Curl of a Vector Field

The curl is one of the two first-order vector derivatives that the del operator builds from a vector field, the twin of the divergence. Where divergence extracts a scalar — the net rate at which field lines spring from or vanish into a point — the curl extracts a vector: how strongly, and about which axis, the field swirls in the immediate neighbourhood of that point. It maps a vector field to another vector field, $\nabla\times\mathbf{F}$ , and it is the operator at the heart of electromagnetism, fluid dynamics, and elasticity — anywhere circulation, rotation, or "twist" is the physics. Operationally the curl is the formal cross product $\nabla\times\mathbf{F}$ , a determinant in the unit vectors, the components of $\nabla$ , and the components of $\mathbf{F}$ . Every entry is an antisymmetric pair of partial derivatives, $\partial_i F_j-\partial_j F_i$ , so the curl sees only the rotational (shear) part of how the field changes — never the uniform stretching the divergence measures. Its deepest meaning, though, is integral: the component of $\nabla\times\mathbf{F}$ along a unit normal $\hat{\mathbf{n}}$ is the circulation of $\mathbf{F}$ per unit area, in the limit of a shrinking loop. That coordinate-free picture is the bridge to Stokes' theorem and the reason the curl appears in Ampère's law, in vorticity, and in Faraday's electromotive force. Two structural identities govern it. The curl of any gradient is zero, and the divergence of any curl is zero — both consequences of the commutativity of mixed partial derivatives. Together they underwrite the Helmholtz decomposition: every well-behaved field splits into an irrotational part and a solenoidal part.

Key takeaways

The curl $\nabla\times\mathbf{F}$ is itself a vector field; at each point it measures the local rotation, or circulation density, of $\mathbf{F}$ about that point.
In Cartesian coordinates it is the antisymmetric combination of partial derivatives $\partial_i F_j-\partial_j F_i$ , written compactly as the determinant of $\nabla$ with $\mathbf{F}$ .
Its direction is the axis of strongest local rotation (right-hand rule); its magnitude is twice the local angular velocity of an infinitesimal paddle wheel.
Stokes' theorem ties curl to circulation: the flux of $\nabla\times\mathbf{F}$ through a surface equals the line integral of $\mathbf{F}$ around its boundary.
Two null identities structure vector calculus: $\nabla\times(\nabla\phi)=\mathbf{0}$ and $\nabla\cdot(\nabla\times\mathbf{F})=0$ , both consequences of symmetric mixed partials.
On a simply connected domain, zero curl everywhere means the field is conservative and derivable from a scalar potential.
Key idea: The curl is circulation density — $(\nabla\times\mathbf{F})\cdot\hat{\mathbf{n}}$ is the limiting ratio of the circulation of $\mathbf{F}$ around an infinitesimal loop perpendicular to $\hat{\mathbf{n}}$ to the loop's area — and it points along the axis of maximal local rotation with magnitude twice the local angular velocity.

The core insight

The curl is circulation density: $(\nabla\times\mathbf{F})\cdot\hat{\mathbf{n}}$ at a point is the limiting ratio of the line integral of $\mathbf{F}$ around an infinitesimal loop in the plane perpendicular to $\hat{\mathbf{n}}$ to the loop's area. The vector curl points along the axis of maximal local rotation, with magnitude twice the local angular velocity.

Where this lives today

The curl is not a museum piece; it is a live measurement and a design constraint. In experimental fluid dynamics, particle-image velocimetry images tracer particles between laser pulses and differentiates the reconstructed velocity field to map the vorticity $\boldsymbol{\zeta}=\nabla\times\mathbf{v}$ directly — the quantity that organizes everything from aircraft wingtip vortices to combustion mixing. In atmospheric science, the curl of the wind (relative vorticity) and its potential-vorticity refinement are the state variables forecasters watch to track cyclones and jet-stream meanders. Most strikingly, the free-vortex subtlety of Worked Example 2 — curl zero everywhere except a singular core — is realized literally in quantum fluids: in superfluid helium and Bose–Einstein condensates the circulation is quantized, $\oint\mathbf{v}\cdot\mathrm{d}\boldsymbol{\ell}=n\,h/m$ , with the entire vorticity confined to microscopic vortex lines while the flow around them is irrotational. The same mathematics governs the quantized flux tubes and Abrikosov lattices of type-II superconductors — a nineteenth-century operator reading out twenty-first-century quantum matter.

The physical picture

Drop a tiny paddle wheel into a flowing fluid whose velocity is $\mathbf{F}$ . If the wheel spins, the field has curl there; the spin axis is the direction of $\nabla\times\mathbf{F}$ and the rate of spin is half its magnitude. Crucially, the wheel responds to differences in flow across its diameter, not to the shape of the streamlines. A field of perfectly straight, parallel arrows whose length grows sideways — a shear — turns the wheel just as surely as a field of nested circles does. This is the single most common stumbling block with the curl, and the first figure puts both cases side by side.

Figure

A paddle wheel reads local rotation. In rigid rotation F = ω × r the wheel spins and ∇×F = 2ω points out of the page; in the shear flow F = (y, 0) the streamlines are perfectly straight yet the wheel still turns, because ∇×F = −(∂Fₓ/∂y) = −1 points into the page

Think of the curl the other way round, as a circulation meter. Trace a small closed loop and add up the tangential component of $\mathbf{F}$ as you walk around it: if the field carries you forward more on one side than it pushes back on the other, the loop accumulates net circulation. Divide by the enclosed area and shrink the loop, and you have recovered the component of $\nabla\times\mathbf{F}$ along the loop's normal. The two null identities then read almost by inspection: a gradient field points "downhill" from a potential, so any closed walk gains exactly what it loses and the circulation is zero; and the curl field, being a field of rotation axes, has no free ends, so its own field lines close and its divergence vanishes. The second figure carries this into a real, quantitative flow — a channel with a parabolic velocity profile whose vorticity is a number you can read off.

Figure

A Poiseuille channel flow v(y) = v₀(1 − (y/h)²) with v₀ = 2 m/s and half-width h = 3 m. The velocity points only along x, yet the transverse shear gives a vorticity ζ = ∇×v = 2v₀y/h². At y = 1.5 m, ζ = 0.667 s⁻¹ along +z, so a suspended wheel spins at ω = ζ/2 = 0.33 rad/s; on the centre-line y = 0 the shear vanishes and the wheel does not turn

The core equations

The curl is the formal cross product of the del operator with the field. In Cartesian coordinates it is most memorably written as a symbolic $3\times3$ determinant — unit vectors on the top row, differential operators in the middle, field components on the bottom.

\nabla\times\mathbf{F}=\begin{vmatrix}\hat{\mathbf{x}}&\hat{\mathbf{y}}&\hat{\mathbf{z}}\\[2pt]\dfrac{\partial}{\partial x}&\dfrac{\partial}{\partial y}&\dfrac{\partial}{\partial z}\\[4pt]F_x&F_y&F_z\end{vmatrix} \qquad (1)

where: $\nabla$ — the del operator $\hat{\mathbf{x}}\,\partial_x+\hat{\mathbf{y}}\,\partial_y+\hat{\mathbf{z}}\,\partial_z$ ; $\mathbf{F}$ — the vector field, with components $F_x,F_y,F_z$ ; $\hat{\mathbf{x}},\hat{\mathbf{y}},\hat{\mathbf{z}}$ — the Cartesian unit basis vectors. The determinant is a mnemonic, not a literal matrix: the "products" are derivatives acting on components. Carrying out the cofactor expansion gives the explicit component form. Each component is a difference of two "cross" derivatives — the antisymmetric pattern that makes the curl blind to uniform stretching and sensitive only to shear.

\nabla\times\mathbf{F}=\left(\frac{\partial F_z}{\partial y}-\frac{\partial F_y}{\partial z}\right)\hat{\mathbf{x}}+\left(\frac{\partial F_x}{\partial z}-\frac{\partial F_z}{\partial x}\right)\hat{\mathbf{y}}+\left(\frac{\partial F_y}{\partial x}-\frac{\partial F_x}{\partial y}\right)\hat{\mathbf{z}} \qquad (2)

where: each $\partial F_i/\partial j$ is the partial derivative of component $F_i$ with respect to coordinate $j$ . Read cyclically: the $x$ -component involves $y$ and $z$ , the $y$ -component involves $z$ and $x$ , the $z$ -component involves $x$ and $y$ . The same object is far more compact in index notation, using the Levi-Civita symbol and the Einstein summation convention. This form is the workhorse for proving vector identities, because the algebraic properties of $\epsilon_{ijk}$ do the reasoning for you.

(\nabla\times\mathbf{F})_i=\epsilon_{ijk}\,\partial_j F_k \qquad (3)

where: $\epsilon_{ijk}$ — the Levi-Civita symbol, $+1$ for even permutations of $(1,2,3)$ , $-1$ for odd, $0$ if any index repeats; $\partial_j=\partial/\partial x_j$ , with an implied sum over the repeated indices $j$ and $k$ . The single free index $i$ labels the component of the result. The definition that captures the meaning of the curl is integral. Project onto any unit normal $\hat{\mathbf{n}}$ and the result is the circulation of $\mathbf{F}$ around a small loop $C$ of area $A$ , per unit area, in the limit that the loop shrinks to a point — the physical, coordinate-free content of the operator.

\hat{\mathbf{n}}\cdot(\nabla\times\mathbf{F})=\lim_{A\to 0}\frac{1}{A}\oint_{C}\mathbf{F}\cdot\mathrm{d}\boldsymbol{\ell} \qquad (4)

where: $\hat{\mathbf{n}}$ — unit normal to the loop $C$ , fixed relative to its orientation by the right-hand rule; $A$ — the area enclosed by $C$ ; $\mathrm{d}\boldsymbol{\ell}$ — the vector line element traversing $C$ . Equation (4) is the local form of Stokes' theorem and the seed of Green's theorem in the plane. Two identities fall out of the antisymmetry of the curl the moment it is composed with a gradient or fed into a divergence. They are the backbone of potential theory and of the Helmholtz decomposition.

\nabla\times(\nabla\phi)=\mathbf{0},\qquad \nabla\cdot(\nabla\times\mathbf{F})=0 \qquad (5)

where: $\phi$ — any twice-differentiable scalar field; $\mathbf{F}$ — any twice-differentiable vector field. The first says gradients are irrotational; the second says curls are solenoidal (source-free). Applying the curl twice produces the identity that generates the electromagnetic wave equation. It trades a hard-to-picture double curl for a gradient of a divergence minus a vector Laplacian.

\nabla\times(\nabla\times\mathbf{F})=\nabla(\nabla\cdot\mathbf{F})-\nabla^2\mathbf{F} \qquad (6)

where: $\nabla^2\mathbf{F}$ — the vector Laplacian, which in Cartesian coordinates acts componentwise. Mnemonically it echoes the "BAC−CAB" rule for the vector triple product $\mathbf{A}\times(\mathbf{B}\times\mathbf{C})$ . Finally, in curved or non-Cartesian problems the curl is written with the scale (Lamé) factors of an orthogonal curvilinear system, so that spherical and cylindrical forms are read off a single template.

\nabla\times\mathbf{F}=\frac{1}{h_1h_2h_3}\begin{vmatrix}h_1\hat{\mathbf{e}}_1&h_2\hat{\mathbf{e}}_2&h_3\hat{\mathbf{e}}_3\\[2pt]\partial_{u_1}&\partial_{u_2}&\partial_{u_3}\\[2pt]h_1F_1&h_2F_2&h_3F_3\end{vmatrix} \qquad (7)

where: $u_i$ — the curvilinear coordinates; $h_i$ — the scale factors, e.g. $(1,r,r\sin\theta)$ in spherical coordinates and $(1,\rho,1)$ in cylindrical; $\hat{\mathbf{e}}_i$ — the orthonormal curvilinear basis. Setting all $h_i=1$ recovers equation (1).

Derivation ACirculation density recovers the z-component of the curl

The goal is to prove that the integral definition (4) and the differential definition (2) are the same object, by computing the circulation of $\mathbf{F}$ around an infinitesimal rectangle in the $xy$ -plane and dividing by its area. If the two definitions agree, we can move freely between "curl as a determinant of derivatives" and "curl as circulation per unit area" — the move that powers Stokes' and Green's theorems. Take a rectangle centred at $(x_0,y_0)$ with sides $\Delta x$ and $\Delta y$ , traversed counterclockwise so that the enclosed normal is $+\hat{\mathbf{z}}$ , and sum $\mathbf{F}\cdot\mathrm{d}\boldsymbol{\ell}$ over its four edges.

Step 1. The bottom edge (run in $+x$ ) and the top edge (run in $-x$ ) contribute only through $F_x$ , and their sum is $\left[F_x(x_0,y_0-\tfrac{\Delta y}{2})-F_x(x_0,y_0+\tfrac{\Delta y}{2})\right]\Delta x$ . Why: the top edge is traversed in the reverse direction, so its integrand carries a minus sign, leaving a difference of $F_x$ taken across a separation $\Delta y$ in the $y$ -direction.

Step 2. That difference is $-\,(\partial F_x/\partial y)\,\Delta y$ to first order, so the top-and-bottom pair gives $-(\partial F_x/\partial y)\,\Delta x\,\Delta y$ . Why: a first-order Taylor expansion in $y$ turns the finite difference of $F_x$ into a derivative times the offset, which is the whole content of "shrink the loop."

Step 3. By the identical argument the right edge (run in $+y$ ) and left edge (run in $-y$ ) contribute only through $F_y$ , and their sum is $+(\partial F_y/\partial x)\,\Delta x\,\Delta y$ . Why: the left edge reverses sign, leaving a difference of $F_y$ across $\Delta x$ , which Taylor-expands to $\partial F_y/\partial x$ times the area element.

Step 4. Adding the two pairs and dividing by the area $\Delta x\,\Delta y$ , then letting $\Delta x,\Delta y\to0$ , gives $(\nabla\times\mathbf{F})_z=\partial F_y/\partial x-\partial F_x/\partial y$ . Why: this is exactly the $z$ -component of the component form (2), so the integral and differential definitions coincide; the same construction in the other two coordinate planes yields the $x$ - and $y$ -components.

Go deeper · Derivation B · 1 minThe curl of a gradient is identically zero

Derivation BThe curl of a gradient is identically zero

The goal is to show that $\nabla\times(\nabla\phi)=\mathbf{0}$ for every twice-differentiable scalar field $\phi$ , using nothing but the antisymmetry of the Levi-Civita symbol played against the symmetry of second partial derivatives. This is not a coincidence of special functions but a structural fact, and it is precisely what lets a curl-free field be written as the gradient of a potential — the foundation of conservative forces and scalar potentials. We work in index notation, where the argument is a single line of algebra.

Step 1. Write the $i$ -th component as $[\nabla\times(\nabla\phi)]_i=\epsilon_{ijk}\,\partial_j(\partial_k\phi)=\epsilon_{ijk}\,\partial_j\partial_k\phi$ . Why: the gradient supplies $(\nabla\phi)_k=\partial_k\phi$ , and the curl acts with $\epsilon_{ijk}\partial_j$ , so the composite is a contraction of $\epsilon_{ijk}$ with a double derivative.

Step 2. For $\phi\in C^2$ , Clairaut's theorem gives $\partial_j\partial_k\phi=\partial_k\partial_j\phi$ , so the object $\partial_j\partial_k\phi$ is symmetric under $j\leftrightarrow k$ . Why: continuity of the second partials lets the order of differentiation be swapped, which is the one analytic hypothesis the whole identity rests on.

Step 3. But $\epsilon_{ijk}=-\epsilon_{ikj}$ is antisymmetric under the same swap, and the contraction of a symmetric tensor with an antisymmetric symbol vanishes term by term. Why: each term $\epsilon_{ijk}\,\partial_j\partial_k\phi$ is exactly cancelled by its partner with $j$ and $k$ exchanged, so the sum over $j,k$ is zero for every free index $i$ .

Step 4. Hence $\epsilon_{ijk}\,\partial_j\partial_k\phi=0$ for all $i$ , i.e. $\nabla\times\nabla\phi=\mathbf{0}$ . Why: the same symmetric-times-antisymmetric argument, applied to $\nabla\cdot(\nabla\times\mathbf{F})=\epsilon_{ijk}\,\partial_i\partial_j F_k$ , proves the companion identity that the divergence of a curl vanishes; the converse — curl-free implies gradient — holds on simply connected domains.

Go deeper · Derivation C · 1 minRigid rotation has curl equal to twice the angular velocity

Derivation CRigid rotation has curl equal to twice the angular velocity

The goal is to compute $\nabla\times(\boldsymbol{\omega}\times\mathbf{r})$ for a constant angular velocity $\boldsymbol{\omega}$ and confirm the paddle-wheel interpretation quantitatively: that the curl of a rigidly rotating velocity field is exactly $2\boldsymbol{\omega}$ , everywhere, independent of position. This is the cleanest possible calibration of the operator — it fixes the numerical factor of two that connects "circulation density" to "angular velocity" and justifies reading the curl as twice a local spin rate. Let $\boldsymbol{\omega}=(\omega_1,\omega_2,\omega_3)$ be constant and $\mathbf{r}=(x,y,z)$ the position vector, so the field is $\mathbf{F}=\boldsymbol{\omega}\times\mathbf{r}$ .

Step 1. Expanding the cross product gives the components $\mathbf{F}=(\omega_2 z-\omega_3 y,\ \omega_3 x-\omega_1 z,\ \omega_1 y-\omega_2 x)$ . Why: these are the standard components of $\boldsymbol{\omega}\times\mathbf{r}$ , and writing them out explicitly is what lets the partial derivatives be taken directly.

Step 2. The $x$ -component of the curl is $(\nabla\times\mathbf{F})_x=\partial_y F_z-\partial_z F_y=\partial_y(\omega_1 y-\omega_2 x)-\partial_z(\omega_3 x-\omega_1 z)$ . Why: this is just the first slot of the component form (2), with $F_z$ and $F_y$ substituted from Step 1.

Step 3. Evaluating the derivatives, $\partial_y(\omega_1 y)=\omega_1$ and $\partial_z(-\omega_1 z)=-\omega_1$ , so $(\nabla\times\mathbf{F})_x=\omega_1-(-\omega_1)=2\omega_1$ . Why: only the terms linear in the differentiated coordinate survive, and the two surviving pieces reinforce rather than cancel, producing the factor of two.

Step 4. Cyclic symmetry gives $(\nabla\times\mathbf{F})_y=2\omega_2$ and $(\nabla\times\mathbf{F})_z=2\omega_3$ , so $\nabla\times(\boldsymbol{\omega}\times\mathbf{r})=2\boldsymbol{\omega}$ . Why: the calculation is invariant under $x\to y\to z\to x$ , so all three components share the same form, and the curl is twice the angular velocity at every point — the exact statement the paddle-wheel picture promises.

Limiting cases

Special configurations strip the curl down to something you can check at a glance:

Case	Result	Meaning
Uniform field	$\mathbf{F}=\mathbf{c}=\text{const}\ \Rightarrow\ \nabla\times\mathbf{F}=\mathbf{0}$	No spatial variation means every partial derivative vanishes: a constant field has no circulation density anywhere.
Rigid rotation	$\mathbf{F}=\boldsymbol{\omega}\times\mathbf{r}\ \Rightarrow\ \nabla\times\mathbf{F}=2\boldsymbol{\omega}$	A field co-rotating at angular velocity $\boldsymbol{\omega}$ has curl equal to twice $\boldsymbol{\omega}$ at every point — the calibration of curl as twice the local spin.
Gradient field	$\mathbf{F}=\nabla\phi\ \Rightarrow\ \nabla\times\mathbf{F}=\mathbf{0}$	Every gradient is irrotational; on a simply connected domain the converse holds and the field is conservative.
Two-dimensional field	$\mathbf{F}=F_x\hat{\mathbf{x}}+F_y\hat{\mathbf{y}}\ \Rightarrow\ \nabla\times\mathbf{F}=(\partial_x F_y-\partial_y F_x)\hat{\mathbf{z}}$	A planar field has curl pointing purely out of the plane; that single component is what Green's theorem integrates.
Free vortex	$\mathbf{F}=\dfrac{-y\hat{\mathbf{x}}+x\hat{\mathbf{y}}}{x^2+y^2}\ \Rightarrow\ \nabla\times\mathbf{F}=\mathbf{0}\ (r\ne0)$	Curl-free everywhere it is defined, yet all the circulation is concentrated on the singular axis: zero curl need not mean conservative on a punctured domain.

Common trapThe curl measures local rotation, not visible curvature of the field lines

The single most common error is to judge the curl by eye — to assume straight field lines mean zero curl and curved lines mean nonzero curl. Both halves are false. A shear of perfectly straight, parallel arrows, $\mathbf{F}=y\,\hat{\mathbf{x}}$ , has curl $\nabla\times\mathbf{F}=-\hat{\mathbf{z}}\ne\mathbf{0}$ , because a paddle wheel feels faster flow on one side than the other and spins. Conversely, the free vortex $\mathbf{F}=(-y\hat{\mathbf{x}}+x\hat{\mathbf{y}})/(x^2+y^2)$ has beautifully curved circular field lines yet $\nabla\times\mathbf{F}=\mathbf{0}$ everywhere except the origin, because the flow slows just fast enough with radius to cancel the turning. The curl is a statement about differential rotation across an infinitesimal element, not the global shape of the streamlines. Always compute it from the partial-derivative formula; never read it off the picture.

Worked examples

1 · Vorticity of a channel flow (fully worked)

Water in a wide channel has the parabolic (Poiseuille) velocity profile $\mathbf{v}=v_0\left(1-\dfrac{y^2}{h^2}\right)\hat{\mathbf{x}}$ , with $v_0=2\ \mathrm{m/s}$ and half-width $h=3\ \mathrm{m}$ ; the walls are at $y=\pm h$ . Find the vorticity $\boldsymbol{\zeta}=\nabla\times\mathbf{v}$ as a function of $y$ , evaluate it at $y=1.5\ \mathrm{m}$ , and give the local spin rate of a small suspended paddle wheel there.

Step 1. The flow is planar and points only along $x$ , so only the $z$ -component of the curl survives: $\zeta_z=\partial_x v_y-\partial_y v_x=-\,\partial_y v_x$ . The transverse gradient of a purely longitudinal velocity is the entire source of vorticity here.

Step 2. Differentiate the profile: $\partial_y v_x=v_0\cdot\dfrac{\mathrm{d}}{\mathrm{d}y}\!\left(1-\dfrac{y^2}{h^2}\right)=-\dfrac{2v_0\,y}{h^2}$ , so $\boldsymbol{\zeta}=\dfrac{2v_0\,y}{h^2}\,\hat{\mathbf{z}}$ . The vorticity is linear in $y$ : zero on the centre-line, greatest at the walls, and oppositely signed in the two halves.

Step 3. Insert the numbers at $y=1.5\ \mathrm{m}$ : $\zeta_z=\dfrac{2(2)(1.5)}{3^2}=\dfrac{6}{9}=\mathbf{0.667\ s^{-1}}$ , directed along $+\hat{\mathbf{z}}$ (out of the page in the figure). A paddle wheel there rotates at $\omega=\tfrac12\zeta_z=\mathbf{0.33\ rad/s}$ , counterclockwise.

Step 4. Sanity checks: on the centre-line $y=0$ the shear vanishes, so $\boldsymbol\zeta=\mathbf 0$ and a wheel drifts without spinning; at the lower wall $y=-h$ the sign flips, so the wheel there spins the opposite way. Vorticity $\boldsymbol{\zeta}=(2v_0y/h^2)\hat{\mathbf{z}}$ ; at $y=1.5$ m it is $0.667\ \mathrm{s^{-1}}$ and $\omega=0.33$ rad/s. Straight streamlines, unmistakable curl — the warning made quantitative.

2 · The magnetic field of a straight wire (curl-free yet circulating)

A long straight wire on the $z$ -axis carries current $I=10\ \mathrm{A}$ . Its field is the free-vortex form $\mathbf{B}=\dfrac{\mu_0 I}{2\pi}\dfrac{-y\hat{\mathbf{x}}+x\hat{\mathbf{y}}}{x^2+y^2}$ . Compute $\nabla\times\mathbf{B}$ away from the wire, find the field magnitude at $r=0.05\ \mathrm{m}$ , and reconcile the vanishing curl with Ampère's law.

Step 1. Write $r^2=x^2+y^2$ and compute the only surviving component, $(\nabla\times\mathbf{B})_z=\partial_x B_y-\partial_y B_x$ . Both partials evaluate to $(y^2-x^2)/r^4$ (times $\mu_0 I/2\pi$ ), so they cancel exactly: $\nabla\times\mathbf{B}=\mathbf{0}$ for every point with $r\ne0$ . The differential form of Ampère's law, $\nabla\times\mathbf{B}=\mu_0\mathbf{J}$ , agrees: outside the wire the current density $\mathbf{J}$ is zero.

Step 2. The magnitude is $B=\dfrac{\mu_0 I}{2\pi r}=\dfrac{(4\pi\times10^{-7})(10)}{2\pi(0.05)}=\mathbf{4.0\times10^{-5}\ T}=40\ \mu\mathrm{T}$ , circling the wire by the right-hand rule.

Step 3. Now the reconciliation. Although the curl vanishes at every point off the axis, the circulation around a loop enclosing the wire is not zero: $\oint\mathbf{B}\cdot\mathrm{d}\boldsymbol{\ell}=B\,(2\pi r)=(4.0\times10^{-5})(2\pi)(0.05)=\mathbf{1.26\times10^{-5}\ T\,m}$ , which equals $\mu_0 I=(4\pi\times10^{-7})(10)=1.26\times10^{-5}\ \mathrm{T\,m}$ exactly.

Step 4. The resolution is topological: the region "all space minus the wire" is not simply connected, so a vanishing curl everywhere does not force zero circulation around a loop that cannot shrink to a point without crossing the wire. All the circulation sits on the current-carrying axis — the same structure as the free vortex above, and the reason Stokes' theorem must lay a piece of surface through the wire to see the source. $\nabla\times\mathbf{B}=0$ off-axis, $B=40\ \mu$ T at 5 cm, and $\oint\mathbf{B}\cdot\mathrm{d}\boldsymbol\ell=\mu_0 I$ .

Physics of the portrait

The curl is where three strands of nineteenth-century physics braided together: Hamilton's algebra of directed quantities, Helmholtz's dynamics of vortices, and Maxwell's field theory of electromagnetism — and it was Maxwell who gave the operator its enduring name.

Origins · 1 minJames Clerk Maxwell and how this was discovered

James Clerk Maxwell

1831–1879 · vector calculus and the electromagnetic field

James Clerk Maxwell did more than anyone to make the curl a physical operator rather than a formal one. Building on William Rowan Hamilton's quaternions of 1843 — from which the operator $\nabla$ first emerged — Maxwell needed a language for the rotational structure of the electromagnetic field. In his 1861–62 papers "On Physical Lines of Force" he pictured the magnetic field as a lattice of spinning molecular vortices, a mechanical model in which the curl of the field is literally the rotation of the medium. When he came to name the vector derivatives, in his 1871 essay on the mathematical classification of physical quantities, he weighed candidates — "rotation", "twist", "twirl" — before settling on curl for $\nabla\times$ , alongside what became divergence and gradient; the names entered his 1873 Treatise and never left. His deeper insight was that two of the four field equations are curl equations — Faraday's law and the Ampère–Maxwell law — and that taking the curl of them, via the double-curl identity, yields a wave travelling at the speed of light. Maxwell's identification of light as an electromagnetic wave is, at its mathematical core, a computation with $\nabla\times$ . The clean symbol we use today was standardized a generation later by Gibbs and Wilson, but the concept, and the word, are his.

History

1843

— Hamilton introduces quaternions, out of whose algebra the operator ∇ first arises

1858

— Helmholtz analyses vortex motion, making ∇×v the central object of rotational fluid dynamics

1861

— Maxwell models the magnetic field as molecular vortices in "On Physical Lines of Force"

1871

— Maxwell proposes the name "curl" for the rotation operator ∇×

1901

— Gibbs and Wilson's "Vector Analysis" fixes the modern ∇×F notation

Going further

Texts. Stone & Goldbart, Mathematics for Physics, §11 (vector calculus) · Arfken & Weber, Mathematical Methods for Physicists, 7th ed., §3.5–3.6 · Hassani, Mathematical Physics, §12–13 (vector analysis and integral theorems) · Riley, Hobson & Bence, Mathematical Methods for Physics and Engineering, 3rd ed., §10.8 · Griffiths, Introduction to Electrodynamics, 4th ed., §1.2.6 · Schey, Div, Grad, Curl, and All That, 4th ed., Ch. III. Key results for reference: the index-notation null identities are $\epsilon_{ijk}\partial_j\partial_k\phi=0$ and $\epsilon_{ijk}\partial_i\partial_j F_k=0$ ; the rigid-rotation result is $\nabla\times(\boldsymbol{\omega}\times\mathbf{r})=2\boldsymbol{\omega}$ ; Worked Example 1 gives $\boldsymbol\zeta=(2v_0y/h^2)\hat{\mathbf{z}}=0.667\ \mathrm{s^{-1}}$ at $y=1.5$ m; Worked Example 2 gives $B=\mu_0I/2\pi r=40\ \mu$ T at $r=5$ cm with $\oint\mathbf{B}\cdot\mathrm{d}\boldsymbol{\ell}=\mu_0I=1.26\times10^{-5}$ T·m. The only physical constant invoked is $\mu_0=4\pi\times10^{-7}\ \mathrm{T\,m/A}$ .

Prerequisites. The Del Operator and Vector Differential Identities · Scalar and Vector Fields · Divergence of a Vector Field · The Gradient Operator. Leads to. Stokes' Theorem · Green's Theorem in the Plane · The Laplacian Operator · Line Integrals.

Ask the tutor: Ask me to work any field of your choosing through the curl — a real velocity profile, a magnetic field, or a tricky "straight lines but nonzero curl" case — and I will compute it step by step and show you what the paddle wheel does.

Is the curl a vector or a scalar?

In three dimensions the curl of a vector field is another vector field — it has to be, because a rotation needs both a rate and an axis, and the axis is a direction in space. That is why $\nabla\times\mathbf{F}$ has three components, one for the circulation in each coordinate plane. In two dimensions the situation collapses: a planar field can only circulate about the out-of-plane axis, so the "curl" is effectively a single scalar, $\partial_x F_y-\partial_y F_x$ , which is the quantity that appears in Green's theorem. The vector character is genuinely special to three dimensions; in higher dimensions the honest generalization is not a vector at all but an antisymmetric tensor (a 2-form), which is why the modern language of differential forms treats the curl, divergence, and gradient as facets of a single exterior derivative.

Why is the factor exactly two in "curl equals twice the angular velocity"?

Because the curl adds two contributions that a naive count would take only once. Consider a rigidly rotating field: at a point off the axis, both the rate at which $F_y$ changes as you move in $x$ and the rate at which $F_x$ changes as you move in $y$ carry information about the same rotation, and they enter the curl with opposite signs, so they reinforce rather than cancel. Working it out — as in the third derivation — the two equal pieces $\partial_x F_y=+\omega$ and $-\partial_y F_x=+\omega$ sum to $2\omega$ . Physically, half of the local velocity gradient is the rotation (the antisymmetric part) and half is pure shear strain (the symmetric part); the curl captures twice the rotational half, which is why the angular velocity of a co-moving element is $\tfrac12\nabla\times\mathbf{v}$ , not the curl itself.

How can a field have zero curl everywhere and still not be conservative?

This happens when the domain has a "hole." The theorem that zero curl implies a potential requires the region to be simply connected — every closed loop must be shrinkable to a point without leaving the domain. The free vortex $\mathbf{F}=(-y\hat{\mathbf{x}}+x\hat{\mathbf{y}})/(x^2+y^2)$ is curl-free everywhere it is defined, but it is undefined on the $z$ -axis, so the domain is the punctured plane, which is not simply connected. A loop encircling the axis cannot be contracted past the singularity, and indeed its circulation is $2\pi\ne0$ . The curl only reports what happens locally; it is blind to the global topology of the domain. This is exactly the situation of the current-carrying wire in Worked Example 2, and it is the physical seed of Aharonov–Bohm phases and of flux quantization.

What is the difference between the curl and the divergence?

They are the two independent first-order things a vector field can do, and they are complementary. The divergence $\nabla\cdot\mathbf{F}$ is a scalar that measures the outflow per unit volume — whether the point acts as a source or a sink — and it is built from the symmetric derivatives $\partial_i F_i$ . The curl $\nabla\times\mathbf{F}$ is a vector that measures the circulation per unit area — whether the field swirls — and it is built from the antisymmetric derivatives $\partial_i F_j-\partial_j F_i$ . A uniform expansion has divergence but no curl; a rigid rotation has curl but no divergence. The Helmholtz theorem makes the complementarity precise: a field that decays at infinity is completely determined by its divergence and its curl together, which is why the two operators between them capture everything Maxwell's equations need to say.

Why do two of Maxwell's equations use the curl?

Because electromagnetism is fundamentally about circulation. Faraday's law, $\nabla\times\mathbf{E}=-\partial\mathbf{B}/\partial t$ , says a changing magnetic field wraps a circulating electric field around itself — the loop of $\mathbf{E}$ is what drives current in a transformer or a generator. The Ampère–Maxwell law, $\nabla\times\mathbf{B}=\mu_0\mathbf{J}+\mu_0\varepsilon_0\,\partial\mathbf{E}/\partial t$ , says electric currents and changing electric fields wrap a circulating magnetic field around themselves — the loop of $\mathbf{B}$ around a wire. The other two Maxwell equations use the divergence, because they are about sources (charge) and the absence of magnetic monopoles. Taking the curl of the two curl equations and applying the double-curl identity (6) decouples $\mathbf{E}$ and $\mathbf{B}$ into wave equations, and out drops the speed of light — the historical triumph that identified light as an electromagnetic wave.

Does the curl formula change in cylindrical or spherical coordinates?

The operator is the same geometric object, but its component expressions pick up the scale factors of the coordinate system, because the basis vectors themselves now vary from point to point. Equation (7) is the master template: insert the Lamé factors $h_i$ and it produces the correct curl in any orthogonal system. In cylindrical coordinates $(\rho,\phi,z)$ the factors are $(1,\rho,1)$ , and in spherical coordinates $(r,\theta,\phi)$ they are $(1,r,r\sin\theta)$ ; the $\rho$ 's and $r\sin\theta$ 's that appear in the familiar spherical curl formula are exactly these factors. The lesson is that the extra terms are not new physics — they are bookkeeping for the fact that "moving one unit in the $\phi$ -direction" covers a different amount of arc depending on where you are. Set every $h_i=1$ and equation (7) collapses back to the Cartesian determinant (1).