1.4. Cutoff Functions and Partitions of Unity 1.4. Unity 的截止函数和分区
In distribution theory one often has to replace a function by one with compact support without changing it on a large compact set. This is done by multiplication with a “cutoff function” as constructed in the following 在分布理论中,人们经常不得不用具有紧凑支座的函数替换函数,而不在大型紧凑集合上更改它。这是通过使用 “截止函数” 进行乘法来完成的,如下所示
Theorem 1.4.1. If XX is an open set in R^(n)\mathbb{R}^{n} and KK is a compact subset, then one can find phi inC_(0)^(oo)(X)\phi \in C_{0}^{\infty}(X) with 0 <= phi <= 10 \leqq \phi \leqq 1 so that phi=1\phi=1 in a neighborhood of KK. 定理 1.4.1.如果 XX 是 中的 R^(n)\mathbb{R}^{n} 开集 并且 KK 是紧子集,则可以在 phi inC_(0)^(oo)(X)\phi \in C_{0}^{\infty}(X) 中找到 , 0 <= phi <= 10 \leqq \phi \leqq 1 因此 phi=1\phi=1 在 KK 的邻域中。
Proof. Choose epsi > 0\varepsilon>0 so small that 证明。选择 epsi > 0\varepsilon>0 如此小的
|x-y| >= 4epsiquad" when "x in K,y inℓX,|x-y| \geqq 4 \varepsilon \quad \text { when } x \in K, y \in \ell X,
and let vv be the characteristic function of ,设 vv 为 的特征函数
K_(2epsi)={y;|x-y| <= 2epsi" for some "x in K}.K_{2 \varepsilon}=\{y ;|x-y| \leqq 2 \varepsilon \text { for some } x \in K\} .
According to Lemma 1.2 .3 we can find a non-negative function chi inC_(0)^(oo)(B)\chi \in C_{0}^{\infty}(B) where BB is the unit ball, such that int chi dx=1\int \chi d x=1. Then chi_(epsi)(x)\chi_{\varepsilon}(x)=epsi^(-n)chi(x//epsi)=\varepsilon^{-n} \chi(x / \varepsilon) has support in the ball {x;|x| < epsi}\{x ;|x|<\varepsilon\} and intchi_(epsi)dx=1\int \chi_{\varepsilon} d x=1, so 根据引理 1.2 .3,我们可以找到一个非负函数 chi inC_(0)^(oo)(B)\chi \in C_{0}^{\infty}(B) ,其中 BB 是单位球,使得 int chi dx=1\int \chi d x=1 。then chi_(epsi)(x)\chi_{\varepsilon}(x)=epsi^(-n)chi(x//epsi)=\varepsilon^{-n} \chi(x / \varepsilon) 在球 {x;|x| < epsi}\{x ;|x|<\varepsilon\} 和 intchi_(epsi)dx=1\int \chi_{\varepsilon} d x=1 中具有支撑,因此
by Theorem 1.3.1 and (1.3.11), and 1-phi=(1-v)**chi_(epsi)1-\phi=(1-v) * \chi_{\varepsilon} vanishes in K_(epsi)K_{\varepsilon} by (1.3.11). This proves the theorem. 由定理 1.3.1 和 (1.3.11) 完成,并在 (1.3.11) 中 1-phi=(1-v)**chi_(epsi)1-\phi=(1-v) * \chi_{\varepsilon}K_(epsi)K_{\varepsilon} 消失。这证明了这个定理。
For future reference we also note that 为了将来参考,我们还注意到
Thus 因此
|del^(alpha)phi| <= int|del^(alpha)chi_(epsi)|dx=epsi^(-|alpha|)int|del^(alpha)chi|dx.\left|\partial^{\alpha} \phi\right| \leqq \int\left|\partial^{\alpha} \chi_{\varepsilon}\right| d x=\varepsilon^{-|\alpha|} \int\left|\partial^{\alpha} \chi\right| d x .
where C_(alpha)C_{\alpha} only depends on alpha,n\alpha, n and the choice of the norm. Using Theorem 1.3.5 it is possible to give a still more precise result which we mention for the sake of completeness since it is sometimes useful. However, the reader can jump to Theorem 1.4 .4 without loss of continuity. 其中 C_(alpha)C_{\alpha} 仅取决于 alpha,n\alpha, n 范数的选择。使用 Theorem 1.3.5 可以给出更精确的结果,为了完整起见,我们提到它,因为它有时很有用。但是,读者可以跳转到 Theorem 1.4 .4 而不会丢失连续性。
Theorem 1.4.2. Let XX be an open set in the nn dimensional vector space VV with norm ||||\|\| and let KK be a compact subset. If 定理 1.4.2.设 XX 为 nn 维向量空间中 VV 具有 norm 的开集 ||||\|\| ,设 KK 为紧子集。如果
d=i n f{||x-y||;x in{X,y in K}d=\inf \{\|x-y\| ; x \in\{X, y \in K\}
and d_(j)d_{j} is a positive decreasing sequence with sum_(1)^(oo)d_(j) < d\sum_{1}^{\infty} d_{j}<d, then one can find phi inC_(0)^(oo)(X)\phi \in C_{0}^{\infty}(X) with 0 <= phi <= 10 \leqq \phi \leqq 1, equal to 1 in a neighborhood of KK, so that 和 d_(j)d_{j} 是 的正递减序列,则 sum_(1)^(oo)d_(j) < d\sum_{1}^{\infty} d_{j}<d 可以在 KK 的邻域中找到 phi inC_(0)^(oo)(X)\phi \in C_{0}^{\infty}(X) ,等于 0 <= phi <= 10 \leqq \phi \leqq 1 1,因此
Here CC depends only on the dimension nn. 此处 CC 仅取决于 维度 nn 。
Proof. Assume first that V=R^(n)V=\mathbb{R}^{n} and that ||x||=max|x_(j)|\|x\|=\max \left|x_{j}\right|. Let uu be the function in Theorem 1.3.5 with a_(j)=d_(j+1)a_{j}=d_{j+1} and set h(t)=u(t+sumd_(j)//2)h(t)=u\left(t+\sum d_{j} / 2\right). Then we have |t| <= sum|d_(j)|//2|t| \leqq \sum\left|d_{j}\right| / 2 if t in supp ht \in \operatorname{supp} h, and for every jj 证明。首先假设 that V=R^(n)V=\mathbb{R}^{n} 和 that ||x||=max|x_(j)|\|x\|=\max \left|x_{j}\right| 。设 uu 为定理 1.3.5 中的函数,其中 a_(j)=d_(j+1)a_{j}=d_{j+1} 和 集 h(t)=u(t+sumd_(j)//2)h(t)=u\left(t+\sum d_{j} / 2\right) 。然后我们有 |t| <= sum|d_(j)|//2|t| \leqq \sum\left|d_{j}\right| / 2 if t in supp ht \in \operatorname{supp} h ,对于每个 jj
int|h^((j))(t)|dt <= 2^(j)//d_(1)dotsd_(j),quad int h(t)dt=1\int\left|h^{(j)}(t)\right| d t \leqq 2^{j} / d_{1} \ldots d_{j}, \quad \int h(t) d t=1
We can now apply the proof of Theorem 1.4 .1 with epsi=1\varepsilon=1 and 我们现在可以用 epsi=1\varepsilon=1 和 来应用定理 1.4 .1 的证明
which proves Theorem 1.4.2 in this case. The passage to a Euclidean norm and from there to an arbitrary norm can be made by the following lemma which gives (1.4.3) with C=2n^(2)C=2 n^{2}. 这证明了 Theorem 1.4.2 在这种情况下。到欧几里得范数的段落,以及从那里到任意范数的段落,可以通过以下引理来完成,它给出 (1.4.3) 和 C=2n^(2)C=2 n^{2} 。
Lemma 1.4.3. If KK is a convex symmetric body in the nn dimensional vector space VV, then one can find an ellipsoid BB with center at 0 and 引理 1.4.3.如果 KK 是 nn 维向量空间 VV 中的凸对称体,则可以找到一个 BB 中心为 0 且
B sub K sub Bsqrtn.B \subset K \subset B \sqrt{n} .
Proof. Let BB be an ellipsoid of maximal volume contained in KK and choose coordinates so that BB is the unit ball. We have to show that KKsub Bsqrtn\subset B \sqrt{n}. To do so we assume that KK contains a point (t,0,dots,0)(t, 0, \ldots, 0) with t > sqrtnt>\sqrt{n} and prove that BB cannot be maximal then. The tangent cone to BB with vertex at (t,0,dots,0)(t, 0, \ldots, 0) touches BB where x_(1)=1//tx_{1}=1 / t, so the part of BB where |x_(1)| > 1//t\left|x_{1}\right|>1 / t is in the interior of KK because of the convexity and symmetry with respect to 0 . Now consider the ellipsoid 证明。设 BB 是一个包含最大体积的椭球 KK 体,并选择坐标, BB 以便作为单位球。我们必须证明 KKsub Bsqrtn\subset B \sqrt{n} .为此,我们假设 contains KK a point (t,0,dots,0)(t, 0, \ldots, 0) with t > sqrtnt>\sqrt{n} 并证明 BB that 不能是最大值。与顶点接触 (t,0,dots,0)(t, 0, \ldots, 0)BB where x_(1)=1//tx_{1}=1 / t 的切圆锥 BB 体 ,因此 where |x_(1)| > 1//t\left|x_{1}\right|>1 / t 的部分 BB 位于 的内部 KK ,因为相对于 0 的凸性和对称性。现在考虑椭球体