Visual walkthrough — Divergence — definition, physical meaning (flux density)

Before we begin, let us agree on the two words everything rests on.

Step 1 — Drop a tiny box on the point

WHAT. Pick the point we care about, call its coordinates $(x,y,z)$. Around it, draw a tiny rectangular box with side-lengths $\Delta x$ (width), $\Delta y$ (depth), $\Delta z$ (height). The little symbol $\Delta$ (Greek "delta") just means "a small step of."

WHY. Divergence is flux per unit volume. To get flux we need a closed surface to count arrows crossing; the simplest closed surface is a box with 6 flat faces. A box is easy because each face points along exactly one axis, so on each face only one component of $\vec{F}$ matters.

PICTURE. The box has volume $V=\underset{\text{width}}{\underbrace{\Delta x}}\cdot \underset{\text{depth}}{\underbrace{\Delta y}}\cdot \underset{\text{height}}{\underbrace{\Delta z}}.$ We will count arrows leaving through all 6 walls, add them up, then divide by $V$.

Step 2 — The outward normal: which way is "out"?

WHAT. On each face we draw a short arrow $\hat{n}$ (read "n-hat") that points straight out of the box, perpendicular to that face, with length exactly $1$. The hat means "length one" — it only carries a direction, no size.

WHY. Flux counts fluid leaving the box. "Leaving" means crossing a wall in the outward direction. So on every face we must know which way is out — that is exactly what $\hat{n}$ records. Fluid moving with $\hat{n}$ counts as leaving (positive); fluid moving against $\hat{n}$ is entering (negative).

PICTURE. For the two faces perpendicular to the $x$-axis:

• Right face (at $x+\Delta x$): $\hat{n}=+\hat{x}$ (points right, out of the box).
• Left face (at $x$): $\hat{n}=-\hat{x}$ (points left, still out of the box).

The tool that measures "how much of $\vec{F}$ points along $\hat{n}$" is the dot product $\vec{F}\cdot \hat{n}$. Why the dot product and not something else? Because the dot product of a vector with a unit direction returns exactly the shadow (projection) of that vector onto the direction — precisely the "part poking through the wall" we defined flux from.

Step 3 — Flux out of the two $x$-faces

WHAT. Only $F_1$ (the rightward part of the arrow) crosses the $x$-faces, because $\hat{n}$ on those faces is $\pm \hat{x}$ and $\vec{F}\cdot (\pm \hat{x})=\pm F_1$. Approximate each face's flux as (field value there) $\times$ (face area). The $x$-faces have area $\Delta y\Delta z$.

• $F_1(x+\Delta x,y,z)$ — the rightward push measured on the right wall.
• $F_1(x,y,z)$ — the rightward push measured on the left wall.
• $\Delta y\Delta z$ — the area of each $x$-face.
• The minus appears because on the left wall $\hat{n}=-\hat{x}$, so $\vec{F}\cdot \hat{n}=-F_1$: rightward fluid there is entering, which subtracts.

WHY. We want net "out minus in" along $x$. Fluid leaves the right and (if $F_1>0$) enters the left. Subtracting gives the true leftover.

PICTURE. Two parallel walls, same area, arrows piercing each. Whatever pierces the right leaves; whatever pierces the left enters and is subtracted.

Step 4 — A difference over a step is a derivative waiting to happen

WHAT. Factor out the common area $\Delta y\Delta z$:

The bracket is the change in $F_1$ as you step a distance $\Delta x$ to the right.

WHY this tool — the derivative. A raw difference like $F_1(x+\Delta x)-F_1(x)$ depends on how big the step $\Delta x$ is; that is awkward. The derivative answers a sharper question: "at what rate does $F_1$ change per unit of $x$, right here?" We turn the difference into a rate by dividing and re-multiplying by $\Delta x$:

PICTURE. Plot $F_1$ against $x$: the bracket is the rise, $\Delta x$ is the run, and rise/run is the slope of the line joining the two wall-values. Shrinking the box makes that line kiss the curve — the slope becomes the tangent slope.

Step 5 — Shrink the box: the slope becomes $\partial F_1/\partial x$

WHAT. Let the box shrink, $\Delta x\to 0$. The slope of that connecting line becomes the exact tangent slope — the partial derivative:

• The curly $\partial$ (read "partial-dee") means: hold $y$ and $z$ frozen, wiggle only $x$. We freeze the others because our two $x$-walls sit at the same $y,z$ — only $x$ differs between them.

So the $x$-contribution to flux is

WHY. In the limit "$V\to 0$" the block picture becomes an exact statement about the point itself, not an approximation for a fat box.

PICTURE. Same two-point plot as Step 4, but the two dots slide together; the secant line rotates onto the tangent, whose steepness is $\partial F_1/\partial x$.

Step 6 — The other four faces, by identical reasoning

WHAT. The $y$-faces (top and bottom) have area $\Delta x\Delta z$ and only feel $F_2$; the $z$-faces (front and back) have area $\Delta x\Delta y$ and only feel $F_3$. Copy Steps 3–5 word for word with the labels rotated:

• Notice the matching indices: $F_1$ pairs with $x$, $F_2$ with $y$, $F_3$ with $z$. Each component is compared across its own pair of walls, which lie along its own axis. That is why cross-terms like $\partial F_1/\partial y$ never appear — the left/right walls differ only in $x$, so only $x$-changes of $F_1$ matter.

WHY. Space treats all three axes the same; there is no privileged direction, so the argument for $x$ transplants unchanged to $y$ and $z$.

PICTURE. The same box, now with the top/bottom pair and front/back pair highlighted, each with its own arrows and its own matching partial derivative.

Step 7 — Add the three, divide by $V$

WHAT. Total outward flux is the sum over all six faces:

Divide by $V$ and take $V\to 0$ (the definition of divergence):

• Each term = net outflow along one axis, per unit volume.
• The sum = total spreading rate at the point.
• The $V$ cancels cleanly because every face-flux carried one factor of $V$.

WHY. "Flux per unit volume" literally means "divide the leftover fluid by the box's size." Doing so erased all mention of the box — the answer belongs to the point alone.

Step 8 — The degenerate case: symmetric flow, zero net divergence

WHAT. Test $\vec{F}=(x^2,0,0)$ at the origin. Its true divergence is $\partial (x^2)/\partial x=2x$, which is $0$ at $x=0$. Let us watch the box confirm it.

For a box $[-h,h{]}^3$ centred at the origin:

• Right wall at $x=+h$: $F_1=(+h{)}^2=h^2$, $\hat{n}=+\hat{x}$, area $(2h{)}^2$ → flux $+h^2(2h{)}^2$.
• Left wall at $x=-h$: $F_1=(-h{)}^2=h^2$, $\hat{n}=-\hat{x}$, area $(2h{)}^2$ → flux $-h^2(2h{)}^2$.

They cancel exactly: net flux $=0$, so divergence $=0$. ✓

WHY show this. It rescues you from the trap "big arrows $\Rightarrow$ big divergence." Here the arrows on both walls are equally strong ($h^2$ each) yet equally leaving both sides — the leftover is nothing. Divergence measures imbalance across the walls, not the size of the arrows. (See the parent mistake list.)

PICTURE. Left and right walls with equal-length arrows both pointing outward — mirror images. Equal out, equal out, but at $x=0$ they balance to no net source.

The one-picture summary

Everything compressed: box → six face-fluxes → three matching differences → three partial derivatives → their sum. Trace the coloured path from "count arrows on the walls" to "add three slopes."

Quick self-check

Connections

• Divergence — definition, physical meaning (flux density) — the parent this page derives.
• Flux through a surface — the surface integral we approximated face-by-face.
• Divergence Theorem (Gauss) — glues infinitely many of these tiny boxes into a global flux law.
• Gradient and the del operator — where the $\nabla$ and $\partial$ notation is born.
• Curl — rotation density — the sibling built from the cross-terms we discarded.
• Continuity equation — where this "fluid born per volume" idea becomes physics.
• Laplacian — divergence of a gradient.