↓ 跳过正文

De Rham Cohomology I: 引言

·12049 字·25 分钟

集合
#

开集
#

R\mathbb{R}的子集UU称为开集, 如果它内部的每一点都能被一个完全落在UU内的开区间包住:

∀x∈U, ∃ε>0, 使得 (x−ε, x+ε)⊆U\forall x \in U,\ \exists \varepsilon > 0,\ \text{使得}\ (x - \varepsilon,\ x + \varepsilon) \subseteq U

抽象代数
#

线性映射
#

设V,WV, W是同一域F\mathbb{F}上的向量空间. 映射φ:V→W\varphi: V \to W称为线性映射, 如果保持加法与数乘:

φ(c1v1+c2v2)=c1φ(v1)+c2φ(v2),∀v1,v2∈V, c1,c2∈F\varphi(c_1 v_1 + c_2 v_2) = c_1 \varphi(v_1) + c_2 \varphi(v_2),\quad \forall v_1, v_2 \in V,\ c_1, c_2 \in \mathbb{F}

同构
#

同态
#

映射φ:G→H\varphi: G \to H称为同态(homomorphism), 如果保持运算:

φ(g1g2)=φ(g1)φ(g2),∀g1,g2∈G\varphi(g_1 g_2) = \varphi(g_1) \varphi(g_2),\quad \forall g_1, g_2 \in G

由此自动得φ(eG)=eH\varphi(e_G) = e_H, φ(g−1)=φ(g)−1\varphi(g^{-1}) = \varphi(g)^{-1}.

同构
#

同态φ:G→H\varphi: G \to H称为同构(isomorphism), 如果它是双射. 此时存在逆同态φ−1:H→G\varphi^{-1}: H \to G, 满足φ−1∘φ=id⁡G\varphi^{-1} \circ \varphi = \operatorname{id}_G, φ∘φ−1=id⁡H\varphi \circ \varphi^{-1} = \operatorname{id}_H. 记作G≅HG \cong H.

直观上, G≅HG \cong H意味着两个群作为代数结构完全相同, 仅元素的标签不同.

示例
#

  • R/Z≅S1\mathbb{R}/\mathbb{Z} \cong S^1: [x]↦e2πix[x] \mapsto e^{2\pi i x}是同构
  • (R,+)≅(R>0,⋅)(\mathbb{R}, +) \cong (\mathbb{R}_{>0}, \cdot): x↦exx \mapsto e^x将加法变成乘法
  • Z/6Z≅Z/2Z×Z/3Z\mathbb{Z}/6\mathbb{Z} \cong \mathbb{Z}/2\mathbb{Z} \times \mathbb{Z}/3\mathbb{Z}(中国剩余定理)

商群
#

商群是把群GG按某个正规子群NN粘合得到的新群.

正规子群
#

子群N≤GN \leq G称为正规子群, 记作N◃GN \triangleleft G, 如果:

gNg−1=N,∀g∈GgNg^{-1} = N,\quad \forall g \in G

等价地, 左陪集等于右陪集: gN=NggN = Ng. Abel群里每个子群自动正规, 此条件只在非交换群中非平凡.

核与像
#

设φ:G→H\varphi: G \to H是群同态.

核是映射到HH的单位元eHe_H的元素全体:

Ker⁡φ={g∈G:φ(g)=eH}\operatorname{Ker}\varphi = \{g \in G : \varphi(g) = e_H\}

像是φ\varphi实际产出的元素全体:

Im⁡φ={φ(g):g∈G}\operatorname{Im}\varphi = \{\varphi(g) : g \in G\}

由同态条件易得Ker⁡φ\operatorname{Ker}\varphi是GG的正规子群, Im⁡φ\operatorname{Im}\varphi是HH的子群.

商群
#

元素g∈Gg \in G对应陪集gNgN, 商群是所有陪集的集合:

G/N={gN:g∈G}G/N = \{gN : g \in G\}

群运算定义为:

(g1N)⋅(g2N)=(g1g2)N(g_1 N) \cdot (g_2 N) = (g_1 g_2) N

运算良定义要求NN正规: 设g1′=g1n1g_1' = g_1 n_1, g2′=g2n2g_2' = g_2 n_2, 则

g1′g2′=g1n1g2n2=g1g2(g2−1n1g2)⏟∈Nn2∈g1g2Ng_1' g_2' = g_1 n_1 g_2 n_2 = g_1 g_2 \underbrace{(g_2^{-1} n_1 g_2)}_{\in N} n_2 \in g_1 g_2 N

需要g2−1Ng2⊆Ng_2^{-1} N g_2 \subseteq N, 这正是正规性的来源.

第一同构定理
#

群同态φ:G→H\varphi: G \to H满足:

G/Ker⁡φ≅Im⁡φG / \operatorname{Ker}\varphi \cong \operatorname{Im}\varphi

证明: 记K=Ker⁡φK = \operatorname{Ker}\varphi, 定义

φ‾:G/K→Im⁡φ,[g]↦φ(g)\overline\varphi: G/K \to \operatorname{Im}\varphi,\quad [g] \mapsto \varphi(g)
  • 良定义: [g1]=[g2]  ⟺  g2=g1k, k∈K  ⟺  φ(g1)=φ(g2)[g_1] = [g_2] \iff g_2 = g_1 k,\ k \in K \iff \varphi(g_1) = \varphi(g_2)
  • 同态: φ‾([g1][g2])=φ(g1g2)=φ(g1)φ(g2)=φ‾([g1])φ‾([g2])\overline\varphi([g_1][g_2]) = \varphi(g_1 g_2) = \varphi(g_1) \varphi(g_2) = \overline\varphi([g_1]) \overline\varphi([g_2])
  • 单射: φ‾([g1])=φ‾([g2])⇒φ(g1)=φ(g2)⇒[g1]=[g2]\overline\varphi([g_1]) = \overline\varphi([g_2]) \Rightarrow \varphi(g_1) = \varphi(g_2) \Rightarrow [g_1] = [g_2]
  • 满射: h=φ(g)=φ‾([g])h = \varphi(g) = \overline\varphi([g])

故φ‾\overline\varphi是群同构.

示例
#

  • Z/nZ\mathbb{Z}/n\mathbb{Z}: 模nn的整数加法群
  • R/Z≅S1\mathbb{R}/\mathbb{Z} \cong S^1: 实数粘合整数, 得到圆周
  • R/2πZ≅S1\mathbb{R}/2\pi\mathbb{Z} \cong S^1: 角度空间

商环
#

商环是把环RR按某个理想II粘合得到的新环, 思想与商群一致, 但需要更强的子结构.

理想
#

子集I⊆RI \subseteq R称为(双边)理想, 如果:

  1. (I,+)(I, +)是(R,+)(R, +)的子群
  2. 吸收性: ∀r∈R, a∈I\forall r \in R,\ a \in I, 有ra∈Ira \in I且ar∈Iar \in I

例如nZ={nk:k∈Z}n\mathbb{Z} = \{nk : k \in \mathbb{Z}\}是Z\mathbb{Z}的理想; (x)={xp(x):p∈R[x]}(x) = \{x p(x) : p \in \mathbb{R}[x]\}是R[x]\mathbb{R}[x]的理想.

商环
#

商环是所有陪集的集合:

R/I={a+I:a∈R}R/I = \{a + I : a \in R\}

加法和乘法定义为:

[a]+[b]=[a+b],[a]⋅[b]=[ab][a] + [b] = [a + b],\qquad [a] \cdot [b] = [ab]

乘法良定义要求II是理想: 设a′=a+ia' = a + i, b′=b+jb' = b + j, 则

a′b′=ab+aj+ib+ij⏟∈Ia'b' = ab + \underbrace{aj + ib + ij}_{\in I}

第一同构定理
#

环同态φ:R→S\varphi: R \to S满足:

R/Ker⁡φ≅Im⁡φR / \operatorname{Ker}\varphi \cong \operatorname{Im}\varphi

证明: 记I=Ker⁡φI = \operatorname{Ker}\varphi, 它是RR的理想, 定义

φ‾:R/I→Im⁡φ,[a]↦φ(a)\overline\varphi: R/I \to \operatorname{Im}\varphi,\quad [a] \mapsto \varphi(a)
  • 良定义: [a1]=[a2]  ⟺  a1−a2∈I  ⟺  φ(a1)=φ(a2)[a_1] = [a_2] \iff a_1 - a_2 \in I \iff \varphi(a_1) = \varphi(a_2)
  • 保加法: φ‾([a1]+[a2])=φ(a1+a2)=φ(a1)+φ(a2)\overline\varphi([a_1] + [a_2]) = \varphi(a_1 + a_2) = \varphi(a_1) + \varphi(a_2)
  • 保乘法: φ‾([a1][a2])=φ(a1a2)=φ(a1)φ(a2)\overline\varphi([a_1] [a_2]) = \varphi(a_1 a_2) = \varphi(a_1) \varphi(a_2)
  • 单射: φ‾([a1])=φ‾([a2])⇒φ(a1)=φ(a2)⇒[a1]=[a2]\overline\varphi([a_1]) = \overline\varphi([a_2]) \Rightarrow \varphi(a_1) = \varphi(a_2) \Rightarrow [a_1] = [a_2]
  • 满射: h=φ(a)=φ‾([a])h = \varphi(a) = \overline\varphi([a])

故φ‾\overline\varphi是环同构.

示例
#

  • Z/nZ\mathbb{Z}/n\mathbb{Z}: 模nn的整数环, 当nn为素数时是域
  • R[x]/(x2+1)≅C\mathbb{R}[x]/(x^2 + 1) \cong \mathbb{C}: 添加关系x2=−1x^2 = -1, [x][x]扮演ii
  • R[x]/(x−a)≅R\mathbb{R}[x]/(x - a) \cong \mathbb{R}: 对应"在x=ax = a处求值"
  • C∞(U)/Ip≅RC^\infty(U)/I_p \cong \mathbb{R}, 其中Ip={f:f(p)=0}I_p = \{f : f(p) = 0\}: 把函数代换成它在pp处的值

微积分
#

设U⊆R3U \subseteq \mathbb{R}^3是开集.

梯度
#

光滑函数f∈C∞(U)f \in C^\infty(U)的梯度(gradient)是向量场:

grad⁡f=∇f=(∂f∂x1, ∂f∂x2, ∂f∂x3)\operatorname{grad} f = \nabla f = \left(\frac{\partial f}{\partial x_1},\ \frac{\partial f}{\partial x_2},\ \frac{\partial f}{\partial x_3}\right)

散度
#

光滑向量场F=(f1,f2,f3)∈C∞(U,R3)F = (f_1, f_2, f_3) \in C^\infty(U, \mathbb{R}^3)的散度(divergence)是标量函数:

div⁡F=∇⋅F=∂f1∂x1+∂f2∂x2+∂f3∂x3\operatorname{div} F = \nabla \cdot F = \frac{\partial f_1}{\partial x_1} + \frac{\partial f_2}{\partial x_2} + \frac{\partial f_3}{\partial x_3}

旋度
#

光滑向量场F=(f1,f2,f3)∈C∞(U,R3)F = (f_1, f_2, f_3) \in C^\infty(U, \mathbb{R}^3)的旋度:

rot⁡F=∇×F=(∂f3∂x2−∂f2∂x3, ∂f1∂x3−∂f3∂x1, ∂f2∂x1−∂f1∂x2)\operatorname{rot} F = \nabla \times F = \left(\frac{\partial f_3}{\partial x_2} - \frac{\partial f_2}{\partial x_3},\ \frac{\partial f_1}{\partial x_3} - \frac{\partial f_3}{\partial x_1},\ \frac{\partial f_2}{\partial x_1} - \frac{\partial f_1}{\partial x_2}\right)

形式上写作行列式:

rot⁡F=∣e1e2e3∂1∂2∂3f1f2f3∣\operatorname{rot} F = \begin{vmatrix} \mathbf{e}_1 & \mathbf{e}_2 & \mathbf{e}_3 \\ \partial_1 & \partial_2 & \partial_3 \\ f_1 & f_2 & f_3 \end{vmatrix}

Stokes公式
#

设Σ⊂R3\Sigma \subset \mathbb{R}^3是带边光滑有向曲面, 单位法向量为n\mathbf{n}, 边界∂Σ\partial\Sigma按右手定则取向, 则对光滑向量场FF有

∬Σrot⁡F⋅n dσ=∫∂Σf1dx1+f2dx2+f3dx3\iint_\Sigma \operatorname{rot} F \cdot \mathbf{n}\ \mathrm{d}\sigma = \int_{\partial\Sigma} f_1 \mathrm{d}x_1 + f_2 \mathrm{d}x_2 + f_3 \mathrm{d}x_3

证明: 思路是先把Σ\Sigma切成许多小曲面片, 在每片上验证公式, 再把所有片的等式拼成全局.

把Σ\Sigma剖分为充分小的曲面片Σ1,…,Σn\Sigma_1, \ldots, \Sigma_n. 曲面积分天然可加, 公式左侧拆为

∬Σrot⁡F⋅n dσ=∑i∬Σirot⁡F⋅n dσ\iint_\Sigma \operatorname{rot} F \cdot \mathbf{n}\ \mathrm{d}\sigma = \sum_i \iint_{\Sigma_i} \operatorname{rot} F \cdot \mathbf{n}\ \mathrm{d}\sigma

若能在每片上验证局部 Stokes

∬Σirot⁡F⋅n dσ=∫∂Σif1dx1+f2dx2+f3dx3\iint_{\Sigma_i} \operatorname{rot} F \cdot \mathbf{n}\ \mathrm{d}\sigma = \int_{\partial\Sigma_i} f_1 \mathrm{d}x_1 + f_2 \mathrm{d}x_2 + f_3 \mathrm{d}x_3

把所有片的等式相加, 右侧每条内部公共边(同时属于两片的边界)在∂Σi,∂Σj\partial\Sigma_i, \partial\Sigma_j中各贡献一次、走向相反, 求和后相消; 不被任何其他片共享的外部边恰好拼成原边界∂Σ\partial\Sigma. 于是问题归约为在每片上证局部 Stokes.

∑i∫∂Σif1dx1+f2dx2+f3dx3=∫∂Σf1dx1+f2dx2+f3dx3\sum_i \int_{\partial\Sigma_i} f_1 \mathrm{d}x_1 + f_2 \mathrm{d}x_2 + f_3 \mathrm{d}x_3 = \int_{\partial\Sigma} f_1 \mathrm{d}x_1 + f_2 \mathrm{d}x_2 + f_3 \mathrm{d}x_3

取小片充分细时, 每片可近似为其切平面内的小矩形, 误差为面积的高阶无穷小, 极限下可视为相等. 故只需在该小矩形上验证公式.

选取TT使Tn=e3T\mathbf{n} = \mathbf{e}_3(这样的TT总存在), 则原切平面被旋转到x3=0x_3 = 0平面, 小矩形落入其中. 故不失一般性, 可在此标准位置上验证.

设旋转后的小矩形为R=[a,b]×[c,d]R = [a, b] \times [c, d]. 此时n=e3\mathbf{n} = \mathbf{e}_3, dσ=dx1dx2\mathrm{d}\sigma = \mathrm{d}x_1 \mathrm{d}x_2, dx3=0\mathrm{d}x_3 = 0沿∂R\partial R, 故只剩f1dx1+f2dx2f_1 \mathrm{d}x_1 + f_2 \mathrm{d}x_2. 沿∂R\partial R逆时针:

∫∂Rf1dx1+f2dx2=∫ab[f1(x1,c)−f1(x1,d)]dx1+∫cd[f2(b,x2)−f2(a,x2)]dx2\int_{\partial R} f_1 \mathrm{d}x_1 + f_2 \mathrm{d}x_2 = \int_a^b \big[f_1(x_1, c) - f_1(x_1, d)\big] \mathrm{d}x_1 + \int_c^d \big[f_2(b, x_2) - f_2(a, x_2)\big] \mathrm{d}x_2

对每项用Newton-Leibniz公式:

=−∬R∂f1∂x2dx1dx2+∬R∂f2∂x1dx1dx2=∬R(∂f2∂x1−∂f1∂x2)dσ= -\iint_R \frac{\partial f_1}{\partial x_2} \mathrm{d}x_1 \mathrm{d}x_2 + \iint_R \frac{\partial f_2}{\partial x_1} \mathrm{d}x_1 \mathrm{d}x_2 = \iint_R \left(\frac{\partial f_2}{\partial x_1} - \frac{\partial f_1}{\partial x_2}\right) \mathrm{d}\sigma

而rot⁡F\operatorname{rot} F的第三分量∂f2∂x1−∂f1∂x2=rot⁡F⋅e3\dfrac{\partial f_2}{\partial x_1} - \dfrac{\partial f_1}{\partial x_2} = \operatorname{rot} F \cdot \mathbf{e}_3, 故等式成立.

上同调
#

U⋆⊆R2U^\star \subseteq \mathbb{R}^2
#

叉积展开为(对任意x\mathbf{x}):

F(x)×x=(f2(x)x3−f3(x)x2, f3(x)x1−f1(x)x3, f1(x)x2−f2(x)x1)F(\mathbf{x}) \times \mathbf{x} = \big(f_2(\mathbf{x}) x_3 - f_3(\mathbf{x}) x_2,\ f_3(\mathbf{x}) x_1 - f_1(\mathbf{x}) x_3,\ f_1(\mathbf{x}) x_2 - f_2(\mathbf{x}) x_1\big)

rot⁡\operatorname{rot}是对空间变量x\mathbf{x}的偏导∂∂xj\frac{\partial}{\partial x_j}的组合, 而∫01⋯dt\int_0^1 \cdots \mathrm{d}t是对参数变量tt的积分, 可交换次序:

rot⁡G(x)=rot⁡∫01F(tx)×tx dt=∫01rot⁡(F(tx)×tx) dt\operatorname{rot} G(\mathbf{x}) = \operatorname{rot} \int_0^1 F(t\mathbf{x}) \times t\mathbf{x} \ \mathrm{d}t = \int_0^1 \operatorname{rot}\big(F(t\mathbf{x}) \times t\mathbf{x}\big) \ \mathrm{d}t

直接计算给出:

rot⁡(F(tx)×tx)=ddt(t2F(tx))\operatorname{rot}\big(F(t\mathbf{x}) \times t\mathbf{x}\big) = \frac{\mathrm{d}}{\mathrm{d}t}\big(t^2 F(t\mathbf{x})\big)

以第一分量为例验证. 由链式法则, fi(tx)f_i(t\mathbf{x})对xjx_j求偏导等于对txjtx_j求偏导再乘以tt: ∂∂xj[fi(tx)]=t∂fi(tx)∂(txj)\dfrac{\partial}{\partial x_j}\big[f_i(t\mathbf{x})\big] = t\dfrac{\partial f_i(t\mathbf{x})}{\partial (tx_j)}. 由叉积展开, F(tx)×txF(t\mathbf{x}) \times t\mathbf{x}的分量为A2=t(f3(tx)x1−f1(tx)x3)A_2 = t\big(f_3(t\mathbf{x}) x_1 - f_1(t\mathbf{x}) x_3\big), A3=t(f1(tx)x2−f2(tx)x1)A_3 = t\big(f_1(t\mathbf{x}) x_2 - f_2(t\mathbf{x}) x_1\big), 故

∂A3∂x2=t[t∂f1(tx)∂(tx2)x2+f1(tx)−t∂f2(tx)∂(tx2)x1]\frac{\partial A_3}{\partial x_2} = t\left[t\frac{\partial f_1(t\mathbf{x})}{\partial (tx_2)} x_2 + f_1(t\mathbf{x}) - t\frac{\partial f_2(t\mathbf{x})}{\partial (tx_2)} x_1\right]∂A2∂x3=t[t∂f3(tx)∂(tx3)x1−t∂f1(tx)∂(tx3)x3−f1(tx)]\frac{\partial A_2}{\partial x_3} = t\left[t\frac{\partial f_3(t\mathbf{x})}{\partial (tx_3)} x_1 - t\frac{\partial f_1(t\mathbf{x})}{\partial (tx_3)} x_3 - f_1(t\mathbf{x})\right]

相减得rot⁡\operatorname{rot}的第一分量

∂A3∂x2−∂A2∂x3=t[2f1(tx)+t∂f1(tx)∂(tx2)x2+t∂f1(tx)∂(tx3)x3−t(∂f2(tx)∂(tx2)+∂f3(tx)∂(tx3))x1]\frac{\partial A_3}{\partial x_2} - \frac{\partial A_2}{\partial x_3} = t\left[2f_1(t\mathbf{x}) + t\frac{\partial f_1(t\mathbf{x})}{\partial (tx_2)} x_2 + t\frac{\partial f_1(t\mathbf{x})}{\partial (tx_3)} x_3 - t\left(\frac{\partial f_2(t\mathbf{x})}{\partial (tx_2)} + \frac{\partial f_3(t\mathbf{x})}{\partial (tx_3)}\right) x_1\right]

由div⁡F=0\operatorname{div} F = 0知∂f2(tx)∂(tx2)+∂f3(tx)∂(tx3)=−∂f1(tx)∂(tx1)\dfrac{\partial f_2(t\mathbf{x})}{\partial (tx_2)} + \dfrac{\partial f_3(t\mathbf{x})}{\partial (tx_3)} = -\dfrac{\partial f_1(t\mathbf{x})}{\partial (tx_1)}, 代入:

=2tf1(tx)+t2[∂f1(tx)∂(tx1)x1+∂f1(tx)∂(tx2)x2+∂f1(tx)∂(tx3)x3]= 2t f_1(t\mathbf{x}) + t^2\left[\frac{\partial f_1(t\mathbf{x})}{\partial (tx_1)} x_1 + \frac{\partial f_1(t\mathbf{x})}{\partial (tx_2)} x_2 + \frac{\partial f_1(t\mathbf{x})}{\partial (tx_3)} x_3\right]

另一方面, 由链式法则ddtf1(tx)=∑j∂f1(tx)∂(txj)xj\dfrac{\mathrm{d}}{\mathrm{d}t} f_1(t\mathbf{x}) = \sum_j \dfrac{\partial f_1(t\mathbf{x})}{\partial (tx_j)} x_j, 可得两式相等:

ddt(t2f1(tx))=2tf1(tx)+t2∑j∂f1(tx)∂(txj)xj\frac{\mathrm{d}}{\mathrm{d}t}\big(t^2 f_1(t\mathbf{x})\big) = 2t f_1(t\mathbf{x}) + t^2 \sum_j \frac{\partial f_1(t\mathbf{x})}{\partial (tx_j)} x_j

代入可得:

rot⁡G(x)=∫01ddt(t2F(tx)) dt=[t2F(tx)]01=12⋅F(x)−0=F(x)\operatorname{rot} G(\mathbf{x}) = \int_0^1 \frac{\mathrm{d}}{\mathrm{d}t}\big(t^2 F(t\mathbf{x})\big) \ \mathrm{d}t = \Big[t^2 F(t\mathbf{x})\Big]_0^1 = 1^2 \cdot F(\mathbf{x}) - 0 = F(\mathbf{x})

故F=rot⁡G∈Im⁡(rot⁡)F = \operatorname{rot} G \in \operatorname{Im}(\operatorname{rot}). 由FF任意, Ker⁡(div⁡)⊆Im⁡(rot⁡)\operatorname{Ker}(\operatorname{div}) \subseteq \operatorname{Im}(\operatorname{rot}); 又div⁡∘rot⁡=0\operatorname{div} \circ \operatorname{rot} = 0给出反向包含, 故Ker⁡(div⁡)=Im⁡(rot⁡)\operatorname{Ker}(\operatorname{div}) = \operatorname{Im}(\operatorname{rot}), 即

H2(U)=Ker⁡(div⁡)/Im⁡(rot⁡)=0H^2(U) = \operatorname{Ker}(\operatorname{div}) / \operatorname{Im}(\operatorname{rot}) = 0

R3∖{x3=0, x12+x22≥1}\mathbb{R}^3 \setminus \{x_3 = 0,\ x_1^2 + x_2^2 \geq 1\}
#

考虑R3\mathbb{R}^3中向量场, 易证rot⁡F(x)=0\operatorname{rot} F(\mathbf{x}) = 0.

F(x)=(−2x1x3x32+(x12+x22−1)2, −2x2x3x32+(x12+x22−1)2, x12+x22−1x32+(x12+x22−1)2)F(\mathbf{x}) = \left(\frac{-2x_1 x_3}{x_3^2 + (x_1^2+x_2^2-1)^2},\ \frac{-2x_2 x_3}{x_3^2 + (x_1^2+x_2^2-1)^2},\ \frac{x_1^2+x_2^2-1}{x_3^2 + (x_1^2+x_2^2-1)^2}\right)

定义集合VV

V=R3∖{(x1,x2,x3):x3=0, x12+x22≥1}V = \mathbb{R}^3 \setminus \{(x_1, x_2, x_3) : x_3 = 0,\ x_1^2 + x_2^2 \geq 1\}

引入辅助变量

u=x12+x22−1,v=x3u = x_1^2 + x_2^2 - 1, \qquad v = x_3

则FF化为

F1=−2x1vu2+v2,F2=−2x2vu2+v2,F3=uu2+v2F_1 = \frac{-2x_1 v}{u^2 + v^2},\quad F_2 = \frac{-2x_2 v}{u^2 + v^2},\quad F_3 = \frac{u}{u^2 + v^2}

设原函数形如G=g(u,v)G = g(u, v), 由链式法则

∂G∂x1=2x1∂g∂u,∂G∂x2=2x2∂g∂u,∂G∂x3=∂g∂v\frac{\partial G}{\partial x_1} = 2x_1 \frac{\partial g}{\partial u},\quad \frac{\partial G}{\partial x_2} = 2x_2 \frac{\partial g}{\partial u},\quad \frac{\partial G}{\partial x_3} = \frac{\partial g}{\partial v}

与FF比对得

∂g∂u=−vu2+v2,∂g∂v=uu2+v2\frac{\partial g}{\partial u} = -\frac{v}{u^2+v^2},\qquad \frac{\partial g}{\partial v} = \frac{u}{u^2+v^2}

根据链式法则

dduarctan⁡uv=1v⋅11+(uv)2=vu2+v2\frac{\mathrm{d}}{\mathrm{d}u}\arctan\frac{u}{v} = \frac{1}{v} \cdot \frac{1}{1+\left(\dfrac{u}{v}\right)^2} = \frac{v}{u^2+v^2}

对uu积分(把vv视为常数):

g(u,v)=∫−vu2+v2 du=−arctan⁡uv+C(v)g(u, v) = \int -\frac{v}{u^2+v^2}\,\mathrm{d}u = -\arctan\frac{u}{v} + C(v)

对vv求偏导匹配另一式:

∂g∂v=−11+(uv)2⋅(−uv2)+C′(v)=uu2+v2+C′(v)=uu2+v2\frac{\partial g}{\partial v} = -\frac{1}{1+\left(\dfrac{u}{v}\right)^2}\cdot\left(-\frac{u}{v^2}\right) + C'(v) = \frac{u}{u^2+v^2} + C'(v) = \frac{u}{u^2+v^2}

得C′(v)=0C'(v) = 0, CC为常数, 故v≠0v \neq 0处

g(u,v)=−arctan⁡uv+Cg(u, v) = -\arctan\frac{u}{v} + C

该表达在v=0v = 0时分母奇异, 不能取值. 但VV包含整个开圆盘{x3=0, x12+x22<1}\{x_3 = 0,\ x_1^2 + x_2^2 < 1\}(对应v=0v = 0), 例如原点OO. 需要把gg替换为在v=0v = 0处也良定义的等价表达.

用极坐标定义x3=0x_3 = 0平面上的u,vu, v, 把θ\theta定义为(u,v)(u, v)与负uu轴的夹角(逆时针为正), 取值范围(−π,π](-\pi, \pi], 即

−u=rcos⁡θ,−v=rsin⁡θ-u = r\cos\theta, \qquad -v = r\sin\theta

把它们当作(r,θ)(r, \theta)对(u,v)(u, v)的隐函数, 两边同时对uu求偏导得

−1=∂r∂ucos⁡θ−rsin⁡θ⋅∂θ∂u,0=∂r∂usin⁡θ+rcos⁡θ⋅∂θ∂u-1 = \frac{\partial r}{\partial u}\cos\theta - r\sin\theta\cdot\frac{\partial \theta}{\partial u},\qquad 0 = \frac{\partial r}{\partial u}\sin\theta + r\cos\theta\cdot\frac{\partial \theta}{\partial u}

两式分别乘sin⁡θ,cos⁡θ\sin\theta, \cos\theta相减消去∂r∂u\dfrac{\partial r}{\partial u}:

−sin⁡θ=−r(sin⁡2θ+cos⁡2θ)∂θ∂u=−r∂θ∂u⇒∂θ∂u=sin⁡θr=−vu2+v2-\sin\theta = -r(\sin^2\theta + \cos^2\theta)\frac{\partial \theta}{\partial u} = -r\frac{\partial \theta}{\partial u} \Rightarrow \frac{\partial \theta}{\partial u} = \frac{\sin\theta}{r} = -\frac{v}{u^2+v^2}

对vv同样处理得∂θ∂v=−cos⁡θr=uu2+v2\dfrac{\partial \theta}{\partial v} = -\dfrac{\cos\theta}{r} = \dfrac{u}{u^2+v^2}. 恰为∂g∂u,∂g∂v\dfrac{\partial g}{\partial u}, \dfrac{\partial g}{\partial v}, 故gg与θ\theta只有常数项不同.

θ\theta只在两处不连续: 夹角无意义的原点(u,v)=(0,0)(u, v) = (0, 0), 对应单位圆SS, 已被VV排除; 夹角在π\pi与−π-\pi之间跳变的正uu轴{u>0,v=0}\{u > 0, v = 0\}, 对应{x3=0, x12+x22>1}\{x_3 = 0,\ x_1^2 + x_2^2 > 1\}, 同样被排除. 故θ\theta在VV上光滑.

用atan2⁡\operatorname{atan2}实现θ\theta, 此时g(u,v)g(u, v)中C=−π2sgn⁡(v)C = -\dfrac{\pi}{2}\operatorname{sgn}(v), 且G(O)=atan2⁡(0,1)=0G(O) = \operatorname{atan2}(0, 1) = 0.

G(x)=atan2⁡(−x3, 1−x12−x22)G(\mathbf{x}) = \operatorname{atan2}\big(-x_3,\ 1 - x_1^2 - x_2^2\big)