I am an undergraduate physics student. I would prefer answers that share more information and clarification rather than try to be concise.

To my understanding, Helmholtz theorem tells us that a vector function can be decomposed into the sum of a gradient of a scalar and curl of a vector function. Thus, if you know the divergence of a vector function is non-zero and also that the curl of the same function is zero, you can say that the vector function is wholly determined by the non-zero quantity. This occurs in electrostatics with the vector function for electric field. I understand this decomposition, but I would like to gain a better intuition as to why both quantities (divergence and curl) are needed to determine the vector field.

Suppose we have a vector function ${\textbf{D}}$ with divergence given by $\nabla \cdot \textbf{D} = \rho_{f}$. Let's say its curl is non-zero: $\nabla \times \textbf{D} = \nabla \times \textbf{P} \not= \textbf{0}$. The symbols aren't important, but they represent electric displacement ($\textbf{D}$), polarization ($\textbf{P}$), and free charge density ($\rho_{f}$) from 4ed Griffith's Introduction to Electrodynamics.

Is it accurate to draw parallels between the relationship of the derivative and integral of a scalar function and the divergence and anti-divergence (if such thing exists) of a vector function?

For example, let $$y = 1$$ Then, $$\frac{dy}{dx} = 0$$ And, $$\int{\frac{dy}{dx}} = C$$ where C is a constant.

Obviously, in the scalar case the anti-derivative does not have/provide enough information for us to go back to the original function $y$. Is this what is happening in the vector function case as well?

It seems I might be conflating boundary conditions (which if provided in the scalar case would give us a unique answer) and something else. Any help would be much appreciated!

  • 2
    $\begingroup$ Yes, the divergence of a vector field is a DERIVATIVE. Just as with the derivative of a number valued function, you lose a constant (vector field). $\endgroup$ May 29 at 0:33
  • 1
    $\begingroup$ @user1046533 Let me know if this is not clarifying. We have the derivative and anti-derivative operators. I am supposing that there is a divergence (true) and an anti-divergence operator analogous in function to the anti-derivative operator. Given such a pair of operators exists, does the analysis I presented in the question on the scalar function 1 carry over to an analysis of some vector function. $\endgroup$ May 29 at 0:45
  • $\begingroup$ If a vector field were determined purely by its divergence, the Maxwell equation $\nabla\cdot B = 0$ would be saying the magnetic field doesn't exist. $\endgroup$ Jun 1 at 13:12
  • $\begingroup$ multivariable-calculus may be a more appropriate tag to use than calculus. $\endgroup$ Jun 1 at 16:07
  • $\begingroup$ Fun fact: the divergence and curl of $F = (y+z,x+z,x+y)$ are 0. $\endgroup$
    – TomKern
    Jun 1 at 16:59

1 Answer 1


$\def\rbf{\mathbf{R}}$For simplicity consider a smooth function $f:\rbf\to\rbf$. Suppose $f'(x)=g(x)$ for all $x$. Then for any constant $C$, one also has $(f+C)'=g$. So the derivative $f'$ does not determine the function $f$ completely. On the other hand, the derivative determines the function "partially". If one knows that $f'=g'$, then one must have $f=g+C$ by the mean value theorem. One can replace the domain $\rbf$ of $f$ with any open interval.

["Divergence" is misread as "gradient" in this part.]
$\def\bg{\mathbf{g}}$Now let $f:\rbf^3\to\rbf$ be a scalar function and suppose $\nabla f(x)=\bg(x)$ for all $x$. Then $\nabla(f+C)=\bg$ for any constant $C$. So the gradient $\nabla f$ does not determine the function $f$ completely. On the other hand, if $\nabla p = \nabla q$ for two scalar functions $p$ and $q$, then similarly, $p=q+C$ for some constant $C$. The result holds if one replaces $\rbf^3$ with any nonempty connected open subset of $\rbf^3$.
In the scalar case, by the fundamental theorem of calculus, an antiderivative of the function $g$ is given by $$ f(x)=\int_0^xg(t)\;dt\;. $$ In the vector-valued function case, an "anti-gradient" of the vector field $\bg$, assuming that $\textrm{curl} (\bg)=0$, can be found as the line integral $$ f(x,y,z)=\int_{(0,0,0)\rightsquigarrow(x,y,z)} \bg\cdot dr $$ where $(0,0,0)\rightsquigarrow(x,y,z)$ denotes any path from $(0,0,0)$ to $(x,y,z)$.

[Added later.]

Let $\lambda:\rbf^3\to\rbf$ be a scalar field on $\rbf^3$ and $F:\rbf^3\to\rbf^3$ a vector field with $$ \nabla F=\lambda\;.\tag{1} $$ Then every vector field $F_1:=F+C$ for some constant $C$ satisfies (1). So divergence alone does not determine the vector field. This answers your question in the title.

A theorem says that a vector field can be constructed with both a specified divergence and a specified curl: $$ \nabla\cdot F=\lambda,\quad \nabla\times F=q\tag{2} $$ where $q$ is a given divergence-free vector field. But since (2) is still invariant up to a constant, $F$ is not determined uniquely even by its divergence and curl. One needs an additional "decay property" of $F$ for uniqueness. That property can be thought of as some "boundary condition" to the system in (2).

  • 1
    $\begingroup$ typo on "∇p=∇q for two scalar functions p and p" $\endgroup$
    – justhalf
    May 29 at 8:54
  • $\begingroup$ Hm. I'm not sure that this quite answers the question. I think it's my fault for not presenting the question clearly. To my understanding, the second part of your answer refers to a conservative vector field which has such path integrals. Maybe my question can be simplified to: Suppose you know the divergence of a vector function: $\nabla \cdot \textbf{D} = \lambda$ where $\lambda$ is some arbitrary quantity. Can the curl of this vector function $\nabla \times \textbf{D}$ be considered a boundary condition for solving for the vector function $\textbf{D}$? $\endgroup$ Jun 1 at 5:02
  • $\begingroup$ If so, could you give a hint (not the answer) as to where the need for two boundary conditions comes from? I haven't taken differential equations yet, but I assume an interpretation can be drawn from there? $\endgroup$ Jun 1 at 5:04
  • $\begingroup$ @SillyGoose: I somehow misread "divergence" as "gradient". Now the answer is edited. $\endgroup$ Jun 1 at 13:10
  • $\begingroup$ If you know $\nabla\cdot D=\lambda$ alone, you would need two more "boundary conditions", counting the degrees of freedom informally. The curl can be regarded as one condition; but for uniqueness, you would need yet another one. $\endgroup$ Jun 1 at 13:12

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.