Stele I · 1976 – 1978

卡拉比猜想The Calabi Conjecture

从零读懂丘成桐的证明Yau's proof, explained from zero
原文Paper · Comm. Pure Appl. Math. 31 (1978) 阅读约 25 分钟~25 min read
TL;DR

1954 年,卡拉比问:在一类叫"凯勒流形"的复几何空间上,能不能随心所欲地指定"里奇曲率",只要它满足一个显然必要的拓扑条件?

1976 年,27 岁的丘成桐证明:能,而且答案唯一。他把这个几何问题化成一个极难的非线性偏微分方程,即复蒙日–安培方程,并且解出了它。

最著名的推论是:存在大量"里奇平坦"的空间,也就是今天的卡拉比–丘流形。它们后来成为弦论中额外维度的标准候选。

In 1954 Eugenio Calabi asked: on a compact Kähler manifold, can you prescribe the Ricci curvature at will, subject only to an obvious topological constraint?

In 1976, at 27, Shing-Tung Yau proved that you can — and that the answer is unique. He turned the geometric question into a hard nonlinear PDE, the complex Monge–Ampère equation, and solved it.

The most famous consequence: an abundance of Ricci-flat spaces — today's Calabi–Yau manifolds, later adopted by string theory as candidate shapes for extra dimensions.

1954
卡拉比在国际数学家大会上提出猜想
Calabi poses the conjecture at the ICM
1976
丘成桐完成证明
Yau completes the proof
1977–78
PNAS 公告,CPAM 全文
PNAS announcement, CPAM full paper
1982
以此为核心工作之一获菲尔兹奖
Fields Medal, with this among the core works
Chapter 1

曲率:空间怎样弯曲Curvature: how space bends

想象一只蚂蚁生活在一个曲面上,它看不到曲面之外的世界,但它可以测量:在自己周围画一个半径为 $r$ 的小圆,量一量周长。平面上周长恰好是 $2\pi r$;球面上会短一点,马鞍面上会长一点。这个偏差就是"曲率",是空间自己能感受到的弯曲。

Picture an ant living on a surface. It cannot see outside, but it can measure: draw a small circle of radius $r$ around itself and measure the circumference. On a plane it is exactly $2\pi r$; on a sphere it comes out a little shorter, on a saddle a little longer. That defect is curvature — bending the space can feel from within.

在高维空间里,弯曲的方式要多得多。里奇曲率是其中一种"平均":它衡量一小团粒子沿测地线一起运动时,体积是收缩还是膨胀。爱因斯坦方程说的正是:物质与能量决定时空的里奇曲率。

In higher dimensions there are many more ways to bend. Ricci curvature is one kind of average: it measures whether a small ball of particles moving along geodesics shrinks or grows in volume. Einstein's equations say exactly this: matter and energy determine the Ricci curvature of spacetime.

◈

"里奇平坦"意味着:没有物质的地方,体积在任何方向上都不被挤压或拉伸。这就是真空中的引力。卡拉比–丘流形就是一种弯曲但里奇平坦的空间,好比没有物质、却依然有形状的宇宙。

"Ricci-flat" means: with no matter present, volumes are neither squeezed nor stretched. This is gravity in vacuum. A Calabi–Yau manifold is curved yet Ricci-flat — a universe with shape but no matter.

卡拉比的问题,就是在问反过来的事:如果我先写下想要的里奇曲率,能不能找到一个空间形状,恰好具有它?

Calabi's question asks the reverse: if I write down the Ricci curvature I want first, can I find a shape of space that has exactly it?

Chapter 2

舞台:凯勒流形与第一陈类The stage: Kähler manifolds and the first Chern class

问题发生在复流形上:局部看起来像 $\mathbb{C}^n$,坐标是复数 $z^1,\dots,z^n$ 的空间。复流形上的凯勒度量是一种和复结构完美协调的度量,可以用一个 2-形式来记录:

The problem lives on a complex manifold: a space that locally looks like $\mathbb{C}^n$, with complex coordinates $z^1,\dots,z^n$. A Kähler metric is a metric perfectly compatible with the complex structure, recorded by a 2-form:

$$\omega=\sqrt{-1}\,g_{j\bar k}\,dz^j\wedge d\bar z^k,\qquad d\omega=0 .$$

"凯勒"的条件 $d\omega=0$ 带来一个惊人的简化:整个里奇曲率可以由一个函数,也就是体积密度的对数算出来:

The Kähler condition $d\omega=0$ brings an astonishing simplification: the whole Ricci curvature is computed from one function — the logarithm of the volume density:

$$\operatorname{Ric}(\omega)=-\sqrt{-1}\,\partial\bar\partial\log\det\big(g_{j\bar k}\big).$$

这个公式是整个故事的钥匙:改变体积密度,就改变了里奇曲率。

This formula is the key to the whole story: change the volume density, and you change the Ricci curvature.

拓扑的约束A topological constraint

陈省身发现,$\operatorname{Ric}(\omega)$ 虽然依赖于度量,它所代表的上同调类却不依赖:无论怎么选凯勒度量,$\operatorname{Ric}(\omega)$ 总代表同一个类 $2\pi c_1(M)$,即流形的第一陈类。这是一个纯拓扑的不变量。

Chern discovered that although $\operatorname{Ric}(\omega)$ depends on the metric, its cohomology class does not: for every Kähler metric, $\operatorname{Ric}(\omega)$ represents the same class $2\pi c_1(M)$ — the first Chern class, a purely topological invariant.

So

如果想让某个 2-形式 $\rho$ 成为里奇曲率,它至少必须满足:$\rho$ 是闭的实 $(1,1)$-形式,并且 $[\rho]=2\pi c_1(M)$。卡拉比猜想说的是:这个显然必要的条件,也是充分的。

For a 2-form $\rho$ to be a Ricci form, it must at least be a closed real $(1,1)$-form with $[\rho]=2\pi c_1(M)$. The Calabi conjecture says: this obviously necessary condition is also sufficient.

Chapter 3

猜想的陈述The statement

卡拉比猜想(丘成桐定理,1976)Calabi Conjecture (Yau's Theorem, 1976)

设 $M$ 是紧致凯勒流形,$\omega$ 是其上一个凯勒形式。若 $\rho$ 是闭的实 $(1,1)$-形式,且 $[\rho]=2\pi c_1(M)$,则在 $\omega$ 的凯勒类 $[\omega]$ 中存在唯一的凯勒形式 $\omega'$,使得

Let $M$ be a compact Kähler manifold with Kähler form $\omega$. If $\rho$ is a closed real $(1,1)$-form with $[\rho]=2\pi c_1(M)$, then there exists a unique Kähler form $\omega'$ in the Kähler class $[\omega]$ with

$$\operatorname{Ric}(\omega')=\rho .$$

最重要的特例是 $c_1(M)=0$:此时可以取 $\rho=0$,于是每个凯勒类里都恰好有一个里奇平坦的凯勒度量。这些流形如今被称作卡拉比–丘流形。

The most important special case is $c_1(M)=0$: take $\rho=0$, and every Kähler class contains exactly one Ricci-flat Kähler metric. These manifolds are now called Calabi–Yau manifolds.

一段插曲An interlude

丘先生在自传中回忆,他起初并不相信这个猜想,觉得它好得不像真的,甚至一度以为找到了反例。在那个"反例"被证明行不通之后,他转而全力证明它,前后花了数年。

In his autobiography Yau recalls that at first he did not believe the conjecture — it seemed too good to be true — and for a while thought he had a counterexample. When that attempt failed, he turned to proving it, which took years.

Chapter 4

从几何到分析:化为一个方程From geometry to analysis: one equation

第一步:只需找一个函数。由 $\partial\bar\partial$-引理,同一凯勒类里的任何凯勒形式都可以写成

Step 1: only one function is unknown. By the $\partial\bar\partial$-lemma, any Kähler form in the same class can be written as

$$\omega_\varphi=\omega+\sqrt{-1}\,\partial\bar\partial\varphi>0,$$

其中 $\varphi$ 是 $M$ 上的实函数。原本要找一整个度量(一个矩阵值的对象),现在只要找一个函数。

for a real function $\varphi$ on $M$. Instead of an entire metric (a matrix-valued object), we seek a single function.

第二步:把目标也写成函数。$\rho$ 和 $\operatorname{Ric}(\omega)$ 在同一个上同调类里,所以它们之差也能写成 $\partial\bar\partial$ 的形式:存在实函数 $F$,使得 $\rho=\operatorname{Ric}(\omega)-\sqrt{-1}\,\partial\bar\partial F$。

Step 2: write the target as a function too. Since $\rho$ and $\operatorname{Ric}(\omega)$ lie in the same class, their difference is also $\partial\bar\partial$-exact: there is a real function $F$ with $\rho=\operatorname{Ric}(\omega)-\sqrt{-1}\,\partial\bar\partial F$.

第三步:用里奇曲率的公式比较两边。由第 2 章的公式,$\operatorname{Ric}(\omega_\varphi)-\operatorname{Ric}(\omega)=-\sqrt{-1}\,\partial\bar\partial\log(\omega_\varphi^n/\omega^n)$。要求它等于 $-\sqrt{-1}\,\partial\bar\partial F$,在紧流形上只需两个函数相差一个常数。把常数吸收进 $F$,就得到:

Step 3: compare using the Ricci formula. From Chapter 2, $\operatorname{Ric}(\omega_\varphi)-\operatorname{Ric}(\omega)=-\sqrt{-1}\,\partial\bar\partial\log(\omega_\varphi^n/\omega^n)$. Setting this equal to $-\sqrt{-1}\,\partial\bar\partial F$ on a compact manifold means the two functions differ by a constant. Absorbing the constant into $F$:

$$\boxed{\;(\omega+\sqrt{-1}\,\partial\bar\partial\varphi)^n=e^{F}\,\omega^n\;}\qquad\int_M e^F\omega^n=\int_M\omega^n .$$

这就是复蒙日–安培方程。在局部坐标里,它写作 $\det(g_{j\bar k}+\varphi_{j\bar k})=e^F\det(g_{j\bar k})$。右边的积分条件是必要的,因为同一凯勒类里的度量总体积相同。

This is the complex Monge–Ampère equation. In local coordinates, $\det(g_{j\bar k}+\varphi_{j\bar k})=e^F\det(g_{j\bar k})$. The integral condition is necessary because metrics in one Kähler class have the same total volume.

◈

这个方程为什么难?它是完全非线性的:未知函数的二阶导数出现在行列式里,而不是线性地出现。而且只有当 $\omega_\varphi$ 保持正定时,方程才是"椭圆的"(性质好的那一类)。解必须时刻待在一个凸锥里,不能越界。

Why is it hard? It is fully nonlinear: second derivatives of the unknown appear inside a determinant, not linearly. And it is "elliptic" (the well-behaved kind) only while $\omega_\varphi$ stays positive — the solution must remain inside a convex cone at all times.

Chapter 5

唯一性:卡拉比的漂亮论证Uniqueness: Calabi's elegant argument

唯一性是卡拉比本人在 1950 年代就证明了的。设 $\varphi_1,\varphi_2$ 都是解,记 $\psi=\varphi_1-\varphi_2$,$\omega_i=\omega+\sqrt{-1}\partial\bar\partial\varphi_i$。两边相减并因式分解:

Uniqueness was proved by Calabi himself in the 1950s. Suppose $\varphi_1,\varphi_2$ both solve the equation; let $\psi=\varphi_1-\varphi_2$ and $\omega_i=\omega+\sqrt{-1}\partial\bar\partial\varphi_i$. Subtract and factor:

$$0=\omega_1^n-\omega_2^n=\sqrt{-1}\,\partial\bar\partial\psi\wedge T,\qquad T=\sum_{k=0}^{n-1}\omega_1^{k}\wedge\omega_2^{\,n-1-k}>0 .$$

两边乘以 $\psi$,在 $M$ 上积分,再分部积分(紧流形没有边界):

Multiply by $\psi$, integrate over $M$, and integrate by parts (a compact manifold has no boundary):

$$0=\int_M\sqrt{-1}\,\partial\psi\wedge\bar\partial\psi\wedge T .$$

被积函数处处非负,所以 $\partial\psi\equiv0$,$\psi$ 是常数,两个解给出同一个度量。

The integrand is pointwise nonnegative, so $\partial\psi\equiv0$: $\psi$ is constant, and both solutions give the same metric.

真正困难的是存在性,这道难题悬置了二十多年。

The real difficulty was existence — open for more than twenty years.

Chapter 6

存在性:连续性方法Existence: the continuity method

直接解方程太难。丘先生使用的策略是连续性方法:把一个容易的方程连续地变形成我们想要的方程,一路追踪解。考虑一族方程,$t\in[0,1]$:

Solving the equation head-on is too hard. Yau's strategy is the continuity method: deform an easy equation continuously into the one we want, tracking the solution along the way. Consider a family, $t\in[0,1]$:

$$(\omega+\sqrt{-1}\,\partial\bar\partial\varphi_t)^n=e^{\,tF+c_t}\,\omega^n ,$$

其中常数 $c_t$ 用来保证总体积不变。$t=0$ 时 $\varphi_0=0$ 显然是解;$t=1$ 就是我们要的方程。令 $S=\{t\in[0,1]:\ \text{方程在 } t \text{ 处有光滑解}\}$。

where the constant $c_t$ keeps the total volume fixed. At $t=0$, $\varphi_0=0$ obviously works; $t=1$ is our equation. Let $S=\{t\in[0,1]:\ \text{the equation at } t \text{ has a smooth solution}\}$.

The logic

$[0,1]$ 是连通的。如果 $S$ 非空、开、并且闭,那么 $S=[0,1]$,特别地 $1\in S$。

$[0,1]$ is connected. If $S$ is nonempty, open, and closed, then $S=[0,1]$ — in particular $1\in S$.

开性:往前走一小步Openness: one small step forward

若在 $t_0$ 处有解,把方程在解附近线性化,得到的是新度量 $\omega_{t_0}$ 的拉普拉斯算子 $\Delta_{t_0}$。在平均为零的函数上,它是可逆的。由隐函数定理(巴拿赫空间版本),对 $t_0$ 附近的 $t$ 也有解。这一步相对标准。

If there is a solution at $t_0$, linearizing the equation there gives the Laplacian $\Delta_{t_0}$ of the new metric $\omega_{t_0}$, which is invertible on mean-zero functions. By the implicit function theorem (Banach-space version), solutions exist for all $t$ near $t_0$. This step is fairly standard.

闭性:真正的战场Closedness: the real battlefield

若 $t_k\to t_\infty$ 且每个 $t_k$ 处都有解 $\varphi_{t_k}$,我们需要它们收敛到 $t_\infty$ 处的一个解。为此需要先验估计:对 $\varphi_t$ 的各阶导数给出与 $t$ 无关的上界。有了一致的上界,阿尔泽拉–阿斯科利定理就能抽出收敛子列。

If $t_k\to t_\infty$ with solutions $\varphi_{t_k}$, we need them to converge to a solution at $t_\infty$. That requires a priori estimates: bounds on $\varphi_t$ and its derivatives that are independent of $t$. With uniform bounds, Arzelà–Ascoli extracts a convergent subsequence.

下面这张图是整个证明的结构,点击方块查看每一步的作用。

Below is the architecture of the whole proof. Click a box to see what each step does.

Chapter 7

先验估计:一层一层的控制A priori estimates: control, layer by layer

把解规范化(例如令 $\sup_M\varphi=0$),需要依次控制:$\varphi$ 本身、二阶导数、三阶导数。顺序上有一个巧妙的依赖关系:二阶估计只依赖于零阶估计,所以关键落在零阶估计上。

Normalize the solution (say $\sup_M\varphi=0$). We must control $\varphi$ itself, its second derivatives, and its third derivatives. There is a clever dependency: the second-order bound depends only on the zeroth-order bound, so everything hinges on the latter.

二阶估计:拉普拉斯控制Second order: controlling the Laplacian

记 $\Delta$ 为原度量的拉普拉斯,$\Delta'$ 为新度量 $\omega_\varphi$ 的拉普拉斯。量 $n+\Delta\varphi=\operatorname{tr}_\omega\omega_\varphi$ 衡量新度量相对旧度量"有多大"。对 $e^{-C\varphi}(n+\Delta\varphi)$ 计算 $\Delta'$,并用到原度量的双截曲率下界,再在最大值点使用极大值原理,可以得到(丘成桐,与 Aubin 的工作独立):

Let $\Delta$ be the Laplacian of the original metric and $\Delta'$ that of $\omega_\varphi$. The quantity $n+\Delta\varphi=\operatorname{tr}_\omega\omega_\varphi$ measures how "large" the new metric is relative to the old. Computing $\Delta'$ of $e^{-C\varphi}(n+\Delta\varphi)$, using a lower bound for the bisectional curvature of the original metric, and applying the maximum principle at the maximum point gives (Yau; independently related work by Aubin):

$$0由于 $\omega_\varphi^n/\omega^n=e^F$ 有上下界,迹有上界就意味着新旧两个度量一致等价。只要知道 $\varphi$ 的振幅 $\sup\varphi-\inf\varphi$ 有界,二阶就全部控制住了。

Since $\omega_\varphi^n/\omega^n=e^F$ is bounded above and below, a bound on the trace means the old and new metrics are uniformly equivalent. As soon as the oscillation $\sup\varphi-\inf\varphi$ is bounded, second order is fully controlled.

零阶估计:丘成桐的决定性一步Zeroth order: Yau's decisive step

这是整个证明最核心、也最新颖的部分:证明 $\sup_M|\varphi|\le C$,而且 $C$ 与 $t$ 无关。困难在于,方程本身并不直接给出 $\varphi$ 的大小信息。

This is the heart of the proof and its most original part: show $\sup_M|\varphi|\le C$ with $C$ independent of $t$. The difficulty is that the equation does not directly say how big $\varphi$ is.

丘先生的做法是:用方程两边相减后的形式 $(\omega_\varphi^n-\omega^n)$,乘以 $|\varphi|^{p}$ 类型的函数并分部积分,得到 $|\varphi|^{p/2}$ 梯度的 $L^2$ 控制;再结合索伯列夫不等式,从 $L^p$ 估计跳到 $L^{\kappa p}$($\kappa>1$)估计,一层层往上迭代,最后取极限得到 $L^\infty$ 估计。这就是莫泽迭代的思想,丘先生把它用到了这个完全非线性的情形。

Yau's approach: use the difference $(\omega_\varphi^n-\omega^n)$, multiply by functions like $|\varphi|^{p}$ and integrate by parts to control the $L^2$ norm of the gradient of $|\varphi|^{p/2}$; combine with the Sobolev inequality to jump from $L^p$ to $L^{\kappa p}$ ($\kappa>1$) bounds, iterate, and pass to the limit to reach an $L^\infty$ bound. This is the idea of Moser iteration, which Yau adapted to this fully nonlinear setting.

◈

打个比方:$L^p$ 范数像是从远处看一座山的平均高度,$p$ 越大越在意山峰。莫泽迭代每一步把 $p$ 放大固定的倍数,同时证明平均高度不会失控;无穷多步之后,就控制住了最高的山峰,也就是 $\sup|\varphi|$。

An analogy: the $L^p$ norm is like viewing a mountain range's average height from afar; the larger $p$, the more you care about peaks. Each Moser step multiplies $p$ by a fixed factor while keeping the average under control; after infinitely many steps you have controlled the highest peak — $\sup|\varphi|$.

三阶估计与正则性Third order and regularity

有了二阶控制,再用卡拉比早年发展的三阶导数估计:对量 $S=|\nabla\partial\bar\partial\varphi|^2$(在新度量下度量)计算 $\Delta' S$,得到 $\Delta' S\ge -CS-C$ 一类的不等式,再用极大值原理得到 $S$ 的上界。到这一步,方程对 $\varphi$ 的二阶导数而言系数是赫尔德连续的,经典的绍德尔理论可以一路自举到 $C^\infty$。

With second order under control, apply the third-derivative estimate Calabi had developed earlier: compute $\Delta' S$ for $S=|\nabla\partial\bar\partial\varphi|^2$ (measured in the new metric), obtain an inequality of the form $\Delta' S\ge -CS-C$, and conclude by the maximum principle that $S$ is bounded. Now the equation has Hölder-continuous coefficients in the second derivatives of $\varphi$, and classical Schauder theory bootstraps all the way to $C^\infty$.

Q.E.D.

所有估计与 $t$ 无关,所以 $S$ 是闭的。$S$ 既开又闭且非空,于是 $1\in S$:复蒙日–安培方程有光滑解,卡拉比猜想成立。

All estimates are independent of $t$, so $S$ is closed. Nonempty, open and closed, $S$ contains $1$: the complex Monge–Ampère equation has a smooth solution, and the Calabi conjecture holds.

Chapter 8

同一篇论文里的另一半:凯勒–爱因斯坦度量The other half: Kähler–Einstein metrics

同样的方法还能求凯勒–爱因斯坦度量,即满足 $\operatorname{Ric}(\omega)=\lambda\,\omega$ 的度量。当 $c_1(M)<0$ 时,问题化为

The same method finds Kähler–Einstein metrics, those with $\operatorname{Ric}(\omega)=\lambda\,\omega$. When $c_1(M)<0$ the problem becomes

$$(\omega+\sqrt{-1}\,\partial\bar\partial\varphi)^n=e^{\,\varphi+F}\,\omega^n ,$$

这里多出的 $e^{\varphi}$ 让零阶估计直接由极大值原理得到,比 $c_1=0$ 的情形容易。丘成桐与 Aubin 独立证明了这种情形下凯勒–爱因斯坦度量存在且唯一。

The extra factor $e^{\varphi}$ makes the zeroth-order estimate follow directly from the maximum principle, easier than the $c_1=0$ case. Yau and Aubin independently proved existence and uniqueness of Kähler–Einstein metrics here.

回到代数几何Back to algebraic geometry

丘先生立刻用这些度量回答了代数几何中的经典问题,这正是 PNAS 公告的标题所说的"一些新结果":

Yau immediately used these metrics to answer classical questions in algebraic geometry — the "new results" in the title of the PNAS announcement:

  • 陈数不等式:一般型代数曲面满足 $c_1^2\le 3c_2$(宫冈洋一用代数方法独立得到),并且丘先生证明,等号成立当且仅当曲面是复球的商。
  • Chern number inequality: surfaces of general type satisfy $c_1^2\le 3c_2$ (obtained independently by Miyaoka by algebraic methods), and Yau showed equality holds exactly for quotients of the complex ball.
  • 塞韦里猜想:任何与复射影平面 $\mathbb{CP}^2$ 同伦等价的紧复曲面,都双全纯等价于 $\mathbb{CP}^2$。
  • Severi conjecture: every compact complex surface homotopy equivalent to $\mathbb{CP}^2$ is biholomorphic to $\mathbb{CP}^2$.
Chapter 9

影响:从几何到弦论Legacy: from geometry to string theory

1985 年,Candelas、Horowitz、Strominger 与 Witten 发现,超弦理论若要把十维时空紧化成我们看到的四维、同时保留一定的超对称,额外的六个维度恰好应该是一个复三维的里奇平坦凯勒流形。丘先生的定理保证了这样的空间大量存在,"卡拉比–丘流形"这个名字由此流传开来。

In 1985 Candelas, Horowitz, Strominger and Witten found that for superstring theory to compactify ten dimensions down to the four we see while preserving some supersymmetry, the six extra dimensions should be precisely a complex three-dimensional Ricci-flat Kähler manifold. Yau's theorem guaranteed such spaces exist in abundance, and the name "Calabi–Yau manifold" spread from there.

首页和本页顶部那个金色的形状,就是一个卡拉比–丘流形(费马五次超曲面)的一个实二维切片在三维中的投影。

The golden shape on the homepage and at the top of this page is a projection to three dimensions of a real two-dimensional slice of a Calabi–Yau manifold (the Fermat quintic).

在数学内部,这篇论文也开启了用非线性偏微分方程研究复几何的道路:复蒙日–安培方程的技术后来延伸到镜像对称、凯勒–里奇流、法诺流形上凯勒–爱因斯坦度量的存在性等一系列方向。

Within mathematics, the paper opened the way to studying complex geometry through nonlinear PDE: complex Monge–Ampère techniques later spread to mirror symmetry, Kähler–Ricci flow, Kähler–Einstein metrics on Fano manifolds, and more.

References

原文与延伸阅读Original papers & further reading

  • 1977S.-T. Yau, Calabi's conjecture and some new results in algebraic geometry, Proc. Natl. Acad. Sci. USA 74(5), 1798–1799. DOI ↗
  • 1978S.-T. Yau, On the Ricci curvature of a compact Kähler manifold and the complex Monge–Ampère equation, I, Comm. Pure Appl. Math. 31(3), 339–411. DOI ↗
  • 1957E. Calabi, On Kähler manifolds with vanishing canonical class, in Algebraic Geometry and Topology: A Symposium in Honor of S. Lefschetz, Princeton Univ. Press.
  • 1985P. Candelas, G. Horowitz, A. Strominger, E. Witten, Vacuum configurations for superstrings, Nuclear Physics B 258, 46–74.
  • 2010S.-T. Yau, S. Nadis, The Shape of Inner Space, Basic Books. (面向大众的讲述)(for general readers)