I'm going to explain my thought process. Question and counter-arguments at the end.

Divergence Theorem:

$$\int_{\mathscr{D}_i\subset \mathbb{R}^3} \nabla \cdot \vec{f} dV_i=\int_{\partial\mathscr{D}_i\subset \mathbb{R}^2} \vec{f} \cdot \hat{n}_i dS_i,\quad i\in\mathbb{N}\quad (1)$$

The volume integral over some domain, ${\mathscr{D}}$, of the divergence of a vector, $\vec{f}$, equals the surface integral of the inner product of, $\vec{f}$, and the unit vector, $\hat{n}$, normal to the boundary of that volume (denoted ${\partial\mathscr{D}}$). The reason I call it the $i^{th}$ domain will be clear in a moment.

Vanishing Theorem:

$$\int_{\forall\mathscr{D}_i\subset \mathbb{R}^n} g \ dx= 0 \rightarrow g\equiv0,\quad i\in\mathbb{N} \quad (2)$$

If the integral of some function, g, over some n-dimensional domain is zero for all domains in $\mathbb{R}^n$, then g can only be zero.

We are generally taught to use (1) on a single volume at a time. If we were interested in two non-intersecting volumes, we would simply solve the system: $$\int_{\mathscr{D}_1\subset \mathbb{R}^3} \nabla \cdot \vec{f} dV_1=\int_{\partial\mathscr{D}_1\subset \mathbb{R}^2} \vec{f} \cdot \hat{n}_1 dS_1\quad (3a)$$ $$\int_{\mathscr{D}_2\subset \mathbb{R}^3} \nabla \cdot \vec{f} dV_2=\int_{\partial\mathscr{D}_2\subset \mathbb{R}^2} \vec{f} \cdot \hat{n}_2 dS_2\quad (3b)$$ Suppose $\vec{f}$ were divergence free ($\nabla \cdot \vec{f}=0$). (3a-b) are now, $$\int_{\partial\mathscr{D}_1\subset \mathbb{R}^2} \vec{f} \cdot \hat{n}_1 dS_1=\int_{\partial\mathscr{D}_2\subset \mathbb{R}^2} \vec{f} \cdot \hat{n}_2 dS_2=0 \quad (4)$$ The nature of the solution depends on many things. I want to draw your attention to $\hat{n}_1$, and $\hat{n}_2$. Above all else, their relative orientation determines the existence and uniqueness of the solution... Now onto my question... Suppose I have an equation like the continuity equation from physics. A generic form of this equation could be: $$\int_{\forall\mathscr{D}_i\subset \mathbb{R}^3} a+\nabla \cdot \vec{b} \ dV= 0\quad (5)$$ Normally authors don't write "$\forall\mathscr{D}_i$". Rather they say that the equation holds for an "arbitrary volume." To me, the concepts are equivalent but I welcome pedantic feedback if you see room for error. If $\nabla \cdot \vec{b}=0$, then by linearity of integration I can simply write, $$\int_{\forall\mathscr{D}_i\subset \mathbb{R}^3} \nabla \cdot \vec{b} \ dV_i= 0\quad (6)$$ By (1) it then follows that, $$\int_{\forall\partial\mathscr{D}_i\subset \mathbb{R}^2} \vec{b} \cdot \hat{n}_i dS_i= 0\quad (7)$$ Here's where my problems start. By (2) it should follow that, $$\int_{\forall\partial\mathscr{D}_i\subset \mathbb{R}^2} \vec{b} \cdot \hat{n}_i dS_i= 0 \, \rightarrow \vec{b}\cdot \hat{n}_i=0 \, \forall \; i\in\mathbb{N} \quad (8)$$ Since "$\forall\partial\mathscr{D}_i$" is an infinitely large set, then from (8) I should be able to conclude that, $$\vec{b}\cdot \hat{n}_i=0\rightarrow\vec{b}\equiv 0\quad (9)$$ Thus if (2) holds, the only divergence-free vector permitted by (5) is the zero vector. Question: Is (9) a valid special case, or a fallacy? I presented this to two math PhD's and got the following counter arguments:

Counter #1: If $\vec{b}$ is constant it is divergence free. My response to that: If $\vec{b}$ is constant, it cannot satisfy (7) and (7) should follow directly from (1), therefore, a constant non-zero vector should not be a valid solution even though it's divergence free.
Counter #2: The divergence theorem (1) does not give us "new" information, therefore, we cannot infer anything about a vector given only the fact that it is divergence free. My response to that: Since (1) is for a single volume, then I agree that for a single volume the inference is impossible. However, I claim that the "new information" comes from (2). By combining (1) and (2) I am saying (1) must hold over all possible volumes, and that "new" information should make (9) valid.

A quick note: all feedback is appreciated but I respectfully request that the responses be pure math responses. Responses based on physics would broaden the scope of the discussion too much. I want to get to the heart of the mathematics here.


Step 8 is incorrect. You know that for any appropriate domain $\mathscr{D}$, $\int_{\partial\mathscr{D}}\vec{b} \cdot \hat{n} dS=0$, where $\hat{n}$ is the normal vector to $\partial\mathscr{D}$. But you cannot conclude from this that the function $\vec{b}\cdot\hat{n}$ is identically zero, because $\hat{n}$ is different for each different choice of $\mathscr{D}$. There is no single function $g$ such that $g=\vec{b}\cdot\hat{n}$ and the integral of $g$ on any domain is $0$. Rather, for each domain, you have a function $\vec{b}\cdot\hat{n}$ depending on that domain (since $\hat{n}$ depends on $\mathscr{D}$) whose integral is $0$. You can't deduce from this that any function vanishes identically.

As your "Counter #1" stated, the conclusion you reach in (9) is in fact not true, since a constant vector field $\vec{b}$ is a counterexample. A constant vector field actually does satisfy (7) (and this is no problem since your step 8 does not follow).

(There are also some more minor issues. For instance, it is not correct to write that $\partial\mathscr{D}\subset\mathbb{R}^2$: the set $\partial\mathscr{D}$ is a surface in $\mathbb{R}^3$, not an open subset of the plane. So you can't apply the vanishing theorem in the form you stated: instead you would need a version of it where you have a function on $\mathbb{R}^3$ and you are integrating it over all possible closed surfaces you can get as boundaries. But this is not an essential point--the key error in your reasoning is in Step 8.)

  • $\begingroup$ Thank you for your answer. I understand now. To put this in terms of the subscripts: $\vec{b}\cdot\hat{n}_1 = g_1$, $\vec{b}\cdot\hat{n}_2 = g_2$, etc. In which case, this is just an infinite list of individual cases and, as you said, you can't make the inference. (2) requires a unique g. $\endgroup$
    – Eric
    Mar 22 '17 at 2:44
  • $\begingroup$ That's exactly right. $\endgroup$ Mar 22 '17 at 2:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.