Vector Differential Operators

Ordinary differentiation measures how one quantity changes along one axis; a field problem asks how a whole scalar or vector field changes in every direction at once. The MSTE paper reaches for the vector differential operators whenever a temperature, a potential, a velocity, or a force field appears, and a single symbol, the del operator, produces all three answers the board can ask for: the gradient of a scalar field, the divergence of a vector field, and its curl. This lesson builds each operator from its components, gives its physical meaning, adds the Laplacian and the two identities worth memorizing, and closes with the conservative and solenoidal fields they define. Every formula is carried to a number with units so you can check your own habits against it.

The del operator

The del operator (also called nabla) is a vector whose components are the three partial-derivative instructions:

del = i partial/partial x + j partial/partial y + k partial/partial z = (partial/partial x, partial/partial y, partial/partial z)

By itself del does nothing; it waits for a field to act on. What makes it powerful is that the three ways a vector can meet another object each produce a different, useful operator. Applied to a scalar field f it gives a vector (the gradient). Dotted into a vector field F it gives a scalar (the divergence). Crossed into a vector field F it gives a vector (the curl). Keeping the input and output types straight is the fastest way to avoid a wrong-shape answer on the board.

Operation	Notation	Input	Output	Measures
Gradient	del f	scalar field	vector	direction and rate of steepest increase
Divergence	del dot F	vector field	scalar	net outflow per unit volume
Curl	del cross F	vector field	vector	local rotation (swirl)
Laplacian	del^2 f = del dot del f	scalar field	scalar	net curvature (divergence of the gradient)

Gradient of a scalar field

The gradient collects the three partial derivatives of a scalar field f(x, y, z) into one vector:

grad f = del f = (partial f/partial x, partial f/partial y, partial f/partial z)

Its meaning is geometric. At any point the gradient points in the direction in which f increases fastest, it is perpendicular (normal) to the level curve or level surface through that point, and its length is the steepness there, that is the maximum rate of increase per unit distance. The rate of change of f in an arbitrary direction is read off with the dot product: the directional derivative in the direction of a unit vector u is D_u f = grad f dot u, which is largest when u lines up with grad f and zero when u is tangent to a level curve.

Worked example: A temperature field is T(x, y, z) = x^2 y + y z^2 in degrees C, with position in meters. Its gradient is grad T = (2xy, x^2 + z^2, 2yz). At the point (1, 2, 1) this is grad T = (2(1)(2), 1^2 + 1^2, 2(2)(1)) = (4, 2, 4) degrees C/m. The maximum rate of increase there is the magnitude

|grad T| = sqrt(4^2 + 2^2 + 4^2) = sqrt(16 + 4 + 16) = sqrt(36) = 6 degrees C/m,

pointing along the unit vector (4, 2, 4)/6 = (2/3, 1/3, 2/3). Adding the components to get 10 or skipping the square root to report 36 are the classic slips.

Worked example (directional derivative): Using the same grad T = (4, 2, 4) degrees C/m, find the rate of change of T in the direction of a = (1, 2, 2). First normalize: |a| = sqrt(1 + 4 + 4) = 3, so u = (1/3, 2/3, 2/3). Then

D_u T = grad T dot u = (4)(1/3) + (2)(2/3) + (4)(2/3) = (4 + 4 + 8)/3 = 16/3 = 5.33 degrees C/m.

Forgetting to divide by |a| and reporting the raw dot product 16 is the standard error, and the answer must never exceed |grad T| = 6 degrees C/m, which is a quick sanity check.

Divergence of a vector field

The divergence of a vector field F = (P, Q, S) is the dot product of del with F, a scalar:

div F = del dot F = partial P/partial x + partial Q/partial y + partial S/partial z

Physically it measures the net outflow of the field per unit volume at a point. A positive divergence marks a source, where more of the field leaves a tiny region than enters it; a negative divergence marks a sink; a zero divergence means whatever flows in also flows out, the hallmark of an incompressible or source-free field. In fluid flow div F is the fractional rate at which a small parcel of fluid expands.

Worked example: For the velocity field F = (x^2 y, y^2 z, z^2 x) in m/s, with position in meters, the divergence is

div F = partial/partial x (x^2 y) + partial/partial y (y^2 z) + partial/partial z (z^2 x) = 2xy + 2yz + 2zx.

At the point (1, 2, 3) this equals 2(1)(2) + 2(2)(3) + 2(3)(1) = 4 + 12 + 6 = 22 per second (units of 1/s, a velocity divided by a length). Dropping the factor of 2 on each term and reporting 11 per second is the common miss.

Curl of a vector field

The curl of F = (P, Q, S) is the cross product of del with F, and like every cross product it is itself a vector. Written out as the i, j, k determinant it is

curl F = del cross F = (partial S/partial y - partial Q/partial z, partial P/partial z - partial S/partial x, partial Q/partial x - partial P/partial y)

Physically the curl measures the local rotation of the field. Drop a tiny paddle wheel into the flow: it spins fastest about the axis pointing along curl F, and the length of curl F is twice the angular speed. The direction follows the right-hand rule, curling the fingers with the swirl so the thumb gives the curl direction. A field whose curl is zero everywhere has no swirl and is called irrotational, which is exactly the condition that makes it conservative.

Worked example: For the same field F = (x^2 y, y^2 z, z^2 x) in m/s, take P = x^2 y, Q = y^2 z, and S = z^2 x. The three components are

i: partial S/partial y - partial Q/partial z = 0 - y^2 = -y^2, j: partial P/partial z - partial S/partial x = 0 - z^2 = -z^2, k: partial Q/partial x - partial P/partial y = 0 - x^2 = -x^2,

so curl F = (-y^2, -z^2, -x^2). At the point (1, 2, 3) this is (-(2^2), -(3^2), -(1^2)) = (-4, -9, -1) per second, with magnitude

|curl F| = sqrt((-4)^2 + (-9)^2 + (-1)^2) = sqrt(16 + 81 + 1) = sqrt(98) = 9.90 per second.

Reversing a subtraction and writing +y^2 for the first component is the usual sign mistake, which flips the whole answer.

The Laplacian

Feeding the gradient of a scalar back into the divergence gives the Laplacian, the most common second-order operator on the board:

del^2 f = div(grad f) = partial^2 f/partial x^2 + partial^2 f/partial y^2 + partial^2 f/partial z^2

It is a scalar that measures the net curvature of f, how much the value at a point differs from the average of its neighbours. A field with del^2 f = 0 everywhere is called harmonic and satisfies Laplace equation; steady-state temperature, electrostatic potential in a charge-free region, and ideal-flow velocity potentials are all harmonic. When a source is present the value is set by Poisson equation, del^2 f = a known source term.

Worked example: For phi = x^2 + y^2 + z^2 in degrees C, with position in meters, the gradient is grad phi = (2x, 2y, 2z), so the Laplacian is

del^2 phi = div(grad phi) = partial/partial x (2x) + partial/partial y (2y) + partial/partial z (2z) = 2 + 2 + 2 = 6 degrees C/m^2,

a constant everywhere, which flags this bowl-shaped field as a distributed source rather than a harmonic one. By contrast V = x^2 - y^2 gives del^2 V = 2 + (-2) + 0 = 0 V/m^2, so that field is harmonic.

Vector identities that save time

Two identities are worth memorizing because they turn a page of algebra into a single word. First, the curl of any gradient is the zero vector, and second, the divergence of any curl is the scalar zero:

curl(grad f) = 0 and div(curl F) = 0

The first says every gradient field is irrotational, which is why testing curl F = 0 tells you whether F could be a gradient. The second says every curl field is source-free, which is why the magnetic field, written as the curl of a vector potential, is automatically solenoidal. Two more are handy: the divergence of a gradient is the Laplacian, div(grad f) = del^2 f, and the product rule div(f F) = f div F + grad f dot F lets you differentiate a scalar times a vector without expanding everything.

Identity	Statement	What it guarantees
Curl of a gradient	curl(grad f) = 0	every gradient field is irrotational (conservative)
Divergence of a curl	div(curl F) = 0	every curl field is solenoidal (source-free)
Divergence of a gradient	div(grad f) = del^2 f	the Laplacian, the net curvature of f
Scalar-times-vector rule	div(f F) = f div F + grad f dot F	splits a product into two simpler pieces

Worked example: Check div(curl F) = 0 on the field from before, whose curl was found to be curl F = (-y^2, -z^2, -x^2). Its divergence is div(curl F) = partial/partial x (-y^2) + partial/partial y (-z^2) + partial/partial z (-x^2) = 0 + 0 + 0 = 0, confirming the identity for this field with no arithmetic left over.

Conservative and solenoidal fields

The two identities carve vector fields into two named families that the board tests directly. A field F is conservative (also called irrotational) when it is the gradient of some scalar potential f, F = grad f. On a simply connected region this happens exactly when curl F = 0, and it is the pleasant case where line integrals depend only on the endpoints: the work from A to B is f(B) - f(A) and every closed-loop integral is zero. A field F is solenoidal (also called divergence-free) when div F = 0, so it has no sources or sinks; by the divergence-of-a-curl identity every such field can be written as F = curl A for some vector potential A, and incompressible fluid flow and the magnetic field are the standard examples.

Field type	Defining test	Also called	Consequence
Conservative	curl F = 0	irrotational, a gradient field	F = grad f, path-independent line integrals
Solenoidal	div F = 0	divergence-free, a curl field	F = curl A, no net flux through any closed surface

Worked example (conservative): Show that F = (2xy, x^2 + z^2, 2yz) in newtons, with position in meters, is conservative, then find the work it does from (1, 2, 1) to (2, 1, 2). With P = 2xy, Q = x^2 + z^2, S = 2yz the curl components are S_y - Q_z = 2z - 2z = 0, P_z - S_x = 0 - 0 = 0, and Q_x - P_y = 2x - 2x = 0, so curl F = (0, 0, 0) and F is conservative. Integrating P in x gives the potential f = x^2 y + y z^2 (check: grad f returns F exactly). The work is then

W = f(2, 1, 2) - f(1, 2, 1) = (2^2)(1) + (1)(2^2) - [(1^2)(2) + (2)(1^2)] = (4 + 4) - (2 + 2) = 8 - 4 = 4 J,

the same for every path between those two points. Swapping the endpoints would flip the sign to -4 J, the standard orientation mistake.

Worked example (solenoidal): Test whether G = (2x, -3y, z) in m/s is solenoidal. Its divergence is div G = partial/partial x (2x) + partial/partial y (-3y) + partial/partial z (z) = 2 - 3 + 1 = 0 per second, so G is solenoidal and represents an incompressible flow, with no net flux out through any closed surface it fills. Miscounting the signs to get 2 + 3 + 1 = 6 per second would wrongly label it a source.

Exam-day strategy

• Track the shape before you compute: grad turns a scalar into a vector, div turns a vector into a scalar, and curl turns a vector into a vector; an answer of the wrong shape is wrong before you check the numbers.
• For a gradient, list the three partial derivatives, evaluate them at the point, and report the vector; the maximum rate of increase is the magnitude |grad f| and its direction is grad f itself, always perpendicular to the level surface.
• For a directional derivative, normalize the direction first (D_u f = grad f dot u with |u| = 1); a raw dot product with an unnormalized vector is the most common slip, and the answer can never exceed |grad f|.
• Read divergence as sources and sinks: div F = P_x + Q_y + S_z (add the matching partials), positive for a source, negative for a sink, zero for solenoidal or incompressible.
• Build the curl from the determinant pattern (S_y - Q_z, P_z - S_x, Q_x - P_y) and mind every minus sign; a single flipped subtraction reverses the whole vector.
• Use the two free identities: curl(grad f) = 0 and div(curl F) = 0 collapse whole expressions to zero on sight, and div(grad f) = del^2 f is the Laplacian, which is zero for a harmonic field.
• Classify the field to pick your method: curl F = 0 means conservative, so use the potential and f(B) - f(A) and expect zero around any loop; div F = 0 means solenoidal, so expect zero net flux through any closed surface.