Fields, Del, Gradient, Divergence & Curl

Fields: a value at every point of space

Step into the world this rung lives in. In Volume I a function ate a single number and gave back a number; even a vector field you met only briefly there. Here the basic object is a field: a rule that pins a value to every point of a region of space. If that value is a single number — temperature in a room, pressure in the air, the height of a hill above each spot of ground — it is a [[scalar-field|scalar field]], written f(x, y, z). Picture the room: at each point there is one number, the temperature, and the whole arrangement is a smooth landscape of warmth filling the space.

But many quantities are not single numbers — they have a direction as well as a size. The velocity of water in a flowing river, the wind, the pull of gravity, the electric force at each point: each of these attaches an arrow, not a number, to every point. That is a [[vector-field-advanced|vector field]], written F(x, y, z) = (P, Q, R), where P, Q, R are themselves ordinary scalar functions of position giving the arrow's three components. Picture the river: at every point sits a little arrow showing which way the water moves and how fast — a whole sea of arrows, sometimes swirling, sometimes streaming straight. Fields of force and flow like these are exactly what the rung's title promised, and they are the reason vector calculus exists.

Del: a vector you can differentiate with

Now meet the one tool that runs the whole show: the [[del-operator|del operator]], written nabla and read 'del'. It is a single, audacious idea — package the three partial derivatives into a vector. In symbols, nabla = (d/dx, d/dy, d/dz): an arrow whose three slots are not numbers but the three instructions 'differentiate with respect to x', 'with respect to y', 'with respect to z'. Recall from Volume I that a partial derivative d/dx measures how a multivariable quantity changes as you nudge only x and freeze the rest; nabla simply gathers all three such nudges into one object that points 'the direction of differentiation' the way an ordinary vector points a direction in space.

Here is why that packaging is so powerful. A vector can be combined with another quantity in exactly three algebraic ways, and nabla obeys all three. You can multiply a vector onto a scalar (scalar times vector); you can take the dot product of two vectors (vector dot vector, giving a number); and you can take the cross product of two vectors (vector cross vector, giving a new vector). Apply each of these to nabla and you generate, in one stroke, the three fundamental derivatives of vector calculus: nabla f is the gradient, nabla dot F is the divergence, and nabla cross F is the curl. Three ways to use a vector, three derivatives — that is the entire architecture, and the next three sections unpack each in turn.

Gradient: the direction of steepest slope

Point nabla at a scalar field and you get the gradient, nabla f = (df/dx, df/dy, df/dz). It takes a landscape and hands back, at every point, an arrow. You met it in Volume I as the gradient of a multivariable function; the picture that makes it stick is the hill. Let f be the height of a hill above each point of ground. Then nabla f at a spot is the arrow pointing in the direction of [[gradient-steepest-ascent|steepest ascent]] — the way a marble would roll if rolling uphill — and its length is exactly how steep that steepest climb is. Where the hill is flat the gradient is the zero vector; where it plunges, the gradient is long.

Two facts make the gradient indispensable, and both flow straight from that picture. First, the gradient always sits perpendicular to the level sets — the contour lines of constant height — because along a contour the height does not change at all, so the steepest-change arrow must point straight across the contours, not along them. Second, the gradient is the master key to all directional slopes: the rate of change of f as you walk in any unit direction u is just the dot product nabla f dot u, a fact the directional derivative made precise in the multivariable rung. The single gradient vector thus contains the slope in every direction at once — dot it with whichever way you choose to walk.

Divergence: how much a flow spreads out

Now dot nabla into a vector field and you get the [[calc-divergence|divergence]], a single scalar at each point: nabla dot F = dP/dx + dQ/dy + dR/dz. It takes a flow and hands back a number that measures spreading. Picture the river again, and zoom in on one tiny imaginary box of water sitting in the current. The divergence at that spot asks: is more water leaving the box than entering it? A positive divergence means net outflow — the point acts like a tiny source, a faucet adding fluid. A negative divergence means net inflow — a sink, a drain swallowing it. Zero divergence means whatever flows in flows back out, the box neither gains nor loses; such a flow is called incompressible.

Notice the type change, and savour it: divergence eats a vector field (an arrow per point) and returns a scalar field (a number per point). That is the dot product doing its job — a dot product of two vectors is always a number. The number it computes is genuinely local: nabla dot F at a point depends only on how the arrows are changing in the immediate neighbourhood of that point, the infinitesimal version of 'flow out minus flow in.' This local 'spreading rate' is the seed of one of the rung's great theorems to come — the divergence theorem, which will add up all this microscopic spreading inside a region to get the total flux out through its surface. For now, simply read divergence as the source-or-sink strength at each point.

Curl: how much a flow spins

The third and last way to use the vector nabla is the cross product, and it gives the [[curl|curl]], nabla cross F — a vector field again, because a cross product of two vectors is a vector. Where divergence measured spreading, curl measures spinning. Drop a tiny paddlewheel into the flow at a point. If the water pushes harder on one side than the other, the paddlewheel turns. The curl at that point is the arrow that records this local rotation: it points along the axis the paddlewheel spins about (by the right-hand rule — curl your right fingers with the spin and the thumb gives the direction), and its length is twice the local rate of turning. Where the flow has no swirl the curl is zero, and the field is called irrotational.

A common trap is worth flagging at once: curl is about LOCAL spinning, not about whether the overall flow goes in a circle. Water swirling tightly down a plughole and water gliding in a wide straight river can both have curl, or both lack it, depending on the shear — whether neighbouring streamlines move at different speeds. A perfectly circular flow can be curl-free if the outer water lags just enough, and a dead-straight flow can have curl if one bank runs faster than the other, twisting the little paddlewheel. The test is always the same: would a freely floating paddlewheel rotate about its own centre? That, not the shape of the streamlines, is what curl detects.

Putting the three together

ONE OPERATOR, THREE DERIVATIVES   ( nabla = (d/dx, d/dy, d/dz) )

  gradient    nabla f       scalar field  ->  vector field   (slope, steepest ascent)
  divergence  nabla . F     vector field  ->  scalar field   (spreading: source/sink)
  curl        nabla x F     vector field  ->  vector field   (rotation: local spin)

  Two identities that are always true (for smooth fields):

    curl of a gradient  =  0        nabla x (nabla f) = 0     gradients never swirl
    div of a curl       =  0        nabla . (nabla x F) = 0   curls never spread

  Divergence of a gradient = the Laplacian:

    nabla . (nabla f) = nabla^2 f = d2f/dx2 + d2f/dy2 + d2f/dz2

Two of the [[vector-identities|vector identities]] in that box deserve a moment, because they are not accidents — they are the dot and cross products of nabla with itself, and they fall out of the same algebra. The curl of any gradient is zero, nabla cross (nabla f) = 0, which echoes that a vector crossed with itself vanishes: a field that is a pure slope can never swirl. The divergence of any curl is zero, nabla dot (nabla cross F) = 0, echoing that a dot-then-cross of one vector vanishes: a field that is pure rotation can never spread. These two facts are the quiet hinges on which the integral theorems and the whole notion of a potential later turn.

There is one more combination, and it is the most important operator in mathematical physics: the divergence of a gradient, nabla dot (nabla f) = nabla^2 f, the Laplacian. It sums the pure second derivatives d2f/dx2 + d2f/dy2 + d2f/dz2 and measures how much a value at a point differs from the average of its neighbours — the heart of diffusion, of steady temperature, of electrostatics. You will meet the related vector Laplacian too. The Laplacian is why these operators matter beyond bookkeeping: it sits at the centre of the great partial differential equations — the heat equation, the wave equation, the Laplace equation — that the rest of this volume is built to solve.

1. Identify the field's type first: a scalar field f (one number per point) or a vector field F (one arrow per point). This decides which operator can even act.
2. For a scalar field, take the gradient nabla f to get the steepest-slope arrow — a vector field perpendicular to the level sets.
3. For a vector field, take the divergence nabla dot F (a scalar: net spreading) and the curl nabla cross F (a vector: local spin).
4. Combine when needed: gradient-then-divergence gives the Laplacian nabla^2 f; and remember curl-of-gradient and divergence-of-curl are always zero.