The Gradient Operator

A scalar field assigns a number to every point in space — temperature, pressure, or electric potential. We often want to know not just the value but how fast and in which direction that value changes. The gradient operator answers exactly this: applied to a scalar field $T$ , it produces a vector field $\nabla T$ that encodes the complete local rate-of-change information at every point. The defining property is the relation $\mathrm{d}T = \nabla T \cdot \mathrm{d}\mathbf{l}$ . Moving an infinitesimal displacement $\mathrm{d}\mathbf{l}$ changes $T$ by an amount that is the projection of $\nabla T$ onto that displacement. Because a dot product is maximized when the two vectors are parallel, $T$ increases fastest when you move along $\nabla T$ , and the rate of that fastest increase is $|\nabla T|$ . Move perpendicular to $\nabla T$ and $T$ does not change at all — which is why the gradient is normal to level surfaces. The gradient is the first of the three derivative operations built from the del operator $\nabla$ ; the divergence and curl, which act on vector fields, come next. In electrostatics the gradient is indispensable: the electric field is conservative, so it can be written as $\mathbf{E} = -\nabla V$ , where $V$ is the scalar potential. This single relation reduces the three-component vector problem of finding $\mathbf{E}$ to the simpler scalar problem of finding $V$ — one of the great computational economies of the subject, and the reason an entire later chapter is devoted to solving for potentials.

Key takeaways

The gradient $\nabla T$ of a scalar field $T(x,y,z)$ is a vector field whose components are the partial derivatives of $T$ along each coordinate axis.
Geometrically, $\nabla T$ points in the direction of steepest increase of $T$ , and its magnitude $|\nabla T|$ equals the maximum rate of change per unit distance.
The infinitesimal change $\mathrm{d}T = \nabla T \cdot \mathrm{d}\mathbf{l}$ packages all directional derivatives into a single dot product.
The gradient is everywhere perpendicular to surfaces of constant $T$ (level surfaces / equipotentials).
In electromagnetism the electric field is (minus) the gradient of the scalar potential, $\mathbf{E} = -\nabla V$ , making the gradient the bridge between potentials and forces.
The gradient is coordinate-independent as a geometric object, but its component formula changes between Cartesian, spherical, and cylindrical systems.
Key idea: The gradient converts a scalar field into the vector field of its steepest ascent: $\nabla T$ points where $T$ rises fastest and has magnitude equal to that maximal slope, with $\mathrm{d}T = \nabla T\cdot\mathrm{d}\mathbf{l}$ tying it to every directional derivative at once.

Where this lives today

Every MRI scanner in every hospital is a gradient machine: three "gradient coils" impose precisely controlled linear variations of the magnetic field strength — a few tens of millitesla per metre — so that spins at different positions precess at different frequencies. The scanner reads position out of frequency, which works only because a linear field $B(\mathbf{r}) = B_0 + \mathbf{G}\cdot\mathbf{r}$ has constant gradient $\mathbf{G}$ , the first row of our limiting-cases table. The rapid-fire switching of those gradients is the loud knocking you hear during a scan. The other ubiquitous incarnation is gradient descent. Training a neural network means minimizing a loss function over billions of parameters, and the algorithm is nothing but "compute $\nabla L$ , step along $-\nabla L$ , repeat" — the steepest-descent geometry of equation (3) in a space of astronomical dimension. Meteorology completes the trio: the pressure-gradient force $-\nabla p/\rho$ is what accelerates every wind on Earth, before rotation (the mechanics track's Coriolis force) bends it along the isobars.

The physical picture

Picture a hill whose height above sea level is $T(x,y)$ . Standing at any point, $\nabla T$ is the arrow pointing straight uphill in the steepest direction, and its length tells you how steep that climb is. Walk along a contour line (constant height) and you feel no change — that direction is perpendicular to $\nabla T$ . Every hiking map you have ever read is a gradient diagram in disguise: where contours crowd together the gradient is long and the terrain is steep; where they spread apart the gradient is short and the ground is nearly flat.

Figure

The gradient on a contour map — always perpendicular to the contours, always pointing uphill, longest where they crowd

The gradient bundles infinitely many directional derivatives into one object. Ask 'how fast does $T$ change if I head northeast?' and the answer is $\nabla T\cdot\hat{\mathbf{n}}_{NE}$ — a single projection. The gradient is the one vector that reproduces every such slope by a dot product. Think of $\nabla T$ as the 'best linear approximation' to $T$ : near a point, $T(\mathbf{r}+\mathrm{d}\mathbf{l})\approx T(\mathbf{r})+\nabla T\cdot\mathrm{d}\mathbf{l}$ — the multivariable analogue of the ordinary derivative $f'(x)$ . In physics the minus sign matters: forces point downhill in potential energy, so $\mathbf{F}=-\nabla U$ and $\mathbf{E}=-\nabla V$ . Charges accelerate toward lower potential energy, in the direction of steepest decrease — opposite to the gradient. This is the E&M payoff to keep in view throughout: once we know the scalar landscape $V$ , the gradient hands us the full electric field, and the entire boundary-value-problem chapter of this track is the art of finding that landscape.

The gradient in components

In Cartesian coordinates the gradient simply collects the three partial derivatives into a vector. If $T$ measures temperature in a room, each component asks "how fast does the temperature change as I move along this axis, holding the others fixed?":

\nabla T = \frac{\partial T}{\partial x}\,\hat{\mathbf{x}} + \frac{\partial T}{\partial y}\,\hat{\mathbf{y}} + \frac{\partial T}{\partial z}\,\hat{\mathbf{z}} \qquad (1)

where: $T$ — scalar field (e.g. potential, temperature); $\nabla T$ — gradient, a vector field; $\hat{\mathbf{x}},\hat{\mathbf{y}},\hat{\mathbf{z}}$ — Cartesian unit basis vectors; $\partial/\partial x$ — partial derivative holding $y,z$ fixed. Note the type change: scalar in, vector out. A field with steep $x$ -variation and no $y$ -variation gets a gradient pointing along $\hat{\mathbf{x}}$ — the formula automatically aims the arrow at the fastest climb.

The master relation

Why is that particular combination of partials the right vector? Because it is exactly what the chain rule produces when you ask how $T$ changes under a small step. Take a displacement $\mathrm{d}\mathbf{l} = \mathrm{d}x\,\hat{\mathbf{x}}+\mathrm{d}y\,\hat{\mathbf{y}}+\mathrm{d}z\,\hat{\mathbf{z}}$ . The total differential of $T(x,y,z)$ is $\mathrm{d}T = (\partial T/\partial x)\mathrm{d}x + (\partial T/\partial y)\mathrm{d}y + (\partial T/\partial z)\mathrm{d}z$ — and that sum is precisely a dot product:

\mathrm{d}T = \nabla T \cdot \mathrm{d}\mathbf{l} \qquad (2)

where: $\mathrm{d}T$ — infinitesimal change in $T$ for the step; $\mathrm{d}\mathbf{l}$ — infinitesimal displacement vector. This is the definition worth memorizing — it is coordinate-free, and every component formula (Cartesian, spherical, cylindrical) is just equation (2) unpacked in a particular coordinate system. Writing the dot product in terms of magnitudes and the angle $\theta$ between the two vectors exposes the geometry:

\mathrm{d}T = |\nabla T|\,|\mathrm{d}\mathbf{l}|\cos\theta \qquad (3)

where: $\theta$ — angle between $\nabla T$ and the step $\mathrm{d}\mathbf{l}$ . For a fixed step length, $\mathrm{d}T$ is largest at $\theta=0$ (walk along the gradient: steepest ascent, rate $|\nabla T|$ ), zero at $\theta=90^\circ$ (walk along a level surface), and most negative at $\theta=180^\circ$ (steepest descent). One equation, the entire hill-walking phenomenology. The rate of change along an arbitrary unit direction $\hat{\mathbf{n}}$ — the directional derivative — is then a single projection:

D_{\hat{\mathbf{n}}}T = \nabla T \cdot \hat{\mathbf{n}} \qquad (4)

where: $\hat{\mathbf{n}}$ — unit vector specifying the direction of travel; $D_{\hat{\mathbf{n}}}T$ — slope of $T$ per unit length along $\hat{\mathbf{n}}$ . Three stored numbers reproduce infinitely many slopes; this economy is why the gradient, not the list of directional derivatives, is the fundamental object.

Figure

One gradient vector, every slope — the directional derivative is a projection

Derivation AThe gradient is normal to level surfaces

We are proving that $\nabla T$ is everywhere perpendicular to the surfaces of constant $T$ — the fact that makes electric field lines cross equipotentials at right angles. The strategy is a one-line application of the master relation (2): choose displacements that stay inside a level surface, note that $T$ cannot change along them, and see what that forces on the dot product. The only subtlety is the final logical step — a vector orthogonal to every tangent direction of a surface has no component left except along the normal.

Step 1. Let $\mathrm{d}\mathbf{l}$ be any infinitesimal displacement lying within the surface $T(x,y,z)=c$ . Since $T$ is constant on that surface,

\mathrm{d}T = 0

Why: the displacement connects two points that both satisfy $T=c$ , so the change in $T$ between them vanishes to first order.

Step 2. Apply the defining relation $\mathrm{d}T=\nabla T\cdot\mathrm{d}\mathbf{l}$ :

0 = \nabla T\cdot\mathrm{d}\mathbf{l} \quad \text{for every tangent } \mathrm{d}\mathbf{l}

Why: equation (2) holds for arbitrary steps; restricting to in-surface steps turns it into an orthogonality statement.

Step 3. A nonzero vector whose dot product with every tangent vector of a surface is zero must point along the surface normal. Hence $\nabla T \perp \{T=c\}$ everywhere. With $\mathbf{E}=-\nabla V$ this is the geometric fact drawn in every electrostatics figure: field lines pierce equipotential surfaces perpendicularly. Verified: standard result, Griffiths §1.2.2; for $T=x^2+y^2+z^2$ (spheres) the gradient $2\mathbf{r}$ is indeed radial.

Beyond Cartesian: where the scale factors come from

The geometric definition (2) is coordinate-free, so in spherical coordinates the gradient must still satisfy $\mathrm{d}T = \nabla T\cdot\mathrm{d}\mathbf{l}$ — but the displacement now involves arc lengths, not bare angle increments. The result is:

\nabla T = \frac{\partial T}{\partial r}\hat{\mathbf{r}} + \frac{1}{r}\frac{\partial T}{\partial\theta}\hat{\boldsymbol{\theta}} + \frac{1}{r\sin\theta}\frac{\partial T}{\partial\phi}\hat{\boldsymbol{\phi}} \qquad (5)

where: $r,\theta,\phi$ — radial distance, polar angle, azimuthal angle; the factors $1/r$ and $1/(r\sin\theta)$ — metric scale factors converting "change per radian" into "change per metre". Forgetting them is the most common curvilinear-coordinates error in the course; the derivation below shows they are forced by dimensional honesty alone.

Go deeper · Derivation B · 1 minThe spherical gradient from the line element

Derivation BThe spherical gradient from the line element

We are deriving the component formula (5), and in particular where the $1/r$ and $1/(r\sin\theta)$ factors come from. The strategy: write the chain-rule expansion of $\mathrm{d}T$ in the spherical variables, write the honest displacement vector $\mathrm{d}\mathbf{l}$ (whose angular pieces are arc lengths, not angles), and then demand that the coordinate-free definition $\mathrm{d}T = \nabla T\cdot\mathrm{d}\mathbf{l}$ hold for arbitrary steps. Watch the units throughout: a gradient component must be a rate of change per unit length, but $\partial T/\partial\theta$ is a rate per radian — the scale factors are exactly the exchange rate between the two.

Step 1. Chain rule in the spherical variables:

\mathrm{d}T = \frac{\partial T}{\partial r}\mathrm{d}r + \frac{\partial T}{\partial\theta}\mathrm{d}\theta + \frac{\partial T}{\partial\phi}\mathrm{d}\phi

Why: $T$ is a function of the three coordinates, so its total differential runs over all three, whatever geometric meaning they carry.

Step 2. Write the displacement for infinitesimal coordinate changes. Increasing $\theta$ by $\mathrm{d}\theta$ moves you along an arc of radius $r$ , and increasing $\phi$ by $\mathrm{d}\phi$ moves you along a circle of radius $r\sin\theta$ :

\mathrm{d}\mathbf{l} = \mathrm{d}r\,\hat{\mathbf{r}} + r\,\mathrm{d}\theta\,\hat{\boldsymbol{\theta}} + r\sin\theta\,\mathrm{d}\phi\,\hat{\boldsymbol{\phi}}

Why: arc length is radius times angle; the azimuthal circle at polar angle $\theta$ has radius $r\sin\theta$ , which is why a step in longitude near the pole covers almost no distance.

Step 3. Demand $\mathrm{d}T=\nabla T\cdot\mathrm{d}\mathbf{l}$ for independent $\mathrm{d}r, \mathrm{d}\theta, \mathrm{d}\phi$ and match coefficients:

(\nabla T)_r=\frac{\partial T}{\partial r},\qquad (\nabla T)_\theta=\frac{1}{r}\frac{\partial T}{\partial\theta},\qquad (\nabla T)_\phi=\frac{1}{r\sin\theta}\frac{\partial T}{\partial\phi}

The scale factors land in the denominators so that each component is a true change-per-unit-length. The same recipe with $\mathrm{d}\mathbf{l} = \mathrm{d}s\,\hat{\mathbf{s}} + s\,\mathrm{d}\phi\,\hat{\boldsymbol{\phi}} + \mathrm{d}z\,\hat{\mathbf{z}}$ yields the cylindrical gradient. Verified: matches the spherical gradient on the Griffiths front cover and the Jackson appendix; for $T=T(r)$ it collapses to $\hat{\mathbf{r}}\,\mathrm{d}T/\mathrm{d}r$ as symmetry demands.

The electrostatic payoff

The single most important gradient in this track converts the potential landscape into the electric field:

\mathbf{E} = -\nabla V \qquad (6)

where: $\mathbf{E}$ — electric field; $V$ — electric scalar potential; the minus sign is the convention that makes $\mathbf{E}$ point from high potential toward low, so that positive charges fall "downhill". One scalar function replaces three coupled components — and because the gradient is normal to level surfaces, equipotential maps and field-line maps are two views of the same data. A gradient field is special in one more structural way. Compute the curl of any gradient and the mixed second partials cancel in pairs, e.g. $[\nabla\times(\nabla T)]_x = \partial_y\partial_z T-\partial_z\partial_y T = 0$ by Clairaut's theorem. Hence the identity

\nabla\times(\nabla T)=\mathbf{0} \qquad (7)

where: $\nabla\times$ — curl operator, developed fully in its own article. Every gradient field is irrotational. Read forward, this is a consistency check: since $\mathbf{E}=-\nabla V$ in electrostatics, we must have $\nabla\times\mathbf{E}=0$ — which is exactly the electrostatic Maxwell equation. Read backward, it is the deeper claim (proved via Stokes's theorem later): any field with vanishing curl can be written as a gradient. That equivalence is what legitimizes the potential in the first place.

Go deeper · Derivation C · 1 minGradient of one over r and the Coulomb field

Derivation CGradient of one over r and the Coulomb field

We are computing $\nabla(1/r)$ , the single most used gradient in electrostatics: with the Coulomb potential attached it produces the inverse-square field, and in the potentials chapter it reappears inside every superposition integral. The route is the chain rule executed carefully in Cartesian components, then reassembled into radial form. The step to watch is $\partial r/\partial x = x/r$ — differentiating the square root correctly — since a slip there corrupts the power of $r$ in the final answer.

Step 1. Differentiate $r=\sqrt{x^2+y^2+z^2}$ itself:

\frac{\partial r}{\partial x} = \frac{x}{r}

Why: from $r^2=x^2+y^2+z^2$ , implicit differentiation gives $2r\,\partial_x r = 2x$ ; the same pattern holds for $y$ and $z$ .

Step 2. Chain rule on $1/r$ :

\frac{\partial}{\partial x}\frac{1}{r} = -\frac{1}{r^2}\frac{\partial r}{\partial x} = -\frac{x}{r^3}

Why: $\mathrm{d}(1/r)/\mathrm{d}r=-1/r^2$ , then insert Step 1; each Cartesian component carries its own coordinate in the numerator.

Step 3. Assemble the three components into a vector:

\nabla\frac{1}{r} = -\frac{x\hat{\mathbf{x}}+y\hat{\mathbf{y}}+z\hat{\mathbf{z}}}{r^3} = -\frac{\mathbf{r}}{r^3} = -\frac{\hat{\mathbf{r}}}{r^2}

Why: the numerator is the position vector $\mathbf{r}=r\hat{\mathbf{r}}$ , so one power of $r$ cancels, leaving the clean inverse-square radial form.

Step 4. Attach the Coulomb potential $V=q/(4\pi\varepsilon_0 r)$ and apply equation (6):

\mathbf{E} = -\nabla\frac{q}{4\pi\varepsilon_0 r} = \frac{1}{4\pi\varepsilon_0}\frac{q}{r^2}\,\hat{\mathbf{r}}

The gradient reproduces Coulomb's law exactly — the first full round trip from scalar landscape to physical force field, with $\varepsilon_0$ the vacuum permittivity. Verified: reproduces Coulomb's law; Griffiths §1.2.3 problems and Jackson §1.5. Quick check via the spherical formula (5): $V\propto 1/r$ depends only on $r$ , so $\nabla V = (\mathrm{d}V/\mathrm{d}r)\hat{\mathbf{r}} = -(q/4\pi\varepsilon_0 r^2)\hat{\mathbf{r}}$ — same answer in one line.

Limiting cases

How the gradient behaves in special configurations:

Case	Result	Meaning
Uniform field	$T = \mathbf{a}\cdot\mathbf{r}+b \;\Rightarrow\; \nabla T = \mathbf{a}$	A linear scalar field has constant gradient — equipotentials are equally spaced parallel planes, as in a parallel-plate capacitor.
Spherically symmetric	$T=T(r) \;\Rightarrow\; \nabla T = \frac{\mathrm{d}T}{\mathrm{d}r}\,\hat{\mathbf{r}}$	The gradient is purely radial; for the Coulomb potential $V\propto 1/r$ this gives $\mathbf{E}\propto\hat{\mathbf{r}}/r^2$ .
At an extremum	$\nabla T = \mathbf{0}$	At a maximum, minimum, or saddle the gradient vanishes — every directional derivative is zero. These are the equilibrium points of $\mathbf{F}=-\nabla U$ .
Constant field	$T=\text{const} \;\Rightarrow\; \nabla T = \mathbf{0}$	No change in any direction means the gradient is identically zero everywhere.
Crowded contours	level surfaces spacing $\Delta s$ at fixed $\Delta T$	$\vert\nabla T\vert \approx \Delta T/\Delta s$ : halving the spacing doubles the gradient — read field strength straight off an equipotential map.

Worked examples

Example 1 — hill walking with numbers (confidence builder). A hill has height $h(x,y) = 50 - 0.5x^2 - 2y^2$ metres, with $x,y$ in metres. You stand at $(x,y) = (3, 1)$ .

Step 1. Compute the gradient components: $\partial h/\partial x = -x$ and $\partial h/\partial y = -4y$ , so at $(3,1)$ ,

\nabla h = -3\,\hat{\mathbf{x}} - 4\,\hat{\mathbf{y}}

Step 2. The steepest ascent direction is along $\nabla h$ : the unit vector $(-3\hat{\mathbf{x}}-4\hat{\mathbf{y}})/5$ , i.e. back toward the summit at the origin. The maximum slope is

|\nabla h| = \sqrt{(-3)^2+(-4)^2} = 5 \quad \text{— a } 5\ \mathrm{m\ rise\ per\ metre\ of\ horizontal\ travel}

Step 3. Heading "northeast", $\hat{\mathbf{n}} = (\hat{\mathbf{x}}+\hat{\mathbf{y}})/\sqrt{2}$ , the slope is the projection

D_{\hat{\mathbf{n}}}h = \nabla h\cdot\hat{\mathbf{n}} = \frac{-3-4}{\sqrt 2} = -4.95

You descend at 4.95 m per metre travelled — nearly the steepest possible descent, because northeast happens to lie almost opposite the gradient here.

Example 2 — electric field from a potential (exam level). In a region of space the potential is $V(x,y,z) = 5x^2y - 3z^2$ volts, with coordinates in metres. Find $\mathbf{E}$ at $P=(1, 2, 1)$ and the rate of change of $V$ at $P$ toward the origin.

Step 1. Take the gradient term by term: $\partial V/\partial x = 10xy$ , $\partial V/\partial y = 5x^2$ , $\partial V/\partial z = -6z$ . Then

\mathbf{E} = -\nabla V = -10xy\,\hat{\mathbf{x}} - 5x^2\,\hat{\mathbf{y}} + 6z\,\hat{\mathbf{z}}

Step 2. Evaluate at $P=(1,2,1)$ :

\mathbf{E}(P) = -20\,\hat{\mathbf{x}} - 5\,\hat{\mathbf{y}} + 6\,\hat{\mathbf{z}}\ \mathrm{V/m}, \qquad |\mathbf{E}| = \sqrt{400+25+36} = \sqrt{461} = 21.5\ \mathrm{V/m}

Step 3. The unit vector from $P$ toward the origin is $\hat{\mathbf{n}} = -(1,2,1)/\sqrt{6}$ . The directional derivative of $V$ (note: of $V$ , so use $\nabla V = +\,(20, 5, -6)$ at $P$ ):

D_{\hat{\mathbf{n}}}V = \nabla V\cdot\hat{\mathbf{n}} = \frac{-(20)(1)-(5)(2)-(-6)(1)}{\sqrt 6} = \frac{-20-10+6}{\sqrt 6} = \frac{-24}{2.449} = -9.80\ \mathrm{V/m}

The potential drops at 9.80 V per metre as you head toward the origin — less steep than the maximal rate $|\nabla V| = 21.5$ V/m, exactly as the projection picture demands.

Applications

Electric field from potential. The cornerstone application: for any electrostatic configuration the field is $\mathbf{E}=-\nabla V$ . Computing the scalar $V$ (one function) and taking its gradient is far easier than directly summing the three vector components of $\mathbf{E}$ . The minus sign means $\mathbf{E}$ points 'downhill' in potential, and field lines cross equipotential surfaces at right angles — a direct consequence of the gradient being normal to level surfaces.

Forces from potential energy. In mechanics a conservative force is $\mathbf{F}=-\nabla U$ . The gradient of the potential energy gives the force, equilibria occur where $\nabla U=\mathbf{0}$ , and the stability of an equilibrium is read off from the second derivatives (the Hessian / Laplacian). Earnshaw's theorem in electrostatics — that no static charge configuration can be in stable equilibrium in free space — follows from properties of $\nabla V$ and $\nabla^2 V=0$ .

Heat and diffusion flow. Fourier's law of heat conduction states that the heat flux is $\mathbf{q}=-k\nabla T$ , proportional to the negative temperature gradient: heat flows from hot to cold, down the steepest temperature drop. The same gradient structure appears in Fick's law of diffusion and in Ohm's law in continuous media, $\mathbf{J}=\sigma\mathbf{E}=-\sigma\nabla V$ .

Optimization and steepest descent. Because $\nabla T$ points toward fastest increase, $-\nabla T$ points toward fastest decrease. Numerical algorithms that minimize a function (gradient descent) repeatedly step in the $-\nabla$ direction. The same geometric fact underlies how charges and fields relax toward minimum-energy configurations.

Origins · 1 minJosiah Willard Gibbs and how this was discovered

Josiah Willard Gibbs

1839–1903 · Yale University, New Haven

Gibbs spent his entire career in a single office at Yale, publishing in an obscure Connecticut journal, and quietly rebuilt the mathematical language of physics twice — once with thermodynamic potentials, once with vector analysis. In the early 1880s, teaching electromagnetism from Maxwell's quaternion-laden Treatise, he decided the quaternions were scaffolding, not structure: what physics actually used were two products (dot and cross) and one differential operator. His privately printed Elements of Vector Analysis (1881–84) organized everything around $\nabla$ — gradient, divergence, curl — in essentially the notation this article uses. Heaviside reached the same system independently in England, and after a spirited pamphlet war with the quaternionists, the Gibbs–Heaviside language won. When his student E. B. Wilson expanded the notes into the 1901 textbook Vector Analysis, the gradient acquired the form every physics student now learns. The steepest-ascent vector on your contour map is Gibbs's tidying of Hamilton's wilder invention.

History

1828

— Green's essay develops potential functions for electricity, making "the function whose derivative is the force" a central object

1843

— Hamilton invents quaternions; buried inside them is the operator that will become del

1846

— Thomson (Kelvin) uses the flow of heat down temperature gradients to model electrostatic problems, cementing the gradient-flux analogy

1873

— Maxwell's Treatise phrases electromagnetism partly in quaternion form, computing what he calls the "slope" of the potential

1881

— Gibbs's Elements of Vector Analysis isolates the gradient as one of three fundamental del operations; Heaviside does the same independently

1901

— Wilson's Vector Analysis, from Gibbs's lectures, fixes the modern notation and makes the gradient standard curriculum

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §1.2.2–1.2.3 · Jackson, Classical Electrodynamics, 3rd ed., §1.5 and front-matter vector identities · Purcell & Morin, Electricity and Magnetism, 3rd ed., §2.3 · Zangwill, Modern Electrodynamics, §1.3

Provenance. All formulas are standard vector-calculus results. A verifier should: (1) confirm the Cartesian and spherical gradient component formulas against Griffiths front cover / §1.2; (2) recompute $\nabla(1/r)=-\hat{\mathbf{r}}/r^2$ and check it yields Coulomb's law with $\varepsilon_0$ ; (3) verify $\nabla\times(\nabla T)=0$ via equality of mixed partials; (4) check the worked example $\nabla(x^2y)=2xy\,\hat{\mathbf{x}}+x^2\,\hat{\mathbf{y}}$ . Cited: Griffiths Introduction to Electrodynamics 4th ed. §1.2.2–1.2.3; Jackson Classical Electrodynamics 3rd ed. §1.5; Purcell & Morin 3rd ed. §2.3; Zangwill §1.3.

Builds on: Scalar and Vector Fields · The Del Operator (Nabla) Leads to: The Fundamental Theorem for Gradients · The Divergence of a Vector Field · The Curl of a Vector Field · Second Derivatives and the Laplacian

Is the gradient the same thing as the derivative?

It is the honest multivariable generalization of it. In one dimension, $f'(x)$ tells you everything about local change. In three dimensions no single number can, because the rate of change depends on direction — so the derivative information gets packaged as a vector. The master relation $\mathrm{d}T=\nabla T\cdot\mathrm{d}\mathbf{l}$ is the 3D twin of $\mathrm{d}f = f'\,\mathrm{d}x$ : gradient dotted with step replaces derivative times step. Every property of the gradient (steepest ascent, normal to level sets) is that one formula unpacked.

Why is there a minus sign in E = −∇V but not in the definition of the gradient?

The gradient itself has no sign convention — it points uphill by construction. The minus sign in $\mathbf{E}=-\nabla V$ (and $\mathbf{F}=-\nabla U$ ) is physics, not mathematics: we want positive charges to be pushed from high potential to low, and systems to feel forces toward lower potential energy. Without the minus sign, "downhill" motion would raise the energy and nothing would be stable. Fourier's law $\mathbf{q}=-k\nabla T$ carries the same sign for the same reason: heat flows from hot to cold.

Does the direction of the gradient depend on the coordinate system I use?

No — and this is the point of defining it through $\mathrm{d}T=\nabla T\cdot\mathrm{d}\mathbf{l}$ rather than through components. The gradient at a point is a geometric arrow: it points along the steepest increase of the field, a fact about the landscape, not about your axes. What changes between Cartesian, spherical and cylindrical systems is only the recipe for computing its components, including those scale factors $1/r$ and $1/(r\sin\theta)$ . Same arrow, different bookkeeping.

What happens to the gradient at the top of the hill, and what does that mean in electrostatics?

At a smooth maximum, minimum, or saddle, all directional derivatives vanish, so $\nabla T=\mathbf{0}$ — there is no "uphill" at the summit. For potentials this marks an equilibrium: a charge placed where $\nabla V=\mathbf{0}$ feels no force. Whether the equilibrium is stable depends on second derivatives, and here electrostatics has a sting: in charge-free space $V$ satisfies Laplace's equation, which forbids true maxima or minima of $V$ — every equilibrium is a saddle. That is Earnshaw's theorem, and it is why you cannot trap a charge with static fields alone.

How do I take the gradient of a function like V(r) that depends only on distance from the origin?

Use the spherical form and let symmetry kill two of the three terms: $\nabla V = (\mathrm{d}V/\mathrm{d}r)\,\hat{\mathbf{r}}$ . No angular derivatives, no scale factors to remember. This one-liner handles the Coulomb potential, the Yukawa potential, and every other central field. The classic mistake is doing it in Cartesian components and mangling the chain rule — if you must, remember $\partial r/\partial x = x/r$ and expect the answer to reassemble into something radial.

Is the directional derivative ever larger than the magnitude of the gradient?

Never. By Cauchy–Schwarz (or just $\cos\theta \le 1$ in equation 3), $|D_{\hat{\mathbf{n}}}T| = |\nabla T\cdot\hat{\mathbf{n}}| \le |\nabla T|$ , with equality only when $\hat{\mathbf{n}}$ is parallel or antiparallel to the gradient. The gradient's magnitude is the envelope of all possible slopes at that point. If an exam answer for a directional derivative exceeds $|\nabla T|$ , a dot product went wrong — usually a forgotten normalization of $\hat{\mathbf{n}}$ .

Why do the spherical gradient components have those 1/r factors when the Cartesian ones do not?

Because coordinates and distances are the same thing in Cartesian systems but not in angular ones. Changing $\theta$ by one radian moves you a distance $r$ — the farther out you are, the more ground a radian covers. A gradient component must report change per metre, so the per-radian derivative $\partial T/\partial\theta$ has to be divided by the metres-per-radian factor $r$ (and $r\sin\theta$ for longitude). Cartesian coordinates already measure metres, so their factors are all 1. If you remember the displacement $\mathrm{d}\mathbf{l} = \mathrm{d}r\,\hat{\mathbf{r}} + r\,\mathrm{d}\theta\,\hat{\boldsymbol\theta} + r\sin\theta\,\mathrm{d}\phi\,\hat{\boldsymbol\phi}$ , you never need to memorize the gradient formula at all.