Curl — definition, physical meaning (rotation): Visual walkthrough

Before we begin, three plain-word promises about the words we will use:

Everything below is a machine for turning that last idea into a formula.

Step 1 — Drop a paddlewheel and ask the right question

WHAT. We place a tiny toy waterwheel in the flow and ask: does it spin, and how fast? Instead of watching the wheel physically, we trace a tiny loop — a small rectangle — and measure how much the water pushes us around that loop.

WHY. Spinning and "being pushed around a loop" are the same thing. If the water carries you all the way around a little square with net help, the same water would spin a wheel sitting there. Turning "spin" into "circulation around a loop" lets us use arithmetic instead of vibes.

PICTURE. The wheel sits at a point; the dashed square is the loop we will walk. We choose to walk it counterclockwise, which by convention is the "positive" direction.

Figure — Curl — definition, physical meaning (rotation)

Step 2 — Name the corners and the four edges

WHAT. Put the bottom-left corner of the square at $(x, y)$ . The bottom side has width $Δ x$ (a tiny step in the $x$ -direction) and the right side has height $Δ y$ (a tiny step in the $y$ -direction). Label the four edges: bottom, right, top, left, walked in that counterclockwise order.

WHY. To add up the push around the loop we must know exactly where we are and which way we are walking on each edge — because the push depends on both. The symbols $Δ x$ and $Δ y$ (read "a little bit of $x$ ", "a little bit of $y$ ") are small positive lengths; we will shrink them to zero at the very end.

PICTURE. Four colored arrows show the walking direction on each edge. Notice bottom goes right, top goes left (we are coming back), right goes up, left goes down.

Step 3 — On each edge, only ONE part of the field pushes you

WHAT. On a horizontal edge you move purely sideways, so only the rightward part $F_{1}$ helps or fights you. On a vertical edge you move purely up/down, so only the upward part $F_{2}$ matters. Multiplying "push in your direction" by "length walked" gives the contribution of that edge.

WHY. Circulation adds up (push-along-you) $\times$ (distance). If you walk right and the field pushes up, that up-push doesn't move you along your path — it does nothing to your circulation. That is why each edge keeps only one component. Formally the little push is the dot product $F \cdot d r$ : it multiplies matching directions and ignores perpendicular ones.

PICTURE. Each edge is annotated with the single term it contributes:

The four contributions, with a walk-through of every symbol:

Step 4 — Add all four and factor into two honest differences

WHAT. Sum the four terms. Group the two $F_{2}$ terms together and the two $F_{1}$ terms together.

WHY. Because a paddlewheel doesn't care about the total push — it cares about the difference in push from side to side. Grouping reveals exactly those differences: right edge vs left edge, top edge vs bottom edge.

$\oint F \cdot d r \approx how much the up-push grew as we moved right [F_{2} (x + Δ x, y) - F_{2} (x, y)] Δ y - how much the right-push grew as we moved up [F_{1} (x, y + Δ y) - F_{1} (x, y)] Δ x$

PICTURE. The two brackets are drawn as two little "difference bars": the right edge's up-push minus the left edge's up-push (green), and the top edge's right-push minus the bottom's (red).

Step 5 — Turn each difference into a derivative

WHAT. A derivative $\partial_{x} F_{2}$ (read "how fast $F_{2}$ changes as $x$ increases") is defined as exactly this: the change in the value divided by the tiny step that caused it. So $F_{2} (x + Δ x, y) - F_{2} (x, y) \approx \partial_{x} F_{2} Δ x, F_{1} (x, y + Δ y) - F_{1} (x, y) \approx \partial_{y} F_{1} Δ y .$

WHY this tool and not another? We need a way to say "how much did the up-push grow per unit of sideways step." That per-unit-of-step rate is precisely what a partial derivative is built to answer — nothing else measures a rate of change of one variable while the other is frozen. Substituting: $\oint F \cdot d r \approx (\partial_{x} F_{2}) Δ x Δ y - (\partial_{y} F_{1}) Δ y Δ x = (\partial_{x} F_{2} - \partial_{y} F_{1}) Δ x Δ y .$

$\partial_{x} F_{2}$ = rate the upward push strengthens as you slide right → tries to spin the wheel counterclockwise.
$\partial_{y} F_{1}$ = rate the rightward push strengthens as you rise → tries to spin the wheel clockwise, hence the minus.
$Δ x Δ y$ = the area of the little square.

PICTURE. A shrinking sequence of squares showing the difference bars becoming exact slopes as $Δ x, Δ y \to 0$ .

Step 6 — Divide by the area: swirl density is born

WHAT. Divide both sides by the area $Δ x Δ y$ and let the square shrink to a point ( $A \to 0$ ).

WHY. The raw circulation shrinks to zero for a tiny loop (small area = small push). To get a number that survives shrinking — a density that describes the point itself — we divide out the area first. Circulation per unit area is the true "swirl at this point."

$(\nabla \times F)_{z} = A \to 0 lim \frac{1}{A} \oint_{C} F \cdot d r = \partial_{x} F_{2} - \partial_{y} F_{1}$

$A$ = area of the loop; $\frac{1}{A} \oint$ = circulation packed into that area.
$lim_{A \to 0}$ = "keep shrinking the loop onto the point" so the answer belongs to the point, not the loop.
The result no longer mentions $Δ x$ or $Δ y$ — it is a pure property of the field at $(x, y)$ .

This limit IS the coordinate-free statement, and it is Stokes' Theorem (and its flat cousin Green's Theorem) shrunk to a single point.

Step 7 — The other two components by symmetry (the full 3D vector)

WHAT. Nothing above used that the loop lay flat in the $x y$ -plane except the names $x, y$ . Run the identical argument on a loop in the $y z$ -plane, then the $z x$ -plane, cycling $x \to y \to z \to x$ :

$\nabla \times F = (loop in y z \partial_{y} F_{3} - \partial_{z} F_{2}, loop in z x \partial_{z} F_{1} - \partial_{x} F_{3}, loop in x y \partial_{x} F_{2} - \partial_{y} F_{1})$

WHY it is a vector, not a number. A spin in 3D needs an axle direction. Each component is the swirl measured around one axis, so curl carries three swirls at once — one per axis — which is exactly what an arrow (vector) stores. (In 2D we only ever have the last one, which is why 2D people call curl "a single number.")

PICTURE. Three mutually perpendicular tiny loops, each colored, feeding one component of the output vector.

Step 8 — Edge & degenerate cases (never hit an unshown scenario)

WHAT & WHY. Let's test the formula on the exact situations that break intuition.

The one-picture summary

Every step collapses into this single diagram: a tiny counterclockwise square, its right-vs-left up-push ( $\partial_{x} F_{2}$ ) racing its top-vs-bottom right-push ( $\partial_{y} F_{1}$ ), divided by area, shrunk to a point.

$spin at a point = lim_{A \to 0} \frac{push around tiny loop}{area} = \partial_{x} F_{2} - \partial_{y} F_{1}$

Recall Feynman retelling — the whole walkthrough in plain words

Toss a tiny toy waterwheel into a river. To find out if it spins, don't stare at the wheel — walk a tiny square around it, counterclockwise, and add up how much the water helps you as you go. Along the bottom and top only the sideways push counts; along the two sides only the up-push counts. When you add it all up, the totals don't matter — only the differences: is the water pushing up harder on the right side than the left? Is it pushing sideways harder on the top than the bottom? Those two differences are the two rates $\partial_{x} F_{2}$ and $\partial_{y} F_{1}$ . Subtract them, divide by the square's area, and shrink the square to a dot. That number is the spin at that dot. The stunning part: even a river flowing perfectly straight can spin the wheel, if one side flows faster than the other. And a field that just squirts outward, no matter how dramatic, spins nothing at all.

Active Recall

Recall Rebuild it yourself

Why do the top and left edges carry minus signs? ::: We walk them against the positive axis, so a push along the axis fights our motion — the dot product $F \cdot d r$ goes negative.
Why divide the circulation by area before taking the limit? ::: Raw circulation of a tiny loop shrinks to zero; dividing by area gives a finite density that belongs to the point.
Which tool converts a difference like $F_{2} (x + Δ x) - F_{2} (x)$ into $\partial_{x} F_{2} Δ x$ , and why that tool? ::: The partial derivative — it is defined as change-per-unit-step, exactly the rate we need.
Compute curl of $(- y, 0)$ and say why it surprises people. ::: $0 - (- 1) = 1$ ; the arrows are straight yet the wheel spins because of shear.
Why is curl a vector while divergence is a scalar? ::: Rotation in 3D needs an axle direction; three perpendicular loops give three swirl numbers, which pack into one arrow.