This is an old revision of the document!
Table of Contents
2020-02-10, Monday
We first finish up the derivation of Laplacian last time. See the lecture note 2020-02-07, Friday. Then, we introduce two vector operations, divergence and curl. Finally, we give a summary of the mathematical treatment of tensor analysis.
Then, we will follow Boas 10.9 and 10.10, to introduce the physic-engineer notations: the nabla operator $\gdef\b{\mathbf} \b \nabla$.
Differential of a function
In Cartesian coordinate, the differential of a function $f$ is $$ df = \sum_i \frac{\df }{\d x_i} dx_i. $$
In general coordinate $(u_1, \cdots, u_n)$, the differential of a function $f$ is $$ df = \sum_i \frac{\df }{\d u_i} d u_i. $$
You can specify the differential of a function directly: $df$ at a point $p \in \R^n$, is a linear function on $T_p \R^n$: it sends an element $\b v \in T_p \R^n$ to $\b v(f)$ the directional derivative of $f$ along the vector $\b v$.
Gradient of a function (is a vector field)
In Cartesian coordinate, the gradient of a function is $$ \gdef\grad{\text{ grad } } \grad f = \sum_i $$
Divergence of a Vector field (is a function)
Let $\R^n$ be the flat space, with standard coordinates $(x_1, \cdots, x_n)$.
Let $\b V$ be a vector field on $\R^n$, that is, for each point $p \in \R^n$, we specify a tangent vector $$ \b V(p) = \sum_{i=1}^n V^i(p) \d_i \in T_p \R^n. $$ We require that $V^i(p)$ varies smoothly with respect to $p$.
The divergence of $\b V$ is a function on $\R^n$, $$ div(\b V) = \sum_{i=1}^n \d_i( V^i ) $$ recall that $V^i$ is a function on $\R^n$, and $\d_i$ is taking the partial derivative with respect to $x_i$.
What does divergence mean? Geometrically, it measure the relative change-rate of the volume of an infinitesimal cube situated at point $p$. Suppose $\Phi^t: \R^n \to \R^n$ is the flow generated by $\b V$ (every point moves as dictated by $\b V$). And let $C= C(p, \epsilon)$ be a cube of side-length $\epsilon$, center at $p$. Then, we have the geometrical interpretation as $$ \gdef\div{\text{div}} \gdef\vol{\text{Vol}} \div(\b V) = \lim_{\epsilon \to 0} \frac{1}{\vol(C)} \frac{d \vol(\Phi^t(C))}{dt} \vert_{t=0}. $$ That is why, if $S \subset \R^n$ is an open domain, we can compute the change-rate of the volume of $S$ by $$ \frac{d \vol(\Phi^t(S)) } {dt}\vert_{t=0} = \int_{S} \div(\b V)(\b x) \, d \vol(\b x). $$
If we are given a curvilinear coordinate, how to compute the divergence?
