Second Derivatives and the Laplacian

Once the first-order vector operators are in hand — the gradient acting on a scalar field, the divergence and curl acting on a vector field — it is natural to ask what happens when we apply $\nabla$ twice. Because $\nabla$ takes a scalar to a vector (gradient) and a vector to either a scalar (divergence) or a vector (curl), the dimensional bookkeeping severely restricts which second derivatives are meaningful. Five distinct combinations survive that bookkeeping, and the single most important among them, the Laplacian, governs an enormous fraction of mathematical physics. The Laplacian $\nabla^2$ is defined as the divergence of the gradient. Acting on a scalar field it produces a scalar; in Cartesian coordinates it is the sum of unmixed second partial derivatives. Its central role stems from the fact that the basic equations of electrostatics, magnetostatics (for the vector potential), heat conduction, diffusion, and even the wave equation are all expressed through it. Understanding the Laplacian therefore unlocks Poisson's and Laplace's equations and, with them, the entire boundary-value machinery of electromagnetism. Of the five second derivatives, two vanish identically as consequences of the equality of mixed partials — the curl of any gradient and the divergence of any curl — and these underpin the scalar potential $V$ and the vector potential $\mathbf{A}$ . Geometrically the Laplacian measures the difference between the value of a field at a point and its average over a tiny surrounding sphere: where $\nabla^2 f > 0$ the point sits below its neighbourhood average (a local valley), and where $\nabla^2 f < 0$ it sits above (a local peak). This averaging interpretation is exactly why harmonic functions — solutions of Laplace's equation — can have no interior maxima or minima, a fact with deep physical consequences such as Earnshaw's theorem.

Key takeaways

From the first-order operators gradient, divergence, and curl, exactly five second-derivative combinations can be formed, but only a few are independent.
The most important is the Laplacian $\nabla^2 = \nabla\cdot\nabla$ , the divergence of the gradient of a scalar field, which appears throughout electromagnetism.
Two of the five combinations vanish identically: the curl of a gradient $\nabla\times(\nabla V)=0$ and the divergence of a curl $\nabla\cdot(\nabla\times\mathbf{A})=0$ .
The curl of a curl is not a new operator: $\nabla\times(\nabla\times\mathbf{A}) = \nabla(\nabla\cdot\mathbf{A}) - \nabla^2\mathbf{A}$ , which defines the vector Laplacian.
Poisson's equation $\nabla^2 V = -\rho/\varepsilon_0$ and Laplace's equation $\nabla^2 V = 0$ are the central field equations of electrostatics, both built from the Laplacian.
Physically $\nabla^2 f$ measures how the value of $f$ at a point compares to the average over an infinitesimal surrounding neighbourhood.
Key idea: The Laplacian $\nabla^2 = \nabla\cdot\nabla$ is the divergence of the gradient; it measures the excess of a field's neighbourhood average over its central value, and the identities $\nabla\times(\nabla V)=0$ and $\nabla\cdot(\nabla\times\mathbf{A})=0$ guarantee the existence of scalar and vector potentials.

Where this lives today

The Laplacian is the workhorse operator of numerical computing. Chip designers extract parasitic capacitances by solving Laplace's equation around millions of interconnects; astrophysics codes solve Poisson's equation for gravity at every timestep of a galaxy simulation, using multigrid methods whose core move — replace each value by a weighted neighbourhood average — is the discrete equation (7) run as an algorithm. The mean-value property is not just a theorem; it is literally the iteration that relaxation solvers execute until convergence. Two more habitats: in computer vision, the Laplacian (and its Laplacian-of-Gaussian refinement) is the classic edge detector, firing exactly where an image's intensity deviates from its neighbourhood average; and in machine learning, the graph Laplacian — the same centre-minus-neighbours meter defined on a network instead of space — powers spectral clustering and graph neural networks. Anywhere "compare a point to its surroundings" is the question, this operator is the answer.

The physical picture

Think of a scalar field like temperature $T(x,y,z)$ . The Laplacian $\nabla^2 T$ compares the temperature at a point to the average temperature of its immediate surroundings. If the point is colder than its neighbourhood, heat flows inward and $\nabla^2 T > 0$ ; the sign of the Laplacian thus drives diffusion toward equilibrium. It is the mathematical embodiment of "regression to the local mean" — and any field whose Laplacian vanishes everywhere is already perfectly relaxed.

Figure

The Laplacian as an averaging meter — below your neighbours means positive, above them means negative, harmonic means exactly average

The five second derivatives come from a simple type-counting game. $\nabla$ raises a scalar to a vector and lowers a vector to a scalar (divergence) or keeps it a vector (curl). Apply $\nabla$ twice and only the type-consistent combinations survive: from a scalar you can form $\nabla\cdot(\nabla V)$ and $\nabla\times(\nabla V)$ ; from a vector you can form $\nabla(\nabla\cdot\mathbf{A})$ , $\nabla\cdot(\nabla\times\mathbf{A})$ , and $\nabla\times(\nabla\times\mathbf{A})$ . Two of those five are forced to vanish because partial derivatives commute. 'The curl of a gradient is zero' and 'the divergence of a curl is zero' are really the statement $\partial_i\partial_j = \partial_j\partial_i$ dressed up in vector clothing — and they are precisely why a curl-free field admits a potential and a divergence-free field admits a vector potential. The averaging picture explains why harmonic functions ( $\nabla^2 V=0$ ) cannot have local maxima or minima inside a region: a maximum would have to exceed its surrounding average, which Laplace's equation forbids. Electrostatic potential in empty space is harmonic, so a charge cannot be held in stable equilibrium by static fields alone (Earnshaw's theorem). The E&M payoff of this whole article is the pair Poisson/Laplace: two chapters from now, "solve the electrostatics problem" will mean "solve $\nabla^2 V = -\rho/\varepsilon_0$ with the right boundary conditions".

The Laplacian, assembled

Compose the two operators you already own: take the gradient of a scalar field (a vector field of steepest-ascent arrows), then take the divergence of that (asking where the arrows pour out or in). In Cartesian coordinates $\nabla V = (\partial_x V, \partial_y V, \partial_z V)$ , and the divergence sums each component's derivative along its own direction:

\nabla^2 V \equiv \nabla\cdot(\nabla V) = \frac{\partial^2 V}{\partial x^2} + \frac{\partial^2 V}{\partial y^2} + \frac{\partial^2 V}{\partial z^2} \qquad (1)

where: $V$ — a scalar field, e.g. the electrostatic potential; $\nabla^2$ — the Laplacian, divergence of the gradient (also written $\Delta$ ). Only unmixed second partials appear — the Laplacian is the three-dimensional generalization of ordinary concavity $f''(x)$ , one curvature per axis, summed. The two vanishing second derivatives deserve their statement here even though their proofs live in the gradient and divergence articles:

\nabla\times(\nabla V) = \mathbf{0}, \qquad \nabla\cdot(\nabla\times\mathbf{A}) = 0 \qquad (2)

where: $V$ — any twice-differentiable scalar field; $\mathbf{A}$ — any twice-differentiable vector field. Both are the commutativity of mixed partials in vector costume. Their electromagnetic dividend: a curl-free $\mathbf{E}$ may be written $-\nabla V$ , and a divergence-free $\mathbf{B}$ may be written $\nabla\times\mathbf{A}$ — the potentials exist because these two expressions are identically zero. The fifth combination folds back onto the others. In Cartesian index notation (proved in the del-operator article via the epsilon-contraction identity):

\nabla\times(\nabla\times\mathbf{A}) = \nabla(\nabla\cdot\mathbf{A}) - \nabla^2\mathbf{A} \qquad (3)

where: $\nabla^2\mathbf{A}$ — the vector Laplacian, componentwise in Cartesian coordinates only. Read as a definition, equation (3) is how the vector Laplacian is defined in curvilinear systems, where componentwise differentiation fails. Read as a tool, it is the crank that converts Maxwell's two curl equations into the electromagnetic wave equation.

Derivation AThe five second derivatives and why only one is new

We are enumerating every meaningful way to apply del twice and classifying the results — the bookkeeping that explains why "second derivatives" in vector calculus means, in practice, "the Laplacian". The strategy is pure type-checking: track whether each first-order output is a scalar or a vector, apply only the operations that accept that type, then eliminate the combinations that vanish identically or reduce to others. Nothing here is computation; it is grammar — but grammar that dictates which field equations can exist.

Step 1. From a scalar $V$ : the only first move is the gradient, $\nabla V$ , a vector. A vector accepts divergence or curl, giving two candidates:

\nabla\cdot(\nabla V) = \nabla^2 V, \qquad \nabla\times(\nabla V)

Why: you cannot dot or cross into a scalar, so $\nabla V$ is the unique gateway out of scalar-land.

Step 2. From a vector $\mathbf{A}$ : the first moves are $\nabla\cdot\mathbf{A}$ (a scalar, which then accepts only a gradient) and $\nabla\times\mathbf{A}$ (a vector, which accepts divergence or curl), giving three more candidates:

\nabla(\nabla\cdot\mathbf{A}), \qquad \nabla\cdot(\nabla\times\mathbf{A}), \qquad \nabla\times(\nabla\times\mathbf{A})

Why: five total combinations — the complete list; anything else ( $\nabla\times$ of a scalar, $\nabla$ of a vector without dot or cross) is grammatically illegal in this calculus.

Step 3. Strike the identical zeros. By equality of mixed partials,

\nabla\times(\nabla V)=\mathbf{0}, \qquad \nabla\cdot(\nabla\times\mathbf{A})=0

Why: each component of the first is $\partial_i\partial_j V - \partial_j\partial_i V$ ; the second is a cyclic sum of the same cancelling pairs. Two candidates gone — but their vanishing is the whole potential formalism of E&M, so they exit with honours.

Step 4. Reduce the curl of a curl using identity (3): it is gradient-of-divergence minus the (vector) Laplacian — a combination of the survivors, not a new operator. The final census: one genuinely new second-order operator, $\nabla^2$ , plus the seldom-named $\nabla(\nabla\cdot\mathbf{A})$ , plus two structural zeros. This is why one operator can dominate electrostatics, magnetostatics, diffusion and waves simultaneously.

Poisson, Laplace, and the point source

Here is the payoff equation of the entire vector-calculus chapter. Take Gauss's law $\nabla\cdot\mathbf{E}=\rho/\varepsilon_0$ , insert the potential representation $\mathbf{E}=-\nabla V$ (legal because electrostatic fields are curl-free), and the two first-order Maxwell statements fuse into one second-order equation:

\nabla^2 V = -\frac{\rho}{\varepsilon_0} \qquad (4)

where: $\rho$ — electric charge density; $\varepsilon_0 = 8.854\times10^{-12}\ \mathrm{F/m}$ — permittivity of free space; $V$ — electrostatic potential. This is Poisson's equation: the local curvature of the potential landscape is set by the local charge. In the averaging language, charge forces the potential at a point above its neighbourhood average (for $\rho>0$ ) — potential mounds sit on positive charge. Wherever the region is empty, the source term dies and we get Laplace's equation:

\nabla^2 V = 0 \qquad (5)

where: solutions are called harmonic functions — fields that equal their neighbourhood average everywhere, maximally smooth interpolators between their boundary values. The whole of the boundary-value-problems chapter (E2) is the craft of solving (4) and (5) in various geometries.

Figure

Poisson where the charge lives, Laplace in the space between — the anatomy of every boundary-value problem

In spherical coordinates — feeding the spherical gradient into the spherical divergence, as done in the del-operator article — the Laplacian reads:

\nabla^2 f = \frac{1}{r^2}\frac{\partial}{\partial r}\!\left(r^2\frac{\partial f}{\partial r}\right) + \frac{1}{r^2\sin\theta}\frac{\partial}{\partial\theta}\!\left(\sin\theta\frac{\partial f}{\partial\theta}\right) + \frac{1}{r^2\sin^2\theta}\frac{\partial^2 f}{\partial\phi^2} \qquad (6)

where: $r,\theta,\phi$ — radial distance, polar angle, azimuth. For spherically symmetric fields only the first term survives — a one-dimensional operator that makes point-charge problems almost trivial, as the derivation below exploits.

Go deeper · Derivation B · 1 minThe Laplacian of one over r and the Green function of electrostatics

Derivation BThe Laplacian of one over r and the Green function of electrostatics

We are proving the single most consequential delta-function identity in the track: $\nabla^2(1/r) = -4\pi\,\delta^3(\mathbf{r})$ . It says the Coulomb potential is harmonic everywhere except the origin, where its Laplacian is an infinitely concentrated spike — and it is the reason the superposition integral $V=\frac{1}{4\pi\varepsilon_0}\int \rho/|\mathbf{r}-\mathbf{r}'|\,\mathrm{d}\tau'$ solves Poisson's equation. The strategy chains two results we already own: the gradient of $1/r$ (gradient article) and the divergence of $\hat{\mathbf{r}}/r^2$ with its hidden delta (divergence article). Watch the signs — the two minus signs conspire to a single overall minus.

Step 1. Write the Laplacian as divergence of gradient and insert the known gradient:

\nabla^2\frac{1}{r} = \nabla\cdot\left(\nabla\frac{1}{r}\right) = \nabla\cdot\left(-\frac{\hat{\mathbf{r}}}{r^2}\right)

Why: $\nabla(1/r)=-\hat{\mathbf{r}}/r^2$ was derived by direct chain rule in the gradient article; the composition is exactly what "divergence of the gradient" means.

Step 2. Check the smooth region with the spherical formula (6), where only the radial term acts on $f(r)=1/r$ :

\nabla^2\frac{1}{r} = \frac{1}{r^2}\frac{\partial}{\partial r}\!\left(r^2\cdot\left(-\frac{1}{r^2}\right)\right) = \frac{1}{r^2}\frac{\partial}{\partial r}(-1) = 0 \qquad (r\neq 0)

Why: $r^2\,\partial_r(1/r) = -1$ is constant — the Coulomb potential is harmonic in empty space, as Laplace's equation demands of a source-free field.

Step 3. Capture the origin using the known divergence identity $\nabla\cdot(\hat{\mathbf{r}}/r^2)=4\pi\,\delta^3(\mathbf{r})$ :

\nabla^2\frac{1}{r} = -\nabla\cdot\frac{\hat{\mathbf{r}}}{r^2} = -4\pi\,\delta^3(\mathbf{r})

Why: all the flux bookkeeping was done in the divergence article — the inverse-square field carries exactly $4\pi$ of flux out of any surface around the origin, which no smooth zero-divergence claim can account for.

Step 4. Read it as a Green function. Multiply by $q/(4\pi\varepsilon_0)$ : the potential $V = q/(4\pi\varepsilon_0 r)$ satisfies $\nabla^2 V = -(q/\varepsilon_0)\,\delta^3(\mathbf{r})$ — Poisson's equation for the point charge density $\rho = q\,\delta^3(\mathbf{r})$ . By superposition, summing shifted copies over a continuous distribution shows

V(\mathbf{r}) = \frac{1}{4\pi\varepsilon_0}\int\frac{\rho(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}^3r'

solves $\nabla^2 V = -\rho/\varepsilon_0$ for any $\rho$ : the function $-1/(4\pi r)$ is the Green function that inverts the Laplacian. Why the result is self-consistent: dimensions check throughout, and integrating $\nabla^2(1/r)$ over any ball centred on the origin reproduces $-4\pi$ via the divergence theorem.

The averaging meter, made exact

The claim that the Laplacian measures "centre versus neighbourhood average" is quantitative, and its precise form is worth having:

\langle f\rangle_{\partial B} - f(\mathbf{r}_0) \approx \frac{R^2}{6}\,\nabla^2 f(\mathbf{r}_0) \qquad (7)

where: $\langle f\rangle_{\partial B}$ — average of $f$ over a small sphere of radius $R$ centred at $\mathbf{r}_0$ ; $f(\mathbf{r}_0)$ — the central value. Positive Laplacian: the point is a local valley relative to its surroundings. Zero Laplacian at every point and every radius: the field is harmonic and equals its spherical averages — the mean-value property.

Go deeper · Derivation C · 1 minThe averaging interpretation from a Taylor expansion

Derivation CThe averaging interpretation from a Taylor expansion

We are deriving equation (7), the exact sense in which $\nabla^2$ is an averaging meter — the source of the mean-value property of harmonic functions and of every relaxation algorithm that solves Laplace's equation numerically. The strategy: Taylor-expand $f$ about the centre point, average term by term over a sphere, and watch symmetry assassinate the odd terms while the quadratic term survives with a universal isotropic coefficient. The step to watch is the average $\langle s_i s_j\rangle$ over the sphere — its $R^2/3$ coefficient is where the famous $R^2/6$ comes from.

Step 1. Expand about $\mathbf{r}_0$ , with $\mathbf{s}$ the displacement to a point on the sphere ( $|\mathbf{s}|=R$ ):

f(\mathbf{r}_0+\mathbf{s}) = f_0 + s_i\,\partial_i f + \tfrac{1}{2}s_i s_j\,\partial_i\partial_j f + O(s^3)

Why: we need only the first few orders because $R$ is small; index notation with implied summation keeps the bookkeeping honest.

Step 2. Average over the sphere's surface. By symmetry every direction and its opposite contribute equally, so

\langle s_i\rangle = 0, \qquad \langle s_i s_j\rangle = \frac{R^2}{3}\,\delta_{ij}

Why: the linear average vanishes since $+\mathbf{s}$ and $-\mathbf{s}$ cancel; the quadratic average must be isotropic (no preferred axis), and its trace fixes the coefficient: $\langle s_is_i\rangle = \langle R^2\rangle = R^2$ forces $3\times$ the coefficient to be $R^2$ .

Step 3. Insert the averages into the expansion:

\langle f\rangle = f_0 + \tfrac{1}{2}\cdot\frac{R^2}{3}\,\delta_{ij}\,\partial_i\partial_j f = f_0 + \frac{R^2}{6}\,\nabla^2 f

Why: the deltas contract the two derivative indices, and $\partial_i\partial_i$ is precisely the Laplacian; odd orders vanish by the same parity argument, so corrections start at $R^4$ .

Step 4. Consequences. If $\nabla^2 f=0$ , then $\langle f\rangle = f_0$ for every small sphere — and, by a bootstrapping argument, every sphere: harmonic functions obey the mean-value property exactly, so they cannot have interior maxima or minima (a peak would beat its own average). Electrostatically: in charge-free space no point of stable equilibrium exists for a test charge, since that would require a genuine minimum of $V$ — Earnshaw's theorem in one line. This result reproduces the mean-value property of potential theory; the discrete 6-point version of equation (7) is checked numerically in the worked examples.

Limiting cases

How the Laplacian and the related identities specialize in important cases:

Case	Result	Meaning
Harmonic field	$\nabla^2 V = 0$	Laplace's equation: $V$ has no local extrema in the interior; its value at any point equals its average over any centred sphere (mean-value property).
Source-driven field	$\nabla^2 V = -\rho/\varepsilon_0$	Poisson's equation of electrostatics: the Laplacian is fixed by the local source.
Curl of a gradient	$\nabla\times(\nabla V) = \mathbf{0}$	Identically zero for any twice-differentiable $V$ ; guarantees that $\mathbf{E}=-\nabla V$ has zero curl, consistent with electrostatics.
Divergence of a curl	$\nabla\cdot(\nabla\times\mathbf{A}) = 0$	Identically zero; guarantees that $\mathbf{B}=\nabla\times\mathbf{A}$ is automatically divergence-free, encoding $\nabla\cdot\mathbf{B}=0$ .
Point source (3D)	$\nabla^2\!\left(\dfrac{1}{r}\right) = -4\pi\,\delta^3(\mathbf{r})$	The Laplacian of the Coulomb potential is a Dirac delta at the origin; this single identity underlies the Green's function of electrostatics.
One dimension	$\nabla^2 f = f''(x)$	Pure concavity: the averaging meter with two neighbours, $\frac{f(x+h)+f(x-h)}{2} - f(x) \approx \frac{h^2}{2}f''$ .

Worked examples

Example 1 — a harmonic field and the discrete averaging meter (confidence builder). Consider $V(x,y,z) = x^2 + y^2 - 2z^2$ (volts, coordinates in metres).

Step 1. Compute the Laplacian term by term: $\partial_x^2 V = 2$ , $\partial_y^2 V = 2$ , $\partial_z^2 V = -4$ :

\nabla^2 V = 2 + 2 - 4 = 0

$V$ is harmonic — it could be an electrostatic potential in a charge-free region (it is, in fact, the potential near the centre of a quadrupole trap).

Step 2. Test the mean-value property numerically at the origin with the six-point discrete average at $h = 0.1\ \mathrm{m}$ . The six neighbours give $V(\pm h,0,0) = 0.01$ , $V(0,\pm h,0) = 0.01$ , $V(0,0,\pm h) = -0.02$ each:

\langle V\rangle_6 = \frac{0.01+0.01+0.01+0.01-0.02-0.02}{6} = \frac{0}{6} = 0 = V(0,0,0)

Centre equals neighbour-average exactly — the discrete mean-value property. Note the physics of Earnshaw hiding in the signs: the potential rises along $x$ and $y$ but falls along $z$ ; a positive charge at the origin is trapped sideways but escapes vertically. No arrangement of static charges can close that exit.

Step 3. Contrast with the non-harmonic $W = x^2+y^2+z^2$ : $\nabla^2 W = 6$ , and the same six-point average gives $\langle W\rangle_6 - W(0) = 6h^2/6 = 0.01 = \frac{h^2}{6}\nabla^2 W$ — the meter reads $R^2/6$ times the Laplacian, exactly as equation (7) promises.

Example 2 — verifying Poisson inside a uniformly charged ball (exam level). A ball of radius $R = 0.10\ \mathrm{m}$ carries uniform charge density $\rho = 2.66\ \mathrm{nC/m^3}$ (the density found in the divergence article's example). The interior potential is claimed to be $V(r) = \dfrac{\rho}{6\varepsilon_0}(3R^2 - r^2)$ . Check Poisson's equation and evaluate the central potential.

Step 1. Radial derivative: $\dfrac{\partial V}{\partial r} = -\dfrac{\rho r}{3\varepsilon_0}$ . (This gives $E_r = -\partial V/\partial r = \rho r/3\varepsilon_0 = A r$ with $A = \rho/3\varepsilon_0$ — precisely the linearly growing interior field of the divergence article. The two examples are one problem seen through two operators.)

Step 2. Apply the spherical Laplacian (6), radial term only:

\nabla^2 V = \frac{1}{r^2}\frac{\partial}{\partial r}\!\left(r^2\cdot\left(-\frac{\rho r}{3\varepsilon_0}\right)\right) = \frac{1}{r^2}\cdot\left(-\frac{\rho}{3\varepsilon_0}\right)\cdot 3r^2 = -\frac{\rho}{\varepsilon_0}

Poisson's equation holds identically at every interior point — the $r^2$ inside the derivative is what makes the check work; the naive $\partial_r^2 V = -\rho/3\varepsilon_0$ alone would be off by a factor of 3.

Step 3. Central value:

V(0) = \frac{\rho R^2}{2\varepsilon_0} = \frac{(2.66\times 10^{-9})(0.10)^2}{2(8.854\times 10^{-12})} = \frac{2.66\times 10^{-11}}{1.771\times 10^{-11}} = 1.50\ \mathrm{V}

The potential mound peaks at 1.50 V over the charge — a maximum sustained only because charge is present there; the moment $\rho\to 0$ , Laplace's equation would flatten it.

Applications

Electrostatic boundary-value problems. In any charge-free region the potential obeys Laplace's equation $\nabla^2 V=0$ ; with charges present it obeys Poisson's equation $\nabla^2 V=-\rho/\varepsilon_0$ . Solving these subject to boundary conditions (specified $V$ or its normal derivative on conductors) is the central computational task of electrostatics, handled by separation of variables, the method of images, or numerical relaxation — all of which exploit the averaging property of the Laplacian.

Automatic satisfaction of Maxwell constraints. Writing $\mathbf{E}=-\nabla V$ uses $\nabla\times(\nabla V)=0$ to guarantee the electrostatic curl-free condition automatically. Writing $\mathbf{B}=\nabla\times\mathbf{A}$ uses $\nabla\cdot(\nabla\times\mathbf{A})=0$ to guarantee $\nabla\cdot\mathbf{B}=0$ automatically. The two vanishing second derivatives are thus the mathematical foundation for introducing the scalar and vector potentials throughout electromagnetism.

The electromagnetic wave equation. Combining Maxwell's equations in vacuum and applying the curl-of-curl identity $\nabla\times(\nabla\times\mathbf{E})=\nabla(\nabla\cdot\mathbf{E})-\nabla^2\mathbf{E}$ with $\nabla\cdot\mathbf{E}=0$ yields $\nabla^2\mathbf{E}=\mu_0\varepsilon_0\,\partial_t^2\mathbf{E}$ , the wave equation. The vector Laplacian therefore sits at the heart of electromagnetic radiation, with wave speed $c=1/\sqrt{\mu_0\varepsilon_0}$ .

Green's functions and the delta-function identity. The identity $\nabla^2(1/r)=-4\pi\delta^3(\mathbf{r})$ identifies $-1/(4\pi r)$ as the Green's function of the Laplacian. It lets one invert Poisson's equation as $V(\mathbf{r})=\frac{1}{4\pi\varepsilon_0}\int\frac{\rho(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}^3r'$ , the integral solution underlying all of potential theory.

Origins · 1 minPierre-Simon Laplace and how this was discovered

Pierre-Simon Laplace

1749–1827 · Paris, École Militaire and the Institut de France

Laplace arrived in Paris a miller's son from Normandy and made himself, by relentless calculation, the Newton of his age. His five-volume Mécanique Céleste set out to prove the solar system stable by analysis alone — famously (perhaps apocryphally) telling Napoleon he had "no need of that hypothesis" when asked where God appeared in it. The operator bearing his name emerged from exactly this work: studying the gravitational attraction of spheroids in the 1780s, he showed the potential outside a body satisfies $\nabla^2 V = 0$ , converting force calculations into a single partial differential equation. He missed the interior case — his former student Poisson supplied the $-\rho/\varepsilon_0$ source term in 1813, completing the pair that now governs electrostatics. Laplace survived the Revolution, the Empire and the Restoration, collecting titles from each; his equation outlasted them all, running today in every chip-design tool and weather model.

History

1782

— Laplace shows the gravitational potential of a spheroid satisfies what is now Laplace's equation

1813

— Poisson adds the source term, extending the equation to the interior of charged and massive bodies

1828

— Green's essay introduces Green functions, inverting the Laplacian with the 1/r kernel

1840

— Gauss proves the mean-value property of harmonic functions, formalizing the averaging interpretation

1842

— Earnshaw deduces that no static charge configuration is stably trapped, a direct corollary of the mean-value property

1873

— Maxwell's Treatise organizes electrostatics around Poisson and Laplace boundary-value problems, the template this track follows

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §1.2.7 (Second Derivatives) and §2.3.3 · Jackson, Classical Electrodynamics, 3rd ed., §1.7 and inside front cover (vector identities) · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 2 and Appendix · Zangwill, Modern Electrodynamics, §1.2 and Appendix C (vector calculus identities) Key cross-checks: the curl-of-curl identity follows in index notation from $\epsilon_{ijk}\epsilon_{klm}=\delta_{il}\delta_{jm}-\delta_{im}\delta_{jl}$ ; the identity $\nabla^2(1/r)=-4\pi\delta^3(\mathbf{r})$ is confirmed by the divergence theorem; and the averaging coefficient $R^2/6$ follows from the spherical average $\langle s_i s_j\rangle=(R^2/3)\delta_{ij}$ . The permittivity value $\varepsilon_0=8.854\times10^{-12}\ \mathrm{F/m}$ is the CODATA value.

Builds on: The Gradient Operator · The Divergence of a Vector Field · The Curl of a Vector Field · The Del Operator (Nabla) · Scalar and Vector Fields Leads to: The Divergence (Gauss) Theorem · Poisson Equation · Laplace Equation · Spherical Coordinates

Is the Laplacian just the 3D version of the second derivative from single-variable calculus?

Essentially yes, with one refinement: it is the sum of the three unmixed second derivatives, one concavity per axis. In 1D, $f''>0$ means the point sits below the chord through its neighbours; the Laplacian says the same with a sphere of neighbours instead of two. What is new in 3D is that the individual concavities can trade off — $x^2+y^2-2z^2$ curves up two ways and down one, netting zero. That trading is exactly why saddle-shaped harmonic potentials exist and why Earnshaw's theorem leaves every static trap with an escape direction.

What is the practical difference between Poisson's and Laplace's equations?

Same operator, different question. Poisson's equation ( $\nabla^2V = -\rho/\varepsilon_0$ ) applies where sources live: the charge distribution dictates the potential's curvature pointwise. Laplace's equation ( $\nabla^2V = 0$ ) applies in the empty space between sources, where the potential is determined entirely by its boundary values — the field just interpolates as smoothly as possible. Most real problems stitch the two: Poisson inside the charged region, Laplace outside, matched at the surface. Solving the empty-space problem given boundary data is the entire craft of chapter E2.

Why do harmonic functions matter so much if real problems always have charges somewhere?

Because the charges usually sit on boundaries — conductor surfaces, electrodes, capacitor plates — leaving the volume where you actually want the field charge-free and hence harmonic. The mean-value property then gives astonishing rigidity: the solution is unique once boundary values are fixed, it attains its extremes only on the boundary, and it can be computed by literal averaging (relaxation). All the uniqueness theorems that make the method of images legitimate rest on this. Harmonicity is not the absence of physics; it is the physics of "in between".

How literally can I use the averaging property to compute potentials?

Completely literally — it is a working algorithm. Discretize space into a grid; the discrete Laplacian at a node is (neighbour average − centre value) times $6/h^2$ . To solve Laplace's equation, repeatedly replace every interior value by the average of its six neighbours while holding boundary values fixed; the iteration converges to the harmonic solution. This is the relaxation (Jacobi/Gauss–Seidel) method, and with the multigrid acceleration trick it is among the fastest PDE solvers known. Example 1's six-point check is the $n=1$ step of exactly this algorithm.

Does the minus sign in Poisson's equation mean charge creates a potential valley?

The opposite — track the signs carefully. For $\rho > 0$ , $\nabla^2 V < 0$ , and by the averaging meter the central value exceeds the neighbourhood average: positive charge sits under a potential mound, and $V$ falls off as you leave it (the $1/r$ hill of a point charge). Valleys of potential form over negative charge. The mnemonic: positive charges are the "mountain sources" of the potential landscape, and $\mathbf{E}=-\nabla V$ sends test charges rolling downhill away from them.

Why does quantum mechanics write the kinetic energy with a minus sign in front of the Laplacian?

The operator $-\hbar^2\nabla^2/2m$ looks alien until you see the averaging meter. A wavefunction that wiggles rapidly deviates strongly from its local average, giving a large $|\nabla^2\psi|$ ; the minus sign (with the sign conventions of the Schrödinger equation) makes rapid wiggling cost energy — short wavelength = high momentum, the de Broglie relation operationalized. Mathematically, $-\nabla^2$ is a positive operator (its eigenvalues $k^2$ are non-negative), which is what lets kinetic energy be bounded below. The same positivity is why diffusion smooths ( $\partial_t T = \alpha\nabla^2 T$ damps wiggles) rather than sharpens.

Is there anything physical in the fifth combination, the gradient of the divergence?

It appears wherever compression matters. In electromagnetism it surfaces inside the curl-of-curl identity and in gauge-fixing choices (the Coulomb gauge kills it by setting $\nabla\cdot\mathbf{A}=0$ ; the Lorenz gauge trades it for a time derivative). In elasticity and acoustics it stands on its own: the restoring force in a compressible medium contains $\nabla(\nabla\cdot\mathbf{u})$ , which is why longitudinal sound waves travel at a different speed from transverse ones. E&M in vacuum, having divergence-free fields, quietly discards it — one reason light is purely transverse.