14. Change of variables formula, Spherical measure and polar coordinates  PDF TEX

Linear transformations

Euclidean spaces

Let \(\mathbb{R}^d\) be the Euclidean \(d\)-dimensional space with the Euclidean norm \[|x|=\bigg(\sum_{j=1}^{d}|x_j|^2\bigg)^{1/2}=\sqrt{\langle x,x \rangle},\] which is induced by the standard inner product \[\langle x,y \rangle= x\cdot y= \bigg(\sum_{j=1}^{d}x_jy_j\bigg)^{1/2},\] for any \(x=(x_1,\ldots,x_d), y=(y_1,\ldots,y_d)\in\mathbb R^d\).

Jacobians

Jacobians If \(E \subseteq \mathbb{R}^d\) is open and \(f:E \to \mathbb{R}^d\) is differentiable at a point \(x \in E\), the determinant of the linear operator \(D_xf=f'(x)\) is called the Jacobian of \(f\) at \(x\in\mathbb R^d\) and denoted by \[J_f(x)=\det D_xf=\det f'(x).\] We shall also use the notation \[\frac{\partial (y_1,\ldots,y_d)}{\partial (x_1,\ldots,x_d)}=J_f(x) =\det \begin{bmatrix} \frac{\partial f_1}{\partial x_1} & \ldots & \frac{\partial f_1}{\partial x_d}\\ \vdots & \ddots & \vdots \\ \frac{\partial f_d}{\partial x_1} & \ldots & \frac{\partial f_d}{\partial x_d} \end{bmatrix}.\] if \((y_1,\ldots,y_d)=(f_1(x_1,\ldots,x_d), \ldots, f_d(x_1,\ldots,x_d))=f(x_1,\ldots,x_d)\).

Some comments

  • In terms of Jacobians, the crucial hypothesis in the inverse function theorem is that \(J_f(x) \neq 0\).

  • We will investigate behavior of the Lebesgue integral under linear transformations. We will identify a linear map \(T:\mathbb{R}^d \to \mathbb{R}^d\) with the matrix \[(T_{ij})_{1\le i, j\le d}=(e_i \cdot Te_j)_{1\le i, j\le d},\] where \(\{e_j: 1\le j\le d\}\) is the standard basis in \(\mathbb{R}^d\).

  • We denote the determinant of this matrix by \(\det T\) and recall that \[\det(T \circ S)=(\det T)(\det S).\]

  • Furthermore, \(\text{GL}(d,\mathbb{R})\) denotes the general linear group of invertible linear transformations of \(\mathbb{R}^d\).

Transformations of elementary types

Recall that every \(T \in {\rm GL}(d,\mathbb{R})\) can be written as the product of finitely many transformations of three elementary types:

  1. \[T_1(x_1,\ldots,x_j,\ldots,x_d)=(x_1,\ldots,cx_j,\ldots,x_d),\]

  2. \[T_2(x_1,\ldots,x_j,\ldots,x_d)=(x_1,\ldots,x_j+cx_k,\ldots,x_d), \quad \text{ for } \quad k \neq j,\]

  3. \[T_3(x_1,\ldots,x_j,\ldots,x_k,\ldots,x_d)=(x_1,\ldots,x_k,\ldots,x_j,\ldots,x_d).\]

That every \(T \in {\rm GL}(d,\mathbb{R})\) is a product of transformations of these three types is simply the fact that every invertible matrix can be row-reduced to the identity matrix.

Change of variables for linear transformations

From now on we shall consider \(\mathbb R^d\) with Lebesgue measure \(\lambda_d\) either on \(\mathcal L(\mathbb R^d)\) or on \({\rm Bor}(\mathbb R^d)=\mathcal B(\mathbb R^d)\). We shall also abbreviate \(d\lambda_d(x)\) to \(dx\).

Theorem.

Suppose \(T \in {\rm GL}(d,\mathbb{R})\).

  1. If \(f\) is a Lebesgue measurable function on \(\mathbb{R}^d\), so is \(f \circ T\). If \(f \geq 0\) or \(f \in L^1(\mathbb R^d)\), then \[\qquad\qquad \int_{\mathbb{R}^d}f(x)dx=|\det T|\int_{\mathbb{R}^d}f \circ T(x)dx. \qquad\qquad {\color{purple}(*)}\]

  2. If \(E \in \mathcal{L}(\mathbb R^d)\), then \(T[E] \in \mathcal{L}(\mathbb R^d)\) and \[\lambda_d(T[E])=|\det T|\lambda_d(E).\]

Proof. First suppose that \(f\) is Borel measurable, then \(f \circ T\) is Borel measurable since \(T\) is continuous.

  • If (*) is true for \(T,S \in {\rm GL}(d,\mathbb{R})\) it is also true for \(T \circ S\) since \[\begin{aligned} \int_{\mathbb{R}^d}f(x)dx=|\det T|\int_{\mathbb{R}^d}f(Tx)dx =|\det T||\det S|\int_{\mathbb{R}^d}f(TSx)dx. \end{aligned}\] It suffices to prove (*) when \(T\in\{T_1,T_2,T_3\}\) described above, which will be a simple consequence of the Fubini–Tonelli theorem.

  • Indeed, for \(T_3\) we interchange the order of integration in the variables \(x_j\) and \(x_k\), and for \(T_1\) and \(T_2\) we integrate first with respect to \(x_j\) and use the one-dimensional formulas \[\int_{\mathbb{R}}f(t)dt=|c|\int_{\mathbb{R}}f(xt)dt, \quad \text{ and } \quad \int_{\mathbb{R}}f(t+a)dt=\int_{\mathbb{R}}f(t)dt.\]

  • It is easily verified that \(\det T_1=c\), \(\det T_2=1\), \(\det T_3=-1\), so the formula (*) is proved.

  • Moreover, if \(E\) is a Borel set, then \(\lambda_d(T[E])=|\det T|\lambda_d(E)\) by taking \(f=\mathbf{1}_{{T[E]}}\) in (*) since \(T[E]\) is a Borel set as well. WHY? In particular, the class of Borel null sets is invariant under \(T\) and \(T^{-1}\).

  • Suppose \(E\in \mathcal{L}(\mathbb R^d)\), since \(\mathcal{L}(\mathbb R^d)\) is a completion of \(\mathcal{B}(\mathbb R^d)\), thus \(E=A\cup B\), where \(A\in \mathcal{L}(\mathbb R^d)\) and \(B\subseteq C\in \mathcal{B}(\mathbb R^d)\) and \(\lambda_d(C)=0\).

  • Then \(\lambda_d(T[B])\le \lambda_d(T[C])=|\det T|\lambda_d(C)=0\) and \(T[B]\in \mathcal{L}(\mathbb R^d)\), since \(\lambda_d\) is complete. Thus \[\lambda_d(T[E])=\lambda_d(T[A])=|\det T|\lambda_d(A)= |\det T|\lambda_d(E),\] since \(\lambda_d(T[E])=\lambda_d(T[A]\cup T[B])=\lambda_d(T[A])\) and \(\lambda_d(A)=\lambda_d(E)\).

  • By the previous argument (*) holds for \(f=\mathbf{1}_{{E}}\) if \(E\in \mathcal{L}(\mathbb R^d)\). By linearity (*) is true for step functions, by the is true for nonnegative \(\mathcal{L}(\mathbb R^d)\)-measurable functions. Consequently, (*) remains true for all \(\mathcal{L}(\mathbb R^d)\)-measurable functions by considering positive and negative parts of the real and imaginary parts of \(f\).$$\tag*{$\blacksquare$}$$

Change of variables for diffeomorphisms

Our goal

Corollary.

Lebesgue measure is invariant under rotations.

Proof. \(T \in {\rm GL}(d,\mathbb{R})\) is a rotation iff \(TT^*=I\), where \(T^{*}\) is the transpose of \(T\). Since \(\det T=\det T^*\) we obtain \(|\det T|=1\).$$\tag*{$\blacksquare$}$$

We now generalize previous theorem to differentiable maps. Let \(\Omega \subseteq \mathbb{R}^d\) be an open set and \(G=(g_1,\ldots,g_n):\Omega\to \mathbb{R}^d\) be a map with \(g_j\in C^1(\Omega)\).

Definition.

\(G:\Omega\to \mathbb{R}^d\) is called a diffeomorphism if \(G\) is injective and \(D_xG\) is invertible for all \(x \in \Omega\). In this case, the inverse function theorem guarantees that \(G^{-1}:G[\Omega] \to \Omega\) is also a \(C^1\) diffeomorphism and that \[D_x(G^{-1})=(D_{G^{-1}(x)}G)^{-1}\quad \text{ for all } \quad x \in G[\Omega].\]

Change of variables for diffeomorphisms

Theorem.

Let \(\Omega\subseteq \mathbb{R}^d\) be open and let \(G:\Omega \to \mathbb{R}^d\) be a \(C^1(\Omega)\) diffeomorphism.

  1. If \(f\) is a Lebesgue measurable function on \(G[\Omega]\) then \(f \circ G\) is Lebesgue measurable on \(\Omega\). If \(f \geq 0\) or \(f \in L^1(G[\Omega],\lambda_d)\), then \[\int_{G[\Omega]}f(x)dx=\int_{\Omega}f \circ G(x)|\det D_x G|dx.\]

  2. If \(E \subseteq \Omega\) and \(E \in \mathcal{L}(\mathbb R^d)\) then \(G[E] \in \mathcal{L}(\mathbb R^d)\) and \[\lambda_d(G[E])=\int_E |\det D_x G|dx.\]

Notation

For \(x \in \mathbb{R}^d\) and \(T=(T_{ij})_{1\le i, j\le d}\in {\rm GL}(d,\mathbb{R})\) we set \[|x|_{\infty}=\max_{1 \leq j \leq d}|x_j|, \qquad \|T\|=\max_{1 \leq i \leq d}\sum_{j=1}^d |T_{ij}|.\] Then it is easy to see that \[|Tx|_{\infty} \leq \|T\| |x|_{\infty}\] and \[a+[-r, r]^d=\{x \in \mathbb{R}^d: |x-a|_{\infty} \leq r\},\] is the cube of side length \(2r\) centered at \(a \in \mathbb{R}^d\).

Proof. It suffices to consider Borel measurable functions and sets. Since \(G\) and \(G^{-1}\) are both continuous, there are no problems with measurability.

  • Let \(Q=\{x \in \mathbb{R}^d: |x-a|_\infty \leq r\}\subseteq \Omega\). By the mean value theorem: \[g_j(x)-g_j(a)=\sum_{i=1}^d(x_i-a_i)\frac{\partial g_j}{\partial x_i}(y)\] for some \(y\) lying on the the line segment joining \(x\) and \(a\), so that \[|G(x)-G(a)|_\infty \leq r\big(\sup_{y \in Q}\|D_y G\|\big)\quad \text{ for } \quad x \in Q.\]

  • In other words, \(G[Q]\) is contained in the cube of side length \(\sup_{y \in Q}\|D_y G\|\) times that of \(Q\), so by the previous theorem, \[\lambda_d(G[Q]) \leq \big(\sup_{y \in Q}\|D_yG\|\big)^d\lambda_d(Q).\]

  • If \(T \in {\rm GL}(d,\mathbb{R})\) we can apply this formula with \(G\) replaced by \(T^{-1} \circ G\) together with the previous theorem to obtain \[\begin{aligned} \qquad \lambda_d(G[Q])&=|\det T|\lambda_d(T^{-1}(G[Q]))\\ &\leq |\det T|\Big(\sup_{y \in Q}\|T^{-1}D_y G\|\Big)^d\lambda_d(Q). \qquad {\color{purple}(**)} \end{aligned}\]

  • Since \(y \mapsto D_y G\) is continuous, for any \(\varepsilon>0\) we choose \(\delta>0\) so that \[\|(D_z G)^{-1}(D_y G)\|^d \leq 1+\varepsilon\] if \(y,z \in Q\) and \(|y-z|_{\infty} \leq \delta.\)

  • Let us subdivide \(Q\) into subcubes \(Q_1,\ldots,Q_N\) centered respectively at \(x_1,\ldots,x_N\), whose interiors are disjoint, and whose side lengths are at most \(\delta\).

  • By (**) with \(Q_j\) in place of \(Q\) and \(T=D_xG\) we see that

    \[\begin{aligned} \lambda_d(G[Q]) &\leq \sum_{j=1}^N \lambda_d(G[Q_j])\\ &\leq \sum_{j=1}^{N}|\det D_{x_j}G|\Big(\sup_{y \in Q_j}\|(D_{x_j}G)^{-1}D_yG\|\Big)^d \lambda_d(Q_j)\\ &\leq (1+\varepsilon)\sum_{j=1}^N |\det D_{x_j}G|\lambda_d(Q_j). \end{aligned}\]

  • This last sum is the integral of \[\sum_{j=1}^{N}|\det D_{x_j}G|\mathbf{1}_{{Q_j}}(x).\]

  • Then we see that \[\sum_{j=1}^N |\det D_{x_j}G|\mathbf{1}_{{Q_j}} \ _{\overrightarrow{\delta\to 0}} \ |\det D_x G|\mathbf{1}_{{Q}}\] converges uniformly, since \(D_x G\) is continuous.

  • Thus letting \(\delta \to 0\) and \(\varepsilon \to 0\) we find that \[\qquad \qquad \lambda_d(G[Q]) \leq \int_Q |\det D_x G|dx. \qquad \qquad {\color{purple}(***)}\]

  • We claim that (***) holds with \(Q\) replaced by any Borel set in \(\Omega\). Indeed, if \(U \subseteq \Omega\) is open we can write \[U=\bigcup_{j=1}^{\infty}Q_j,\] where \(Q_j\)’s are cubes with disjoint interiors.

  • Since the boundaries of the cubes have Lebesgue measure zero, hence by (***) we have \[\lambda_d(G[U])=\sum_{j=1}^{\infty}\lambda_d(G[Q_j]) \leq \sum_{j=1}^{\infty}\int_{Q_j}|\det D_x G|dx=\int_{U}|\det D_x G|dx.\]

  • Next, let \[W_k=\{x \in \Omega: |x|_{\infty}<k \text{ and } |\det D_x G|<k\},\] and note that \(W_k\subseteq W_{k+1}\) for \(k\in\mathbb N\).

  • If \(E\) is a Borel subset of \(W_k\) there is a decreasing sequence of open sets \(U_j \subseteq W_{k+1}\) such that \[E \subseteq \bigcap_{j=1}^\infty U_j \quad \text{ and } \quad \lambda_d\Big(\bigcap_{j=1}^\infty U_j \setminus E\Big)=0.\]

  • By the proceeding estimate and the dominated convergence theorem we have

    \[\begin{aligned} \lambda_d(G[E])& \leq \lambda_d\bigg(G\Big[\bigcap_{j=1}^{\infty}U_j\Big]\bigg) =\lim_{j \to \infty}\lambda_d(G[U_j])\\ &\leq \lim_{j \to \infty}\int_{U_j}|\det D_x G|dx=\int_{E}|\det D_xG|dx. \end{aligned}\]

  • Finally, if \(E\) is any Borel subset of \(\Omega\) we apply this argument to \(E \cap W_k\) let \(k \to \infty\) and conclude via the monotone convergence theorem that \[\lambda_d(G[E]) \leq \int_E |\det D_x G|dx.\]

  • If \(f=\sum_{j=1}^M a_j\mathbf{1}_{{A_j}}\) is a nonnegative simple function on \(G[\Omega]\) we therefore have \[\begin{gathered} \int_{G[\Omega]}f(x)dx=\sum_{j=1}^M a_j\lambda_d(A_k) \leq \sum_{j=1}^M a_j\int_{G[A_j]}|\det D_x G|dx\\ =\int_{\Omega}f \circ G(x) |\det D_x G|dx. \end{gathered}\]

  • By the monotone convergence we conclude that the same is true for any nonnegative measurable function \(f\), i.e. \[\begin{gathered} \int_{G[\Omega]}f(x)dx\le \int_{\Omega}f \circ G(x) |\det D_x G|dx. \end{gathered}\]

  • The same reasoning applied with \(G\) replaced by \(G^{-1}\) and \(f\) replaced by \(f \circ G\), yields that

    \[\begin{gathered} \int_{\Omega}f \circ G(x)|\det D_x G|dx \\ \leq \int_{G[\Omega]}f \circ G \circ G^{-1}|\det D_{G^{-1}(x)}G||\det D_x G^{-1}|dx\\ =\int_{G[\Omega]}f(x)dx. \end{gathered}\]

  • This establishes Theorem (a) for all \(f \geq 0\).

  • The case \(f \in L^1(\mathbb R^d)\) follows immediately.

  • Theorem (b) is a special case of Theorem (a) where \(f=\mathbf{1}_{{G[E]}}\).

  • The case if \(f\) is Lebesgue measurable can be handled in much the same ways as the previous theorem, we omit the details. $$\tag*{$\blacksquare$}$$

Example

We show that \[\int_{\mathbb{R}}e^{-x^2}dx=\sqrt{\pi}.\]

Consider \[\left(\int_{\mathbb{R}}e^{-x^2}dx\right)^2 =\int_{\mathbb{R}}\int_{\mathbb{R}}e^{-x^2-y^2}dxdy\] and a mapping \(G:(0,\infty)\times [0,2\pi) \to \mathbb{R}^2 \setminus [0,+\infty)\) given by \[G(r,\theta)=(r\cos(\theta),r\sin(\theta))\] Then we have \[D_{r,\theta}G= \begin{bmatrix} \cos(\theta) & \sin(\theta) \\ -r\sin(\theta) & r\cos(\theta) \end{bmatrix}, \quad \det D_{(r,\theta)}=r\cos(\theta)^2+r\sin(\theta)^2=r.\]

Thus \[\begin{aligned} \int_{\mathbb{R}}\int_{\mathbb{R}}e^{-x^2-y^2}dxdy&=\int_0^\infty \int_0^{2\pi}e^{-r^2}r d\theta dr\\ &=2\pi \int_0^{\infty}re^{-r^2}dr\\ &=\pi \int_0^{\infty}2re^{-r^2}dr\\ &=\pi \int_0^\infty e^{-t}dt=\pi. \end{aligned}\] Thus \[\qquad {\color{red}\int_{\mathbb{R}}e^{-x^2}dx =\left(\int_{\mathbb{R}}\int_{\mathbb{R}}e^{-(x^2+y^2)}dxdy\right)^{1/2}=\sqrt{\pi}.}\qquad\tag*{$\blacksquare$}\]

Integration in polar coordinates

  • The most important nonlinear coordinate system in \(\mathbb{R}^2\) and \(\mathbb{R}^3\) are polar coordinates \[\begin{gathered} {\color{red}(r,\theta) \longmapsto (r\cos(\theta),r\sin(\theta))},\\ {\color{blue}(r,\phi,\theta) \longmapsto (r\sin(\phi)\cos(\theta),r\sin(\phi)\sin(\theta),r\cos(\theta))}. \end{gathered}\]

  • The multi-dimensional change of variables formula yields \[\begin{gathered} {\color{red}dx\,dy=r\,dr\,d\theta},\\ {\color{blue}dx\,dy\,dz=r^2\sin(\phi)\,dr\,d\theta\,d\phi}. \end{gathered}\]

  • Similar coordinate systems exist in higher dimensions, but they become more complicated as the dimension increases.

Polar coordinates

  • For most purposes, however, it is sufficient to know that Lebesgue measure is effectively the product of the measure \({\color{purple}r^{n-1}dr}\) on \(\mathbb R_+=(0,\infty)\) and a certain "surface measure \(\sigma\)" on the unit sphere (\(d\sigma(\theta)=d\theta\) for \(d=2\); and \(d\sigma(\theta, \phi)=\sin(\phi)\,d\theta\,d\phi\) for \(d=3\)).

  • Our construction of this measure is motivated by a familiar fact from the plane geometry. Namely, if \(S_{\theta}\) is a sector of a disc of radius \(r>0\) with an angle \(\theta\), the area \(\lambda_2(S_{\theta})\) is proportional to \(\theta\). In fact, we have \[\lambda_2(S_{\theta})=\frac{1}{2}r^2\theta.\]

  • The equation above can be solved for \(\theta\) and used to define the angular measure \(\theta\) in terms of the area of \(\lambda_2(S_{\theta})\). The same works in \(\mathbb{R}^d\) for \(d \geq 2\).

Our construction

  • Let \(\mathbb S^{d-1}=\{x \in \mathbb{R}^d:|x|=1\}\) be the unit sphere in \(\mathbb{R}^d\). If \(x \in \mathbb{R}^d \setminus \{0\}\) the polar coordinates of \(x\) are \[r=|x| \in (0,\infty) \quad \text{ and } \quad x'=\frac{x}{|x|} \in \mathbb S^{d-1}.\]

  • The map \(\Phi(x)=(r,x')\) is a continuous bijection from \(\mathbb{R}^d \setminus \{0\}\) to \((0,\infty) \times \mathbb S^{d-1}\), whose continuous inverse is \[\Phi^{-1}(r,x')=rx'.\]

  • We denote by \(m_{d}\) the Borel measure on \((0,\infty) \times \mathbb S^{d-1}\) induced by \(\Phi\) from Lebesgue measure on \(\mathbb{R}^d\), that is \[m_d(E)=\lambda_d(\Phi^{-1}[E]).\]

  • Moreover, we define the measure \(\rho=\rho_d\) on \((0,\infty)\) by \[{\color{red}\rho(E)=\int_Er^{d-1}dr}.\]

Surface measure

Theorem.

There is a unique Borel measure \(\sigma=\sigma_{d-1}\) on \(\mathbb S^{d-1}\) such that \(\lambda_d=\rho \times \sigma\). If \(f\) is Borel measurable on \(\mathbb{R}^d\) and \(f \geq 0\) or \(f \in L^1(\mathbb R^d)\) then \[\qquad \qquad \int_{\mathbb{R}^d}f(x)dx =\int_0^{\infty}\int_{\mathbb S^{d-1}}f(rx')r^{d-1}d\sigma(x')dr. \qquad \qquad{\color{purple}(*)}\]

Remark. This theorem can be extended to Lebesgue measurable functions by considering the completion of the measure \(\sigma\).

Proof. Equation (*) when \(f\) is a characteristic function of a set is a restatement of the equation \(m_{d}=\rho \times \sigma\) and it follows for general \(f\) by the usual linearity and approximation arguments. Hence we need only to construct \(\sigma\).

  • If \(E\) is a Borel set in \(\mathbb S^{d-1}\) for \(a>0\) let \[E_{a}=\Phi^{-1}((0,a] \times E)=\{rx': (r, x')\in (0,a] \times E\}.\] If (*) is to hold with \(f=\mathbf{1}_{{E}}\), we must have \[\lambda_d(E_1)=\int_0^1 \int_E r^{d-1}d\sigma(x')dr =\sigma(E)\int_0^1 r^{d-1}dr =\frac{\sigma(E)}{d}.\]

  • We therefore define

    Definition.

    \[{\color{red}\sigma(E)=d \cdot \lambda_d(E_1)}.\]

  • Since the map \(E \mapsto E_1\) takes Borel sets to Borel sets and commutes with unions, intersections, and complements, it is clear that \(\sigma\) is a Borel measure on \(\mathbb S^{d-1}\). Also, since \(E_a\) is the image of \(E_1\) under the map \(x \mapsto ax\), it follows that \(\lambda_d(E_a)=a^d\lambda_d(E_1)\).

  • Fix \(E \in \mathcal{B}(\mathbb S^{d-1})\) and let \(\mathcal{A}_E\) be a the collection of finite disjoint unions of sets of the form \((a,b] \times E\). We know that \(\mathcal{A}_E\) is an algebra on \(\mathbb R_+ \times E\) that generates the \(\sigma\)-algebra \(\mathcal{M}_E=\{A \times E: A \in \mathcal{B}(\mathbb R_+)\}\).

  • We show that \(m_d=\rho \times \sigma\) on \(\mathcal{A}_E\). If \(0<a<b\), then we have \[\begin{aligned} m_d((a,b] \times E)&=\lambda_d(E_b \setminus E_a)=\frac{b^d-a^d}{d}\sigma(E)\\ &=\sigma(E)\int_a^b r^{d-1}dr={\color{blue}\rho \times \sigma((a,b] \times E).} \end{aligned}\]

  • Since \(m_d=\rho \times \sigma\) on \(\mathcal{A}_E\), hence by the uniqueness assertion of Caratheodory’s extension theorem we have \(m_d=\rho \times \sigma\) on \(\mathcal{M}_E\).

  • But \(\bigcup_{E \in \mathcal{B}(\mathbb S^{d-1})}\mathcal{M}_E\) is precisely the set of Borel rectangles in \(\mathbb R_+ \times \mathbb S^{d-1}\) so another application of the uniqueness theorem shows that \(m_d=\rho \times \sigma\) on all Borel sets.$$\tag*{$\blacksquare$}$$

Remark and Corollary

Corollary.

If \(f:\mathbb{R}^d\to \mathbb C\) is a measurable function nonnegative or integrable, such that \(f(x)=g(|x|)\) for some measurable function \(g:\mathbb{R}_+\to \mathbb C\), then \[\int_{\mathbb{R}^d}f(x)dx =\sigma(\mathbb S^{d-1})\int_0^{\infty}g(r)r^{d-1}dr.\]

Proof. By the polar decomposition \[\begin{aligned} \int_{\mathbb{R}^d}f(x)dx&=\int_0^{\infty} \int_{\mathbb S^{d-1}}f(rx')d\sigma(x')dr\\ &=\int_0^\infty \int_{\mathbb S^{d-1}}g(r|x'|)r^{d-1}d\sigma(x')dr\\ &=\sigma(\mathbb S^{d-1})\int_{0}^{\infty}g(r)r^{d-1}dr. \qquad\qquad {\blacksquare} \end{aligned}\]

Proposition

Proposition. For \(a>0\) we have \[I_d=\int_{\mathbb{R}^d}e^{-a|x|^2}dx=\left(\frac{\pi}{a}\right)^{d/2}.\]

Proof. For \(d=2\) we obtain \[I_2=2\pi\int_0^{\infty}re^{-ar^2}dr=\frac{\pi}{a},\] since \(\sigma(\mathbb S^1)=2\pi\). Observe now that \(e^{-a|x|^2}=\prod_{j=1}^de^{-ax_j^2}.\) Thus, by Tonelli’s theorem, \(I_d=(I_1)^d.\) In particular, \(I_1=I_2^{1/2}\), so \[\qquad\qquad I_d=(I_2)^{d/2}={\color{red}\left(\frac{\pi}{a}\right)^{d/2}}.\qquad\qquad \tag*{$\blacksquare$}\]

Surface area of \(\mathbb S^{d-1}\)

Proposition. Recall that \(\Gamma(x)=\int_0^{\infty}t^{x-1}e^{-t}dt\). Then one has \[\sigma(\mathbb S^{d-1})=\frac{2\pi^{d/2}}{\Gamma(d/2)}.\]

Proof. By the previous proposition we have \[\begin{aligned} \pi^{d/2}&=\int_{\mathbb{R}^d}e^{-|x|^2}dx=\sigma(\mathbb S^{d-1})\int_0^\infty e^{-r^2}r^{d-1}dr\\ &=\frac{1}{2} \sigma(\mathbb S^{d-1}) \int_0^{\infty}s^{d/2-1}e^{-s}ds =\frac{1}{2}\sigma(\mathbb S^{d-1})\Gamma(d/2). \end{aligned}\] Thus \[\qquad\qquad{\color{blue}2\pi^{d/2}\Gamma(d/2)^{-1}=\sigma(\mathbb S^{d-1})}.\qquad\qquad\tag*{$\blacksquare$}\]

Volume of the unit ball

Proposition. One has \[\lambda_d(B(0, 1))=\frac{\pi^{d/2}}{\Gamma(d/2+1)},\] where \(B(0, 1)=\{x \in \mathbb{R}^d: |x|<1\}.\)

Proof. Since \(x\Gamma(x)=\Gamma(x+1)\) we have \[\begin{aligned} \qquad\qquad\lambda_d(B(0, 1))& =\int_{\mathbb{R}^d}\mathbf{1}_{{B(0, 1)}}(x)dx\\ &=\sigma(\mathbb S^{d-1})\int_0^1 r^{d-1}dr\\ &=\frac{1}{d}\sigma(\mathbb S^{d-1})\\ &=\frac{2\pi^{d/2}}{d\Gamma(d/2)} =\frac{\pi^{d/2}}{\Gamma(d/2+1)}.\qquad\qquad{\blacksquare} \end{aligned}\]

Top