Helmholtz Decomposition of Vector Fields

Every vector field you will meet in electrodynamics is secretly two fields wearing one coat. There is a fan — flow that streams radially away from sources, the way the electric field streams away from a positive charge — and there is a swirl — flow that circulates in closed loops, the way the magnetic field wraps around a current-carrying wire. The Helmholtz decomposition is the statement that this is not a metaphor but an exact theorem: any well-behaved vector field $\mathbf{F}$ splits, uniquely, into an irrotational piece $-\nabla\varphi$ that carries all of the divergence and a solenoidal piece $\nabla\times\mathbf{A}$ that carries all of the curl. Nothing is left over. The previous article proved the Helmholtz theorem — that a field's divergence and curl, plus decay at infinity, pin it down completely. This article is about the decomposition itself as a working tool: how to actually split a given field into its two pieces, why the two pieces are orthogonal to each other in a precise integral sense, what the split looks like in Fourier space (where it becomes the longitudinal/transverse separation that runs radiation theory and quantum field theory), and what extra "harmonic" piece sneaks in when you work in a finite box instead of all of space. Mastering the split buys you the entire organizational chart of electrodynamics. Electrostatics is the pure-fan sector: $\nabla\times\mathbf{E}=\mathbf{0}$ , so $\mathbf{E}=-\nabla V$ . Magnetostatics is the pure-swirl sector: $\nabla\cdot\mathbf{B}=0$ , so $\mathbf{B}=\nabla\times\mathbf{A}$ . Radiation lives in the transverse part alone. One page of vector calculus, four chapters of physics.

Key takeaways

Any vector field that decays faster than $1/r$ at infinity splits uniquely as $\mathbf{F}=-\nabla\varphi+\nabla\times\mathbf{A}$ : an irrotational (curl-free) part plus a solenoidal (divergence-free) part.
The split works because of two identities: $\nabla\times(\nabla\varphi)=\mathbf{0}$ makes the gradient piece automatically curl-free, and $\nabla\cdot(\nabla\times\mathbf{A})=0$ makes the curl piece automatically divergence-free.
Each potential is sourced by one interrogation of the field: $\nabla^2\varphi=-\nabla\cdot\mathbf{F}$ and $\nabla^2\mathbf{A}=-\nabla\times\mathbf{F}$ (with $\nabla\cdot\mathbf{A}=0$ ), both solved by the same $1/4\pi|\mathbf{r}-\mathbf{r}'|$ kernel.
The two pieces are orthogonal: $\int(-\nabla\varphi)\cdot(\nabla\times\mathbf{A})\,\mathrm{d}\tau=0$ for decaying fields — the decomposition is a projection onto perpendicular subspaces of field space.
In Fourier space the split is algebraic: the irrotational part is the component of $\tilde{\mathbf{F}}(\mathbf{k})$ parallel to $\mathbf{k}$ (longitudinal), the solenoidal part the component perpendicular to $\mathbf{k}$ (transverse).
On a bounded domain a third, harmonic piece $\mathbf{H}$ (curl-free and divergence-free) appears, pinned only by boundary data — the Helmholtz–Hodge decomposition.
Key idea: every well-behaved vector field is exactly one fan plus one swirl — a gradient carrying all the divergence plus a curl carrying all the rotation — and the two pieces do not talk to each other.

The core insight

A well-behaved vector field equals minus the gradient of a scalar potential plus the curl of a vector potential; the scalar potential is sourced by the field's divergence and the vector potential by its curl, so knowing $\nabla\cdot\mathbf{F}$ and $\nabla\times\mathbf{F}$ (with decay or boundary data) determines $\mathbf{F}$ completely.

Where this lives today

Every time a weather model or an incompressible-flow solver takes a time step, it runs a Helmholtz decomposition. The provisional velocity field that comes out of the momentum update is not divergence-free; the solver projects it back onto the solenoidal subspace by solving one Poisson equation for a pressure-like scalar and subtracting its gradient — literally computing $-\nabla\varphi$ and discarding it. Chorin's projection method and its descendants, running inside every production CFD code and every film-studio smoke simulation, are equation (3) executed billions of times per second. On the physics side, the longitudinal/transverse split is the standard first move of quantum electrodynamics in Coulomb gauge: the longitudinal electric field is eliminated in favor of instantaneous Coulomb interactions between charges, and only the two transverse polarizations get quantized into photons. When a plasma physicist separates "electrostatic" from "electromagnetic" turbulence in a tokamak, or a geodesist splits crustal deformation into dilatation and shear, they are all asking a field the same two questions this page asked: where do you fan, and where do you swirl?

The physical picture

Think of the divergence as the field's source-charge density and the curl as its vortex density. The Helmholtz decomposition says any flow you can draw is the superposition of exactly two archetypes: the flow generated by spread-out sources and sinks — streamlines that begin and end, never loop — and the flow generated by spread-out vortices — streamlines that loop, never end. The electric field of static charges is a pure specimen of the first kind. The magnetic field of steady currents is a pure specimen of the second. A generic field, like the velocity of real water or the electric field during radiation, is a blend, and the decomposition is the machine that unblends it.

Figure

Any field unblends into a fan plus a swirl — the divergence sources one piece, the curl the other

Why can the two archetypes never impersonate each other? Because each one is structurally deaf to one of the two questions you can ask a field. A gradient field $-\nabla\varphi$ has zero curl identically — take its circulation around any loop and the fundamental theorem for gradients kills it. A curl field $\nabla\times\mathbf{A}$ has zero divergence identically — its flux through any closed surface vanishes. So when you interrogate a blended field, the answers separate cleanly: all of the divergence you measure belongs to the fan, and all of the curl belongs to the swirl. The two interrogations never cross-contaminate, which is precisely why the split is unambiguous. There is one fine-print clause, and it matters. On all of space, with the field dying faster than $1/r$ , fan-plus-swirl is the whole story. But confine yourself to a finite box and a third character walks in: a harmonic field, simultaneously curl-free and divergence-free — think of a perfectly uniform flow crossing the box. It answers "zero" to both interrogations, so no interior measurement can detect it; only boundary data pins it down. That third piece is why electrostatics inside a cavity needs boundary conditions and not just the charge density.

The decomposition, piece by piece

The master statement is a single line. For a vector field $\mathbf{F}$ that is smooth and falls off faster than $1/r$ as $r\to\infty$ ,

\mathbf{F}(\mathbf{r}) \;=\; -\nabla\varphi(\mathbf{r}) \;+\; \nabla\times\mathbf{A}(\mathbf{r}) \tag{1}

where: $\varphi$ is the scalar potential (its gradient piece is called irrotational or longitudinal) and $\mathbf{A}$ is the vector potential (its curl piece is called solenoidal or transverse). The minus sign on the gradient is pure convention, chosen so that in electrostatics $\varphi$ becomes the familiar potential $V$ with $\mathbf{E}=-\nabla V$ . Equation (1) is an identity about all well-behaved fields, not a special property of $\mathbf{E}$ or $\mathbf{B}$ — that generality is what makes it an organizing principle rather than a trick. The split works because of two identities you proved with index notation earlier in this chapter, and it is worth staring at them as a matched pair:

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

where: the first says gradients are curl-blind, the second says curls are divergence-blind. Consequently, taking the divergence of (1) hits only the first term, and taking the curl of (1) hits only the second: each interrogation reads out exactly one potential. These two identities are the load-bearing beams; everything else on this page is furniture. Apply $\nabla\cdot$ and $\nabla\times$ to (1) and you get the equations that define the potentials in terms of the field's sources:

\nabla^2\varphi = -\nabla\cdot\mathbf{F}, \qquad \nabla^2\mathbf{A} = -\nabla\times\mathbf{F} \quad (\text{with } \nabla\cdot\mathbf{A}=0) \tag{3}

where: each is a Poisson equation — the same equation electrostatics solves for $V$ given $\rho$ . The side condition $\nabla\cdot\mathbf{A}=0$ uses up the freedom to shift $\mathbf{A}$ by a gradient (which would not change $\nabla\times\mathbf{A}$ ); it is your first meeting with a gauge choice, and it is what reduces the vector equation to three independent scalar copies in Cartesian components. Because both equations in (3) are inverted by the same Green's function $1/4\pi|\mathbf{r}-\mathbf{r}'|$ , the potentials are explicit integrals over the sources:

\varphi(\mathbf{r}) = \frac{1}{4\pi}\int\frac{\nabla'\cdot\mathbf{F}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}\tau', \qquad \mathbf{A}(\mathbf{r}) = \frac{1}{4\pi}\int\frac{\nabla'\times\mathbf{F}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}\tau' \tag{4}

where: primed coordinates label source points and the integrals run over all space. The construction of (4) — representing $\mathbf{F}$ with a delta function, inserting the Green's identity, and splitting with the curl-of-curl identity — is the proof of the Helmholtz theorem, done in full in the previous article; here we take it as established machinery. What we own on this page is what the split means: orthogonality, the Fourier picture, and the boundary fine print. When $\mathbf{F}=\mathbf{E}$ in statics, the first integral in (4) with $\nabla\cdot\mathbf{E}=\rho/\varepsilon_0$ is exactly the Coulomb potential; when $\mathbf{F}=\mathbf{B}$ , the second with $\nabla\times\mathbf{B}=\mu_0\mathbf{J}$ is exactly the vector potential of magnetostatics. Which formula answers which question:

you are given	you want	reach for
$\mathbf{F}$ everywhere	its irrotational/solenoidal pieces	compute $\nabla\cdot\mathbf{F}$ , $\nabla\times\mathbf{F}$ , then (4), then (1)
$\nabla\cdot\mathbf{F}$ and $\nabla\times\mathbf{F}$ , decay	$\mathbf{F}$ itself	(4) then (1) — reconstruction
a curl-free field	a compact description	$\mathbf{F}=-\nabla\varphi$ : one scalar function
a divergence-free field	a compact description	$\mathbf{F}=\nabla\times\mathbf{A}$ : a vector potential
a field in a finite box	the full split	Helmholtz–Hodge: add the harmonic piece $\mathbf{H}$ , fix it with boundary data

Derivation AThe two pieces are orthogonal — the split is a projection

We want to show that the fan and the swirl are perpendicular as vectors in field space: $\int(-\nabla\varphi)\cdot(\nabla\times\mathbf{A})\,\mathrm{d}\tau=0$ when the integral runs over all space and the fields decay. The strategy is a single integration by parts: move the gradient off $\varphi$ and onto the curl term, where the identity $\nabla\cdot(\nabla\times\mathbf{A})=0$ annihilates it. The point of the exercise is conceptual — orthogonality is what makes the decomposition behave like resolving a vector onto perpendicular axes, with each piece obtainable by projection and an energy (Pythagoras) theorem to go with it. Watch the surface term: the whole argument lives or dies on the decay condition.

Step 1. Rewrite the integrand as a total divergence minus a correction, using the product rule $\nabla\cdot(\varphi\,\mathbf{G})=\varphi\,(\nabla\cdot\mathbf{G})+(\nabla\varphi)\cdot\mathbf{G}$ with $\mathbf{G}=\nabla\times\mathbf{A}$ :

(-\nabla\varphi)\cdot(\nabla\times\mathbf{A}) = -\nabla\cdot\big[\varphi\,(\nabla\times\mathbf{A})\big] + \varphi\,\nabla\cdot(\nabla\times\mathbf{A})

Why this rearrangement? It isolates the only structure we can integrate exactly — a pure divergence — and parks the rest in a term containing $\nabla\cdot(\nabla\times\mathbf{A})$ , which we know to be identically zero.

Step 2. The second term dies on the spot: $\nabla\cdot(\nabla\times\mathbf{A})=0$ is identity (2). What remains is a single total divergence:

\int_V(-\nabla\varphi)\cdot(\nabla\times\mathbf{A})\,\mathrm{d}\tau = -\int_V \nabla\cdot\big[\varphi\,(\nabla\times\mathbf{A})\big]\,\mathrm{d}\tau

Step 3. Convert the volume integral to a surface integral with the divergence theorem, taking $V$ to be a ball of radius $R\to\infty$ :

-\int_V \nabla\cdot\big[\varphi\,(\nabla\times\mathbf{A})\big]\,\mathrm{d}\tau = -\oint_{S_R} \varphi\,(\nabla\times\mathbf{A})\cdot\mathrm{d}\mathbf{a}

Why does the surface term vanish? Count powers of $R$ . For a source distribution of finite extent, $\varphi\sim 1/R$ and $|\nabla\times\mathbf{A}|\sim 1/R^2$ at large $R$ (each potential inherits Coulomb-like decay from its $1/|\mathbf{r}-\mathbf{r}'|$ kernel), while the sphere's area grows only as $R^2$ . The product scales as $R^2\cdot(1/R)\cdot(1/R^2)=1/R\to0$ .

Step 4. Conclude orthogonality, and harvest the Pythagoras corollary. Writing $\mathbf{F}_\parallel=-\nabla\varphi$ and $\mathbf{F}_\perp=\nabla\times\mathbf{A}$ , expanding $|\mathbf{F}|^2=|\mathbf{F}_\parallel+\mathbf{F}_\perp|^2$ and using the vanished cross term:

\int|\mathbf{F}|^2\,\mathrm{d}\tau = \int|\mathbf{F}_\parallel|^2\,\mathrm{d}\tau + \int|\mathbf{F}_\perp|^2\,\mathrm{d}\tau

Why care? This is the statement that "field energy" separates cleanly — in electrodynamics the energy in the Coulomb field and the energy in the radiative field add with no interference term, which is what lets you count them as independent degrees of freedom. Verified: the same by-parts argument with the roles swapped (moving the derivative off $\mathbf{A}$ ) gives the identical conclusion, and the result matches Zangwill §1.9 and the standard $L^2$ -orthogonality of the Hodge decomposition.

The Fourier view: longitudinal and transverse

The decomposition looks like hard integral calculus in position space, but in Fourier space it collapses to freshman geometry. Write the field as a superposition of plane waves, $\mathbf{F}(\mathbf{r})=\int\tilde{\mathbf{F}}(\mathbf{k})\,e^{i\mathbf{k}\cdot\mathbf{r}}\,\mathrm{d}^3k/(2\pi)^3$ . Every derivative becomes an algebraic factor: $\nabla\to i\mathbf{k}$ . The two interrogations become dot and cross products with $\mathbf{k}$ , and the split becomes: resolve the vector $\tilde{\mathbf{F}}(\mathbf{k})$ into its components parallel and perpendicular to $\mathbf{k}$ , one wavevector at a time.

Go deeper · Derivation B · 1 minThe split is algebra in Fourier space

Derivation BThe split is algebra in Fourier space

We want to show that the irrotational piece of $\mathbf{F}$ is exactly the part of $\tilde{\mathbf{F}}(\mathbf{k})$ parallel to $\mathbf{k}$ (the longitudinal component) and the solenoidal piece is the part perpendicular to $\mathbf{k}$ (the transverse component). The strategy: translate the defining conditions $\nabla\times\mathbf{F}_\parallel=\mathbf{0}$ and $\nabla\cdot\mathbf{F}_\perp=0$ into $\mathbf{k}$ -space, where they become statements about cross and dot products, then read off the projection operators. This derivation is the bridge between this chapter and radiation theory: "transverse field" in every later chapter means precisely the second projection.

Step 1. Translate the derivatives. Under the Fourier transform, $\nabla\cdot\mathbf{F}\to i\mathbf{k}\cdot\tilde{\mathbf{F}}$ and $\nabla\times\mathbf{F}\to i\mathbf{k}\times\tilde{\mathbf{F}}$ . So "curl-free" means $\mathbf{k}\times\tilde{\mathbf{F}}_\parallel=\mathbf{0}$ for every $\mathbf{k}$ — i.e. $\tilde{\mathbf{F}}_\parallel$ is parallel to $\mathbf{k}$ — and "divergence-free" means $\mathbf{k}\cdot\tilde{\mathbf{F}}_\perp=0$ — i.e. $\tilde{\mathbf{F}}_\perp$ is perpendicular to $\mathbf{k}$ .

Step 2. Write the projections. Any vector resolves along and across a given direction $\hat{\mathbf{k}}$ :

\tilde{\mathbf{F}}_\parallel(\mathbf{k}) = \hat{\mathbf{k}}\,\big(\hat{\mathbf{k}}\cdot\tilde{\mathbf{F}}\big), \qquad \tilde{\mathbf{F}}_\perp(\mathbf{k}) = \tilde{\mathbf{F}} - \hat{\mathbf{k}}\,\big(\hat{\mathbf{k}}\cdot\tilde{\mathbf{F}}\big)

Why is this the same split as (1)? Because the two pieces satisfy exactly the defining conditions: $\mathbf{k}\times\tilde{\mathbf{F}}_\parallel=\mathbf{0}$ by construction (parallel vectors have zero cross product), and $\mathbf{k}\cdot\tilde{\mathbf{F}}_\perp=\mathbf{k}\cdot\tilde{\mathbf{F}}-k\,(\hat{\mathbf{k}}\cdot\tilde{\mathbf{F}})=0$ . Uniqueness of the Helmholtz split (previous article) guarantees these are the transforms of $-\nabla\varphi$ and $\nabla\times\mathbf{A}$ .

Step 3. Notice what became of the hard analysis. Orthogonality is now the triviality $\tilde{\mathbf{F}}_\parallel\cdot\tilde{\mathbf{F}}_\perp^{*}=0$ at each $\mathbf{k}$ (a vector along $\hat{\mathbf{k}}$ dotted with one perpendicular to it), and Parseval's theorem converts that pointwise statement back into the integral orthogonality of Derivation A. The $1/4\pi r$ kernel of (4) is nothing but the Fourier transform of the algebraic factor $1/k^2$ that appears when you invert $\nabla^2\to-k^2$ .

Step 4. Read off the physics dictionary. In electrodynamics the longitudinal electric field ( $\parallel\mathbf{k}$ ) is the instantaneous Coulomb field tied to the charges, while the transverse field ( $\perp\mathbf{k}$ ) is the two-polarization radiative field that detaches and propagates. Counting components: 3 per wavevector split as 1 longitudinal + 2 transverse — those 2 are the photon's two polarization states. Verified: applying the projectors to a test field $\tilde{\mathbf{F}}=(F_x,F_y,F_z)$ with $\mathbf{k}=k\hat{\mathbf{z}}$ gives $\tilde{\mathbf{F}}_\parallel=(0,0,F_z)$ , $\tilde{\mathbf{F}}_\perp=(F_x,F_y,0)$ , matching $i\mathbf{k}\times\tilde{\mathbf{F}}_\parallel=\mathbf{0}$ and $i\mathbf{k}\cdot\tilde{\mathbf{F}}_\perp=0$ directly.

Figure

In Fourier space the split is geometry — resolve each plane-wave amplitude along and across k

Limiting cases

case	result	meaning
$\nabla\times\mathbf{F}=\mathbf{0}$ (electrostatics)	$\mathbf{F}=-\nabla\varphi$ only	pure fan: one scalar function describes the whole field — this is $\mathbf{E}=-\nabla V$
$\nabla\cdot\mathbf{F}=0$ (magnetostatics)	$\mathbf{F}=\nabla\times\mathbf{A}$ only	pure swirl: this is $\mathbf{B}=\nabla\times\mathbf{A}$
$\nabla\cdot\mathbf{F}=0$ and $\nabla\times\mathbf{F}=\mathbf{0}$ , finite box	$\mathbf{F}=\mathbf{H}$ , harmonic	invisible to both interrogations; fixed only by boundary values
$\mathbf{F}\sim1/r$ or slower at infinity	integrals (4) diverge or split is non-unique	decay faster than $1/r$ is a hypothesis, not decoration
constant field $\mathbf{F}_0$ on all of space	violates decay; $\mathbf{F}_0=-\nabla(-\mathbf{F}_0\cdot\mathbf{r})$ or a curl	the standard counterexample to uniqueness without decay
$\mathbf{k}$ -space, single mode	$\tilde{\mathbf{F}}=\tilde{\mathbf{F}}_\parallel+\tilde{\mathbf{F}}_\perp$	the split is one vector resolution per wavevector

Common trapDivergence and curl inside a finite region tell you everything, right?

No — and the failure mode is exactly the harmonic piece. Inside a bounded volume you can always add a field $\mathbf{H}$ with $\nabla\cdot\mathbf{H}=0$ and $\nabla\times\mathbf{H}=\mathbf{0}$ (a uniform field is the simplest example) without changing any interior source measurement. The interior divergence and curl are silent about it; only boundary data speaks. That is why the all-space theorem carries the decay clause "faster than $1/r$ ," and why on a finite domain the honest statement is the Helmholtz–Hodge decomposition $\mathbf{F}=-\nabla\varphi+\nabla\times\mathbf{A}+\mathbf{H}$ with $\mathbf{H}$ pinned by boundary conditions. In practice this is not pathology but physics: solving Laplace's equation in a cavity is the hunt for the harmonic piece consistent with the walls. If a problem hands you a finite region, your checklist is sources plus boundary values — never sources alone.

Worked examples

1 · Unblending a planar field by inspection (confidence builder)

Take the field $\mathbf{F}(x,y,z) = (x-y)\,\hat{\mathbf{x}} + (x+y)\,\hat{\mathbf{y}}$ . Interrogate it twice.

Step 1. Divergence: $\nabla\cdot\mathbf{F} = \partial_x(x-y) + \partial_y(x+y) = 1+1 = 2$ . Curl: the only surviving component is $(\nabla\times\mathbf{F})_z = \partial_x(x+y)-\partial_y(x-y) = 1-(-1) = 2$ . So $\nabla\times\mathbf{F}=2\,\hat{\mathbf{z}}$ . Both sources are nonzero — this is a genuine blend.

Step 2. Guess the fan. The field $(x,y,0)$ has divergence $2$ and curl $\mathbf{0}$ ; indeed $(x,y,0) = -\nabla\varphi$ with $\varphi = -\tfrac{1}{2}(x^2+y^2)$ . Check: $-\nabla\varphi = (x, y, 0)$ . ✓

Step 3. Guess the swirl. The remainder $\mathbf{F}-(x,y,0)=(-y,x,0)$ has divergence $\partial_x(-y)+\partial_y(x)=0$ and curl $2\hat{\mathbf{z}}$ — it is the rigid-rotation field. It is a curl: with $\mathbf{A} = -\tfrac{1}{2}(x^2+y^2)\,\hat{\mathbf{z}}$ , the curl formula gives $\nabla\times\mathbf{A} = (\partial_y A_z)\,\hat{\mathbf{x}} - (\partial_x A_z)\,\hat{\mathbf{y}} = (-y)\,\hat{\mathbf{x}} + x\,\hat{\mathbf{y}}$ . ✓

Step 4. Audit at the point $(2,1,0)$ . Full field: $\mathbf{F}=(1,3,0)$ . Fan piece: $(2,1,0)$ . Swirl piece: $(-1,2,0)$ . Sum: $(2-1,\,1+2,\,0)=(1,3,0)$ . ✓ Notice $(2,1,0)\cdot(-1,2,0) = -2+2 = 0$ — for this special field the two pieces happen to be perpendicular at every point, a bonus accident (rotate $(x,y)$ by $90°$ and you get $(-y,x)$ ). The theorem only promises orthogonality of the integrals, not pointwise.

Answer: $\mathbf{F} = \underbrace{(x,y,0)}_{-\nabla\varphi,\ \varphi=-\frac{1}{2}(x^2+y^2)} + \underbrace{(-y,x,0)}_{\nabla\times\mathbf{A},\ \mathbf{A}=-\frac{1}{2}(x^2+y^2)\hat{\mathbf{z}}}$ , carrying divergence $2$ and curl $2\hat{\mathbf{z}}$ respectively.

2 · A longitudinal/transverse split with numbers (exam level)

A single plane-wave mode has wavevector $\mathbf{k} = (0,0,4)\ \mathrm{m}^{-1}$ and amplitude $\tilde{\mathbf{F}} = (3, 0, 2)$ (units of the field, say V/m). Split it, then confirm each piece passes its interrogation.

Step 1. Unit wavevector: $\hat{\mathbf{k}} = (0,0,1)$ . Longitudinal projection: $\hat{\mathbf{k}}\cdot\tilde{\mathbf{F}} = 2$ , so $\tilde{\mathbf{F}}_\parallel = 2\,(0,0,1) = (0,0,2)$ V/m.

Step 2. Transverse remainder: $\tilde{\mathbf{F}}_\perp = (3,0,2)-(0,0,2) = (3,0,0)$ V/m.

Step 3. Interrogate. Divergence test on the transverse piece: $i\mathbf{k}\cdot\tilde{\mathbf{F}}_\perp = i\,(0,0,4)\cdot(3,0,0) = 0$ . ✓ Curl test on the longitudinal piece: $i\mathbf{k}\times\tilde{\mathbf{F}}_\parallel = i\,(0,0,4)\times(0,0,2) = \mathbf{0}$ (parallel vectors). ✓ The full field fails both tests — $i\mathbf{k}\cdot\tilde{\mathbf{F}} = 8i \neq 0$ and $i\mathbf{k}\times\tilde{\mathbf{F}} = i\,(0\cdot2-4\cdot0,\ 4\cdot3-0\cdot2,\ 0) = 12i\,\hat{\mathbf{y}} \neq \mathbf{0}$ — so it genuinely needed splitting.

Step 4. Energy bookkeeping (Pythagoras). $|\tilde{\mathbf{F}}|^2 = 9+4 = 13$ ; $|\tilde{\mathbf{F}}_\parallel|^2 + |\tilde{\mathbf{F}}_\perp|^2 = 4 + 9 = 13$ . ✓ The cross term is absent, exactly as Derivation A promised. If this mode were an electric field in vacuum radiation, only the transverse $(3,0,0)$ part could propagate; the longitudinal $(0,0,2)$ part would be tied to charge density through Gauss's law — and in genuinely empty space Gauss's law would force it to vanish.

Answer: $\tilde{\mathbf{F}}_\parallel = (0,0,2)$ V/m (longitudinal), $\tilde{\mathbf{F}}_\perp = (3,0,0)$ V/m (transverse), with $|\tilde{\mathbf{F}}|^2 = 13 = 4+9$ splitting cleanly.

Origins · 1 minHermann von Helmholtz and how this was discovered

Hermann von Helmholtz

1821–1894 · Berlin, Physiological Institute to Physics Chair

Helmholtz came to this theorem not from electricity but from fluids. His 1858 paper on vortex motion asked how the velocity field of a fluid is determined by its rotation, and the decomposition of a flow into source-driven and vortex-driven parts fell out as the natural bookkeeping. A physiologist by training who measured the speed of nerve impulses and invented the ophthalmoscope, he moved to physics chairs in Heidelberg and Berlin and became the great unifier of nineteenth-century science — energy conservation, acoustics, electrodynamics, hydrodynamics. His vortex theorems so impressed his friend William Thomson (Kelvin) that Thomson spent a decade trying to build atoms out of vortex rings. The atoms did not survive; the decomposition did, and today it organizes everything from Maxwell's equations to weather-model initialization and computer-graphics fluid solvers.

History

1849

— Stokes, studying diffraction and fluid motion, separates a vector function into gradient-like and rotational parts — the germ of the split

1858

— Helmholtz's "On integrals of the hydrodynamical equations which express vortex-motion" states the decomposition and the reconstruction of a flow from its divergence and curl

1873

— Maxwell's Treatise leans on potentials

V

and

\mathbf{A}

throughout, embedding the decomposition in electromagnetic practice

1930

s — Hodge generalizes the split to differential forms on curved manifolds; the harmonic piece becomes a topological fingerprint

1967

— Chorin's projection method turns the decomposition into the workhorse algorithm of computational fluid dynamics, splitting velocity updates into solenoidal flow plus pressure gradient

Going further

References. Griffiths, Introduction to Electrodynamics, 4th ed., §1.6 and Appendix B (Helmholtz Theorem) · Jackson, Classical Electrodynamics, 3rd ed., §6.3 and §5.4 · Zangwill, Modern Electrodynamics, §1.9 (Helmholtz's Theorem) · Purcell & Morin, Electricity and Magnetism, 3rd ed., Ch. 2 · Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., §3.9

Provenance. Equations and identities follow Griffiths Introduction to Electrodynamics 4th ed. Appendix B and §1.6, Zangwill Modern Electrodynamics §1.9, and Jackson Classical Electrodynamics 3rd ed. A verifier should recompute: (1) the vector identities curl-of-grad = 0 and div-of-curl = 0 via index notation with the Levi-Civita symbol; (2) the Green's-function identity $-\nabla^2[1/(4\pi r)]=\delta^3$ ; (3) the curl-curl identity $\nabla\times(\nabla\times\mathbf{W})=\nabla(\nabla\cdot\mathbf{W})-\nabla^2\mathbf{W}$ ; and (4) the uniqueness argument via Green's first identity. No empirical numbers are quoted; all relations are exact mathematical identities. Worked-example numbers on this page (divergence 2 and curl $2\hat{\mathbf{z}}$ ; the audit $(1,3,0)=(2,1,0)+(-1,2,0)$ ; the split $(3,0,2)=(0,0,2)+(3,0,0)$ with $13=4+9$ ) are illustrative arithmetic recomputed from the boxed formulas.

Builds on: The Divergence of a Vector Field · The Curl of a Vector Field · Second Derivatives and the Laplacian · Volume Integrals · The Divergence (Gauss) Theorem · The Helmholtz Theorem Leads to: Electrostatic Potential · Magnetic Vector Potential · Gauge Transformations · Maxwell's Equations

What is the difference between this article and the Helmholtz theorem article — aren't they the same result?

Same mathematical core, two different jobs. The theorem article answers "do divergence and curl determine a field?" — existence and uniqueness, with the reconstruction integrals as proof machinery. This article treats the decomposition as an operation you perform: how to split a given field, why the pieces are orthogonal (so energies add), how the split becomes a per-wavevector projection in Fourier space, and what the harmonic third piece means on bounded domains. In practice you will quote the theorem once per course but run the decomposition constantly — every potential you introduce, every "transverse field" in radiation theory, every projection step in a fluid code is this page.

Why is there a minus sign on the gradient but not on the curl?

Pure convention, chosen for physics compatibility. Fields point "downhill" from high potential to low, so writing $-\nabla\varphi$ makes $\varphi$ match the electrostatic potential $V$ with $\mathbf{E}=-\nabla V$ and positive charge rolling toward lower $V$ . No analogous convention pressure exists for the curl piece, so it stays bare. Some math texts write $+\nabla\varphi$ ; the theorem is indifferent — absorb the sign into $\varphi$ .

Is the vector potential A unique? I can apparently add things to it without changing the field.

Correct, and the freedom is exactly gradients: $\mathbf{A}\to\mathbf{A}+\nabla\chi$ leaves $\nabla\times\mathbf{A}$ unchanged because curl-of-grad vanishes. The condition $\nabla\cdot\mathbf{A}=0$ quoted in equation (3) is a choice — the Coulomb gauge — that spends this freedom to make the math cleanest. The scalar potential has a humbler cousin freedom: $\varphi\to\varphi+\text{const}$ . Neither ambiguity touches the physical field $\mathbf{F}$ ; managing this redundancy is the subject of gauge transformations later in the track.

Does the decomposition require the field to go to zero at infinity, or just to be bounded?

Stronger than bounded: it must decay faster than $1/r$ . Two things go wrong otherwise. First, the reconstruction integrals (4) can diverge — the kernel $1/|\mathbf{r}-\mathbf{r}'|$ only tames sources that thin out. Second, uniqueness fails: a uniform field $\mathbf{F}_0$ has zero divergence and zero curl, yet can be written as a pure gradient, so it can be silently added to either piece. Physical fields of localized sources decay like $1/r^2$ or faster, comfortably inside the hypothesis — which is why the theorem works so smoothly in electrostatics and magnetostatics.

What exactly is "longitudinal" and "transverse" — longitudinal relative to what?

Relative to the wavevector $\mathbf{k}$ of each Fourier mode, not to any lab direction. Decompose the field into plane waves; at each $\mathbf{k}$ , resolve the amplitude vector along $\hat{\mathbf{k}}$ (longitudinal) and perpendicular to it (transverse). A field is "transverse" when every one of its Fourier amplitudes is perpendicular to its own wavevector — which by the dictionary $\nabla\to i\mathbf{k}$ is the same as saying its divergence vanishes. Radio engineers' "transverse electromagnetic wave" and this chapter's "solenoidal field" are the same statement in different dialects.

If the two pieces are orthogonal, does that mean they don't interact physically?

It means their quadratic integrals separate: total $\int|\mathbf{F}|^2\mathrm{d}\tau$ is the sum of the two pieces' contributions with no cross term, so energy bookkeeping treats them as independent ledgers. It does not mean the underlying dynamics never couples them — in time-dependent electrodynamics, charge motion (longitudinal sector) generates radiation (transverse sector). The orthogonality is a statement about the instantaneous geometry of field space, and it is what makes "energy in the Coulomb field" and "energy in radiation" separately meaningful quantities.

How do I actually compute a decomposition in practice — do I really have to do those two 3D integrals?

Usually not. Three escalating strategies. (a) Inspection: compute $\nabla\cdot\mathbf{F}$ and $\nabla\times\mathbf{F}$ ; if you can guess a curl-free field with the right divergence, the remainder is automatically your solenoidal piece (worked example 1). (b) Fourier: transform, project along and across $\hat{\mathbf{k}}$ per mode, transform back — this is what numerical codes do with FFTs, at cost $\mathcal{O}(N\log N)$ . (c) Poisson solve: solve $\nabla^2\varphi=-\nabla\cdot\mathbf{F}$ with your favorite elliptic solver and set the solenoidal part to $\mathbf{F}+\nabla\varphi$ — the projection method of CFD. The integrals (4) are the proof that these all work; they are rarely the algorithm.

Why does knowing div E and curl E for both fields "justify" Maxwell's equations as complete?

Because divergence and curl exhaust the independent local first-derivative information a vector field carries, and the Helmholtz theorem says that with decay conditions they determine the field. Maxwell's four equations specify precisely $\nabla\cdot\mathbf{E}$ , $\nabla\times\mathbf{E}$ , $\nabla\cdot\mathbf{B}$ , $\nabla\times\mathbf{B}$ . There is no missing fifth equation — nothing else could be independently specified about the fields' local structure. That is why the count is four, and why the decomposition is often called the "fundamental theorem of vector calculus" in electrodynamics texts.