Mar 172015
 

烟花不堪剪写在豆瓣的文章, 原文标题”一些值得一读的数学论文”,   http://www.douban.com/group/topic/38870457/

Weil 建议年轻人读经典论文, 但是我认为经典的定义也会随时代而有所不同. 在Weil那个时代, Gauss, Riemann, Hilbert, Poincare 是经典. Weil 受 Gauss 启发提出了Weil 猜想, Ahlfors 受 Riemann 启发创立了covering surface 理论, Mumford 受 Hilbert 启发创立了GIT, Griffiths 受 Poincare 启发创立了variation of Hodge structure 理论··· 但是现在 Weil 时代的经典已经过于古老从而无法从中发现新东西了, 因此这个时代应该阅读适合于这个时代的经典.下面我就根据自己的兴趣罗列一些这个时代的经典著作, 由于我推荐的文章都是自己大体上读过的, 因此不可能很多. 并且因为是经典, 都是1940-2000年之间的文章. 今后读到好文章或者想到遗漏的文章时可以继续补充此贴.

1. Chern, A Simple Intrinsic Proof of the Gauss-Bonnet Formula for Closed Riemannian Manifolds
我在几何上读的第一篇文章,在这里我学到了什么是整体微分几何。Chern把局部定义的微分形式paste到一起得到了Gauss-Bonnet integrand,计算积分时又巧妙地拉回到切从上利用Poincare-Hopf指标定理来产生Euler数。此文可谓微言大义,之后Chern在整体微分几何上的几乎所有工作都能从这里找到源头。

2. Mathai-Quillen, Superconnections, Thom classes, and equivariant differential forms
对于equivariant cohomology得到了Cartan model和Weil model之间的isomorphism,这被Bott & Tu称为Mathai-Quillen isomorphism。构造了equivariant Thom class。此文最神奇之处莫过于他们给出的Thom form的精巧的几何构造,此Thom form不是compactly supported,而是Gaussian的,即在无穷远处快速衰减。Mathai-Quillen’s Thom form拉回到base manifold上依然给出bundle的Euler form,因此事实上他们的构造给出了classical Thom form的一个推广。这一现象可以用vector bundle [;E;]和它的unit ball bundle \(\Bbb BE\) 之间的diffeomorphism解释,利用此diffeomorphism pullback Mathai-Quillen’s Thom form到 \(\Bbb BE\)上,由于Mathai-Quillen’s Thom form在\(E_x\)上衰减很快,这导致pullback之后它在\(\Bbb BE\)上是紧支的,我们就得到了classical Thom form。

3. Atiyah-Bott, The moment map and equivariant cohomology
此文利用elegant的topological argument证明了所谓的Atiyah-Bott localization formula,它是当时已知的最广泛的localization formula。Atiyah和Bott发现了localization的本质就是交换代数中的module localization。之后的诸多发展都和Atiyah-Bott的发现有关,比如刘克峰给出的Witten刚性定理的简单证明,Lian-Liu-Yau的mirror conjecture证明,Pandharipahde发展的virtual localization等等。此文我已经写了相关note:http://www.douban.com/note/235066024/

4. Atiyah-Bott, The Yang-Mills equations over Riemann surfaces
Atiyah和Bott的许多工作都是对已有事实的reinterpretation,但是这些reinterpretation最终被表明非常关键。这就是Atiyah和Singer的那个故事中Atiyah面对已知的事实说“我还没有理解它”的缘故。上面的localization formula就是这样,此文是又一个例子。Atiyah-Bott在Yang-Mills connection的语言下重新formulate了Narasimhan-Seshadri定理。这导致了Kobayashi和Hitchin对于一般的Kahler manifold上的全纯向量丛提出了他们的猜想。此猜想被Donaldson-Uhlenbeck-Yau证明。此文包含了许多方面,并且写得很详细,是学习Morse-Bott理论,equivariant cohomology,gauge theory等等知识的好场所。

5. Atiyah-Bott-Shapiro, Clifford modules
此文是Cliffrod algebra的标准参考文献,Lawson-Michelsohn的Spin Geometry第一章即用来解释本文。此文在Clifford algebra的基本性质方面写得很详细,而\(K\)-theory的部分则采用了Grothendieck的语言来处理。他们的主要结果是一些由Clifford module的Grothendieck群定义出的群 \(A_\ast\),\(A_\ast^\Bbb C\)和 \(K\) 群, \(KO\)群之间的同构关系: \(A_\ast\simeq\bigoplus_{k\geq0}KO^{-k}(pt)\), \(A_\ast^\mathbb{C}\simeq\bigoplus_{k\geq0}K^{-k}(pt)\), 这就将表示和拓扑联系到一起。他们的证明用到了\(K\)和\(KO\)的Bott periodicity并且注意到他们的结果实际上推广了Bott periodicity. 他们猜测应该有一个不必用到Bott periodicity的证明, 此证明已经被找到。

6. Chern-Simons, Characteristic forms and geometric invariants
事实上,他们的工作起源于要对带边流形证明Hirzebruch signature theorem,这时boundary term并不能用拓扑不变量来表示。考虑一个4-manifold with boundary,则boundary就是一个3-fold,而boundary term就是Chern-Simons invariant。当Simons把这个现象告诉Chern时,Chern立刻意识到这实际上通过是在主丛上超度一个over determined characteristic form得到的secondary characteristic class。事实上,这也是 \(\eta\)-invariant的动机所在。带边流形上的Hirzebruch signature theorem,包括带边流形上的index theorem,最终被Atiyah-Patodi-Singer证明,boundary term的出现使得他们要用到相关的geometric invariant,因此Atiyah-Patodi-Singer的指标定理称为geometric index theorem。更多的解释见:http://www.douban.com/note/272854801/

7. Cheeger-Simons, Differential characters and geometric invariants
为了要在底流形上得到Chern-Simons class的类比,他们引进了differential character的概念。事实上,Differential character可以解释被超度的形式不为\(0\) 的一般情形的超度现象。因此,这等价于将超度现象公理化。利用differential character的概念,他们推广了Chern-Weil同态并且解释了Atiyah-Patodi-Singer的geometric index theorem和foliation的示性类。更多解释见:http://www.douban.com/note/272854801/

8. Li-Yau, On the parabolic kernel of the Schrodinger operator
Li-Yau将研究elliptic equation的方法应用于parabolic equation,自此由maximum principle得到gradient estimate,再由gradient estimate在minimizing geodesic上积分得到Harnack inequality的办法成为研究parabolic equation的固定模式. 尤其是Li-Yau method被Hamilton同来研究Ricci flow,Hamiton找到了正确的类比从而为Perelman证明几何化猜想打下了基础。从这个意义上说,此文代表了几何分析上的最高成就。此文中最关键的结果是Li-Yau的heat kernel estimate:

\[H(x,y,t)\leq C(\varepsilon,n)V_x^{-\frac12}(\sqrt{t})V_y^{-\frac12}(\sqrt{t})\exp\big(\frac{-r^2(x,y)}{(4+\varepsilon)t}+C_1\varepsilon Rt\big)\]

其中\(-(n-1)R\) 是Ricci曲率的下界. \(H(x,y,t)\) 的上界估计是类似的. 令人惊奇的是, 这个估计中居然在 \(\exp\) 中的分母上出现了 \(4\), 这让人联想到\(\mathbb{R}^n\)中的热核, 可见这个估计的精确性是无与伦比的. 事实上, 这个 \(4\) 最早出现在 parabolic Harnack inequality 中, 不难想到用 Harnack 不等式来证明 Li-Yau 估计. 利用这一估计, 他们可以部分地 recover 此前 Gromov 得到的 higher eigenvalue estimate 和 Betti number estimate.

9. Mok-Siu-Yau, The Poincare-Lelong equations on complete Kahler manifolds
很典型的几何分析文章,所使用的工具是基本的gradient estimate和Moser的elliptic Harnack inequality。详细的说明见:http://www.douban.com/note/272854801/

10. Bott-Chern, Hermitian vector bundles and the equidistribution of the zeroes of their holomorphic sections
他们观察到Nevanlinna’s first main theorem等价于对double transgression做两次积分,并应用一些基本的积分不等式。他们的这个想法是Nevanlinna理论几何化的开始。此外,此工作定义了所谓的refined Chern class和Bott-Chern form,这在日后影响深远。详细的介绍见:http://www.douban.com/note/272854801/http://www.douban.com/note/227346029/

11. Chern-Moser, Real hypersurfaces in complex manifolds
CR几何的奠基之作,完全解决了高维CR结构的equivalence problem,相应的obstruction称为Chern-Moser invariants。Chern-Moser invariants应该类比于Riemann几何上的curvature tensor,因此可以希望有一整套相应的Chern-Weil理论。事实上,此情形下Chern-Simons的类比已经被作出:http://www.douban.com/group/topic/37328263/。他们解决equivalence problem的方法是把CR结构的等价性问题转化为 \(G\)-结构的等价性问题,本质上和Tanaka prolongation是一致的,相关介绍见:http://www.douban.com/note/257579475/

12. Serre, Faisceaux Algébriques Cohérents
文章写得非常详细,可以说作为代数几何初学者的入门也不为过。此文应该和Cartan’s theorem A and B放在一起理解,正是因为要理解coherent algebraic sheaf和coherent analytic sheaf之间的联系,Serre证明了GAGA。有关介绍参考:http://www.douban.com/note/264543532/

13. Cornalba-Griffiths, Analytic Cycles and Vector Bundles on Non-Compact Algebraic Varieties
一言以蔽之,此文要对Stein manifold上证明的结果在affine variety上给出相应的定量结果,所以实际上可以看做是GAGA principle的refinement。本文的叙述非常详细,对Serre’s theorem A and B,Grothendieck’s algebraic de Rham theorem等等重要的经典结果都有介绍。本文发展的transcendental cycle的定量理论和Oka principle with growth condition都发展甚微,可见尚有潜力。详细的介绍见:http://www.douban.com/note/264543532/

14. Griffiths-King, Nevanlinna theory and holomorphic mappings between algebraic varieties
此文是equi-dimensional value distribution theory方面最重要的文章,关键在于要在projective variety \(V\)上构造singular volume form,singular points的locus是具有simple normal crossing的divisor \(D\). 他们的文章写得非常详细,是想要知道代数几何怎样用于函数论的同学的上佳读物。关于这一工作我已经多次提及,见:http://www.douban.com/note/227346029/http://book.douban.com/review/5755301/http://www.douban.com/note/260177163/http://book.douban.com/review/5768238/

15. Yau, A general Schwarz lemma for Kahler manifolds
此文是Yau关于Ricci曲率有下界的Riemann流形上gradient estimate工作的直接推论。此Schwarz lemma事实上是Yau‘s Liouville theorem,即非负Ricci曲率的完备Riemann流形上不存在非常数的有界调和函数的推论。关于完备Riemann流形上的调和函数有一大堆有趣的结果,可以参考Peter Li写的综述:http://site.douban.com/204378/widget/forum/12552929/discussion/52845693/。Ahlfors的工作表明Schwarz lemma在curve case就是Nevanlinna第二基本定理,见:http://book.douban.com/review/5755301/

16. Mckean-Singer, Curvature and the eigenvalues of the Laplacian
Mckean-Singer证明了公式 \(\textrm{ind}(\mathcal{D})=\textrm{Str}(e^{-t\mathcal{D}^2})\),其中\(\mathcal{D}\)是Dirac operator而\(\textrm{Str}\)是supertrace。事实上,他们的原始工作只讨论了\(\mathcal{D}=d+d^\ast\)的情形,这时左边就是Euler characteristic。这时Mckean和Singer猜想右边应该能够用曲率展开,得到Gauss-Bonnet integrand,这将给出Gauss-Bonnet公式的另一个证明。他们验证了此猜想在 \(n=2\)的情形,得到了古典的曲面上的Gauss-Bonnet公式。此文将分析和拓扑相联系,是heat equation方面fundamental的好工作之一。此文应该视为将Gauss-Bonnet公式解释为指标定理的一种尝试。他们的猜想4年之后被Patodi证明了。

17. Patodi, Curvature and the eigenforms of the Laplace operator
此文是对Mckean-Singer猜想的第一个证明。事实山,大量的张量计算非常复杂。不过此文中通过复杂的计算证明了张量分析上最deep的定理,即Patodi’s cancellation theorem。一个简单证明见Gilkey的书:http://book.douban.com/subject/3749059/

18. Atiyah, Convexity and commuting Hamiltonians
利用Morse-Bott理论证明了moment map的convexity。事实上,假设\(M\)不止是symplectic,而且是Kahler的,Atiyah还给出了此定理的一个refinement。事实上,Atiyah的证明等价于使symplecto-morphism的fixed locus和某个Morse-Bott function的critical submanifolds相一致,因此这个方法应该和Atiyah-Bott localization的证明方法相类比。据此,我曾经建议考察使group action的fixed locus和其他submanifold coincide是否会导致其他重要结果,特别地我关心比如当这个submanifold是 \(M\) 的某个Gromov-Hausdorff limit的情形.

19. Griffiths, Variations on a theorem of Abel
详细地讲述了古典的Abel椭圆积分理论及其推广,这揭示了Hodge理论的本来面目。Hodge理论是Griffiths一辈子工作的中心所在,因此此文应该视为Ahlfors对自己的工作动机的解释。Griffiths还得到了Abel定理的两个推广,这能够帮助我们更好地理解具有higher codimension的algebraic cycle。此文有一篇重要的续作:http://www.douban.com/group/topic/39084301/

20. Clemens-Griffiths, The Intermediate Jacobian of the Cubic Threefold
这是Griffiths一辈子最伟大的工作。事实上,Griffiths一开始对于Hodge structure的研究都是formal的,但是他慢慢发现对于cubic 3-fold,相应的period map可以很好地被理解。因为cubic 3-fold实际上是cubic curve,即elliptic curve的类比。因此,他猜想在cubic 3-fold上对curve所用的研究方法应该能够得到很好的推广,相应的global Torelli theorem应该成立。因此他和Clemens具体研究了cubic 3-fold的polarized Hodge structure,结果令人惊奇:他们不但对cubic 3-fold证明了global Torelli theorem,而且通过考察theta divisor证明了它不是rational的,这是长期以来古典代数几何中最困难的猜想。详细的介绍见:http://www.douban.com/note/264543532/

21. Kobayashi, Kähler-Einstein metric on an open algebraic manifold
利用Carlson-Griffiths构造的singular volume form,Kobayashi找到了正确的initial metric并对它在Kahler class内做deformation得到了相应的Kahler-Einstein metric。于是在affine variety \(A=V\setminus D\)上存在使\(c_1(M)<0\)得Kahler-Einstein度量。据此,Kobayashi对\(A\)推广了Miyaoka-Yau不等式。详细的介绍见:http://book.douban.com/review/5768238/

22. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge-Ampère equation
Yau证明Calabi-Yau流形存在性的文章,此文的意义不需要任何解释。Yau的方法是continuity method,因此他必须找到相应的\(C^2\)-estimate。Yau的厉害之处在于用一大堆复杂的估计把这个\(C^2\)-estimate转化为一个\(C^0\)-estimate,而\(C^0\)-estimate就简单多了。在得到了\(c_1(M)=0\)的Kahler-Einstein度量存在性之后, \(c_1(M)<0\) 的情形立刻就变得trivial了. Yau的力量在此文中表露无遗, 有志于几何分析的学生应该多了解在复几何情形怎么使用分析方法, 本文是很好的出发点.

23. Carleson, On convergence and growth of partial sums of Fourier series
Carleson证明Luzin猜想的文章,也是调和分析上的最高成就。事实上, \(L^2\) 函数的Fourier级数几乎处处收敛这个结论本身是如此令人惊奇, 这是我本科阶段最想要弄明白的定理. Carleson的方法是使用dyadic partition,然后利用一个interated process来估计Carleson operator。对证明过程的概括性解释见:http://book.douban.com/review/4884107/。事实上,此定理目前已经有了简单证明,尤其是Lacey和Thiele利用Hilbert bilinear transform的证明,可以说大大简化了Carleson的证明方式。但是我始终更喜欢Carleson的证明,他的方法是如此神圣,教会我什么是一个分析学家。

24. Carleson, Two remarks on \(H^1\) and \(BMO\)
Calrson证明了\(BMO\)中函数的一个表示定理,这个深刻的结果导致了\(H^1\) 和 \(BMO\) 对偶关系最自然的证明.\(H^1\) 和 \(BMO\) 的对偶之所以令人惊奇是因为对于很多自然出现的函数空间我们事实上根本无法找出它的对偶,例如\(L^1\) 的对偶空间in general不是 \(L^\infty\)。因此,很难想象 \(L^1\) 的类比 \(H^1\) 会有一个如此自然的对偶空间。利用此定理可以recover Hardy space中的诸多古典结果,方法是把问题转化到\(BMO\)中解决。例如著名的Hardy定理,假如

\[f(z)=\sum_{n=0}^\infty c_nz^n\in H^1,\]

则 \(\sum\limits_{n=0}^\infty|c_n|n^{-1}<\infty\).

25. Griffiths, Periods of integrals on algebraic manifolds: summary of main results and discussion of open problems
这是Griffiths对自己variation of Hodge structure理论的阶段性总结。此文中他描述了自己的动机,得到的主要结果和一些猜想及问题。就像我们多次提及的,Griffiths的工作的原始形式就是Abel椭圆积分理论。但是注意到Griffiths在这方面的第一篇论文出现在1968年,他显然还受到了Kodaira-Spencer的deformation theory的影响,这导致他定义了Hodge bundle。事实上,这就是locally constant sheaf \(R^nf_\ast\mathbb{C}\),其中\(f:\mathcal{X}\rightarrow S\). Hodge bundle导致我们能够理解对于一族Kahler流形,其Hodge结构是怎样变化的。事实上,设\(\mathcal{F}^n\subset\cdot\cdot\cdot\subset\mathcal{F}^0=\mathcal{H}\) 是Hodge bundle的Hodge filtration,\(\nabla\) 是Gauss-Manin connection,则我们有Griffiths transversality theorem \(\nabla\mathcal{F}^p\subset\mathcal{F}^{p-1}\otimes\Omega_S^1\). 此结果的non-trivialness在于它描述了Hodge结构的变化方式。据此Griffiths能够给出variation of Hodge structure的formal definition,这影响深远,它出现在Simpson关于Higgs bundle和Saito关于primitive form的工作中。90年代mirror symmetry出现之后,Hodge理论受到了前所未有的重视。

26. Chern, Geometry of  \(G\)-structures
Chern定义 \(G\)-结构的动机是equivalence problem,他后来意识到 Chern-Weil 理论最general的setting也应该是 \(G\)-结构。此文中他利用\(G\)-结构来分类各种几何,并且介绍了到1966年为止各个方面的主要成就,可以说对初学者而言是相当好的综述,可以理清领域的基本图景。补充介绍见:http://www.douban.com/note/257579475/

27. Atiyah, Vector bundles over an elliptic curve
Grothendieck分类了\(\mathbb{P}^1\)上的向量丛,证明了任意的向量丛 \(E\rightarrow\mathbb{P}^1\) 都可以分解为线丛的直和,这是对向量丛模空间的最早研究。事实上,Grothendieck的结果应该理解为\(\Bbb{P}^1\)上indecomposable的向量丛都是线丛。由于\(\mathbb{P}^1\) 的线丛被它们的Chern class决定,因此\(\mathbb{P}^1\) 上indecomposable的向量丛的模空间的维数是0。在genus大于0情形这是不对的,因为Jacobi inversion theorem表明\(\textrm{Pic}^d(C)=J(C)\),其中\(J(C)\)是代数曲线\(C\)的Jacobian variety, \(d\)是线丛的degree。Atiyah对于\(C\)是elliptic curve的情形研究了 \(C\)上不可约向量丛的模空间,这推广了Grothendieck的工作。用\(\textrm{Bun}(r,d)\)表示rank \(r\),degree \(d\)的向量丛的模空间,则 \(\textrm{Bun}(1,d)\simeq J(C)\simeq C\). 通过take transition function的determinant我们可以定义映射 \(\det:\textrm{Bun}(r,d)\rightarrow\textrm{Bun}(1,d)\)。Atiyah的主要结果是证明了映射\(\det\)的mapping degree是\(\gcd(r,d)\),因此当 \(\gcd(r,d)=1\) 时我们有 \(\textrm{Bun}(r,d)\simeq\textrm{Bun}(1,d)\simeq C\).

28. Guuning-Narasimhan, Immersion of open Riemann surfaces
此文是为了证明任意的open Riemann surface可以immerse到 \(\mathbb{C}\) 中,因此可以被realize成一个Riemann domain。事实上,他们的证明是open Riemann surface诸多良好性质的结果:任何一个open Riemann surface都能被嵌入到\(\mathbb{C}^3\)里,特别地,它们都是Stein manifold。open Riemann surface上任意的holomorphic vector bundle都是trivial的,因此\(H^1(X,\mathcal{O}_X^\ast)=0\)。另一方面,根据Cartan’s theorem B, \(H^1(X,\mathcal{O}_X)=0\). 因此利用exponential sheaf sequence我们得到\(H^2(X,\mathbb{Z})=0\). 这个事实对他们的证明是关键的。

29. Donaldson, Remarks on gauge theory, complex geometry and 4-manifold topology
作者试图将Hermitian-Einstein问题和 \(c_1(M)>0\) 情形的Kahler-Einstein问题纳入到同一框架。事实上,注意到\(\mu:\bar{\partial}\rightarrow\Lambda F(\bar{\partial})\)定义了unitary connection的无限维空间\(\mathcal{C}\)上的一个moment map,Kobayashi-Hitchin猜想等价于说\(\mathcal{C}_s/\mathcal{G}^\mathbb{C}\simeq\mu^{-1}(0)/\mathcal{G}\),其中\(\mathcal{C}_s\) 表示\(\mathcal{C}\)  的stable locus,\(\mathcal{G}\) 是gauge group。注意这个事实在finite dimensional case 就是 Kempf-Ness 定理,Kobayashi-Hitchin 猜测它在这个particular infinite dimensional case成立. Donaldson 试图对 Kahler-Einstein 情形 reformulate 出同样的表述. 另外这时 stability 的定义涉及到\(\mathcal{G}^\mathbb{C}\)-equivariant line bundle \(\mathcal{L}\), 它是一个determinant line bundle。这时我们应该考虑紧致辛流形\((M,\omega)\)上的compatible的近复结构的无限维空间\(\mathcal{J}\),\(\mathcal{G}\) 是symplectomorphism的identity component。Donaldson的主要结果是观察到这时可以定义推广的scalar curvature(在integrable case就是scalar curvature: \(\textrm{Sc}\omega^n=n!i\Omega\wedge\omega^{n-1}\), 并且证明了\(J\rightarrow\textrm{Sc}(J)\) 定义了\(\mathcal{J}\) 上的moment map。这样看来,Kahler-Einstein问题似乎应该对almost complex case推广。在integrable case,\(\mathcal{J}^{\textrm{int}}\subset\mathcal{J}\) 上有一自然的foliation,它的任意一个leaf可以看做固定的Kahler class中的Kahler metric。

30. Gross, Topological Mirror Symmetry
在SYZ框架下构造mirror pair的第一篇重要文章,作者topologically构造出了quintic 3-fold及其mirror。事实上,start with integral affine manifold \(B\), SYZ将mirror pair视为dual Lagrangian torus fibration \(T_B/\Lambda\) 和\(T^\ast_B/\check{\Lambda}\), 其中\(\Lambda\) 和\(\check{\Lambda}\) 是 dual lattices. 假如要得到有意义的例子,那么这两个 Lagrangian torus fibration 应该允许 singular fiber. Gross证明了对于quintic 3-fold case, 假如允许codimension 2 的singular locus, 那么 topologically 可以从 affine \(S^3\) with singularity 构造出quintic及quintic mirror. 事实上 quintic 及其 mirror 在拓扑上的根本区别在于交换了 positive fiber 和 negative fiber 的位置, 这些欧拉示性数不为 \(0\) 的 fiber 出现在一些分散的点上.

 Posted by at 5:47 am
Aug 062013
 

质数 \(k\)-tuples 猜想和 \(\pi(m+n)\leqslant\pi(m)+\pi(n)\) 是 Hardy 和 Littlewood 提出的两个关于质数分布的猜测. 习惯上, 人们也把前一个猜想称为第一 Hardy-Littlewood 猜想(Prime \(k\)-tuple), 后一个称为第二 Hardy-Littlewood 猜想(Second Hardy–Littlewood conjecture). 这两个猜想都还没有解决, 但数学家们倾向于认为质数 \(k\)-tuples 猜想是正确的, 并且存在无穷多组正整数 \(m,n\), 使得 \(\pi(m+n)\gt\pi(m)+\pi(n)\).

质数 \(k\)-tuples 猜想

整数 \(k_0\geqslant1\), \(k_0\)-tuples

\[\mathcal H=(h_1,h_2,\dotsc,h_{k_0}),\]

这里 \(h_1,h_2,\dotsc,h_{k_0}\) 是 \(k_0\) 个互不相同的整数, 并且 \(h_1\lt h_2\lt \dotsb\lt h_{k_0}\). 那么, 是否存在无穷多个 \(n\), 使得 \(n+\mathcal H=(n+h_1,n+h_2,\dotsc,n+h_{k_0})\) 全部由质数组成? \(k_0=1\) 就是质数的无限性. 至于 \(k_0=2\), 就很困难了, \(\mathcal H=(0,2)\) 即为当前搅翻数学界的孪生质数猜想.

显然的, 不能指望对于任意的 \(\mathcal H\), 都有这么好的结果. 就拿 \(\mathcal H=(0,1)\) 来说, 每个 \(n+\mathcal H=(n,n+1)\) 由两个相邻的整数组成, 只有 \((2,3)\) 包含两个质数. 一般来说, 如果存在质数 \(p\), 使得可以从 \(\mathcal H\) 中选出 \(p\) 个数作为 \(\bmod  p\) 的完系, 那么对于任意的正整数 \(n\), \(n+\mathcal H\) 至少包含一个 \(p\) 的倍数. 于是, 只可能有有限个 \(n\), 使得 \(n+\mathcal H\) 全部由质数组成.

于此, 我们必须对 \(\mathcal H\) 添加一些限制条件, 来避免这种情况. \(k_0\)-tuples \(\mathcal H\) 称为允许的(admissible), 如果对于任意质数 \(p\), 总存在 \(\bmod  p\) 的至少一个剩余类, 使得 \(\mathcal H\) 不包括这剩余类的任何数. 作为例子, \((0,2), (0,2,6)\) 都是允许的, 但 \(0,2,4\) 不被允许, 因为 \(0,2,4\) 是 \(\bmod  3\) 的完系.

现在, 我们可以正式的把 Hardy-Littlewood 质数 tuples 猜想陈述如下:

Hardy-Littlewood prime tuples conjecture  如果 \(k_0\)-tuples \(\mathcal H\) 是允许的, 那么, 存在无穷多个正整数 \(n\), 使得 \(n+\mathcal H\) 全部由质数组成.

这个猜想太困难了. 实际上, 数学家还没有找到任何一个允许的 \(k_0\)-tuples \(\mathcal H\)\((k_0\geqslant2)\), 来证明这猜想是成立的. 仅仅是前不久, 张益唐的工作的横空出世, 我们才得知, 存在一个正整数 \(h\), 满足 \(0\lt h\lt 70,000,000\), 使得这猜想对于 \((0,h)\) 是成立的. 虽然, Tao(陶哲轩), Ben Green 等人接下来的工作, 已经把 \(70,000,000\) 大大降低, 目前是 \(5414\), 但我们仍然不清楚 \(h\) 到底是多少.

Second Hardy–Littlewood conjecture

1923 年, Hardy 和 Littlewood 发表了一篇论文[1]. 这篇长达 \(70\) 页, 已经是数论史上的经典, 的论文提出, 对任意整数 \(m,n\geqslant2\),

\[\pi(m+n)\leqslant\pi(m)+\pi(n).\]

这猜想, 如今被冠名为第二 Hardy–Littlewood 猜想.

质数 tuples 猜想与第二 Hardy–Littlewood 猜想不能同时成立

1974 年, Ian Richards 和他的博士研究生 Douglas Hensley 指出[2], Hardy-Littlewood 的两个猜想, 是不相容的. 也就是说, 这两个猜想, 至少有一个是不成立的.

张益唐的工作与此有关. Engelsma 指出, 如果质数 tuples 猜想为真, 那么, 存在无穷多个正整数 \(n\), 使得

\[\pi(n+3159)-\pi(n)=447\gt\pi(3159)=446.\]

Annotations

  1. 第一部分, 对于质数 tuples 猜想的介绍, 参考了 Tao 的一篇 blog.

References

  1. G. H. Hardy and J. E. Littlewood, On some problems of “partitio numerorum III: On the expression of a number as a sum of primes. Acta Math, 1923, 44: 1–70.
  2. D. Hensley and I. Richards, Primes in intervals. Acta Arith. 25 (1974), pp. 375-391.
  3. conjectures concerning primes; a discussion of the use of computers in attacking a theoretical problem. Bulletin of the American Mathematical Society 80:3 (1974), pp. 419-438.
Oct 032012
 

Number theory will be understood, not as a collection of tricks and isolated results, but as a coherent and interconnected theory.

算术基本定理最早的准确表述与证明, 应该是出自 Gauss 的名著算术研究. 但是, 在这之前很久, 人们似乎就已经知道这个定理的具体内容, 并且已经广泛使用.

很明显, 这个结果是重要的! \(\sqrt2\) 是无理数的经典证法也依仗它! 当然, 定理也是漂亮的!

Green(与 Tao 合作 Green–Tao theorem 的那位)的(博士)导师 Gowers, 写了两篇关于此定理的文章: Why isn’t the fundamental theorem of arithmetic obvious?  和 Proving the fundamental theorem of arithmetic.

定理不是显然的. 流行的证明, 大家都知道, 是利用 Bézout 恒等式还有归纳法. 罕见的一个证明来自英国大数学家 Hardy 的数论导引, 如今名为哈代数论, 的第二章的最后: 利用最小数原理.

陈省身说, 好的数学就是可以引导出很多后来发展的数学! 所以, 题外话是: 竞赛数学不是好的数学!

对于算术基本定理来说, 很遗憾, 在一般的数域中, 在更广泛的范围, 并不成立. 20世纪最伟大的数学家 Hilbert 曾经举个一个例子: 唯一分解在

\[H=\{1,5,9,13,17,\dotsc\}\]

中不成立. 众所周知的一个事情是: 如果唯一分解总是成立的的话, 著名的 Fermat 大定理就太 easy 了.

算术基本定理诱导了如唯一分解整环, Euclid 整环等等概念. Gauss 为了描述唯一分解成立的程度, 引进了类数的概念. 大致上, 类数越大, 一个数被分解成质数的分法个数就越多. 虚二次域 \(Q( \sqrt D)\) 的 Gauss 猜想已经解决, 但实二次域的情形, 现在还是木有证明的事情.

关于理想 ideal, 倒是有一个相应的唯一分解定理: \(O_k\) 的任意非零真理想可以唯一写成质理想的积, 不考虑顺序的话.

 Posted by at 11:44 am
Aug 212012
 

If \(p,q\) are distinct odd primes,  then

\[\left( \frac pq\right) \left( \frac qp\right) = (-1)^{\frac{(p-1)(q-1)}4},\]

where \(\left( \frac{}{}\right)\) is the Legendre symbol. 这就是被 Gauss 称为”数论酵母” 的二次互反律.

自 Legendre 的那个没有完成的证明以来, 据 Reciprocity Laws: From Euler to Eisenstein 的作者 Franz Lemmermeyer 统计, 发表的证明是 \(240\) 个. 可以预见的到, 这个数字还会不断增加. 这些证明的作者, 发表年份,使用的方法以及文献的详细列表可在 Lemmermeyer 的个人主页找到. 其中, \(1889\) 这一年就有 \(6\) 个证明公布, \(1893,1951,1961\) 年也各有 \(5\) 个证明发表在不同的期刊上. 仅看从 \(1950\) 年到今天的这 \(60\) 多年, 只有 \(1956,1959,1968,1970,1975,1977,1982,1986,1988,1996,2002\) 这 \(11\) 年没有证明发表.不过, Lemmermeyer 的统计好像有小错误, 例如 Wouter Castryck 在 \(2008\) 年发表了一篇文章, 办法类似于 V. A. Lebesgue 在 \(1838\) 年的论证, 但 Lemmermeyer 的统计认为 Castryck 的证明在 \(2007\) 年给出.

这么多的证明, 想要全部分门别类, 整理好写出来, 肯定是困难的. 文献多, 不容易都找到. 即便都找齐了, 也因为是不同的语言, 也不能都看懂. 这些方法各有繁简. 哪个证明才是最简单的呢? 显然, Proofs from THE BOOK 给出的两个途径, 是比较繁琐的; Jean-Pierre Serre 在他的 A Couse in Arithmetic 第一章给出的都依靠 Gauss 引理的两个证明, 也不算太简单. 那么, 到底哪个才是最简单的呢? 各人的看法可能有不同. 但, 美妙的, 能给人以深刻印象, 令人荡气回肠的证明, 应该有一些共同的特征, 或者说一个证明要能成为好的证明, 美丽的证明, 应该有一定的门槛, 满足一定的条件. 我们试着列出这些条件:

  • 简单是首要条件. 基于简单的想法, 能揭示问题的本质, 加深对事物的理解.
  • 美. 构思巧妙, 论证精妙方能展现出深刻. 当然, 简单其实也是一种美.
  • 自然. 方法能用在更广泛的地方, 解决更多的问题.

同时满足这些要求的证明不容易找到, Castryck 的证明大概可以满足前两条. 下面是他的详细论证:

For any odd \(n \in\Bbb N\), denote by \(N_n\) the number of solutions in \(\left( \Bbb Z / (q)\right)^n\) to the equation

\[x_1^2-x_2^2+x_3^2-\dotsb + x_n^2 = 1.\]

If we substitute \(x_1 \gets x_1+x_2\), then we get

\[x_1^2+x_3^2-\dotsb + x_n^2-1 = -2x_1x_2.\]

For any non-zero \(x_1\)-value and any values of \(x_3,\dotsc,x_n\), there is a unique corresponding \(x_2\)-value. If \(x_1=0\), there are no solutions, except if \(x_3^2-\dotsb + x_n^2 = 1\) (which happens in \(N_{n-2}\) cases): then all possible values of \(x_2\) do the job. We find that

\[N_n = q^{n-2}(q-1)+qN_{n-2}.\]

and hence \(N_n = q^{n-1}+q^{\frac{n-1}{2}}(N_1-1) = q^{n-1}+q^{\frac{n-1}{2}}.\) In particular,

\begin{equation}N_p\equiv1+\left(\frac qp\right) \pmod p.\end{equation}

Next, \(N_p\) can be classically determined as

\[\sum_{t_1+\dotsb + t_p = 1}N(x_1^2 = t_1)N(x_2^2 = -t_2)N(x_3^2=t_3)\dotsm N (x_p^2=t_p),\]

where the \(t_i\) are in \(\Bbb Z/(q)\) and \(N(\cdots)\) denotes the number of solutions to the corresponding univariate equation. This can be rewritten as

\[\sum_{t_1+\dotsb+t_p=1}\left(1+\left(\frac{t_1}q\right)\right)\left( 1+\left( \frac{-t_2}q\right)\right)\left( 1+\left( \frac{t_3}q\right)\right)\dotsm \left( 1+\left( \frac{t_p}q\right)\right),\]

When expanding out the product, only the terms \(1\cdot 1 \cdot 1 \cdots 1\) and \(\left(\frac{t_1}q\right)\cdot\left(\frac{-t_2}q\right)\cdot\left( \frac{t_3}q\right)\cdots\left( \frac{t_p}q\right)\) should be taken into consideration; the other terms disappear because Legendre symbols sum up to zero: \(\sum\limits_{t\in\Bbb Z/(q)}\left(\frac tq\right) = 0.\) Therefore, the above expression simplifies to

\[q^{p-1}+\left( \frac{(-1)^{\frac{p-1}2}}q\right)\sum_{t_1+\dotsb+t_p = 1}\left(\frac{t_1t_2t_3\dotsm t_p}q\right).\]

Modulo \(p\), the latter sum almost completely vanishes, since the tuples \((t_1,\dotsc,t_p)\) satisfying \(t_1+\dotsb + t_p = 1\) with not all \(t_i\) equal to \(p^{-1}\) can be collected in groups of size \(p\) by cyclic permutation. Note that \(p\) is indeed a multiplicative unit in \(\Bbb Z/(q)\). We thus obtain

\begin{equation}N_p\equiv1+\left( \frac{(-1)^{\frac{p-1}2}}q\right) \left( \frac{p^{-p}}q\right) \equiv1+(-1)^{\frac{p-1}2\frac{q-1}2}\left( \frac pq\right) \pmod p.\end{equation}

The last congruence follows from Euler’s criterion \(\left(\frac aq\right) \equiv a^{\frac{q-1}2} \pmod q\) and the observation that \(p^{-p}\) is a square in \(\Bbb Z/(q)\) if and only if \(p\) is a square in \(\Bbb Z/(q)\).

Comparing \((1)\) and \((2)\), the reciprocitylaw follows. \(\Box\)

接下来的论证, 是 Aurelien Bessard (2010) 改编自 V. A. Lebesgue 和 Eisenstein的一个证明:

Let \(p = 2m+1\) and \(q\) be distinct odd primes and let \(N\) denote the number of solutions of the equation

\[ x_1^2 + \ldots + x_p^2 = 1 \]

in the finite field \(\Bbb F_q\).

The group \(\Bbb Z/p\Bbb Z\) acts on the solution space \(X\) by shifting indices: if \((x_1, \ldots, x_p) \in X\), then so is \((x_a,x_{a+1}, \dotsc)\) for each \(a \in {\mathbb Z}/p{\mathbb Z}\), where the indices have to be read modulo \(p\). Each orbit has exactly \(p\) elements except if there is an \(x\) with \((x,x,\ldots,x) \in X\): the orbit of this element has \(1\) element. Now \((x,x,\ldots,x) \in X\) if and only if \(px^2 = 1\) is solvable in \(\Bbb F_q\), hence

\begin{equation} N \equiv \Big( \frac pq \Big) + 1 \pmod p.\end{equation}

We make a change of variables to transform the diagonal equation into an equation where counting the number of solutions is easier. To this end, consider the matrix

\[ A = \left( \begin{matrix} 0 & 1 & & & & & & \\ 1 & 0 & & & & & & \\ & & 0 & 1 & & & & \\ & & 1 & 0 & & & & \\ & & & & \ddots & & & \\ & & & & & 0 & 1 & \\& & & & & 1 & 0 & \\& & & & & & & a\end{matrix} \right) \]

with \(a = (-1)^{\frac{p-1}2}\). Since \(\det A = 1\), this matrix is congruent to the unit matrix, hence \(X\) and the solution spaces \(X^\prime\) of the equation \(x^T A x = 1\),i.e.,recall that \(p=2m+1\), of

\[2(y_1z_1+\dotsb+y_mz_m)+ax_p^2=1\]

are isomorphic.

For counting the number of solutions of \(X^\prime\), observe that if \((y_1, \dotsc, y_m) = 0\), we have \(q^m(1+a^{\frac{q-1}2})\) possibilities for choosing \(z_1, \dotsc, z_m\) and \(x_p\).

If \(y = (y_1, \dotsc, y_m) \ne 0\), on the other hand, then for each choice of \(y\) and \(x_p\) we have to count the number of points on a hyperplane of dimension \(m\); there are \(q^{m-1}\) points on such a hyperplane, and the number of overall possibilities in this case is \((q^m-1) \cdot q \cdot q^{m-1} = q^m(q^m-1)\).

Thus we find

\begin{equation}\begin{split} N & = q^m (1+a^{\frac{q-1}2})+q^m(q^m-1) \\&=q^m(q^m+(-1)^{\frac{p-1}2\frac{q-1}2})\\ & \equiv\Big(\frac qp\Big)\bigg(\Big(\frac qp\Big)+(-1)^{\frac{p-1}2\frac{q-1}2}\bigg)\\&\equiv1+(-1)^{\frac{p-1}2\frac{q-1}2}\Big(\frac qp\Big)\pmod p.\end{split}\end{equation}

Comparing \((3)\) and \((4)\), gives the quadratic reciprocity law. \(\Box\)

Aug 092012
 

依赖相对迹公式方面的最新成果, 同余数(congruent number)最近有所进展.

其实, 我第一次从不定方程的书上了解到何为同余数的时候, 并不称为同余数, 而被冠名合同数. 同余数就是这样的 \(n\in\Bbb N^+\), 存在一个边长为有理数的直角三角形, 其面积为 \(n\). 边长为有理数的直角三角形被定义为有理三角形(有理三角形在不同的环境有不同的定义,比如有些作者不要求是直角三角形,也有人把直角三角形这个条件换成面积是有理数). 当有理三角形的边长都是整数的时候, 又称为勾股三角形.

哪些 \(n\) 是同余数?  有无简单的判定方法? 如果 \(n\) 是同余数, 请给出一个面积为 \(n\) 的有理三角形. 这些问题古老而困难, 目前仅有部分结果. 同余数和椭圆曲线, BSD猜想(Birch and Swinnerton-Dyer conjecture)联系甚大.

第一个结果是 André Weil 的 Number Theory: An Approach Through History From Hammurapi to Legendre (1984) 第二章 \(10\) 的主题.

定理 \(1\)  任意 \(n\in\Bbb N^+, n^2,\, 2n^2\) 不是同余数.

Jul 222012
 

意大利几何学家 L. Mascherom 于 \(1797\) 年证明了, 凡是能用尺规作出的几何图形都可以仅仅使用圆规来完成. \(1833\) 年, Jakob Steiner 根据 Poncelet 的想法, 指出: 如果给定一个圆和它的中心, 那么, 能用尺规作出的几何图形都能单独用直尺来作出.

我一直很好奇, 这些结论究竟是如何证明的? Ross Honsberger 在他的书, “Ingenuity in Mathematics”, 的第 \(15\) 节提供了一个详尽的证明, 这一节的标题是: Mascheroni and Steiner. \(1994\) 年, Norbert Hungerbuhler 给出了 Mascherom 的结果, 所谓的Mohr–Mascheroni定理, 的一个简单的证明.

Jul 062012
 

开篇

Zsigmondy’s theorem states that if \(a>b>0\) and \(n>1\) are positive integers, then

  1. \(a^n+b^n\) has at least one prime factor that does not divide \(a^k+b^k\) for all positive integers \(k<n\), with the exception of \(2^3+1^3\);
  2. if \((a,b)=1\), then there is a prime number that divides \(a^n-b^n\) and does not divide \(a^k-b^k\) for all positive integer \(k<n\) unless
  •   \(a=2,b=1\) and \(n=6\); or
  •   \(a+b\) is a power of \(2\) and \(n=2\).

这就是如今人们常常谈到的 Zsigmondy 定理, \(2\) 实际上更为常见.  Zsigmondy 当初其实允许 \(b<0\), 此时的例外当然要稍作修改, 比如 \(2\) 要增加 \(a=2,b=-1,n=3\).

此定理经常在群论(group theory)中大展拳脚, 在数论中当然也有很多应用.

这个定理的完整证明很难找到, 容易搜索到的是 \(b=1\) 时, 使用分圆多项式(cyclotomic polynomial)对\(2\)的证明.

历史

在 Zsigmondy\(1892\)年的论文出现之前, Bang已经在\(1886\) 年证明了\(b=1\) 的情形, 即所谓的 Bang定理. 至于 Bang 当年的文章,是否包括了 Zsigmondy 定理的两种情形, 我无法得知, 因为没有看到原文. 后来, 又有不少人重新发现 Zsigmondy 定理或者 Bang 的结果, 也可能是某种特殊情况下的结论, 当然也有人给出新的证明. Birkhoff 和 Vandiver \(1904\) 年的论文 \([1]\) 用初等的办法证明了 \(2\). 这个证明, 稍后我们会详细谈到.

Carmichael \(1913\) 年的文章 \([2]\) 很值得一看, 他实际上证明了对 Lucas 序列(Lucas sequence)

\begin{equation}U_n=\frac{a^n-b^n}{a-b},\,V_n=a^n+b^n\end{equation}

而言, 结论一样成立. 尽管 \(a^n-b^n\) 与 \(U_n\) 确有许多相似, 但是毕竟差别还是有的, \(U_n\) 的本原质因数并不见得就是 \(a^n-b^n\) 的本原质因数. 例如, 当 \(a=5,\,b=2,\,n=3\) 时, \(U_3=\frac{5^3-2^3}{5-2}\) 的本原质因数 \(3\) 并不是 \(5^3-2^3\) 的本原质因数.

Artin 在研究线性群(linear group)的时候, 用初等办法建立了Zsigmondy定理的 \(2\), 并且允许 \(a,b\) 可以为负, 见 \([3]\).

准备工作

继续下文之前, 先看几个简单的事实.

引理\(1\) 当 \(a,b\) 互质, 也即 \((a,b)=1\) 之时,

\begin{equation}(a^m-b^m,\,a^n-b^n)=a^{(m,n)}-b^{(m,n)}\end{equation}

成立, 这里 \(m,n\in\Bbb N^+.\)

\((2)\) 虽然简单, 却从未在哪本书上见过. 数论书上只有当 \(b=1\) 时的特殊情形, 所以这里算是一个推广. 至于证明, 可以归纳完成, 也可以选择 Bézout 恒等式(Bézout’s identity), 与 \(b=1\) 没有本质不同.

下面来探讨阶的性质.

设正整数 \(m>1,a\in\Bbb Z,(a,m)=1.\) 记 \(a\) 模 \(m\) 的阶为 \(\delta_m(a).\) 正整数 \(r\) 使得

\begin{equation}a^r\equiv-1\pmod m\end{equation}

成立.

引理\(2\) 设满足 \((3)\) 的最小正整数为 \(r\), 那么

  • 若 \(a^n\equiv-1\pmod m, n\in\Bbb N^+\), 则 \(r|n;\)
  • \(\delta_m(a)=\begin{cases}\;r,\quad m=2;\\2r, \quad m\geqslant3.\end{cases}\)

引理\(3\)  \(p\) 为质数, \(a,b,n\in\Bbb Z, n>0, p\nmid ab, p^\alpha\parallel n, p^\beta\parallel (a-b), \alpha,\beta\in\Bbb N.\) 如果在 \(p=2\) 时, \(\beta\geqslant2\); 在 \(p\geqslant3\) 时, \(\beta\geqslant1,\) 那么 \(p^{\alpha+\beta}\parallel (a^n-b^n).\)

设 \(a=b+tp^\beta, p\nmid t,\) 则

\[a^n-b^n=(b+tp^\beta)^n-b^n=\sum_{i=1}^n{n\choose i}b^{n-i}(tp^\beta)^i.\]

当 \(i\geqslant2\) 时, 由 \({n\choose i}(tp^\beta)^i=\dfrac ni{n-1\choose i-1}(tp^\beta)^i, (p^\beta)^{i-1}\geqslant3^{i-1}>i,\) 可得 \(p^{\alpha+\beta+1}|{n\choose i}b^{n-i}(tp^\beta)^i\); \(p^{\alpha+\beta}\parallel{n\choose1}b^{n-1}(tp^\beta).\) 综合起来, 我们的证明得以完成.

本原因子的定义与性质

若 \(a^n\pm b^n\) 的某质因数不整除 \(a^k\pm b^k(k=1,\,2,\,\dotsc,\,n-1),\) 我们把这样的质因子叫做本原质因数. 也可以考察 \(a^n\pm b^n\) 的不是质数的因数, 这个时候, 要把整除不整除换成互质不互质: 若 \(a^n\pm b^n\) 的某因数与 \(a^k\pm b^k( k=1,\,2,\,\dotsc,\,n-1)\) 都互质, 我们把这样的因数叫做本原因数.

本原因数要求与 \(a^k\pm b^k, k<n\) 都互质, 所有的 \(n-1\) 个数 \(a\pm b, a^2\pm b^2,\dotsc,a^{n-1}\pm b^{n-1}\) 一个不落下. 能否把这个条件换成表面上更弱,实际上等价的条件, 使得利用这个定义的时候, 更加得心应手?

借助阶的性质以及引理 \(2\), 很容易就得到了第一个定理.

定理 \(1\)  记 \(n\) 的 \(d(n)\)个因数是 \(d_1=1,d_2,\dotsc,d_{d(n)}=n\). 如果 \(a^n\pm b^n\) 的一个因数与 \(a^k\pm b^k( k=d_1,d_2,\dotsc,d_{d(n)-1})\) 都互质, 那么这个因数是 \(a^n\pm b^n\) 的本原因数.

引理 \(1\) 也可以很方便的完成就 \(W_n\) 这种情况的证明, 这里

\begin{equation}W_n=a^n-b^n.\end{equation}

下一条性质出自 Euler.

定理\(2\)   \(d\) 是 \(W_n\) 的本原因数, 则 \(d\equiv1\pmod n.\)

\(2\) 的初等证明

参考文献

  1. Geo. D. Birkhoff and H. S. Vandiver, On the Integral Divisors of \(a^n – b^n \), The Annals of Mathematics, \(1904, 173-180\).
  2. Carmichael, On the Numerical Factors of  the Arithmetic Forms \(a^n \pm b^n \), The Annals of Mathematics, \(1913, 30-70\).
  3. Artin, The orders of the linear groups, Communications on pure and applied mathematics, vol 7, \(1955, 355-366\).
Jun 262012
 

                                      定义

定义 1[离散形式] 称 \( p=(p_1,p_2,\dotsc,p_n)\)Majorization(控制) \( q=(q_1,q_2,\dotsc,q_n)\), 记为 \(p \succ q\), 如果\( \overline{p}=(\overline{p_1} ,\overline{p_2} ,\dotsc, \overline{p_n}) \) 与 \(\overline{q}=(\overline{q_1} ,\overline{q_2} ,\dotsc, \overline{q_n})\) 分别是 \(p\) 与 \(q\)  的重新排序, 使得 \( \overline{p_1} \geq \overline{p_2} \geq \dotsb \geq \overline{p_n},  \overline{q_1} \geq \overline{q_2}\geq\dotsb \geq\overline{q_n}\), 并且满足如下两个条件:

  • \(\sum\limits_{i=1}^k \overline{p_i} \geq \sum\limits_{i=1}^k \overline{q_i} \), 当 \(k=1,2,\dotsc,n-1\)时;
  •  \(\sum\limits_{i=1}^n p_i = \sum\limits_{i=1}^n q_i \).

定义 2[积分形式] 设 \( f(x), g(x)\) 是区间 \([a,b]\) 上的递增函数,称 \(f \) Majorization(控制) \(g\), 记为 \(f \succ g\),  如果

  •  \(\int_a^xf(t)\mathrm{d}t \geq \int_a^x g(t)\mathrm{d}t \), 当\(a<x<b\)时;
  •  \(\int_a^bf(t)\mathrm{d}t = \int_a^bg(t)\mathrm{d}t \).

定义 3[\(p\)-平均] 设 \( x_1,x_2,\dotsc,x_n\) 为正实数, \(p=(p_1,p_2,\dotsc,p_n)\in \Bbb R^n.  x_1,x_2,\dotsc,x_n\) 的 \(p\)-平均 定义为

\[ [p]= \frac{1}{n!} \sum_{\sigma \in S_n}\prod_{i=1}^nx_{\sigma(i)}^{p_i}, \]

这里 \(S_n\) 是 \(1,2,\dotsc,n\) 的所有排列组成的集合.

 

                                     主要结果

Majorization inequality (控制不等式)

  1. 离散形式   设 \( p=(p_1,p_2,\dotsc,p_n), q=(q_1,q_2,\dotsc,q_n)\),  所有的 \(p_i,q_i \in (a,b)\). 若 \(p \succ q, \varphi(x)\) 为区间 \((a,b)\) 内的凸函数(Convex function),则下式为真 \begin{equation}\sum_{i=1}^n\varphi(p_i)\geq \sum_{i=1}^n\varphi(q_i).\end{equation}
  2. 积分形式    \( f(x), g(x)\) 是区间 \([a,b]\) 上的递增函数, \(f \succ g, \varphi(x)\) 在区间 \(a,b]\) 上是连续凸函数,则下式为真 \begin{equation}\int_a^b\varphi(f(x))\mathrm{d}x \geq \int_a^b\varphi(g(x))\mathrm{d}x.\end{equation}

Muirhead’s Inequality   设 \( x_1,x_2,\dotsc,x_n\) 为正实数, \(p,q \in \Bbb R^n \). 如果 \(p \succ q\), 则 \([p] \geq [q]\); 并且当 \(p \ne q\) 时, 等号成立当且仅当 \(x_1=x_2= \dotsc =x_n\).

 

                                  主要结果的证明

下面是这些定理的证明,先从最简单的 Majorization inequality开始!

Majorization inequality 的离散形式证明  记 \(r_i=\frac{\varphi(q_i)-\varphi(p_i)}{q_i-p_i}\), 则 \(r_i\) 递减,这是因为 \(\varphi(x)\) 是凸函数.

\begin{equation}P_k=\sum_{i=1}^kp_i,Q_k=\sum_{i=1}^kq_i, k=1,2,\dotsc, n; P_0=0,Q_0=0.\end{equation}

于是, 当 \(i=1,2,\dotsc,n-1\) 时, \(P_i \geq Q_i\), 而 \(P_n=Q_n\).

由于

\begin{equation}\begin{split}\sum_{i=1}^n\varphi(p_i)-\sum_{i=1}^n\varphi(q_i) & = \sum_{i=1}^n(\varphi(p_i)-\varphi(q_i) ) \\& = \sum_{i=1}^n r_i (p_i-q_i) \\& = \sum_{i=1}^n r_i (P_i-P_{i-1}-Q_i+Q_{i-1}) \\& = \sum_{i=1}^n r_i (P_i-Q_i) – \sum_{i=1}^n r_i (P_{i-1}-Q_{i-1}) \\& =\sum_{i=1}^{n-1} r_i (P_i-Q_i) – \sum_{i=0}^{n-1} r_{i+1} (P_i-Q_i)\\&= \sum_{i=1}^{n-1}(r_i- r_{i+1})(P_i-Q_i),\end{split}\end{equation}

注意 \(i=1,2,\dotsc,n-1\) 时, \(r_i \geq  r_{i+1}\), 因之我们的证明得以完成.

 

Muirhead’s Inequality 的证明不少, 先看一个比较传统的:

Muirhead’s Inequality 的归纳证明   先看 \(n=2\)的情况.

我们来指出: 若 \((n,m) \succ (p,q), x,y > 0\), 则

\begin{equation}x^ny^m+x^my^n \geq x^py^q + x^qy^p,\end{equation}

等号成立当且仅当 \(x=y\) 或者 \( n=p,m=q\).

事实上, 记 \(w= \frac{n-q}{n-m}\), 因为 \(n\geq  p \geq q \geq  m\), 于是 \(0 \leq w \leq1, 0 \leq  1-w \leq  1\).

由 Generalized mean inequality(幂平均不等式),得

\begin{equation}\begin{split}x^ny^m+x^my^n & = wx^ny^m + (1-w)x^my^n +wx^my^n +(1-w)x^ny^m \\& \geq x^{wn} y^{wm}x^{(1-w)m}y^{(1-w)m} +x^{wm}y^{wn}x^{(1-w)n}y^{(1-w)m} \\& = x^{wn +(1-w)m}y^{wm+ (1-w)m}+ x^{wm+(1-w)n}y^{wn+(1-w)m} \\& = x^py^q + x^qy^p .\end{split}\end{equation}

接下来,要说明的是:如果