Helmholtz Decomposition Theorem

Hand a physicist an arbitrary vector field $\mathbf{F}(\mathbf{r})$ and there are exactly two questions worth asking at each point: how much is it spreading out, and how much is it winding around. The first is the divergence $\nabla\cdot\mathbf{F}$ , the second the curl $\nabla\times\mathbf{F}$ . The Helmholtz decomposition theorem is the precise statement that these two are independent and complete: they carry all the local information there is, and from them (plus a boundary or decay condition) the entire field can be rebuilt. Concretely, $\mathbf{F}$ splits into a part that owns all the divergence and has no curl, and a part that owns all the curl and has no divergence — the irrotational (longitudinal) and solenoidal (transverse) fields. Because a curl-free field is a gradient and a divergence-free field is a curl, the split takes the working form $\mathbf{F}=-\nabla\phi+\nabla\times\mathbf{A}$ , with a scalar potential $\phi$ fed by the divergence and a vector potential $\mathbf{A}$ fed by the curl. The construction is explicit, not merely an existence claim: invert the Laplacian with the Coulomb Green's function and the potentials fall out as volume integrals of the sources. This one theorem scaffolds electromagnetic potentials, the potential-plus-vortex split of a fluid, the transverse photon of QED, and — on domains with holes — the harmonic remainder that seeds Hodge theory: the master key turning "a field" into "its sources plus its topology."

Key takeaways

Any sufficiently smooth vector field that decays fast enough at infinity splits uniquely as $\mathbf{F}=-\nabla\phi+\nabla\times\mathbf{A}$ : an irrotational (curl-free) gradient plus a solenoidal (divergence-free) curl.
The two pieces are fixed entirely by the field's sources — its divergence $\nabla\cdot\mathbf{F}$ builds the scalar potential $\phi$ , its curl $\nabla\times\mathbf{F}$ the vector potential $\mathbf{A}$ .
The proof inverts Poisson's equation with the Newtonian kernel $\nabla^2\!\big(1/|\mathbf{r}-\mathbf{r}'|\big)=-4\pi\,\delta^3(\mathbf{r}-\mathbf{r}')$ , so a single superpotential $\mathbf{W}$ with $\nabla^2\mathbf{W}=-\mathbf{F}$ delivers both potentials at once.
The longitudinal and transverse subspaces are $L^2$ -orthogonal — why the split is a genuine projection, unique on all of $\mathbb{R}^3$ .
On a bounded or multiply-connected domain, uniqueness fails by exactly the harmonic fields — the Helmholtz–Hodge refinement, whose count is a topological invariant.
Key idea: divergence and curl are the two independent local invariants of a vector field; knowing both, plus decay or boundary data, reconstructs the field uniquely as $-\nabla\phi+\nabla\times\mathbf{A}$ , with $\phi$ inverting $\nabla\cdot\mathbf{F}$ and $\mathbf{A}$ inverting $\nabla\times\mathbf{F}$ through the Laplacian's Green function.

The core insight

Divergence and curl are the two independent local invariants of a vector field; knowing both — with decay or boundary data — reconstructs the field uniquely as $\mathbf{F}=-\nabla\phi+\nabla\times\mathbf{A}$ , where $\phi$ inverts $\nabla\cdot\mathbf{F}$ and $\mathbf{A}$ inverts $\nabla\times\mathbf{F}$ through the Laplacian's Green function.

Where this lives today

The Helmholtz–Hodge decomposition is a daily workhorse far beyond its fluid-dynamics birthplace. In computational fluid dynamics, Chorin's projection method (1968) advances an incompressible flow by computing the Helmholtz gradient part of an intermediate velocity from a pressure Poisson equation and subtracting it to enforce $\nabla\cdot\mathbf{u}=0$ each timestep — every unsteady incompressible solver, from weather models to visual-effects smoke, runs a Helmholtz projection in its inner loop. In medical imaging, 4D-flow MRI measures the full velocity field of blood in the heart, and a discrete Helmholtz–Hodge decomposition isolates the vortical component that flags valve disease and aneurysm risk. In quantum electrodynamics, the transverse projector $\delta_{ij}-\hat k_i\hat k_j$ of equation (5) sits in the photon propagator and in Coulomb-gauge quantization, isolating the two physical transverse photons from the non-dynamical longitudinal and scalar modes. And in data analysis, discrete exterior calculus splits fields sampled on meshes — ocean currents, brain connectivity, robot swarms — into curl-free, divergence-free, and harmonic parts, the last counting the domain's loops and voids and linking to persistent homology.

The physical picture

Picture a vector field as carrying two kinds of "source charge." Divergence-charge lives where field lines are born or die — a source sprays lines outward, a sink swallows them. Curl-charge lives where field lines wind — a vortex bends them into closed loops. These are geometrically distinct: you can spray without winding, wind without spraying, or do both. Helmholtz's theorem says they are the whole story: integrate the field's own divergence and curl against the Coulomb kernel $1/(4\pi|\mathbf{r}-\mathbf{r}'|)$ — the potential of a unit point source — and you have rebuilt $\mathbf{F}$ exactly. Nothing survives that is not already encoded in where it spreads and where it winds.

Figure

A vector field carries two independent kinds of source: a divergence part that spreads (left, irrotational, the −∇φ piece with ∇·F ≠ 0 and ∇×F = 0) and a curl part that circulates (right, solenoidal, the ∇×A piece with ∇·F = 0 and ∇×F ≠ 0)

In Fourier space the independence is even sharper: at each wavevector the component along $\mathbf{k}$ is longitudinal (curl-free, since $\mathbf{k}\times\mathbf{k}=0$ ) and the components across $\mathbf{k}$ are transverse (divergence-free, since $\mathbf{k}\cdot(\mathbf{k}\times\cdots)=0$ ), and because an orthogonal projection is unique, so is the whole decomposition. Geometrically this is a completeness statement — gradients and curls together span every decaying field, falling short only by the harmonic fields whose number counts the domain's topology. The next figure makes the split concrete on a computable field.

Figure

One field, two sources — the spiral F = (x+y, y−x) has uniform ∇·F = 2 and uniform ∇×F = −2 ẑ everywhere; at P = (2, 1) it separates into a radial part F∥ = (2, 1) from −∇φ and an azimuthal part F⊥ = (1, −2) from ∇×A, orthogonal so the squared magnitude splits as 10 = 5 + 5

The core equations

The theorem is a single line; the sign on $\phi$ is the electrostatics convention $\mathbf{E}=-\nabla\phi$ . The curl-free piece is a gradient because $\nabla\times(\nabla\phi)=\mathbf{0}$ identically, and the divergence-free piece a curl because $\nabla\cdot(\nabla\times\mathbf{A})=\mathbf{0}$ identically — the two identities that make "irrotational" and "solenoidal" self-consistent.

\mathbf{F}(\mathbf{r}) = -\nabla\phi(\mathbf{r}) + \nabla\times\mathbf{A}(\mathbf{r}) \qquad (1)

where: $\mathbf{F}$ — the given smooth, decaying vector field; $\phi$ — scalar potential, generates the irrotational/longitudinal part; $\mathbf{A}$ — vector potential, generates the solenoidal/transverse part; $-\nabla\phi=\mathbf{F}_{\parallel}$ — the curl-free piece; $\nabla\times\mathbf{A}=\mathbf{F}_{\perp}$ — the divergence-free piece. Reading it: every field is a gradient plus a curl, carrying its divergence and curl. The scalar potential is the divergence of $\mathbf{F}$ smeared by the Coulomb kernel, plus a boundary term that survives only on a finite region. Take the divergence of (1): since $\nabla\cdot(\nabla\times\mathbf{A})=0$ , $\nabla\cdot\mathbf{F}=-\nabla^2\phi$ — a Poisson equation for $\phi$ sourced by the divergence.

\phi(\mathbf{r}) = \frac{1}{4\pi}\int_{V}\frac{\nabla'\!\cdot\mathbf{F}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}^3r' \;-\; \frac{1}{4\pi}\oint_{S}\frac{\mathbf{F}(\mathbf{r}')\cdot\mathrm{d}\mathbf{S}'}{|\mathbf{r}-\mathbf{r}'|} \qquad (2)

where: $\nabla'\!\cdot\mathbf{F}$ — divergence of $\mathbf{F}$ at the source point $\mathbf{r}'$ ; $V,\,S$ — source volume and its bounding surface; $\mathrm{d}\mathbf{S}'$ — outward area element on $S$ ; $1/(4\pi|\mathbf{r}-\mathbf{r}'|)$ — Newtonian/Coulomb Green function of $-\nabla^2$ . Reading it: $\phi$ is the running total of nearby divergence, discounted by distance; the surface term dies for decaying fields. Symmetrically, the vector potential is the curl of $\mathbf{F}$ smeared by the same kernel. Taking the curl of (1), $\nabla\times(\nabla\phi)=\mathbf{0}$ and the double-curl identity give a vector Poisson equation $\nabla\times\mathbf{F}=-\nabla^2\mathbf{A}$ in the transverse (Coulomb) gauge $\nabla\cdot\mathbf{A}=0$ .

\mathbf{A}(\mathbf{r}) = \frac{1}{4\pi}\int_{V}\frac{\nabla'\!\times\mathbf{F}(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}^3r' \;+\; \frac{1}{4\pi}\oint_{S}\frac{\mathbf{F}(\mathbf{r}')\times\mathrm{d}\mathbf{S}'}{|\mathbf{r}-\mathbf{r}'|} \qquad (3)

where: $\nabla'\!\times\mathbf{F}$ — curl of $\mathbf{F}$ at the source point $\mathbf{r}'$ ; $\mathbf{F}\times\mathrm{d}\mathbf{S}'$ — boundary contribution, vanishing for decaying fields. Reading it: $\mathbf{A}$ is to the curl what $\phi$ is to the divergence — a distance-weighted census of vorticity. Both formulas rest on one fact: the Coulomb kernel is the Green function of the Laplacian — harmonic everywhere except the origin, where it hides a delta of strength $-4\pi$ .

\nabla^2\,\frac{1}{|\mathbf{r}-\mathbf{r}'|} = -4\pi\,\delta^3(\mathbf{r}-\mathbf{r}') \qquad (4)

where: $\delta^3$ — three-dimensional Dirac delta; $|\mathbf{r}-\mathbf{r}'|$ — distance between field and source points. Reading it: inverting $-\nabla^2$ means convolving with $1/(4\pi r)$ — why every potential wears that kernel. The cleanest existence proof packs both potentials into one superpotential $\mathbf{W}$ solving $\nabla^2\mathbf{W}=-\mathbf{F}$ , reading the split off the double-curl identity as $\phi=\nabla\cdot\mathbf{W}$ , $\mathbf{A}=\nabla\times\mathbf{W}$ (Derivation 1). Finally, the split is transparent in Fourier space, where derivatives become multiplication by $\mathbf{k}$ and it becomes an orthogonal projection at each wavevector.

\widehat{\mathbf{F}}(\mathbf{k})=\underbrace{(\hat{\mathbf{k}}\cdot\widehat{\mathbf{F}})\,\hat{\mathbf{k}}}_{\text{longitudinal}}+\underbrace{\big(\mathbf{1}-\hat{\mathbf{k}}\hat{\mathbf{k}}\big)\cdot\widehat{\mathbf{F}}}_{\text{transverse}} \qquad (5)

where: $\widehat{\mathbf{F}}(\mathbf{k})$ — Fourier transform of $\mathbf{F}$ ; $\hat{\mathbf{k}}=\mathbf{k}/|\mathbf{k}|$ — unit wavevector; $\hat{\mathbf{k}}\hat{\mathbf{k}}=\hat k_i\hat k_j$ — longitudinal projector; $\mathbf{1}-\hat{\mathbf{k}}\hat{\mathbf{k}}=\delta_{ij}-\hat k_i\hat k_j$ — transverse projector. Reading it: at every $\mathbf{k}$ the field splits into its shadow along and across $\mathbf{k}$ ; both projectors are idempotent, so the split is unique mode by mode.

Derivation AConstruction via the Laplacian Green function

The goal is to prove the decomposition exists and to hand back explicit formulas for both potentials, not just an abstract guarantee. The strategy: invert the Laplacian once on the whole field to get a vector superpotential $\mathbf{W}$ , then feed it through the identity $\nabla\times(\nabla\times\mathbf{W})=\nabla(\nabla\cdot\mathbf{W})-\nabla^2\mathbf{W}$ . Rearranging separates $\mathbf{F}$ into a pure gradient and a pure curl, and translating $\mathbf{W}$ back into the sources expresses each potential as a Coulomb-weighted integral of divergence or curl.

Step 1. Solve $\nabla^2\mathbf{W}=-\mathbf{F}$ componentwise with the Green function (4), giving $\mathbf{W}(\mathbf{r})=\tfrac{1}{4\pi}\int \mathbf{F}(\mathbf{r}')/|\mathbf{r}-\mathbf{r}'|\,\mathrm{d}^3r'$ . Why: each Cartesian component obeys a scalar Poisson equation, and convolving with $1/(4\pi r)$ inverts $-\nabla^2$ — the integral converges when $\mathbf{F}$ decays faster than $1/r^2$ .

Step 2. Apply the double-curl identity to this $\mathbf{W}$ and substitute $\nabla^2\mathbf{W}=-\mathbf{F}$ : $\mathbf{F}=-\nabla^2\mathbf{W}=-\nabla(\nabla\cdot\mathbf{W})+\nabla\times(\nabla\times\mathbf{W})$ . Why: the identity relating the vector Laplacian to grad-div and curl-curl automatically presents $\mathbf{F}$ as a gradient plus a curl — the algebraic heart of the theorem.

Step 3. Read off $\phi=\nabla\cdot\mathbf{W}$ and $\mathbf{A}=\nabla\times\mathbf{W}$ , so $\mathbf{F}=-\nabla\phi+\nabla\times\mathbf{A}$ . Why: the first term is then manifestly irrotational ( $\nabla\times\nabla\phi=0$ ) and the second manifestly solenoidal ( $\nabla\cdot\nabla\times\mathbf{A}=0$ ) — the labels are guaranteed, not hoped for.

Step 4. Move the derivatives from the kernel onto $\mathbf{F}$ by integration by parts, converting $\nabla\cdot\mathbf{W}$ and $\nabla\times\mathbf{W}$ into $\phi=\tfrac{1}{4\pi}\int \nabla'\!\cdot\mathbf{F}/|\mathbf{r}-\mathbf{r}'|\,\mathrm{d}^3r'$ and $\mathbf{A}=\tfrac{1}{4\pi}\int \nabla'\!\times\mathbf{F}/|\mathbf{r}-\mathbf{r}'|\,\mathrm{d}^3r'$ . Why: $\nabla(1/|\mathbf{r}-\mathbf{r}'|)=-\nabla'(1/|\mathbf{r}-\mathbf{r}'|)$ transfers the gradient to the primed variable; the boundary terms vanish for decaying fields, exposing divergence and curl as the sources — equations (2) and (3).

Step 5. Verify by acting back: $\nabla\cdot\mathbf{F}=-\nabla^2\phi$ and $\nabla\times\mathbf{F}=-\nabla^2\mathbf{A}$ recover the two sources, the consistency identities killing the cross terms. Why: reproducing the original $\nabla\cdot\mathbf{F}$ and $\nabla\times\mathbf{F}$ confirms the inversion is correct — cross-checked against Arfken–Weber §3.9 and Jackson §6.3.

Go deeper · Derivation B · 1 minUniqueness on all of space under decay

Derivation BUniqueness on all of space under decay

The goal is to show that on $\mathbb{R}^3$ , with the field falling off fast enough at infinity, the decomposition just built is the only one. The method is to assume two valid decompositions, subtract them to form a difference field with zero divergence and zero curl, and prove it must vanish. The proof turns on one energy integral from Green's first identity, whose boundary term dies exactly when decay holds — pinpointing decay as the single hypothesis doing the work.

Step 1. Let $\mathbf{D}=\mathbf{F}_1-\mathbf{F}_2$ be the difference of two decompositions of the same $\mathbf{F}$ ; then $\nabla\cdot\mathbf{D}=0$ and $\nabla\times\mathbf{D}=\mathbf{0}$ everywhere. Why: their difference has no net divergence and no net curl — a harmonic vector field, the only possible ambiguity.

Step 2. Curl-free on the simply connected $\mathbb{R}^3$ forces $\mathbf{D}=\nabla\psi$ , and divergence-free then forces $\nabla^2\psi=0$ , so $\psi$ is harmonic. Why: a vanishing curl is the integrability condition for a potential, and imposing $\nabla\cdot\mathbf{D}=0$ turns Poisson into Laplace.

Step 3. Apply Green's first identity to $\psi$ over a large ball: $\int|\nabla\psi|^2\,\mathrm{d}^3r=\oint \psi\,\nabla\psi\cdot\mathrm{d}\mathbf{S}-\int \psi\,\nabla^2\psi\,\mathrm{d}^3r$ . Why: it converts the volume integral of the nonnegative $|\nabla\psi|^2$ into a boundary flux plus a term that is zero because $\psi$ is harmonic.

Step 4. The volume term vanishes ( $\nabla^2\psi=0$ ) and the surface term dies as the ball grows if $\mathbf{F}$ — hence $\psi$ and $\nabla\psi$ — decays, leaving $\int|\nabla\psi|^2\,\mathrm{d}^3r=0$ , so $\nabla\psi=\mathbf{D}=\mathbf{0}$ . Why: a nonnegative integrand with zero integral is identically zero, giving $\mathbf{F}_1=\mathbf{F}_2$ ; on a bounded domain the surface term survives, an additive harmonic field remains, and uniqueness returns only after boundary data are fixed — the Helmholtz–Hodge case.

Go deeper · Derivation C · 1 minWhy the two parts are orthogonal in the L2 sense

Derivation CWhy the two parts are orthogonal in the L2 sense

The goal is to explain, beyond the componentwise Fourier argument, why the longitudinal and transverse subspaces meet only at zero, so the decomposition is a true orthogonal projection. The plan: compute the $L^2$ inner product of a gradient field with a curl field and show it vanishes by turning the integrand into a total divergence, whose integral is a surface flux that decay annihilates — making each part the closest gradient (or curl) to $\mathbf{F}$ .

Step 1. Write the integrand of $\langle-\nabla\phi,\ \nabla\times\mathbf{A}\rangle$ by the product rule as $(\nabla\phi)\cdot(\nabla\times\mathbf{A})=\nabla\cdot(\phi\,\nabla\times\mathbf{A})-\phi\,\nabla\cdot(\nabla\times\mathbf{A})$ , and drop the second term since $\nabla\cdot(\nabla\times\mathbf{A})=0$ . Why: this reduces the dot product to a single total divergence, all that a divergence-of-a-curl leaves behind.

Step 2. Integrate over $\mathbb{R}^3$ and apply Gauss's theorem: the volume integral becomes $\oint_S \phi\,(\nabla\times\mathbf{A})\cdot\mathrm{d}\mathbf{S}$ , vanishing at infinity for decaying fields, so $\langle\mathbf{F}_{\parallel},\mathbf{F}_{\perp}\rangle=0$ . Why: with the inner product zero the subspaces are perpendicular, the Pythagorean identity $\|\mathbf{F}\|^2=\|\mathbf{F}_{\parallel}\|^2+\|\mathbf{F}_{\perp}\|^2$ holds, and no nonzero field lies in both — uniqueness stated as geometry.

Limiting cases

The decomposition collapses to familiar special cases the moment one source switches off, and degenerates instructively when neither is present.

Case	Result	Meaning
Irrotational field	$\nabla\times\mathbf{F}=\mathbf{0}\ \Rightarrow\ \mathbf{F}=-\nabla\phi$	Pure conservative (longitudinal) field; the vector potential contributes nothing. Electrostatics, gravity, potential flow.
Solenoidal field	$\nabla\cdot\mathbf{F}=0\ \Rightarrow\ \mathbf{F}=\nabla\times\mathbf{A}$	Pure divergence-free (transverse) field; the scalar potential is constant. Magnetostatic $\mathbf{B}$ , incompressible flow.
Harmonic field on a domain	$\nabla\cdot\mathbf{H}=0,\ \nabla\times\mathbf{H}=\mathbf{0},\ \mathbf{H}\neq\mathbf{0}$	Neither source present yet the field is nonzero on bounded or multiply-connected regions; the ambiguity that boundary data or topology must fix.
Fourier / mode picture	$\widehat{\mathbf{F}}_{\parallel}=(\hat{\mathbf{k}}\cdot\widehat{\mathbf{F}})\hat{\mathbf{k}},\ \ \widehat{\mathbf{F}}_{\perp}=\widehat{\mathbf{F}}-\widehat{\mathbf{F}}_{\parallel}$	Per-wavevector orthogonal projection; longitudinal parallel to $\hat{\mathbf{k}}$ , transverse perpendicular to it.
Two dimensions	$\mathbf{F}=\nabla\phi+\nabla\times(\psi\,\hat{\mathbf z})$	The vector potential collapses to a single scalar stream function $\psi$ , conjugate-harmonic to $\phi$ for ideal flow.

Common trapThe split is automatic and unique on unbounded space, but not on a finite region

The clean, unique decomposition is a theorem about all of $\mathbb{R}^3$ with a field decaying faster than $1/r$ . Chop the domain to a finite box and two things break. First, uniqueness fails: you can always add a harmonic field $\mathbf{H}$ with $\nabla\cdot\mathbf{H}=0$ and $\nabla\times\mathbf{H}=\mathbf{0}$ , shifting $\mathbf{F}_{\parallel}$ and $\mathbf{F}_{\perp}$ without changing $\mathbf{F}$ — you must supply normal or tangential boundary data (the Helmholtz–Hodge decomposition) to pin it down. Second, the source integrals (2) and (3) grow surface terms that no longer vanish. A separate, permanent caveat: even on $\mathbb{R}^3$ the potentials $\phi$ and $\mathbf{A}$ are non-unique — $\mathbf{A}\to\mathbf{A}+\nabla\chi$ leaves $\nabla\times\mathbf{A}$ untouched (gauge freedom). Only the physical fields $\mathbf{F}_{\parallel}$ and $\mathbf{F}_{\perp}$ are unique.

Worked examples

1 · A point-source field is purely irrotational (confidence builder)

Take the inverse-square field $\mathbf{F}=\hat{\mathbf r}/r^2=\mathbf{r}/r^3$ , the field of a unit point charge in Gaussian units. Show it is entirely longitudinal, find its scalar potential directly, and confirm the source integral (2) returns the same answer.

Step 1. Compute the curl. For any radial field $f(r)\,\hat{\mathbf r}$ the curl vanishes identically, and here $\nabla\times(\mathbf{r}/r^3)=\mathbf{0}$ for $r\neq0$ . So $\mathbf{F}_{\perp}=\mathbf{0}$ and $\mathbf{F}=-\nabla\phi$ alone — a purely irrotational field.

Step 2. Read off the potential by inspection: $-\nabla(1/r)=\mathbf{r}/r^3=\mathbf{F}$ , so $\phi=1/r$ — the Coulomb potential of a point charge.

Step 3. Now get $\phi$ the hard way, from the divergence. By identity (4), $\nabla\cdot(\mathbf{r}/r^3)=4\pi\,\delta^3(\mathbf{r})$ — all the divergence sits at the origin. Feed this into (2): $\phi(\mathbf{r})=\tfrac{1}{4\pi}\int \dfrac{4\pi\,\delta^3(\mathbf{r}')}{|\mathbf{r}-\mathbf{r}'|}\,\mathrm{d}^3r'=\dfrac{1}{r}$ .

Step 4. Both routes agree: $\boldsymbol{\phi=1/r}$ . The divergence — a single delta of weight $4\pi$ — is the complete source, reconstructed exactly. The field is 100 % longitudinal, $\mathbf{F}_{\perp}=\mathbf{0}$ , $\phi=1/r$ .

2 · Decomposing a spiral field by hand (fully worked)

Split the planar field $\mathbf{F}(x,y,z)=(x+y,\ y-x,\ 0)$ into its irrotational and solenoidal parts, evaluate at $P=(2,1,0)$ , and verify the $L^2$ geometry. This is the field drawn in the second figure.

Step 1. Measure the two sources. The divergence is $\nabla\cdot\mathbf{F}=\partial_x(x+y)+\partial_y(y-x)=1+1=\mathbf{2}$ , a uniform source. The curl has only a $z$ -component, $(\nabla\times\mathbf{F})_z=\partial_x(y-x)-\partial_y(x+y)=-1-1=\mathbf{-2}$ , a uniform vortex — both constant, so the field spreads and winds identically everywhere.

Step 2. Read the potentials off the structure. The radial part $(x,y,0)$ carries divergence $2$ and no curl; the azimuthal part $(y,-x,0)$ carries curl $-2\,\hat{\mathbf z}$ and no divergence. Matching $\mathbf{F}=-\nabla\phi+\nabla\times\mathbf{A}$ : take $\phi=-\tfrac12(x^2+y^2)$ , so $-\nabla\phi=(x,y,0)=\mathbf{F}_{\parallel}$ ; and $\mathbf{A}=\tfrac12(x^2+y^2)\,\hat{\mathbf z}$ , so $\nabla\times\mathbf{A}=(y,-x,0)=\mathbf{F}_{\perp}$ .

Step 3. Check the split reproduces $\mathbf{F}$ : $\mathbf{F}_{\parallel}+\mathbf{F}_{\perp}=(x,y,0)+(y,-x,0)=\mathbf{F}$ . And each part owns one source: $\nabla\cdot\mathbf{F}_{\parallel}=2$ , $\nabla\times\mathbf{F}_{\parallel}=\mathbf{0}$ ; $\nabla\times\mathbf{F}_{\perp}=-2\,\hat{\mathbf z}$ , $\nabla\cdot\mathbf{F}_{\perp}=0$ .

Step 4. Evaluate at $P=(2,1,0)$ : $\mathbf{F}=\mathbf{(3,-1,0)}$ ; $\mathbf{F}_{\parallel}=\mathbf{(2,1,0)}$ ; $\mathbf{F}_{\perp}=\mathbf{(1,-2,0)}$ . The two parts are perpendicular everywhere, since $\mathbf{F}_{\parallel}\cdot\mathbf{F}_{\perp}=xy+y(-x)=0$ — at $P$ , $2\cdot1+1\cdot(-2)=\mathbf{0}$ .

Step 5. Confirm Pythagoras: $|\mathbf{F}|^2=3^2+(-1)^2=\mathbf{10}$ , while $|\mathbf{F}_{\parallel}|^2+|\mathbf{F}_{\perp}|^2=(2^2+1^2)+(1^2+2^2)=5+5=\mathbf{10}$ . The energy splits with no cross term — the orthogonality of Derivation 3 made concrete. At $P$ : $\mathbf{F}=(3,-1,0)=(2,1,0)+(1,-2,0)$ , orthogonal, $|\mathbf{F}|^2=10=5+5$ . A caveat: this field grows at infinity, so the source integrals (2) and (3) diverge. The algebraic split still exists and is what we used — closed-form potentials need decay, but the local decomposition is always available.

Physics of the portrait

The decomposition is named for the physicist who, chasing vortices in an ideal fluid, first proved a field is nothing more than its sources.

Origins · 1 minHermann von Helmholtz and how this was discovered

Hermann von Helmholtz

1821–1894 · the field decomposition and the vortex theorems of hydrodynamics

Hermann von Helmholtz was the last great universalist of physics — an army surgeon who reshaped physiology, acoustics, optics, and electrodynamics, and gave one of the first sharp statements of energy conservation (1847). The result carrying his name into vector calculus came in 1858, in his study of vortex motion in an ideal fluid: to follow how vorticity moves he separated a velocity field into a part driven by expansion and a part driven by rotation, proving the vortex theorems — vortex lines move with the fluid, their strength conserved — that later inspired Kelvin's "vortex atom" and all of modern vortex dynamics. The clean modern statement, any decaying field is a gradient plus a curl fixed by its divergence and curl, distils that analysis. Stokes had used the same split in 1849 for elastic waves, Gibbs recast it in $\nabla$ notation, Blumenthal proved existence rigorously in 1905, and Hodge extended it to curved spaces. Helmholtz's own move — invert the Laplacian, read a field off its sources — is what physicists still make whenever they introduce a scalar and vector potential.

History

1849

— Stokes splits an elastic displacement field into irrotational and solenoidal parts

1858

— Helmholtz proves the vortex theorems and the field decomposition in hydrodynamics

1881

— Gibbs recasts the theorem in modern vector (nabla) notation in his lecture notes

1905

— Blumenthal gives the first rigorous existence proof for general vector fields

1941

— Hodge extends the decomposition to differential forms on curved manifolds

Going further

Texts. Arfken, Weber & Harris, Mathematical Methods for Physicists, 7th ed., §3.9 (Helmholtz's theorem) · Hassani, Mathematical Physics, §15 (Green functions and the Laplacian) · Stone & Goldbart, Mathematics for Physics, §10 (vector calculus and the Helmholtz decomposition) · Nakahara, Geometry, Topology and Physics, Ch. 7 (Hodge theory, the differential-forms sequel) · Byron & Fuller, Mathematics of Classical and Quantum Physics, Ch. 1 · Jackson, Classical Electrodynamics, 3rd ed., §6.3.

Provenance. All equations are textbook-standard. A verifier should confirm: (1) $\nabla^2(1/|\mathbf{r}-\mathbf{r}'|)=-4\pi\,\delta^3$ by the small-ball flux argument (Jackson §1.7); (2) the double-curl identity $\nabla\times\nabla\times=\nabla(\nabla\cdot)-\nabla^2$ componentwise; (3) that taking $\nabla\cdot$ and $\nabla\times$ of the boxed Helmholtz result returns $\nabla\cdot\mathbf{F}$ and $\nabla\times\mathbf{F}$ ; (4) the uniqueness argument via Green's first identity; (5) Worked Example 2 — $\nabla\cdot\mathbf{F}=2$ , $\nabla\times\mathbf{F}=-2\,\hat{\mathbf z}$ , and at $P=(2,1,0)$ , $(3,-1,0)=(2,1,0)+(1,-2,0)$ with $|\mathbf{F}|^2=10=5+5$ . The only numerical constant beyond the geometric $4\pi$ is exact.

Prerequisites. Divergence of a Vector Field · Curl of a Vector Field · The Laplacian Operator · The Del Operator and Vector Differential Identities · The Divergence (Gauss) Theorem · Volume Integrals. Leads to. Stokes' Theorem · Surface Integrals and Flux · The Gradient Operator.

Ask the tutor: Want to see the Helmholtz–Hodge harmonic term appear explicitly on a torus, or watch a CFD solver project out the gradient part? Just ask.

Is the Helmholtz decomposition the same as saying "every field is a gradient plus a curl"

Almost — that slogan is the conclusion, but it hides two hypotheses. The field must be smooth enough to differentiate twice, and on unbounded space it must decay faster than $1/r$ so the source integrals converge and the boundary flux dies. Granted those, yes: $\mathbf{F}=-\nabla\phi+\nabla\times\mathbf{A}$ , gradient part owning the divergence, curl part owning the curl. On a finite region the slogan needs a third term — a harmonic field — which is where Hodge theory and topology enter.

What is the difference between "irrotational" and "solenoidal"

They name the two halves by what each lacks. An irrotational field has zero curl everywhere; it does not circulate, is a gradient $-\nabla\phi$ , and its line integral around any closed loop is zero — electrostatic $\mathbf{E}$ , gravity. A solenoidal field has zero divergence; its lines never begin or end but close on themselves, it is a curl $\nabla\times\mathbf{A}$ , and its flux through any closed surface is zero — magnetic $\mathbf{B}$ , incompressible flow. Helmholtz says any field is one of each, stacked together.

Why does a surface term appear in the potential formulas

Because inverting the Laplacian is an integration by parts, and that always leaves a boundary piece. Moving the derivative off the Coulomb kernel and onto $\mathbf{F}$ , Gauss's theorem hands you a surface integral over the region's boundary. On all of $\mathbb{R}^3$ that boundary is the sphere at infinity, and if $\mathbf{F}$ decays fast enough the term vanishes, leaving the clean volume integrals. On a finite domain the boundary is real and the surface term stays — precisely the information the boundary conditions must supply.

How is this related to gauge freedom in electromagnetism

Directly. The magnetic field $\mathbf{B}$ is solenoidal, so Helmholtz writes it as $\nabla\times\mathbf{A}$ — but $\mathbf{A}$ is not unique, since adding any gradient $\nabla\chi$ changes nothing: $\nabla\times(\mathbf{A}+\nabla\chi)=\nabla\times\mathbf{A}$ . That ambiguity is gauge freedom. The Coulomb gauge $\nabla\cdot\mathbf{A}=0$ makes $\mathbf{A}$ purely transverse — the Helmholtz solenoidal potential with no longitudinal contamination. Gauge and decomposition are the same story: the physical field $\mathbf{F}_{\perp}$ is unique, its potential is not.

Where do topology and Hodge theory come in

The moment the domain has holes. On simply connected unbounded space, curl-free implies gradient and divergence-free implies curl, so gradients and curls span everything. But on a region with a tunnel or cavity — a solid torus, say — there exist harmonic fields, curl-free and divergence-free, that are not globally a gradient or a curl. Their number is a topological invariant (a Betti number), the harmonic remainder in the Helmholtz–Hodge decomposition. Recast in differential forms this becomes Hodge's theorem, the analytic heart of de Rham cohomology — geometry reading topology off a field.

Can I always find the potentials in closed form

Rarely, and that is fine. The theorem guarantees $\phi$ and $\mathbf{A}$ exist and gives them as convolutions of the sources with $1/(4\pi r)$ , but those integrals close up only for simple, symmetric sources — a point charge, a line of current, the spiral in Worked Example 2. In practice one solves the Poisson equations $\nabla^2\phi=-\nabla\cdot\mathbf{F}$ and $\nabla^2\mathbf{A}=-\nabla\times\mathbf{F}$ numerically, exactly what a CFD pressure-projection step does. The theorem's value is structural: it says the sources are all you need, and which equation to solve.