Visual walkthrough — Surface integrals — scalar and vector (flux)

We will build up to this one line: ${\iint }_S\mathbf{F}\cdot d\mathbf{S}={\iint }_D\mathbf{F}(\mathbf{r}(u,v))\cdot ({\mathbf{r}}_u\times {\mathbf{r}}_v)dudv.$ Don't worry about any symbol in it yet — including $d\mathbf{S}$, which we build carefully in Step 7. We meet them one at a time.

---

Step 1 — A flat parameter rectangle is the only thing we understand

WHAT. Start on a flat plane whose two directions we call $u$ (horizontal) and $v$ (vertical). Pick a region $D$ inside it — think of it as a sheet of graph paper. This is our home base.

WHY. A curved surface floating in 3D is hopeless to integrate directly — there is no clean "left-to-right, bottom-to-top" order on a curved sheet. But a flat rectangle we already know how to sweep across: that is exactly a double integral. So the whole plan is: do all the hard work on the flat sheet, then correct for how the surface distorts it.

PICTURE. The green square below is one tiny tile of the flat sheet. Its width is a tiny amount of $u$ we write $du$, and its height a tiny amount of $v$ we write $dv$. "$d$" just means a very small piece of. Its area is honestly $dudv$ — a plain rectangle.

---

Step 2 — The map $\mathbf{r}$ lifts the flat sheet onto the curved surface

WHAT. A parametrization is a rule $\mathbf{r}(u,v)$ that takes each flat point $(u,v)$ and reports where it lands in 3D space: $\mathbf{r}(u,v)=(\underset{\text{left–right}}{\underbrace{x(u,v)}},\underset{\text{forward–back}}{\underbrace{y(u,v)}},\underset{\text{up–down}}{\underbrace{z(u,v)}}).$ Each of $x,y,z$ is a plain number that depends on the two dials $u$ and $v$.

WHY. This is the bridge. Turning the two dials $u,v$ visits every point of the surface $S$. The bold $\mathbf{r}$ ("position vector") is just an arrow from the origin out to that landed point — three coordinates bundled together.

PICTURE. The flat green tile on the left gets carried by $\mathbf{r}$ onto a warped patch on the curved surface on the right. It is no longer a flat rectangle — it has been bent and stretched.

---

Step 3 — Two tangent arrows show which way the tile got stretched

WHAT. Sit at a point on the surface. Nudge only $u$ (keep $v$ frozen): the landing point slides along a curve. Its velocity — the direction and speed it moves per unit of $u$ — is the partial derivative ${\mathbf{r}}_u=\frac{\partial \mathbf{r}}{\partial u}=(\frac{\partial x}{\partial u},\frac{\partial y}{\partial u},\frac{\partial z}{\partial u}).$ Likewise nudging only $v$ gives ${\mathbf{r}}_v=\partial \mathbf{r}/\partial v$.

WHY a derivative here? We need to know how the map distorts an infinitesimal tile. "Rate of change of position as I turn one dial" is exactly the job description of a derivative. The word partial (the curly $\partial$) just means "differentiate with respect to one dial while freezing the other." We pick the derivative — not, say, an average — because distortion is a local, instantaneous thing.

PICTURE. ${\mathbf{r}}_u$ (magenta) points along the $u$-curve; ${\mathbf{r}}_v$ (violet) points along the $v$-curve. They are the two edges the little tile got mapped to — both are tangent to the surface (they graze it, never leave it).

---

Step 4 — The tile becomes a tiny parallelogram, and we need its area

WHAT. The mapped tile is (to first order) the parallelogram with edge vectors ${\mathbf{r}}_{u}du$ and ${\mathbf{r}}_{v}dv$. Here ${\mathbf{r}}_{u}du$ means "the tangent arrow ${\mathbf{r}}_u$ shrunk to the length that a step $du$ actually produces."

WHY. Over a tiny enough tile, the curved patch is indistinguishable from the flat parallelogram spanned by its two edge arrows. Curvature only matters at second order — smaller than tiny — so we ignore it. Our whole question — "how much real area does one flat tile become?" — is now "what is the area of this parallelogram?"

PICTURE. Flat tile of area $dudv$ on the left; its image, a slanted parallelogram, on the right. The two are the same number of tiles but different areas — that gap is the whole story.

---

Step 5 — The cross product measures that parallelogram's area

WHAT. The area of a parallelogram with edge vectors $\mathbf{a}$ and $\mathbf{b}$ is the length of their cross product: $\text{area}=\Vert \mathbf{a}\times \mathbf{b}\Vert =\Vert \mathbf{a}\Vert \Vert \mathbf{b}\Vert \sin \theta .$ Here $\theta$ is the angle between the edges, and $\Vert \cdot \Vert$ means "length of."

WHY the cross product, of all tools? We need two things at once: the parallelogram's area and, later, the direction that points straight out of it. The Cross product delivers both in one object — its length is the area, its direction is perpendicular to both edges. No other product does both. The $\sin \theta$ is exactly right: it is (one edge) $\times$ (the height the other edge reaches when tilted by $\theta$) = base $\times$ height.

Feed in our edges $\mathbf{a}={\mathbf{r}}_{u}du$, $\mathbf{b}={\mathbf{r}}_{v}dv$. Because $du,dv$ are plain positive numbers they slide out: $\Vert {\mathbf{r}}_{u}du\times {\mathbf{r}}_{v}dv\Vert =\underset{\text{stretch factor}}{\underbrace{\Vert {\mathbf{r}}_u\times {\mathbf{r}}_v\Vert }}\underset{\text{flat tile area}}{\underbrace{dudv}}.$

PICTURE. The cross product ${\mathbf{r}}_u\times {\mathbf{r}}_v$ (orange) stands perpendicular to the patch; the shaded parallelogram's area equals its length. When the two edges are nearly parallel ($\theta \to 0$), $\sin \theta \to 0$ and the area collapses — see the thin sliver on the right.

---

Step 6 — Turn the arrow into a unit normal and ask "how much flow goes through?"

WHAT. Divide the orange arrow by its own length to get a pure-direction arrow of length 1, the unit normal: $\hat{\mathbf{n}}=\frac{{\mathbf{r}}_u\times {\mathbf{r}}_v}{\Vert {\mathbf{r}}_u\times {\mathbf{r}}_v\Vert }.$ The little hat ($\hat{}$) always means "length one."

WHY. Now bring in a vector field $\mathbf{F}$ — at every point of space it is an arrow, say the velocity of a fluid. Through our tiny patch, only the part of $\mathbf{F}$ that points along $\hat{\mathbf{n}}$ actually crosses the surface. Sideways flow just slides along the patch and passes through nothing. To extract "the part along $\hat{\mathbf{n}}$" we use the dot product $\mathbf{F}\cdot \hat{\mathbf{n}}$ — the one tool that projects one arrow onto another's direction (it returns $\Vert \mathbf{F}\Vert \cos \alpha$, where $\alpha$ is the angle between $\mathbf{F}$ and $\hat{\mathbf{n}}$).

PICTURE. $\mathbf{F}$ (magenta) hits the patch at an angle. Its shadow onto $\hat{\mathbf{n}}$ (orange, length 1) is the through-part; the leftover violet component slides sideways and carries nothing through.

---

Step 7 — The square root cancels: the punchline

WHAT. First name the object we are building. The vector area element $d\mathbf{S}$ bundles the patch's area and its outward direction into a single arrow:

Now multiply the two pieces from Step 6. Write $[\cdot ]$ for the vector dot product (vector-dot-vector) and keep ordinary scalar multiplication written by juxtaposition: $(\mathbf{F}\cdot \hat{\mathbf{n}})dS=[\mathbf{F}\cdot \frac{{\mathbf{r}}_u\times {\mathbf{r}}_v}{\Vert {\mathbf{r}}_u\times {\mathbf{r}}_v\Vert }](\Vert {\mathbf{r}}_u\times {\mathbf{r}}_v\Vert dudv).$ In this line the $\cdot$ inside the brackets is a dot product of two arrows (giving a number); the bracket is then multiplied (plain number times number) by the factor $\Vert {\mathbf{r}}_u\times {\mathbf{r}}_v\Vert dudv$. That scalar factor sits in the bottom of $\hat{\mathbf{n}}$ and in the top of $dS$, so it cancels exactly: $(\mathbf{F}\cdot \hat{\mathbf{n}})dS=\mathbf{F}\cdot ({\mathbf{r}}_u\times {\mathbf{r}}_v)dudv.$

WHY it is beautiful. For a scalar surface integral you are stuck computing the messy root $\Vert {\mathbf{r}}_u\times {\mathbf{r}}_v\Vert =\sqrt{\cdots }$. For flux it evaporates — you never take a square root. You just dot $\mathbf{F}$ with the raw cross product. So the vector area element simplifies to $d\mathbf{S}=\hat{\mathbf{n}}dS=({\mathbf{r}}_u\times {\mathbf{r}}_v)dudv,$ carrying area and direction with no root in sight.

PICTURE. Left: $\hat{\mathbf{n}}$ (length 1) times a fat patch. Right: after cancellation, one honest arrow ${\mathbf{r}}_u\times {\mathbf{r}}_v$ whose length is the area and whose direction is the orientation — no denominator survives.

---

Step 8 — Sanity check on the unit sphere (all pieces at once)

Setup — the spherical parameters $\varphi ,\theta$. For a sphere the two dials $u,v$ are renamed to the two natural spherical angles:

• $\varphi$ (polar angle, "how far down from the north pole"), ranging $\varphi \in [0,\pi ]$: $\varphi =0$ is the north pole, $\varphi =\pi$ the south pole.
• $\theta$ (azimuth, "how far around the equator"), ranging $\theta \in [0,2\pi ]$.

So here $(u,v)=(\varphi ,\theta )$, and the tangent arrows of Steps 3–5 become ${\mathbf{r}}_{\varphi }=\partial \mathbf{r}/\partial \varphi$ (points southward along a meridian) and ${\mathbf{r}}_{\theta }=\partial \mathbf{r}/\partial \theta$ (points eastward along a circle of latitude). Everything else is exactly the machine we just built.

WHAT. Take $\mathbf{F}=(x,y,z)$ and the outward unit sphere. Parametrize $\mathbf{r}(\varphi ,\theta )=(\sin \varphi \cos \theta ,\sin \varphi \sin \theta ,\cos \varphi )$. One computes ${\mathbf{r}}_{\varphi }\times {\mathbf{r}}_{\theta }=\sin \varphi \mathbf{r}$. Then, since $\mathbf{F}=\mathbf{r}$ on the sphere and $\Vert \mathbf{r}\Vert =1$, $\mathbf{F}\cdot ({\mathbf{r}}_{\varphi }\times {\mathbf{r}}_{\theta })=\mathbf{r}\cdot (\sin \varphi \mathbf{r})=\sin \varphi \Vert \mathbf{r}{\Vert }^2=\sin \varphi .$ ${\iint }_S\mathbf{F}\cdot d\mathbf{S}={\int }_0^{2\pi }{\int }_0^{\pi }\sin \varphi d\varphi d\theta =2\pi \cdot 2=4\pi .$

WHY this example. Here $\mathbf{F}$ points radially outward, straight through the surface everywhere, so every patch contributes fully — flux should be maximal and positive. It is $4\pi$, exactly the sphere's surface area, which makes sense: a unit-speed outward flow through unit-area = area. Cross-check with the Divergence Theorem: $\nabla \cdot \mathbf{F}=3$ times the ball volume $\frac{4}{3}\pi$ gives $4\pi$. ✔ (The order $\varphi$ then $\theta$ makes ${\mathbf{r}}_{\varphi }\times {\mathbf{r}}_{\theta }$ point outward — the right-hand rule confirms the orientation the problem wants.)

PICTURE. Radial arrows $\mathbf{F}$ piercing the sphere dead-on, each normal $\hat{\mathbf{n}}$ aligned with the flow — nothing slides sideways.

---

The one-picture summary

One glance: a flat tile $dudv$ → mapped by $\mathbf{r}$ → two tangent edges ${\mathbf{r}}_u,{\mathbf{r}}_v$ → cross product gives area and direction → project the field with a dot → the root cancels → integrate over flat $D$.

Recall Feynman retelling — explain the whole walkthrough to a friend

Imagine holding a bumpy fishing net under a waterfall and asking "how much water gets through per second?"

The net is curved and awful to reason about, so I lay a sheet of flat graph paper next to it and pretend each little square on the paper corresponds to one bumpy patch on the net (the map $\mathbf{r}$). But a flat square lands as a stretched, tilted patch, so I measure its two edges — the arrows showing how far the patch moved when I slid one graph-paper direction at a time (the tangent vectors ${\mathbf{r}}_u,{\mathbf{r}}_v$). To get that tilted patch's real area and the direction it faces in one shot, I cross the two edges (cross product): its length is the area, its arrow points straight out of the net. Now the waterfall: only the water aimed along that outward arrow gets through — sideways splash slides off — so I shadow the water's velocity onto the outward direction (dot product). When I write it all out, the "shrink to length one" division and the "multiply by the area" both use the same length, and they cancel — so flux is just velocity dotted with the raw cross product, summed over every square of graph paper. No square roots, and if I flipped the net over the answer flips sign.

Recall Quick checks

Why a derivative for the tangents? ::: It measures the instantaneous stretch of the map as one parameter changes. Why the cross product and not something else? ::: It gives area (its length) and outward direction (its arrow) simultaneously. Why does the root cancel for flux but not for scalar integrals? ::: $\hat{\mathbf{n}}$ divides by the length; $dS$ multiplies by it — they meet only in flux. What happens at a singular point where ${\mathbf{r}}_u\parallel {\mathbf{r}}_v$? ::: The cross product is zero; the tile collapses and contributes no area. How do you pick ${\mathbf{r}}_u\times {\mathbf{r}}_v$ vs ${\mathbf{r}}_v\times {\mathbf{r}}_u$? ::: Right-hand rule: fingers along increasing $u$ curl to increasing $v$, thumb is the normal; choose the parameter order so the thumb points the way the problem asks.