Orthogonal Curvilinear Coordinates

Almost every hard problem in physics is easy in the right coordinates and hopeless in the wrong ones. A hydrogen atom, a coaxial cable, a spinning star, a drumhead — each carries a symmetry, and the art is to draw a coordinate grid that bends to match it. Orthogonal curvilinear coordinates make this precise: you replace the flat Cartesian triple $(x,y,z)$ with a smooth invertible relabelling $(u_1,u_2,u_3)$ whose coordinate surfaces meet everywhere at right angles — spheres and cones for spherical coordinates, cylinders and planes for cylindrical, and a zoo of parabolic, elliptic, and toroidal grids beyond. The entire geometry of such a grid is compressed into three numbers at each point — the scale factors (Lamé coefficients) $h_i=|\partial\mathbf{r}/\partial u_i|$ . Each $h_i$ is a local ruler-conversion: how many metres of real distance you cross when you nudge the label $u_i$ by one unit and hold the others fixed. Because the $u_i$ can be angles, radii, or dimensionless parameters, the $h_i$ carry all the dimensional bookkeeping and all the local stretching of space. Once you know them, every metric quantity — arc length, area, volume, and the whole first-order arsenal of vector calculus — is fixed by universal master formulas. The pivot is orthogonality: it diagonalizes the metric tensor, $g_{ij}=h_i^2\,\delta_{ij}$ , and a diagonal metric is exactly what lets $\nabla$ , $\nabla\cdot$ , $\nabla\times$ , and $\nabla^2$ stay transparent instead of dissolving into the covariant–contravariant machinery of general tensor calculus. This article is the geometric engine room beneath the operator articles that follow: it builds the scale factors from scratch, derives the master operators, and specializes them to the cylindrical and spherical forms you will use for life.

Key takeaways

Orthogonal curvilinear coordinates $(u_1,u_2,u_3)$ replace Cartesian $(x,y,z)$ with coordinate surfaces that meet at right angles everywhere, so the local frame $\{\hat{\mathbf{e}}_1,\hat{\mathbf{e}}_2,\hat{\mathbf{e}}_3\}$ is orthonormal but turns from point to point.
All the geometry lives in three scale factors $h_i=\left|\partial\mathbf{r}/\partial u_i\right|$ , which convert a coordinate step $\mathrm{d}u_i$ into a physical length $h_i\,\mathrm{d}u_i$ .
Lengths, areas, and volume follow at once: $\mathrm{d}s^2=\sum_i h_i^2\,\mathrm{d}u_i^2$ and $\mathrm{d}V=h_1h_2h_3\,\mathrm{d}u_1\mathrm{d}u_2\mathrm{d}u_3$ .
Grad, div, curl, and the Laplacian all pick up scale-factor weights; the master formulas reproduce the cylindrical and spherical operators as special cases.
Orthogonality makes the metric diagonal, $g_{ij}=h_i^2\delta_{ij}$ , which is precisely why the calculus stays simple compared with general (non-orthogonal) systems.
Key idea: Everything about an orthogonal curvilinear system — lengths, areas, volumes, and all of vector calculus — is generated by the three scale factors $h_i=\left|\partial\mathbf{r}/\partial u_i\right|$ , because orthogonality makes the metric diagonal, $g_{ij}=h_i^2\delta_{ij}$ .

Where this lives today

The scale-factor machinery is quietly load-bearing across modern computation. Every finite-element and finite-volume solver — the tools that design jet engines, model blood flow, and simulate tokamak plasmas — maps a distorted physical mesh to a reference cube, and the Jacobian $h_1h_2h_3=\sqrt{g}$ is the weight that makes the discretized divergence conserve flux cell by cell; get the metric wrong and the simulation leaks mass. In numerical relativity the same $\sqrt{g}$ generalizes to the spacetime metric, and orthogonal-coordinate intuition survives in the "3+1" split used to evolve merging black holes for LIGO–Virgo templates. Separation of variables still generates the special functions of applied physics: spherical harmonics for the cosmic microwave background multipoles measured by Planck, Bessel modes for optical-fibre and cavity design, spheroidal wavefunctions for antennas. Exactly eleven coordinate systems separate the 3D Helmholtz equation, each matching a symmetry a real device can exploit. Whenever a modern solver respects a geometry, Lamé's scale factors are underneath.

The physical picture

Picture the labels $u_i$ as markings on a stretchy, curved grid laid over space. The scale factor $h_i$ is the local ruler-conversion: a coordinate step $\mathrm{d}u_i$ corresponds to a true distance $h_i\,\mathrm{d}u_i$ in metres. In cylindrical coordinates a step $\mathrm{d}\phi$ near the axis sweeps almost no distance, while the same $\mathrm{d}\phi$ far out sweeps a long arc — because $h_\phi=\rho$ grows with radius. The first figure makes that literal: equal angular steps trace unequal physical arcs, and the local frame stays orthonormal throughout.

Figure

A plane-polar curvilinear grid where circles ρ = const and rays φ = const meet at right angles, and the local frame ê_ρ ê_φ is orthonormal at every point. The same angular step dφ = 0.44 rad traces an arc of physical length ρ dφ — about 0.44 at ρ = 1 but 1.31 at ρ = 3 — so the azimuthal scale factor is h_φ = ρ.

Divergence and curl look busier than in Cartesian coordinates because the unit vectors themselves rotate as you move: $\partial\hat{\mathbf{e}}_\rho/\partial\phi=\hat{\mathbf{e}}_\phi\neq0$ . Some of what looks like a change in a field is really the basis turning underneath it, and the scale factors separate the two. The clean way to see this is an infinitesimal coordinate box: its edge lengths are $h_i\,\mathrm{d}u_i$ , its face areas $h_jh_k\,\mathrm{d}u_j\mathrm{d}u_k$ , and its volume $h_1h_2h_3\,\mathrm{d}u_1\mathrm{d}u_2\mathrm{d}u_3$ . The second figure builds that box with real numbers.

Figure

The infinitesimal coordinate box has physical edge lengths h₁ du₁, h₂ du₂, h₃ du₃, so its volume is dV = h₁h₂h₃ du₁du₂du₃. Specialized to spherical coordinates at r = 2, θ = 60°, the edges are dr, r dθ = 2 dθ, and r sinθ dφ = 1.73 dφ, giving dV = r² sinθ dr dθ dφ = 3.46 dr dθ dφ.

The core equations

Everything begins with one definition: differentiate $\mathbf{r}(u_1,u_2,u_3)$ along a single coordinate curve. The length of that derivative is the scale factor, its direction is the local unit vector, and these two objects generate the whole apparatus.

h_i=\left|\frac{\partial\mathbf{r}}{\partial u_i}\right|,\qquad \hat{\mathbf{e}}_i=\frac{1}{h_i}\frac{\partial\mathbf{r}}{\partial u_i} \qquad (1)

where: $\mathbf{r}$ — position vector $(x,y,z)$ as a function of $(u_1,u_2,u_3)$ ; $u_i$ — the $i$ -th curvilinear coordinate; $h_i$ — scale factor (Lamé coefficient) for $u_i$ ; $\hat{\mathbf{e}}_i$ — local unit tangent to the $u_i$ curve. Reading: $h_i$ is the metres-per-unit- $u_i$ conversion; dividing by it renormalizes the tangent to unit length. The system is orthogonal exactly when those tangent directions are mutually perpendicular at every point — the single condition that diagonalizes the metric tensor, whose entries are the dot products of the coordinate tangents.

\hat{\mathbf{e}}_i\cdot\hat{\mathbf{e}}_j=\delta_{ij},\qquad g_{ij}=\frac{\partial\mathbf{r}}{\partial u_i}\cdot\frac{\partial\mathbf{r}}{\partial u_j}=h_i^2\,\delta_{ij} \qquad (2)

where: $\delta_{ij}$ — Kronecker delta (1 if $i=j$ , else 0); $g_{ij}$ — metric tensor, diagonal precisely because the system is orthogonal. Reading: the off-diagonal entries vanish, so $g=\det(g_{ij})=h_1^2h_2^2h_3^2$ and $\sqrt{g}=h_1h_2h_3$ . Because the metric is diagonal, the squared distance is a term-by-term Pythagorean sum with no cross terms — the line element from which the area and volume elements drop out immediately.

\mathrm{d}s^2=h_1^2\,\mathrm{d}u_1^2+h_2^2\,\mathrm{d}u_2^2+h_3^2\,\mathrm{d}u_3^2,\qquad \mathrm{d}V=h_1h_2h_3\,\mathrm{d}u_1\,\mathrm{d}u_2\,\mathrm{d}u_3 \qquad (3)

where: $\mathrm{d}s$ — physical infinitesimal arc length; $\mathrm{d}V=\sqrt{g}\,\mathrm{d}u_1\mathrm{d}u_2\mathrm{d}u_3$ — physical volume of the coordinate box, with $\sqrt{g}=h_1h_2h_3$ the Jacobian of the transformation. The gradient measures physical rate of change, so each label-derivative is divided by its ruler-conversion $h_i$ to turn per-unit- $u_i$ into per-metre.

\nabla f=\frac{1}{h_1}\frac{\partial f}{\partial u_1}\hat{\mathbf{e}}_1+\frac{1}{h_2}\frac{\partial f}{\partial u_2}\hat{\mathbf{e}}_2+\frac{1}{h_3}\frac{\partial f}{\partial u_3}\hat{\mathbf{e}}_3 \qquad (4)

where: $f$ — a scalar field; $\nabla f$ — its gradient. The divergence and curl are heavier because they must account for the coordinate face areas changing from place to place, so the scale factors sit inside the derivatives, not outside.

\nabla\cdot\mathbf{A}=\frac{1}{h_1h_2h_3}\left[\frac{\partial}{\partial u_1}(h_2h_3A_1)+\frac{\partial}{\partial u_2}(h_3h_1A_2)+\frac{\partial}{\partial u_3}(h_1h_2A_3)\right] \qquad (5)

where: $\mathbf{A}=\sum_i A_i\hat{\mathbf{e}}_i$ — a vector field; $A_i$ — its physical component along $\hat{\mathbf{e}}_i$ .

\nabla\times\mathbf{A}=\frac{1}{h_1h_2h_3}\begin{vmatrix} h_1\hat{\mathbf{e}}_1 & h_2\hat{\mathbf{e}}_2 & h_3\hat{\mathbf{e}}_3 \\ \dfrac{\partial}{\partial u_1} & \dfrac{\partial}{\partial u_2} & \dfrac{\partial}{\partial u_3} \\ h_1A_1 & h_2A_2 & h_3A_3 \end{vmatrix} \qquad (6)

where: $h_iA_i$ — the covariant components that appear because the curl integrates $\mathbf{A}$ along coordinate edges of length $h_i\,\mathrm{d}u_i$ . The Laplacian is the composition $\nabla\cdot(\nabla f)$ — the single most-used operator in the subject, the heart of Laplace's, Poisson's, Helmholtz's, and Schrödinger's equations.

\nabla^2 f=\frac{1}{h_1h_2h_3}\left[\frac{\partial}{\partial u_1}\!\left(\frac{h_2h_3}{h_1}\frac{\partial f}{\partial u_1}\right)+\frac{\partial}{\partial u_2}\!\left(\frac{h_3h_1}{h_2}\frac{\partial f}{\partial u_2}\right)+\frac{\partial}{\partial u_3}\!\left(\frac{h_1h_2}{h_3}\frac{\partial f}{\partial u_3}\right)\right] \qquad (7)

where: $\nabla^2 f$ — Laplacian of $f$ . Reading: each term carries the weight $h_1h_2h_3/h_i^2=h_jh_k/h_i$ ; feed in the spherical scale factors and the radial $\tfrac{1}{r^2}\partial_r(r^2\partial_r)$ falls out.

Derivation AThe diagonal metric and the Pythagorean line element

The goal is to show why orthogonality is the one property that keeps curvilinear vector calculus tractable. It forces the metric to be diagonal, $g_{ij}=h_i^2\delta_{ij}$ , and a diagonal metric collapses the general quadratic form for squared distance into a clean term-by-term Pythagorean sum $\mathrm{d}s^2=\sum_i h_i^2\,\mathrm{d}u_i^2$ with every cross term gone. We assemble the displacement $\mathrm{d}\mathbf{r}$ from the chain rule, dot it with itself, and watch orthonormality do the killing.

Step 1. By the chain rule the displacement is $\mathrm{d}\mathbf{r}=\sum_i(\partial\mathbf{r}/\partial u_i)\,\mathrm{d}u_i=\sum_i h_i\hat{\mathbf{e}}_i\,\mathrm{d}u_i$ , using definition (1) to trade each tangent for its scale factor times a unit vector. Why: this re-expresses a step in abstract label-space as a sum of physical displacements along the local orthonormal frame.

Step 2. Dot the increment with itself: $\mathrm{d}s^2=\mathrm{d}\mathbf{r}\cdot\mathrm{d}\mathbf{r}=\sum_{i,j}h_ih_j(\hat{\mathbf{e}}_i\cdot\hat{\mathbf{e}}_j)\,\mathrm{d}u_i\,\mathrm{d}u_j$ . Why: the squared arc length is the metric acting on the displacement, and here the metric appears explicitly as the array of frame dot products $\hat{\mathbf{e}}_i\cdot\hat{\mathbf{e}}_j$ .

Step 3. Impose orthonormality $\hat{\mathbf{e}}_i\cdot\hat{\mathbf{e}}_j=\delta_{ij}$ , so only the $i=j$ terms survive: $\mathrm{d}s^2=\sum_i h_i^2\,\mathrm{d}u_i^2$ . Why: this is the whole payoff — a non-orthogonal system keeps the off-diagonal $2g_{ij}\,\mathrm{d}u_i\mathrm{d}u_j$ cross terms and forces the full covariant/contravariant apparatus, which orthogonality spares us. *As a quick check: cylindrical $h_\rho=1,h_\phi=\rho,h_z=1$ gives $\mathrm{d}s^2=\mathrm{d}\rho^2+\rho^2\mathrm{d}\phi^2+\mathrm{d}z^2$ ; spherical $h_r=1,h_\theta=r,h_\phi=r\sin\theta$ gives $\mathrm{d}s^2=\mathrm{d}r^2+r^2\mathrm{d}\theta^2+r^2\sin^2\theta\,\mathrm{d}\phi^2$ .*

Go deeper · Derivation B · 1 minThe divergence from flux through a coordinate box

Derivation BThe divergence from flux through a coordinate box

The aim is to build the scale-factor divergence directly from its geometric definition, $\nabla\cdot\mathbf{A}=\lim_{V\to0}V^{-1}\oint\mathbf{A}\cdot\mathrm{d}\mathbf{S}$ , by computing the net outward flux through an infinitesimal coordinate box and dividing by its volume. The subtle point — the reason the scale factors end up inside the derivatives — is that the box-face areas are themselves functions of the coordinates, so they cannot be pulled out of the difference between opposite faces.

Step 1. Take a box with edges $\mathrm{d}u_1,\mathrm{d}u_2,\mathrm{d}u_3$ at $(u_1,u_2,u_3)$ . The two faces of constant $u_1$ have area $h_2h_3\,\mathrm{d}u_2\,\mathrm{d}u_3$ , and only $A_1$ pierces them, so the net flux is $\Phi_1=[(A_1h_2h_3)|_{u_1+\mathrm{d}u_1}-(A_1h_2h_3)|_{u_1}]\,\mathrm{d}u_2\,\mathrm{d}u_3$ . Why: the flux is field-times-area, and because $h_2h_3$ varies with $u_1$ it must stay locked inside the face-to-face difference.

Step 2. Taylor-expand the difference to first order: $\Phi_1=\dfrac{\partial}{\partial u_1}(A_1h_2h_3)\,\mathrm{d}u_1\,\mathrm{d}u_2\,\mathrm{d}u_3$ , and the $u_2,u_3$ face-pairs give the cyclic partners. Why: the whole $u_1$ -derivative acts on the product $A_1h_2h_3$ , capturing both the changing field and the changing face area in one term.

Step 3. Sum the three pairs and divide by the box volume $\mathrm{d}V=h_1h_2h_3\,\mathrm{d}u_1\mathrm{d}u_2\mathrm{d}u_3$ ; the coordinate differentials cancel and leave $\nabla\cdot\mathbf{A}=\dfrac{1}{h_1h_2h_3}\sum_i\dfrac{\partial}{\partial u_i}\!\left(\dfrac{h_1h_2h_3}{h_i}A_i\right)$ . Why: this is equation (5) in compact form — the flux definition has manufactured the scale-factor weights automatically. *As a check: inserting $h_\rho=1,h_\phi=\rho,h_z=1$ recovers $\nabla\cdot\mathbf{A}=\tfrac1\rho\partial_\rho(\rho A_\rho)+\tfrac1\rho\partial_\phi A_\phi+\partial_z A_z$ , the standard cylindrical divergence.*

Go deeper · Derivation C · 1 minThe Laplacian and its spherical specialization

Derivation CThe Laplacian and its spherical specialization

The purpose is to assemble the master Laplacian by composing gradient and divergence, then specialize it to spherical coordinates and recover the operator that governs the hydrogen atom, multipole expansions, and every central-potential problem. The composition is mechanical once the two operators are in hand; the specialization inserts $h_r=1,h_\theta=r,h_\phi=r\sin\theta$ and exposes where the centrifugal $1/r^2$ factors come from.

Step 1. Take the gradient components $(\nabla f)_i=\tfrac{1}{h_i}\partial_{u_i}f$ from equation (4), then feed them as the vector $\mathbf{A}$ into the divergence (5): $\nabla^2 f=\nabla\cdot(\nabla f)=\dfrac{1}{h_1h_2h_3}\sum_i\partial_{u_i}\!\big(\tfrac{h_1h_2h_3}{h_i^2}\partial_{u_i}f\big)$ . Why: each term inherits one $1/h_i$ from the gradient and one from the flux weight $h_1h_2h_3/h_i$ , combining to $h_1h_2h_3/h_i^2=h_jh_k/h_i$ .

Step 2. Insert the spherical scale factors, so $h_rh_\theta h_\phi=r^2\sin\theta$ . The radial term is $\tfrac{1}{r^2\sin\theta}\partial_r(r^2\sin\theta\,\partial_r f)=\tfrac{1}{r^2}\partial_r(r^2\partial_r f)$ , since $h_\theta h_\phi/h_r=r^2\sin\theta$ cancels the prefactor's $\sin\theta$ . Why: this is the radial kinetic term whose eigenfunctions build the hydrogenic radial wavefunctions.

Step 3. The polar term uses $h_\phi h_r/h_\theta=\sin\theta$ , giving $\tfrac{1}{r^2\sin\theta}\partial_\theta(\sin\theta\,\partial_\theta f)$ ; the azimuthal term uses $h_rh_\theta/h_\phi=1/\sin\theta$ , giving $\tfrac{1}{r^2\sin^2\theta}\partial_\phi^2 f$ . Together, $\nabla^2 f=\tfrac{1}{r^2}\partial_r(r^2\partial_r f)+\tfrac{1}{r^2\sin\theta}\partial_\theta(\sin\theta\,\partial_\theta f)+\tfrac{1}{r^2\sin^2\theta}\partial_\phi^2 f$ . Why: the angular block is exactly $-\hat{L}^2/(\hbar^2 r^2)$ in quantum mechanics, so the $1/r^2$ prefactor is the origin of the centrifugal barrier $\ell(\ell+1)/r^2$ . *The angular part of the spherical Laplacian reproduces the $-\hat{L}^2/\hbar^2$ operator of central-force quantum mechanics.*

Limiting cases

The master formulas collapse to every standard operator by inserting the right scale factors — the table below is the whole family of everyday systems as one parametrization.

Case	Result	Meaning
Cartesian	$h_x=h_y=h_z=1$	all scale factors unity; $\nabla^2 f=\partial_x^2 f+\partial_y^2 f+\partial_z^2 f$ — the flat, undistorted reference
Cylindrical $(\rho,\phi,z)$	$h_\rho=1,\ h_\phi=\rho,\ h_z=1$	$\mathrm{d}V=\rho\,\mathrm{d}\rho\,\mathrm{d}\phi\,\mathrm{d}z$ ; the $\rho$ in $h_\phi$ says angular steps span longer arcs farther out
Spherical $(r,\theta,\phi)$	$h_r=1,\ h_\theta=r,\ h_\phi=r\sin\theta$	$\mathrm{d}V=r^2\sin\theta\,\mathrm{d}r\,\mathrm{d}\theta\,\mathrm{d}\phi$ ; recovers the $\tfrac{1}{r^2}\partial_r(r^2\partial_r)$ Laplacian of atomic physics
Conformal / isothermal (2D)	$h_1=h_2=\lambda(u_1,u_2)$	equal scale factors make the map angle-preserving; the 2D Laplacian becomes $\lambda^{-2}(\partial_{u_1}^2+\partial_{u_2}^2)$ — the basis of complex-analytic methods

Common trapthe unit vectors are not constant

The single most common error is to treat $\hat{\mathbf{e}}_\rho,\hat{\mathbf{e}}_\theta,\hat{\mathbf{e}}_\phi$ like the fixed Cartesian $\hat{\mathbf{x}},\hat{\mathbf{y}},\hat{\mathbf{z}}$ and pull them outside derivatives. They are position-dependent: $\partial\hat{\mathbf{e}}_\rho/\partial\phi=\hat{\mathbf{e}}_\phi$ and $\partial\hat{\mathbf{e}}_\phi/\partial\phi=-\hat{\mathbf{e}}_\rho$ . Ignoring this produces wrong divergences, missing centrifugal and Coriolis-type terms in Navier–Stokes, and a broken curl. The master formulas carry the $h_i$ factors precisely so you never differentiate the basis by hand — but the moment you write out components by hand, remember the frame turns under you.

Worked examples

1 · Building the cylindrical scale factors from scratch (confidence builder)

From the map $x=\rho\cos\phi,\ y=\rho\sin\phi,\ z=z$ , compute the three scale factors, confirm orthogonality, and write the line and volume elements.

Step 1. With $\mathbf{r}=(\rho\cos\phi,\rho\sin\phi,z)$ , differentiate along each coordinate. Radially, $\partial\mathbf{r}/\partial\rho=(\cos\phi,\sin\phi,0)$ , whose length is $h_\rho=\sqrt{\cos^2\phi+\sin^2\phi}=\mathbf{1}$ : a unit of $\rho$ is a metre, radial motion is isometric.

Step 2. Azimuthally, $\partial\mathbf{r}/\partial\phi=(-\rho\sin\phi,\rho\cos\phi,0)$ , length $h_\phi=\sqrt{\rho^2\sin^2\phi+\rho^2\cos^2\phi}=\boldsymbol{\rho}$ : the dimensionless angle $\mathrm{d}\phi$ becomes arc length $\rho\,\mathrm{d}\phi$ , the $h_\phi=\rho$ the first figure measured. Axially, $\partial\mathbf{r}/\partial z=(0,0,1)$ gives $h_z=\mathbf{1}$ .

Step 3. Check orthogonality: $(\cos\phi,\sin\phi,0)\cdot(-\rho\sin\phi,\rho\cos\phi,0)=0$ , and both are perpendicular to $(0,0,1)$ . All cross dot products vanish, so the metric is diagonal and the system is orthogonal.

Step 4. Assemble the geometry: $\mathrm{d}s^2=\mathrm{d}\rho^2+\rho^2\mathrm{d}\phi^2+\mathrm{d}z^2$ and $\mathrm{d}V=\rho\,\mathrm{d}\rho\,\mathrm{d}\phi\,\mathrm{d}z$ — the standard cylindrical volume element, from nothing but the map and definition (1). The scale factors are $h_\rho=1,\ h_\phi=\rho,\ h_z=1$ , and $\mathrm{d}V=\rho\,\mathrm{d}\rho\,\mathrm{d}\phi\,\mathrm{d}z$ .

2 · Divergence theorem for a radial field (fully worked, with numbers)

Take $\mathbf{A}=r\,\hat{\mathbf{e}}_r$ in spherical coordinates. Compute its divergence, then verify the divergence theorem on a ball of radius $R=2$ via both the volume integral and the surface flux.

Step 1. The only nonzero component is $A_r=r$ , and $h_rh_\theta h_\phi=r^2\sin\theta$ . From the divergence (5), $\nabla\cdot\mathbf{A}=\dfrac{1}{r^2\sin\theta}\partial_r\big(h_\theta h_\phi\,A_r\big)=\dfrac{1}{r^2\sin\theta}\partial_r\big(r\cdot r\sin\theta\cdot r\big)=\dfrac{1}{r^2\sin\theta}\partial_r(r^3\sin\theta)$ .

Step 2. Differentiate: $\partial_r(r^3\sin\theta)=3r^2\sin\theta$ , so $\nabla\cdot\mathbf{A}=\dfrac{3r^2\sin\theta}{r^2\sin\theta}=\mathbf{3}$ , constant everywhere. (In Cartesian form $\mathbf{A}=(x,y,z)$ gives $\nabla\cdot\mathbf{A}=3$ — a clean cross-check.)

Step 3. Volume integral over the ball $r\le R$ : the integrand is constant, so $\displaystyle\int\nabla\cdot\mathbf{A}\,\mathrm{d}V=3\cdot\tfrac{4}{3}\pi R^3=4\pi R^3=32\pi\approx\mathbf{100.5}$ at $R=2$ .

Step 4. Surface flux over the sphere $r=R$ : the outward area element is $\mathrm{d}\mathbf{S}=R^2\sin\theta\,\mathrm{d}\theta\,\mathrm{d}\phi\,\hat{\mathbf{e}}_r$ and $\mathbf{A}\cdot\hat{\mathbf{e}}_r=R$ there, so $\displaystyle\oint\mathbf{A}\cdot\mathrm{d}\mathbf{S}=R\cdot(4\pi R^2)=4\pi R^3=32\pi\approx\mathbf{100.5}$ .

Step 5. The two agree exactly: $\int\nabla\cdot\mathbf{A}\,\mathrm{d}V=\oint\mathbf{A}\cdot\mathrm{d}\mathbf{S}=4\pi R^3$ . The curvilinear area and volume elements, both built from the scale factors, are exactly what make Gauss's theorem consistent. Both sides equal $32\pi\approx100.5$ , confirming $\nabla\cdot\mathbf{A}=3$ and the divergence theorem.

Physics of the portrait

Orthogonal curvilinear coordinates are where nineteenth-century French mathematical physics learned to bend space to a problem's symmetry. The person who systematized the idea — and whose name still labels the scale factors — was an engineer as much as a mathematician.

Origins · 1 minGabriel Lamé and how this was discovered

Gabriel Lamé

1795–1870 · founder of curvilinear-coordinate methods, from heat conduction to elasticity

Gabriel Lamé trained at the École Polytechnique and spent his formative decade in Russia, teaching and building roads and bridges in St. Petersburg — a working engineer's grounding that shaped his taste for coordinates fitted to real geometry. Back in Paris he held chairs at the Polytechnique and the Sorbonne. His deepest contribution here is the systematic theory of general curvilinear coordinates: studying heat flow through anisotropic solids, he needed coordinate surfaces adapted to the body, and developed the machinery of what we now call the Lamé coefficients — the scale factors $h_i$ that convert coordinate steps into physical lengths. His 1859 Leçons sur les coordonnées curvilignes is the first unified treatment, introducing the ellipsoidal-harmonic Lamé functions along the way. The same instinct runs through his elasticity theory, where the Lamé parameters $\lambda$ and $\mu$ still specify every isotropic solid. Gauss reportedly judged him the foremost French mathematician of his time, and his name is one of the seventy-two on the Eiffel Tower. What he bequeathed was a habit of mind: choose the coordinates the physics wants, and let the scale factors carry it.

History

1827

— Gauss defines the metric, the first fundamental form, of a curved surface

1833

— Lamé develops orthogonal curvilinear coordinates for heat conduction in solids

1859

— Lamé's Leçons sur les coordonnées curvilignes systematizes the scale factors

1873

— Maxwell's Treatise casts field theory in coordinate-free vector operators

1901

— Ricci and Levi-Civita's absolute differential calculus generalizes to arbitrary metrics

Going further

Texts. Arfken and Weber, Mathematical Methods for Physicists, ch. 2 (Curvilinear Coordinates, Tensors) · Hassani, Mathematical Physics, ch. on differential operators in curvilinear coordinates · Stone and Goldbart, Mathematics for Physics, ch. on curvilinear coordinates and the calculus of variations · Riley, Hobson and Bence, Mathematical Methods for Physics and Engineering, ch. 10 · Boas, Mathematical Methods in the Physical Sciences, ch. 10 · Morse and Feshbach, Methods of Theoretical Physics, ch. 5 (separable coordinate systems) · Byron and Fuller, Mathematics of Classical and Quantum Physics, ch. 1. The scale factors used throughout — Cartesian $(1,1,1)$ , cylindrical $(1,\rho,1)$ , spherical $(1,r,r\sin\theta)$ — follow directly from $h_i=|\partial\mathbf{r}/\partial u_i|$ applied to the explicit coordinate maps. The flux-box derivation rests on $\nabla\cdot\mathbf{A}=\lim_{V\to0}V^{-1}\oint\mathbf{A}\cdot\mathrm{d}\mathbf{S}$ , and Worked Example 2 gives $\nabla\cdot(r\hat{\mathbf{e}}_r)=3$ with both sides of Gauss's theorem equal to $32\pi\approx100.5$ at $R=2$ . The count of exactly eleven separable coordinate systems for the 3D Helmholtz equation is Eisenhart's classification, catalogued in Morse and Feshbach.

Builds on. Scalar and Vector Fields · The Gradient Operator · The Del Operator and Vector Differential Identities. Leads to. The Laplacian Operator · Divergence of a Vector Field · Curl of a Vector Field · The Divergence (Gauss) Theorem.

Ask the tutor: Ask me to derive the spherical curl term by term, or to work out the scale factors for parabolic or elliptic-cylinder coordinates and separate the Helmholtz equation in them.

What does orthogonality actually buy you?

It buys a diagonal metric, $g_{ij}=h_i^2\delta_{ij}$ , and everything simple flows from that: squared distance becomes a pure Pythagorean sum with no cross terms, physical vector components need no covariant-versus-contravariant split, and $\nabla$ , $\nabla\cdot$ , $\nabla\times$ , $\nabla^2$ collapse to the compact scale-factor formulas. A general non-orthogonal system gains off-diagonal $g_{ij}$ terms and forces you to track upper and lower indices. Orthogonality is the sweet spot: curved enough to match a sphere or cylinder, flat enough that the calculus stays readable.

Why is h_φ equal to ρ and not to 1?

Because $\phi$ is an angle, not a length. The scale factor is the conversion between a coordinate step and the physical distance it spans, and one radian of azimuth sweeps an arc whose length grows with how far you are from the axis: $\mathrm{d}(\text{arc})=\rho\,\mathrm{d}\phi$ . So $h_\phi=\rho$ . This is not a convention — it drops straight out of $h_\phi=|\partial\mathbf{r}/\partial\phi|$ , which evaluates to $\rho$ . The same logic gives the spherical $h_\phi=r\sin\theta$ : azimuthal arcs shrink to nothing at the poles ( $\sin\theta\to0$ ) and are widest at the equator. Whenever a coordinate is dimensionless, its scale factor carries the length dimension for it.

Are the unit vectors really different at every point?

Yes, and this is the deepest difference from Cartesian coordinates. The Cartesian $\hat{\mathbf{x}},\hat{\mathbf{y}},\hat{\mathbf{z}}$ point the same way everywhere, but $\hat{\mathbf{e}}_\rho$ points outward from the axis, so it swings around as you change $\phi$ : $\partial\hat{\mathbf{e}}_\rho/\partial\phi=\hat{\mathbf{e}}_\phi$ . Because the frame rotates, differentiating a vector field means differentiating both its components and its basis vectors. The master operator formulas are constructed exactly so that all of that basis-turning is folded into the scale factors, sparing you from tracking $\partial\hat{\mathbf{e}}_i/\partial u_j$ by hand. But if you ever expand a vector operation manually, you must include those derivatives or you will drop real physical terms.

How many coordinate systems let you separate variables, and why does it matter?

Exactly eleven coordinate systems separate the 3D Helmholtz equation $\nabla^2\psi+k^2\psi=0$ : Cartesian, cylindrical, spherical, and eight more (parabolic, elliptic, prolate and oblate spheroidal, conical, ellipsoidal, paraboloidal). Catalogued in Morse and Feshbach, this matters because separation turns one hard PDE into three ordinary ones, each solvable in known special functions. Separability requires the coordinate surfaces to align with the problem's boundaries — which is why you reach for spherical coordinates for an atom and elliptic-cylinder coordinates for an elliptical waveguide. Outside these eleven you generally solve the PDE numerically.

Is h₁h₂h₃ the same thing as the Jacobian of the coordinate change?

Yes, for an orthogonal system they are identical. The Jacobian of the map $(u_1,u_2,u_3)\to(x,y,z)$ is $J=|\det(\partial(x,y,z)/\partial(u_1,u_2,u_3))|$ , and it is the factor that appears when you change variables in a volume integral. Because the metric is $g_{ij}=h_i^2\delta_{ij}$ , its determinant is $g=h_1^2h_2^2h_3^2$ , so $\sqrt{g}=h_1h_2h_3=J$ . That single product is simultaneously the volume of the coordinate box, the square root of the metric determinant, and the change-of-variables factor: $\iiint f\,\mathrm{d}V=\iiint f\,h_1h_2h_3\,\mathrm{d}u_1\mathrm{d}u_2\mathrm{d}u_3$ . It is the most reused number in the whole framework.

When do I actually need the full tensor-calculus machinery instead?

Whenever the coordinate surfaces do not meet at right angles — an oblique or sheared grid — the metric acquires off-diagonal terms and the tidy scale-factor formulas no longer apply. Then you need the general apparatus: the full symmetric metric $g_{ij}$ , its inverse $g^{ij}$ , the covariant-versus-contravariant distinction, and Christoffel symbols to differentiate vectors. You also cross into full tensor calculus the moment space itself is curved, as in general relativity. Orthogonal curvilinear coordinates are the largest class for which the intuitive physical-component formulas survive — which is why they are worth mastering before the general theory.