χ(L (D)) = 12 D.(D − K).(2D − K) + 12
1 1
D.c2 + 24
1
c1 c2 . In the case
D = 0, we have 1 − pa = χ(X) = 24 c1 c2 . So we have χ(L (D)) = 12 D.(D − K).(2D − K) + 12 D.c2 + 1 − pa .
1 1 1

Solution Appx.A.6.8. By theorem 8.13. in chapter II, we have 0 → O → O(1)4 → T → 0, we have ct (T ) =

ct (O(1))4 /ct (O) = (1 + ht)4 = 1 + 4ht + 6h2 t2 + 4h3 t3 , where h ∈ A1 (P4 ) is the class of hyperplane. So td(T ) =
1+2h+ 116 h2 +h3 . Since ch(E ) = 2+c1 + 12 (c21 −2c2 )+ 16 (c31 −3c1 c2 ), we have χ(E ) = 8+6c1 +2(c21 −2c2 )+ 16 (c31 −3c1 c2 ).
Since χ(E ) is an integer, we know c31 − 3c1 c2 ≡ 0 mod 6, hence c1 c2 ≡ 0 mod 2.

Solution Appx.A.6.9 (Surfaces in P4 ). (a) For the rational cubic scroll X in example 2.19.1. in chapter V, we
have d = 3, H.K = (C0 + 2 f ).(−2C0 − 3 f ) = −5 and K 2 = 8. So 12pa = −d2 + 10d + 5H.K + 2K 2 − 12 =
−9 + 30 − 25 + 16 − 12 = 0, i.e. pa = 0, which is what we have.
(b) By definition of K3 surface, we have K = 0, pa = pg = 1, we have d2 − 10d − 0 − 0 + 12 + 12 = 0, i.e.
d − 10d + 24 = 0, so d = 4 or 6.
2

(c) By theorem 6.3. in chapter V, we have 12K = 0, pa = −1 and pg = 1. So we have d2 −10d−0−0+12−12 = 0,

i.e. d2 − 10d = 0. Since d > 0, we have d = 10.
(d) Suppose H ∼ aC0 + b f is the very ample divisor of X determining the embedding X → P4 , so we have
H = 2ab − a2 e = a(2b − ae) = 4. Since K = −2C0 − (e + 2) f by corollary 2.11. in chapter V, we have H.K =
2

2ae−(2b+ae+2a) = −2a−2b+ae, and K 2 = −4e+4(e+2) = 8. So we have 16−40+10a−5ae+10b−16+12+0 = 0.

So 10a + 10b − 5ae = 28, which is impossible, since 5 - 28.

Solution Appx.A.6.10. For an abelian 3‑fold X, since TX is free, we have TX  OX3 . So ct (TX ) = ct (OX )4 = 1, i.e.
c1 = c2 = c3 = 0. So td(TX ) = 1. So if we have 0 → TX → TP5 → N → 0, we must have 1·(1+c1 (N )t+c2 (N )t2 +
c3 (N )t3 ) = 1+6Ht+15H 2 t2 +20H 3 t3 , i.e. c1 (N ) = 6H, c2 (N ) = 15H 2 and c3 (N ) = 20H 3 . If N = L (D) for some
divisor D of X, then c1 (N ) = D, c2 (N ) = D2 and c3 (N ) = D3 . But D3 = D2 .D = 15H 2 .6H = 90H 3 , 20H 3 = D3 ,
which is impossible.

B Transcendental Methods
Solution Appx.B.6.1. If the unit disc X = {|z| < 1} is algebraic, i.e. X = Xh for some scheme X of finite type
over C, we have dim X = 1. Then X must has an affine open subscheme Y with dimension 1, and Yh ⊂ Xh .
We may assume Y = C[x1 , . . . , xn ]/I for some ideal I, then there exists a sequence of points {yn } ⊂ Yh such that
|yn | → ∞, which contradicts to the definition of the disc.

Solution Appx.B.6.2. If there exists a sheaf of ideal I = I˜ in Spec C[z] such that I ⊂ (z − zi ) for all i, since
C[z] is a PID, I has a generator f . Then f has zeros at all zi , which contradicts to the fundamental theorem
of algebra. By theorem 4.1. in chapter V of Stein’s Complex Analysis, there exists a holomorphic function f
vanishing on all zi , and other holomorphic function vanishing on all zi has the form f (z) · exp(g(z)) for some
entire function g(z). Hence we can define a C[z]‑module generated by f (z), and F = M̃, then we have Fh  I.

Solution Appx.B.6.3. We may consider a global section s of L, such that s|(z,0) = ez , and s|(w,0) = e−1/w . Then
s on C2 has only one singular point at (0, 0). But for any invertible sheaf L of A2 − {(0, 0)}, then there exists
an open subset U ⊂ A2 near (0, 0), and L |U−(0,0)  OU−(0,0) . Then for any s ∈ L |U−(0,0) , s has the form gf such

Yang Pi‑Yeh 117 Hartshorne Solutions

Appendix C The Weil Conjectures

that f, g are algebraic function on U − (0, 0). So the pole of s must be a curve of codimension 1 in A2 − {(0, 0)}.
Hence there does not exists a section on U −{(0, 0)} such that sh  s|U−(0,0) . So there does not exists an invertible
sheaf L on A2 − {(0, 0)} such that Lh  L.

Solution Appx.B.6.4. Clearly α : H 0 (X, OX ) → H 0 (Xh , OXh ) is injective since reduced. And since X is proper, Xh
is compact, so by Liouville’s theorem, H 0 (Xh , OXh ) = C. Clearly C ⊂ H 0 (X, OX ), hence isomorphic. Conversely,
for any X not proper, then Xh is not complete. Hence there exists an entire exponent function f on Xh which
has an essential pole in the compactification of Xh . So f is not algebraic, hence H 0 (X, OX ) , H 0 (Xh , OXh ).

Solution Appx.B.6.5. Since Xh  Xh0 , we know there exists a bijection between the set of closed points of X
and X 0 . Since the Zariski topology on curve defines the open subset as the whole set minus finite points, so
the topology on closed points of X and X 0 are the same. Since X = {closed points of X} ∪ {generic point η}, and
same for X 0 , and moreover η and η0 are not closed points, and any open subset of X or X 0 contains η or η0 , we
know the topology of X and X 0 are the same, hence X  X 0 .

Solution Appx.B.6.6. Since Y are projective, we may consider the closed embedding Y → Pn for some n, it
induces a morphism Yh → Pnh . So Xh → Yh → Phn . If there exists morphisms f : X → Pn and g : Y → Pn , such
that Im f ⊂ Img = Y, so actually, f is mapping to Y, it is a morphism from X to Y, and fh = f.
So we only need to consider the case Y = Pn . Since f : Xh → Pnf is a morphism, we may consider the
invertible analytic sheaf L = f∗ O(1) on Xh . Since X is projective, by GAGA, there exists an invertible sheaf L
on X such that L = Lh . Since L is very ample, there exists global sections s0 , . . . , sn such that they generates L.
Then there exists global sections s0 , . . . , sn corresponding to all si , such that si,h = si . Hence s0 , . . . , sn generates
L , so by theorem 7.1. in chapter II, L is very ample, and determines a closed embedding f : X → Pn . Clearly
fh = f.

C The Weil Conjectures

` P P P P
Solution Appx.C.5.1. If X = i Xi , we have Nr (X) = i Nr (Xi ), then Z(X, t) = exp( ∞ tr
r=1 Nr (X) r ) = exp( r
tr
i Nr (Xi ) r ) =
P P r Q P r Q
exp( i r Nr (Xi ) tr ) = i exp( r Nr (Xi ) tr ) = i Z(Xi , t).

Solution Appx.C.5.2. Since Pn = An ∪ An−1 ∪ . . . ∪ A1 ∪ {∞}, we have Nr (Pn ) = 1 + qr + . . . + qnr . So Z(Pn , t) =

r Q P i r Q
exp(Nr (Pn ) tr ) = ni=0 exp( r (q rt) ) = (1 − qi t)−1 . The Rationality and the RH are clearly. For the functional
L 0 0
equation, since ct (Pn × Pn ) = ct (Pn )2 , we have chn (Pn × Pn ) = i+ j=n
Pi × P j . Clearly, (Pi × P j ).(Pi × P j ) = 1 if
i + i0 = j + j0 = n and 0 otherwise. So by Appx.A.6.6., we have E = ∆.∆ = n + 1. Then,

Yn Yn Qn i
1 i−n −1 −1 1 −1 i=0 (q t) qn(n+1)/2 tn+1
Z(Pn , ) = (1 − q t ) = (1 − ) = Q = (−1)n+1 Qn = (−1)n+1 qnE/s t E Z(Pn , t)
n
qt i=0 i=0
qi t n
i=0 (q i t − 1) i=0 (1 − q i t)

P2n
For Betti number, we have H i (Pn , Z) = Z if i is even, or 0 if i is odd. So i
i=0 (−1) Bi = n + 1 = E.
P r P r
Solution Appx.C.5.3. Clearly Z(X × A1 , t) = exp( r Nr (X × A1 ) tr ) = exp( r qr Nr (X) tr ) = Z(X, qt).

Solution Appx.C.5.4. Denote Nr (x) as the number of points in X(Fqr ) corresponding to closed point x ∈ X.
Q Q P P P P∞ −s r deg x −1
Then clearly ζX (x) = x (1 − N(x)−s )−1 = x (1 − q−s deg x )−1 = exp( x ∞ r=1 q
−sr deg x −1
r = exp( x r=1 (q ) r =
P P∞ (deg x)(q−s )r deg x P P∞ Nn (x)(q−s )n P −ns −1 −s
exp( x r=1 r deg x
= exp( x n=1 n
) = exp( N
n n q n ) = Z(X, q ).

Solution Appx.C.5.5. By Weil conjecture, we have dim H 1 (X, Q` ) = B1 = deg P1 (t) = 2g. So we may assume
Q
P1 (t) = 2g
i=1 (1 − αi t). Then

Q2g ∞  
− αi t) X  tr qr tr X 2g
tr 
i=1 (1
Z(X, t) = = exp   + − αri 
(1 − t)(1 − qt) r=1
r r i=1
r

Yang Pi‑Yeh 118 Hartshorne Solutions

Appendix C The Weil Conjectures

P r P
Since Z(X, t) = exp( r Nr tr ), we have Nr = 1 + qr − 2g
i=1 αi for all r ≥ 1. Moreover, by functional equation, we
r

have Z(1/qt) = ±q t Z(t), i.e.

1−g 2−2g

Q2g αi Q2g √ α
− t2−2g q1−g i=1 ( qt −
√i )
i=1 (1 qt
) q
=
(1 − 1
qt
)(1 − t)1 (1 − t)(1 − qt)

Q Q2g √ α √
So P1 (t) = 2g
i=1 (1 − αi t) = ± i=1 ( qt − q ). Comparing the coefficient, and by the fact |αi | =
√i q, we know all
αi are pairwise have αi α2g+1−i = ±q and the ± are determined by functional equation. So if we know N1 , . . . , Nr
and ±α1 α2g = . . . = ±αg αg+1 = q, they determine all αi . Hence all Nr ’s are determined by N1 , . . . , Nr .

Solution Appx.C.5.6. By 4.4.16., we have Nr = qr − ( f r + fˆr ) + 1, and f fˆ = q, a = f + fˆ ∈ Z. Then Z(E, t) =

P r P r f t)(1− fˆt) 1−at+qt2
exp( r Nr tr ) = exp( r (qr − ( f r + fˆr ) + 1) tr ) = (1−
(1−t)(1−qt)
= (1−t)(1−qt) . So P1 (t) = 1 − at + qt2 . The functional equation
√ √
is clearly by calculation. And from this, we have P1 (t) = (1 − f t)(1 − fˆt) = ±( qt − √fq )( qt − √fq ), we have
√
|a| = | f + fˆ| ≤ 2q, hence by Appx.C.5.7.(b), we have | f | = | fˆ| = q.
Q2g
(1−αi t) P P r r −1 P2g r
Solution Appx.C.5.7. (a) Since Z(X, t) = (1−t)(1−qt)
i=1
= exp( r (1− 2g i=1 αi +q )t r ), then we have Nr = 1− i=1 αi +
r
P
qr = 1 − ar + qr , hence ar = 2g i=1 αi .
r
P2g αi t P
(b) (⇐) Trivial. (⇒) Since i=1 1−αi t = ∞ r=1 ar t , and |ar | ≤ 2gq , the right‑hand side is holomorphic in
r r/2

|t| < q . So if |αi | > q for some i, the left‑hand side has a pole at t = α−1
−1/2 1/2
i in the domain |t| < q
−1/2
, which
makes a contradiction.
Q Q2g √ α −1 −1
(c) Since P1 (t) = 2gi=1 (1 − αi t) = ± i=1 ( qt − q ), we have {α1 , . . . , α2g } = {α1 /q, . . . , α2g /q}. So we may
√i
√ √ √
assume αi α2g+1−i = q. Then the assumption |αi | ≤ q means |αi | ≥ q, i.e. |αi | = q.

Yang Pi‑Yeh 119 Hartshorne Solutions