Divergence — definition, physical meaning (flux density): Question bank

Before the traps, a short visual refresher so every symbol below is earned.

Recall The two faces of divergence (tap to reveal)

Geometric: $div F (P) = V \to 0 lim \frac{1}{V} \iint_{\partial V} F \cdot \overset{n}{^} d S$ — outward flux per tiny volume. Here $V$ is a small blob around the point $P$ , $\partial V$ is its skin, and $\overset{n}{^}$ (a unit-length arrow, "the normal") points straight out of the skin. Computational: $div F = \nabla \cdot F = \frac{\partial F _{1}}{\partial x} + \frac{\partial F _{2}}{\partial y} + \frac{\partial F _{3}}{\partial z}$ — the sum of the diagonal partial derivatives, a single number.

The picture behind the definition

Look at the tiny cube below. It sits at $(x, y, z)$ with sides $Δ x, Δ y, Δ z$ , so its volume is $V = Δ x Δ y Δ z$ . Each of its six faces has an outward normal $\overset{n}{^}$ (the red arrows) poking straight out. Flux through a face is $F \cdot \overset{n}{^}$ times the face's area — only the part of the flow along the normal escapes.

Figure — Divergence — definition, physical meaning (flux density)

Deriving the formula from the cube (the "why", step by step)

Focus on just the two faces perpendicular to the $x$ -axis and look at the figure below. The right face (at $x + Δ x$ ) has normal $+ \overset{x}{^}$ ; the left face (at $x$ ) has normal $- \overset{x}{^}$ .

Step 1 — what escapes each face. On the right face $F \cdot \overset{n}{^} = + F_{1} (x + Δ x, y, z)$ ; on the left face $F \cdot \overset{n}{^} = - F_{1} (x, y, z)$ (that minus is the whole trap — the outward normal points the opposite way). Multiply each by the face area $Δ y Δ z$ and add: $Flux_{x} \approx [F_{1} (x + Δ x, y, z) - F_{1} (x, y, z)] Δ y Δ z .$

Step 2 — why this is a derivative in disguise. The bracket is a difference of $F_{1}$ over a step $Δ x$ . Multiply and divide by $Δ x$ : $Flux_{x} \approx \frac{F _{1} ( x + Δ x , y , z ) - F _{1} ( x , y , z )}{Δ x} = V Δ x Δ y Δ z Δ x \to 0 \frac{\partial F _{1}}{\partial x} V .$

Step 3 — do the same for the other two axis-pairs. By identical reasoning the $y$ -faces give $\frac{\partial F _{2}}{\partial y} V$ and the $z$ -faces give $\frac{\partial F _{3}}{\partial z} V$ .

Step 4 — sum all six faces, divide by $V$ . Total flux $\approx (\frac{\partial F _{1}}{\partial x} + \frac{\partial F _{2}}{\partial y} + \frac{\partial F _{3}}{\partial z}) V$ , so $\frac{flux}{V} \to \frac{\partial F _{1}}{\partial x} + \frac{\partial F _{2}}{\partial y} + \frac{\partial F _{3}}{\partial z} = div F .$ The geometric definition forced the diagonal-partials formula into being — and because of the shape-independence above, that formula is the divergence no matter what blob you imagined.

True or false — justify

Every answer must give a reason, not just T/F.

F = (5, 0, 0)

(a uniform field, all arrows equal) has zero divergence.

True. The same amount of fluid enters the left face as leaves the right face; nothing is created.

\partial (5) / \partial x = 0

, so div

= 0

— big arrows, no spreading.

F

has large magnitude everywhere, its divergence must be large.

False. Divergence measures change in the flow across a point, not the flow's size. The swirl

(- y, x, 0)

has huge arrows far out yet divergence exactly

0

Divergence is a vector because both

\nabla

and

F

are vectors.

False. The operation is the dot product

\nabla \cdot F

, which collapses two vectors into one number. Vector-in, scalar-out. (Curl uses the cross product and does return a vector.)

A field can have positive divergence at one point and negative at another.

True. Divergence is a point-by-point scalar field. For

F = (x^{2}, 0, 0)

div = 2 x

is negative for

x < 0

(a sink there) and positive for

x > 0

(a source there).

The term

\partial F_{2} / \partial x

contributes to divergence.

False. Only matching-index (diagonal) partials appear:

\partial F_{1} / \partial x

\partial F_{2} / \partial y

\partial F_{3} / \partial z

. Cross-terms like

\partial F_{2} / \partial x

describe rotation and live in curl, not divergence.

If a fluid is incompressible (its density is carried along unchanged), then

div v = 0

everywhere.

True — for incompressible flow in general, steady or unsteady. Incompressibility means each fluid parcel keeps its volume, i.e. "in equals out" at every point at every instant: exactly

div v = 0

. This is the divergence-free condition behind the Continuity equation.

Divergence and curl are just two names for the same measurement of a field.

False. Divergence (dot product) measures net outflow per volume; curl (cross product) measures rotation per area. The swirl

(- y, x, 0)

has zero divergence but nonzero Curl — rotation density.

The units of

div F

differ from the units of

F

True. Divergence is flux-per-volume: you divide the field's units by a length (from the

\partial / \partial x

). If

F

is a velocity (m/s), div is per-second (1/s).

A field pointing radially inward everywhere always has negative divergence.

Mostly true, but check the field. Inward-pointing

(- x, - y, - z)

gives

- 3 < 0

, a sink. But the inverse-square inward field can be zero away from the origin — "pointing inward" alone doesn't fix the sign; the rate of change does.

div F = 0

means no fluid is moving.

False. It means no fluid is created or destroyed at that point — flow can be vigorous. Zero divergence is "as much leaves as enters," not "nothing happens."

Spot the error

Each line contains a flawed statement or step. Reveal the correction.

div (x, y, z)

is a vector

(1, 1, 1)

Error: stopping before adding. Divergence sums the diagonal partials into one number:

1 + 1 + 1 = 3

, a scalar. You listed the terms but never took the dot product.

div F = \frac{\partial F _{1}}{\partial x} + \frac{\partial F _{1}}{\partial y} + \frac{\partial F _{1}}{\partial z}

Error: all three terms use

F_{1}

. It must be

\partial F_{1} / \partial x + \partial F_{2} / \partial y + \partial F_{3} / \partial z

— component

i

differentiated by variable

i

, matching the face each flux crosses.

"For the swirl

(- y, x, 0)

\partial (- y) / \partial x = - 1

, so divergence is negative."

Error:

- y

contains no

x

, so

\partial (- y) / \partial x = 0

, not

- 1

. Likewise

\partial (x) / \partial y = 0

. Divergence is

0

; the

- 1

and

+ 1

that do appear belong to the curl.

"On the left face the flux is

+ F_{1} Δ y Δ z

Error: the outward normal on the left face points in

- \overset{x}{^}

, so

F \cdot \overset{n}{^} = - F_{1}

. The flux there is

- F_{1} Δ y Δ z

; forgetting this sign destroys the whole derivation (it is the Step 1 minus sign above).

"Divergence Theorem says

\iint_{\partial V} F \cdot \overset{n}{^} d S = div F

Error: the right side must be the integral of divergence over the volume:

∭_{V} div F d V

. See Divergence Theorem (Gauss) — a global flux equals summed local flux densities, not a single value.

"Since flux is a surface integral, divergence must also be measured over a surface."

Error: divergence is flux per unit volume in the limit of zero volume — a point quantity. The surface integral appears only as the numerator before you divide by

V

and shrink.

"The Laplacian

\nabla^{2} f

is a vector because it comes from divergence."

Error:

\nabla^{2} f = div (grad f)

takes the divergence of a vector field, so the output is a scalar. See Laplacian.

Why questions

Why does divergence use the dot product rather than plain multiplication?

Because flux through a face is

F \cdot \overset{n}{^}

— only the component of the field along the outward normal pushes fluid through. The dot product extracts exactly that projection; other directions slide along the face and carry nothing out.

Why does only the derivative along an axis (not across it) count for that axis's faces?

The

x

-faces are crossed by the

x

-component of flow; net outflow there is "how much faster the

x

-flow is on the right than the left" — that is

\partial F_{1} / \partial x

. A change in

F_{1}

as you move in

y

shifts flow sideways along the face, adding nothing to what exits.

Why is divergence a scalar while curl is a vector?

Net spreading has no direction — it's a single balance of "out minus in," so one number suffices. Rotation happens about an axis, so you need a direction (the axis) plus a rate — hence a vector for Curl — rotation density.

Why do we shrink the volume to zero in the definition?

To get a local rate at a single point

P

instead of a total over a chunk. Dividing by

V

turns total flux into flux density; taking

V \to 0

removes any averaging so the number belongs to the point, not the blob.

Why may we use a cube to derive the formula when the definition allows any shape?

Because the limit is shape-independent for smooth fields: any small region gives (the same local rate)

\times V

plus corrections that die faster than

V

. So the cube's answer equals a ball's answer — we just pick the shape with the easiest faces.

Why does

\nabla

come from the gradient yet also build divergence?

\nabla = (\partial / \partial x, \partial / \partial y, \partial / \partial z)

is a "vector of derivative slots." Applied to a scalar it fills the slots (gradient); dotted with a vector field it pairs each slot with the matching component (divergence). Same operator, different multiplication — see Gradient and the del operator.

Why can two fields with identical arrows at a point still have different divergences?

Divergence depends on the neighbourhood, not the single arrow. It reads how the field is changing around the point, so two fields that agree at

P

but differ nearby will disagree in divergence.

Why does a positive

\partial F_{1} / \partial x

mean net outflow in the

x

-direction?

It means the outgoing (right-face) flow

F_{1} (x + Δ x)

exceeds the incoming (left-face) flow

F_{1} (x)

. More leaves than enters along

x

, so fluid is being sourced in that direction.

Edge cases

What is the divergence of the zero field

F = (0, 0, 0)

Exactly

0

. All partials are

0

; nothing flows, so nothing is created or destroyed — the trivial "no source" case.

Can divergence be defined at a point where the field blows up, like the origin of

F = (x, y, z) / r^{3}

Not by the ordinary formula — the field is undefined at the origin. Away from the origin its divergence is exactly

0

, yet all the source is concentrated at the origin: in the distributional sense

\nabla \cdot (r / r^{3}) = 4 π δ (r)

(a Dirac delta spike), which the smooth pointwise formula cannot see. This is the mathematics behind a point charge's field.

If a box has equal outward flow through the right and left faces, what is the

x

-contribution to divergence?

Zero. Example:

F = (x^{2}, 0, 0)

x = 0

— both faces leak equally, so

\partial F_{1} / \partial x = 2 x = 0

there. Symmetric outflow gives no net source.

What does zero divergence at every point tell you about the total flux out of any closed surface?

It is zero, by Divergence Theorem (Gauss):

\iint F \cdot \overset{n}{^} d S = ∭ 0 d V = 0

. Whatever enters a closed region must leave it.

At a boundary between a source region (div

> 0

) and a sink region (div

< 0

), what happens to divergence?

It passes smoothly through

0

(for a continuous field): there is a surface where creation exactly balances destruction. E.g.

div (x^{2}, 0, 0) = 2 x

crosses zero at

x = 0

Is divergence defined for a 2D field, and does the box argument still work?

Yes. In 2D use a tiny rectangle:

div F = \partial F_{1} / \partial x + \partial F_{2} / \partial y

, and "flux per volume" becomes "flux per area" through the rectangle's four edges — the same derivation with one fewer axis.

Does the shape of the shrinking volume

V

change the value of the divergence?

No. For a smooth field the limit is shape-independent — a cube, ball, or ellipsoid all give the same number — which is exactly why the definition can leave "

V \to 0

" unspecified in shape.

Connections

Divergence — definition, physical meaning (flux density) (the parent this bank drills)
Curl — rotation density (the cross-product sibling that catches the vector-vs-scalar traps)
Divergence Theorem (Gauss) (behind every "total flux" edge case)
Gradient and the del operator (why $\nabla$ builds both grad and div)
Flux through a surface (the integral being averaged in the definition)
Continuity equation (incompressibility = zero divergence)
Laplacian (divergence of a gradient — a scalar)