How can you best explain divergence and curl?

www.quora.com /How-can-you-best-explain-divergence-and-curl

tl;dr
You and three friends float down a river, each marking a corner of a square. If your square is getting bigger, the
river has positive divergence. If it's shrinking, negative divergence.
Next, you and your friends are rigidly connected so your square can't change shape. If the square starts rotating
like a frisbee as it goes along, the river has curl. Positive curl is counterclockwise rotation. Negative curl is
clockwise.
This answer assumes a good knowledge of calculus, including partial derivatives, vectors, and the way we talk
about these things in an introductory calculus-based physics course. I'll also assume you know the kinematics of
rotations and how to approximate multi-variable functions about an arbitrary point with the first-order terms of its
Taylor series.

Tubing Down a River

Most students learn the divergence and curl because they're important in Maxwell's equations of
electrodynamics. These concepts apply to any vector field, though. Here, let's just visualize you and some
friends floating down a river on inner tubes. The vector field is the field giving the velocity of the river's flow. The
divergence and curl describe what happens to you and your friends as you float down the river together.
Suppose you are tubing down a river with three friends. You position yourselves into a square formation.
Assuming the river flows perfectly-evenly, you'll all float along together and stay in that perfect square.
However, if the river moves faster in some places than others, some of you might get ahead or fall behind.
Maybe the flow picks up sideways components and you drift apart. Your square gets messed up.
The black arrows show some sort of messy, complicated flow. The red square that is your initial positions distorts
into the pointed shape.
Mathematically, we will treat a person in an inner tube as a point moving in two dimensions. The flow of the river
is a vector field - a vector for the velocity of the flow at each point. We imagine that people flow perfectly down
the river; their velocity always matches the vector field exactly. (In real life, this is not exactly how it happens; the
water can move underneath you.) Also, the people themselves are just test particles - bits of dust floating on the
surface that don't affect the river's flow in any way. Because we always know their velocity, we can integrate their
motion forward and find their trajectories.
Finally, to avoid complications, imagine that the vector field describing the river's flow is constant in time.
Different parts can flow at different speeds (so you could speed up or slow down as you tube along), but the
water passing by some particular fixed point always has the same speed and direction.
We'll denote the flow's velocity vector as a function of position as
[math]\mathbf{V}(x,y) = V_x(x,y) \hat{x} + V_y(x,y) \hat{y}[/math]
[math]V_x[/math] and [math]V_y[/math] are real-valued functions of two variables. [math]\hat{x}[/math] and
[math]\hat{y}[/math] are unit vectors in the x and y directions. To save space, I'll sometimes just write
[math]\mathbf{V}[/math] instead of [math]\mathbf{V}(x,y)[/math], so just keep in mind that it's a function of
position.

Area and Divergence

Let's take the shape made by you and your three friends and call that your span. Originally, your span is a
square, but as you go along it may distort. Divergence describes how fast the area of your span is changing.
For example, imagine that the river gets faster and faster the further you go downstream. Then your friends in
front of you will keep getting further and further ahead, and your span stretches out. This is an example of a
positive divergence. This is illustrated below. The shape of the span is accurately calculated for this particular
vector field.
On the other hand, if places further up move more slowly so you scrunch in closer to your friends, that's negative
divergence; the area of your span goes down.
If you are all floating apart from each other, that's positive divergence.
source: The idea of the divergence of a vector field
In this vector field, your square would get bigger in all directions at once.
The opposite, where all the arrows come pointing in, would be negative divergence.
If the river is simultaneously getting faster and narrower, your square gets longer and skinnier. If it does it just
right, the "getting longer" part could cancel the "getting skinnier" part so that the area stays the same. This would
be zero divergence. Here's an example I constructed:

Preliminary Definition of Divergence

We would like to start calculating the divergence. It seems like we should get something like "such-and-such
many square meters per second" for our answer, because the divergence has to do with the rate that an area is
changing.
This has a problem, though, because the number will be different if we start with a tiny span than if we start with
a large span. We want the divergence to refer to a property of the mathematical vector field. It captures the
"spreading apart" nature of the vectors themselves. Its value shouldn't depend on some particular span size we
choose.
The solution is to use the percentage area gained per second, which is the same for large and small spans. This
could also be called the fractional rate of change. The fractional rate of change of the area is the rate of change
of the area in absolute terms divided by the area itself. This our definition of the divergence, which we will
denote by [math]\mathrm{Div}[/math]
[math]\mathrm{Div}(\mathbf{V(x,y)}) = \frac{\text{rate of change of area of span}}{\text{area of span}}[/math]
The units of the area cancel out, and we're left with units of inverse seconds, as is normal for rates.
If the divergence is [math]2 \mathrm{s}^{-1}[/math], then the span is adding area at a rate of two times its own
area once per second. So it would take about 0.005 seconds to increase its area by 1%.
(Side note: if you understand calculus well, you can see that this means that after one second, the span would
be [math]e^2[/math] times as large as it originally was.)
We can now calculate a divergence. Let's imagine the river's flow is described by
[math]V(x,y) = 0.04 x\, \mathrm{s}^{-1} \, \hat{x} [/math]

This is saying that the flow is all in the x-direction and that it gets faster and faster as you go further along in that
direction. If you went out 100 meters, the flow rate would be
[math]0.04 * 100 \mathrm{m} \, \mathrm{s}^{-1} = 4 \mathrm{m/s}[/math].
You're [math]100 \mathrm{m}[/math] out, going [math]4 \mathrm{m/s}[/math]. Let's say you have a span that's a
square [math]1\mathrm{m}[/math] on a side. Your two friends in front of you are [math]1\mathrm{m}[/math]
ahead, so they're going [math]4.04 \mathrm{m/s}[/math]. That means they're pulling away from you at [math]0.04
\mathrm{m/s}[/math]. That means the square is growing at a rate of
[math]0.04 \mathrm{m^2/s}[/math].
If you didn't follow where all the numbers came from in the previous paragraph, you should go back and work
them out. Putting it together, the divergence is
[math]\mathrm{Div}(\mathbf{V}) = \frac{0.04 \mathrm{m^2/s}}{1 \mathrm{m}^2} = 0.04 \mathrm{s}^{-1}[/math]
You can repeat this calculation putting yourself at any position and using any size span (and rotating it if you
want). You'll get the same answer.
If you want to test yourself, suppose the flow is given by
[math]V(x,y) = (a x + by + e) \hat{x} + (c x + dy + f) \hat{y}[/math]
Give yourself a starting span somewhere with a certain size. Figure out the rate that its area is changing, and
find the divergence. You should find that the divergence is
[math]\mathrm{Div}(\mathbf{V}) = a + d[/math]
This result actually makes intuitive sense. [math]e[/math] and [math]f[/math] are constants added to the velocity.
They just move the entire square along without changing its shape, so they don't cause any divergence. We're
only concerned with the relative motion between the corners. Therefore, we can simply take the bottom-left
corner and consider it to be fixed.
[math]a[/math] represents how fast the right hand side of the box is moving out. This adds to the box's area, and
so is part of the divergence.
[math]b[/math] represents how fast the top part moves to the right.
This causes your square to shear, turning it into a parallelogram, like this:
source: Shear stress
It still has the same base and height, so the area doesn't change. That's why [math]b[/math] doesn't come into
the formula.
The same reasoning applies to [math]c[/math] (the right hand side sliding up and down, more shear) and
[math]d[/math] (moving the top upward, adding to the area, and affecting the divergence).

Calculus
For the simple example I gave, the divergence is the same no matter where you put your span. However, in a
complicated flow, the flow will be doing different things at different places. It may be coming together (negative
divergence) in some places and moving apart (positive divergence) in others. Evidently, the divergence needs to
be a function of [math]x[/math] and [math]y[/math].
This presents a problem, because now the size of the span is going to make a difference. If the divergence is

different from spot to spot, then it's different at different spots inside your span, but we're just trying to get a single
correct answer.
This is very similar to the problem of finding the slope of a line in calculus. The slope of the line changes from
point to point, so if you calculate [math]\frac{rise}{run}[/math] between two points, you keep getting different
answers depending on how far apart the points are. Slope shouldn't depend on how far apart the points you
choose are, and divergence shouldn't depend on how large an original span you choose.
The solution to both problems is the same - we take a limit. In calculus, you take the limit as the distance
between your two points on the function goes to zero. In vector calculus, we take the limit as the size of our
starting span goes to zero.
In single-variable calculus you'll get the right answer for the slope using a finite-size interval as long as you're
measuring the slope of a straight line. That's essentially why the examples I gave in the previous section work;
the vector field I gave is the equivalent of a straight line.

Expression for the Divergence

When you zoom in close to a function, it looks locally like a straight line. Likewise, when you zoom in close to a
vector field, it looks locally like our equivalent to a straight line, for which we already calculated the divergence.
To be specific, a function of one variable looks locally like
[math]f(x) \approx f(x_0) + \left. \frac{\mathrm{d}f}{\mathrm{d}x}\right|_{x_0} (x - x_0)[/math]
remember that our vector field is
[math]\mathbf{V}(x,y) = V_x (x,y) \hat x + V_y(x,y) \hat{y}[/math]
We can expand the functions [math]V_x[/math] and [math]V_y[/math] the same way we expand single-variable
functions, so long as we use two partial derivatives.
[math]V_x \approx V_x(x_0,y_0) \!+ \!\left.\frac{\partial V_x}{\partial x }\right|_{x_0,y_0}\!\!\!\!\!\!\!\!\!(x - x_0) \!+\!
\left.\frac{\partial V_x}{\partial y}\right|_{x_0,y_0}\!\!\!\!\!\!\!\!\!(y - y_0)[/math]
and likewise for [math]V_y[/math]. Plugging these zoomed-in functions for [math]V_x[/math] and
[math]V_y[/math] into [math]\mathbf{V}[/math] gives
[math]\mathbf{V}(x,y) = \mathbf{V}(x_0 , y_0) + [/math]
[math]\left(\left.\frac{\partial V_x}{\partial x}\right|_{x_0,y_0} (x - x_0) + \left.\frac{\partial V_x}{\partial
y}\right|_{x_0,y_0} (y - y_0)\right)\hat{x} +[/math]
[math] \left(\left.\frac{\partial V_y}{\partial x}\right|_{x_0,y_0} (x - x_0) + \left.\frac{\partial V_y}{\partial
y}\right|_{x_0,y_0} (y - y_0)\right)\hat{y} [/math]
This is the same as our earlier vector field, with partial derivatives standing in for [math]a,b,c,d[/math], and the
initial value
[math]\mathbf{V}(x_0,y_0)[/math] standing in for [math]e[/math] and [math]f[/math], so by analogy we can say
that the divergence is
[math]\mathrm{Div}(V) = \frac{\partial V_x}{\partial x} + \frac{\partial V_y}{\partial y}[/math]
This is the divergence in two dimensions. Remember that it's a function of [math]x[/math] and [math]y[/math].
A good test of whether you understand the divergence is to find its formula in polar coordinates when the vector

field is
[math]\mathbf{V}(r,\theta) = V_r(r,\theta) \hat{r} + V_\theta(r,\theta) \hat{\theta}[/math]
You can do this in exactly the same way - draw a span, find how fast its area changes, and divide by the area.
The square is a little awkward here; use a span that looks like this:
source: Math 251 diary, fall 2010
The answer is
[math]\mathrm{Div}(\mathbf{V}) = \frac{1}{r}\frac{\partial V_\theta}{\partial \theta} + \frac{\partial V_r}{\partial r} +
\frac{V_r}{r}[/math]

Continuity Equation
Suppose that the flow of water in the river is two dimensional, meaning no water moves upwards or downwards.
If there were positive divergence, that would mean the water was taking up more and more space. But that in
turn means that it would be thinning out, getting less dense. In turn, if the were negative divergence, it would
mean all the water in some area was getting smashed together to become more dense.
For water, this is essentially impossible. The pressure it takes to compress water appreciably is too high to reach
in a river. You similarly can't thin it out; the water would just collapse. This means that for flowing water the
divergence is zero!
More generally, if the water is diverging at 1% per second, that means the same mass is getting spread over 1%
more volume, so the density is going down by 1% per second. This relationship between the divergence and
density is called the continuity equation. If the density is [math]\rho[/math], then
[math]\frac{\mathrm{d}\rho}{\mathrm{d} t} = -\mathrm{Div}(\rho \mathbf{V})[/math]
This equation works because mass is conserved. You'll see a similar equation for any conservation law. For
example, starting from Maxwell's equations, you can derive a continuity equation like this for electric charge to
show that charge is conserved. Starting from Schrodinger's equation in quantum mechanics you can derive a
continuity equation to show that probability is conserved - the total probability to find a particle anywhere always
adds up to 100%.

Three Dimensions
In three dimensions, everything works the same way. You can think of eight dust particles forming a cube that
drifts through the air in your room. The fractional rate of change of the volume of the cube is the divergence in
three dimensions. It is
[math]\mathrm{Div}(\mathbf{V}) = \frac{\partial V_x}{\partial x} + \frac{\partial V_y}{\partial y} + \frac{\partial V_z}
{\partial z}[/math]

Eulerian vs Lagrangian Pictures

What I've described above might be called the "Lagrangian picture" of the divergence. We imagine following
along as you float down the river and talk about what happens to you.
For one reason or another, it is more popular to describe the divergence in the "Eulerian picture", in which we
imagine staying fixed in one spot and looking at properties of the flow right there. Ultimately, there is no
difference between the two, but you might hear people describe the divergence in a way that sounds

superficially different. It's actually the same. You can learn that picture if you want, or stick with this one.

Curl
We now imagine a device made of rods and floaters like this one floating down the river:
It is basically the same thing as before, except rigid. You and your three friends have been replaced by floaters
at the end of the rods, and there's a central hub.
If the device spins, the flow has curl. For example, in this flow, the top part of the device is in faster water and so
gets ahead. The bottom part lags behind, and all together this causes spin.
If the flow is perfectly uniform, the device won't spin and there's no curl.
In detail, the way we imagine this device working is that for the top/bottom floaters, what matters is the horizontal
component of the flow velocity. The vertical component of flow velocity can only push or pull against the vertical
rods, and so has no effect. For the left/right floaters, what matters is the vertical components of the flow rate
(zero in the above picture).
Let's say the rods have length [math]l[/math]. The point [math](x_0,y_0)[/math] is at the center of the device. The
velocity in the x-direction at the top floater is
[math]V_x \approx V_x(x_0,y_0) + \left.\frac{\partial V_x}{\partial y}\right|_{x_0,y_0}l[/math]
On the other hand, if the device is moving along as a whole at
[math]\mathbf{V}(x_0,y_0)[/math] and simultaneously rotating at angular rate [math]\omega[/math], the velocity of
the top point would be
[math]V_x \approx V_x(x_0,y_0) + \omega l[/math]
So it looks like the thing wants to rotate at a rate
[math]\omega = \frac{\partial V_x}{\partial y}[/math]
We get the same answer if we look at the bottom floater.
If we look at the floater on the right and do the same analysis, it wants the device to rotate at
[math]\omega = -\frac{\partial V_y}{\partial x}[/math]
(You should draw this out to picture it.)
The device can only have one rotation speed, which we will say is the average of the speed the top/bottom
floaters want to go and speed the left/right floaters want to go. The curl is then defined to be
[math]\mathrm{Curl}(\mathbf{V}) = \frac{\partial V_x}{\partial y} - \frac{\partial V_y}{\partial x}[/math]
It's twice the rotation rate. There's no special reason to make it twice as fast. It just works out better that way
when you think of some other possible ways to define the curl. Those other ways are equivalent; you can read
about them anywhere. This is just the picture I have.

Three Dimensions
What we found above is the z-component of the curl. That's because the thing is spinning around the z-axis.
(The z-axis comes up out of your monitor.)
Imagine putting a toy jack-shaped device with six rods and floaters in the water, one in each direction in 3D. It

would spin just the same, but the two extra floaters would contribute some extra torque. The device picks up
some spin around the x-axis and y-axis in addition to the familiar z-axis spin.
Adding up the spins around each axis (which you can do because angular velocities are vectors) you find that the
entire jack is spinning around some particular axis in 3D. This axis is the direction of the curl in three
dimensions. So the direction of the curl is just the direction of the axis around which the device spins. The
magnitude of the curl is still how fast it spins.

Further Reading
Div, Grad, Curl, and All That , Schey (An Informal Text on Vector Calculus (Fourth Edition): H. M. Schey:
9780393925166: Amazon.com: Books)
If you have some time to answer a few questions about the effectiveness of this answer, please check out
Dan and I Diverge by Mark Eichenlaub on Painting the Cathedral
I spent a few years trying and failing to understand the concepts of divergence and curl. After failing my physics
courses a few times, I eventually gave up. Later I went on to study differential geometry, differential forms, and
de Rham cohomology, at which point I understood divergence and curl -- not only in some formal sense, but in
the intuitive sense that people had been trying to explain informally all along.
I've talked to at least a dozen people who had exactly the same experience. [1] I'm solidly convinced that, at
least for myself and many others, it literally would have been impossible to arrive at this intuitive understanding
without passing through the formalities first. Some people can just read something like Nicholas Ma's answer
and have it make sense to them the first time, and it looks perfectly transparent to me now, but I remember trying
and failing to understand that kind of explanation too many times to believe that if you just slightly altered it
somehow it would work for everyone.
So, I would explain divergence and curl to someone by first teaching this person topology, linear and multilinear
algebra, and differential geometry (probably including a rehash of calculus since people usually don't learn it the
right way). I estimate this would probably take me about two years to do.
Probably not the answer you wanted, but it's the only answer I can honestly give.
BONUS:
Comments on My Investigation of Divergence and Curl , a nice article by mathematician Lee Lady on his
attempts to explain divergence and curl to his satisfaction
[1] See also
Jesse Farmer's answer to Should you study advanced calculus before real analysis?
I would delighted to write an answer but the best advice I can give would be to refer to the mathinsight page:
Multivariable calculus
Whoever created the site has gone to great lengths to explain the math and the meaning behind the math. There
are plenty of animations and applets to mess around with. Take a look for instance at the detail in explaining the
components of the curl: The components of the curl
There is not a lot in the way of practical applications, though I think a strong understanding of the mathematics
underlying the application will make it that much easier to understand the context.
I see Mark Eichenlaub's name, who is an excellent writer and recommend reading his answer.
Now go and learn some math!
So consider a fluid, and a field that represents the velocity of the fluid at each point. If we imagine a very small

(infinitesimal) cube, the divergence at the center of the cube represents how much fluid is entering or leaving the
cube per unit time. Similarly, the curl represents how much fluid is going around the cube. Neither of these ideas
is very precise mathematically, but it gives an intuitive explanation of their interpretation.
In particular, the divergence law is easily understood using this idea. If we divide a volume into many small cells,
and add up their divergences, the flows from one cell into another will cancel out, and so we're left with the flow
in and out of the volume. You can conceptualize Stokes' theorem in a similar way, with the internal rotation
canceling out, and only the rotation on the boundary contributing to the integral.
Both divergence and curl have numerous applications in vector calculus, particularly in regard to fluid mechanics
and electromagnetism. If we write the electric and magnetic fields as vector fields and the charge density as a
scalar field, then Maxwell's equations relate them using divergence and curl.
Feynman Lectures on Physics, volume II