Curvilinear Coordinates

For centuries, the Cartesian grid has been our faithful guide for mapping space, its perpendicular lines offering a simple and reliable structure. Yet, the universe rarely conforms to such rigidity. From the gravitational field warping spacetime around a star to the swirling flow of air over a wing, nature’s phenomena are rife with curves, symmetries, and complexities that make the Cartesian system feel inadequate and clumsy. To truly describe and understand these systems, we need a language that speaks their native geometry—a language of flexible, adaptive grids. This is the domain of curvilinear coordinates.

This article addresses the fundamental challenge of leaving the Cartesian world behind: How do we construct a consistent framework for measurement and calculus when our grid lines can bend and stretch? We will embark on a journey to build this new language from first principles, discovering how concepts like direction, distance, and rates of change are redefined in a more general and powerful way.

In the following chapters, you will develop a deep understanding of this essential mathematical tool. The "Principles and Mechanisms" section will lay the groundwork, introducing you to the core concepts of basis vectors, scale factors, and the all-important metric tensor. We will see how these tools allow us to reformulate the operators of vector calculus—gradient, divergence, and curl—for any coordinate system. Then, in "Applications and Interdisciplinary Connections," we will witness this machinery in action. We will see how curvilinear coordinates are not just a mathematical curiosity but an indispensable tool for solving crucial problems in electromagnetism, fluid dynamics, quantum mechanics, and engineering, revealing the profound and elegant connection between geometry and the laws of physics.

So, we've decided to abandon the familiar, rigid grid of Cartesian coordinates. It’s a bold move! But the universe, in its elegant complexity, rarely lays itself out on a perfect checkerboard. From the swirling electric field around a wire to the gravitational pull of a planet, the phenomena of nature often display symmetries—cylindrical, spherical, or something more exotic—that make a Cartesian grid feel clumsy and unnatural. To truly understand these phenomena, we need a language that speaks their geometry. This is the world of curvilinear coordinates. But how do we build this new language from the ground up? How do we talk about directions, distances, and rates of change in a world where our grid lines can bend, stretch, and warp?

Let's start with the most basic question: in a new coordinate system, say with coordinates $(u, v)$, what do "direction" and "movement" even mean? In the Cartesian world, the directions "along x" and "along y" are constant everywhere. But on a curved surface, or with bent coordinate lines, the local "north" at one point is different from the "north" a mile away.

The most natural way to define our new "axes" is to see what happens when we move. Imagine our position in space is described by a vector $\mathbf{r}$ which is a function of our new coordinates, say $\mathbf{r}(u, v, w)$. If we take a tiny step by changing only the first coordinate, $u$, while keeping $v$ and $w$ fixed, we trace a path along a $u$-coordinate curve. The tangent to this path, an arrow pointing in the direction of our motion, is simply the partial derivative of the position vector: $\frac{\partial \mathbf{r}}{\partial u}$.

This is it! These partial derivatives are our new ​​basis vectors​​. For a coordinate system $(u, v)$, we have two basis vectors at every point in space: $\mathbf{e}_u = \frac{\partial \mathbf{r}}{\partial u}, \quad \mathbf{e}_v = \frac{\partial \mathbf{r}}{\partial v}$ Unlike the steadfast $\hat{\mathbf{i}}$ and $\hat{\mathbf{j}}$ of the Cartesian world, these basis vectors are themselves functions of position. They change direction and, as we'll see, length from point to point, perfectly adapting to the local geometry.

A crucial question immediately arises: are these new basis vectors perpendicular to each other? In many useful systems, like cylindrical and spherical coordinates, they are. We call these ​​orthogonal​​ coordinate systems. But it's not a given. We can always check by computing their dot product. If $\mathbf{e}_u \cdot \mathbf{e}_v = 0$ at a point, the coordinate system is orthogonal there. If the dot product is non-zero, our local axes are skewed, creating a non-orthogonal system. This might seem like a complication, but sometimes the physics of a problem, like the shearing of a material, is best described by just such a skewed grid.

Now for the next puzzle: measuring distance. If we take a step of size $du$ along the $u$-axis, how far have we actually traveled? If our coordinate is longitude, a one-degree step near the equator is a much larger distance than a one-degree step near the North Pole. The relationship between a change in a coordinate, $du$, and the actual physical distance, $ds_u$, depends on a local conversion factor. This factor is simply the length, or magnitude, of our basis vector.

We call this magnitude the ​​scale factor​​, denoted by $h_u$: $h_u = |\mathbf{e}_u| = \left| \frac{\partial \mathbf{r}}{\partial u} \right|$ So, the infinitesimal distance traveled along the $u$-curve is $ds_u = h_u du$. The same logic applies to all other coordinates. These scale factors are the rulers of our new coordinate system, telling us how to convert coordinate-steps into real-world meters.

For an orthogonal system, the scale factors are all we need to describe the local geometry. But what about a non-orthogonal, or general, system? We need a more powerful tool. This tool is the ​​metric tensor​​, $g_{ij}$. It's a collection of numbers (a matrix) at each point in space, defined by taking all possible dot products of the basis vectors: $g_{ij} = \mathbf{e}_i \cdot \mathbf{e}_j$ Notice what this means. The diagonal components, like $g_{11} = \mathbf{e}_1 \cdot \mathbf{e}_1 = |\mathbf{e}_1|^2 = h_1^2$, tell us the squared lengths of our basis vectors (and thus give us the scale factors). The off-diagonal components, like $g_{12} = \mathbf{e}_1 \cdot \mathbf{e}_2$, tell us the dot products between different basis vectors, which encodes the angle between them. If the system is orthogonal, all off-diagonal components are zero, and the metric tensor is a simple diagonal matrix. If it's not orthogonal, the non-zero off-diagonal terms tell us exactly how skewed our axes are.

The metric tensor is the central character in our story. It's the ultimate geometric rulebook for our chosen coordinate system, beautifully encoding all the information about lengths and angles at every point in space.

Armed with the metric tensor, we can now write down a master formula for the distance between any two infinitesimally close points. In Cartesian coordinates, this is just the Pythagorean theorem: $ds^2 = dx^2 + dy^2 + dz^2$. How does this generalize?

The total displacement vector $d\mathbf{r}$ is the sum of displacements along each coordinate direction: $d\mathbf{r} = \frac{\partial \mathbf{r}}{\partial q^1}dq^1 + \frac{\partial \mathbf{r}}{\partial q^2}dq^2 + \dots = \sum_i \mathbf{e}_i dq^i$. The squared length is $ds^2 = d\mathbf{r} \cdot d\mathbf{r}$. Let's expand this: $ds^2 = \left( \sum_i \mathbf{e}_i dq^i \right) \cdot \left( \sum_j \mathbf{e}_j dq^j \right) = \sum_{i,j} (\mathbf{e}_i \cdot \mathbf{e}_j) dq^i dq^j$ Recognizing the definition of the metric tensor, we arrive at the magnificent formula for the ​​line element​​: $ds^2 = \sum_{i,j} g_{ij} dq^i dq^j$ This single equation contains all the geometry. For an orthogonal system where $g_{ij}$ is diagonal with $g_{ii} = h_i^2$, this simplifies to the generalized Pythagorean theorem we saw earlier: $ds^2 = (h_1 dq^1)^2 + (h_2 dq^2)^2 + (h_3 dq^3)^2$. For a non-orthogonal system, the cross-terms with $i \neq j$ are essential.

This isn't just for distances. The volume of an infinitesimal box is no longer just $dx dy dz$. It becomes $dV = \sqrt{\det(g_{ij})} dq^1 dq^2 dq^3$. For an orthogonal system, this simplifies beautifully to $dV = h_1 h_2 h_3 dq^1 dq^2 dq^3$. That product of scale factors, $h_1 h_2 h_3$, is precisely the Jacobian determinant of the coordinate transformation, which represents the local volume expansion factor.

Now we come to a subtle but profound point, one that is key to understanding the full power of this machinery. When we have a vector, say velocity $\mathbf{v}$, how do we describe its components? It turns out there are two natural ways to do it, which are identical in a simple Cartesian system but different in general curvilinear coordinates.

We can express our vector $\mathbf{v}$ as a sum of our basis vectors $\mathbf{e}_i$: $\mathbf{v} = v^1 \mathbf{e}_1 + v^2 \mathbf{e}_2 + v^3 \mathbf{e}_3 = \sum_i v^i \mathbf{e}_i$ The coefficients $v^i$ (written with an upper index by convention) are called the ​​contravariant components​​. Think of them this way: if you stretch your basis vector $\mathbf{e}_i$ so it's twice as long, you only need half the component $v^i$ to represent the same vector $\mathbf{v}$. The component varies contrary to the basis vector.

But there's another way. We can also use a "dual basis" $\mathbf{e}^j$ (a concept we won't detail here, but it's related to our original basis via the metric). We can write: $\mathbf{v} = v_1 \mathbf{e}^1 + v_2 \mathbf{e}^2 + v_3 \mathbf{e}^3 = \sum_j v_j \mathbf{e}^j$ The coefficients $v_j$ (with a lower index) are the ​​covariant components​​. These are obtained by projecting the vector $\mathbf{v}$ onto the basis vectors, $v_j = \mathbf{v} \cdot \mathbf{e}_j$. These components vary with the basis vectors.

The metric tensor $g_{ij}$ is the magical bridge that connects these two descriptions: $v_i = \sum_j g_{ij} v^j$. What about the "physical components" we might measure in an experiment? Those are typically the projections of the vector onto unit vectors in the coordinate directions. In an orthogonal system, the physical component along the $i$-th direction is $v_{\hat{i}} = v^i h_i = v_i / h_i$. So, neither the covariant nor the contravariant components are, in general, the same as the "physical" components, but they are all precisely related through the scale factors (the metric).

This dual description seems complicated, but it is the source of the immense flexibility and power of tensor calculus. The dot product of two vectors, for instance, has a beautiful and simple form using this notation: $\mathbf{u} \cdot \mathbf{v} = \sum_{i,j} g_{ij} u^i v^j = \sum_i u_i v^i$ The first form, explicitly using the metric, is the most direct way to compute a dot product when you have the contravariant components and a non-orthogonal system.

We have built a powerful dictionary for describing the geometry of space. Now, let's do some physics. The core operators of vector calculus—gradient, divergence, and curl—are physical concepts, independent of coordinates. A temperature gradient points in the direction of fastest heat increase, regardless of whether you describe it with $x,y,z$ or $\rho,\phi,z$. But their mathematical formulas must change to accommodate our new language of scale factors.

These general formulas, which can be derived from the fundamental definitions of the operators, are jewels of mathematical physics. For any orthogonal system, they are:

• ​​Gradient (of a scalar $\phi$):​​ The vector that points in the direction of the steepest ascent. $\nabla \phi = \sum_{i=1}^3 \frac{1}{h_i} \frac{\partial \phi}{\partial u_i} \hat{\mathbf{e}}_i$ Notice the scale factors $h_i$ in the denominator. A change in a coordinate value $\partial \phi / \partial u_i$ must be scaled to represent a true physical gradient.

• ​​Divergence (of a vector $\mathbf{A}$):​​ The measure of how much a vector field "spreads out" or acts as a source. $\nabla \cdot \mathbf{A} = \frac{1}{h_1 h_2 h_3} \sum_{i=1}^3 \frac{\partial}{\partial u_i} \left( \frac{h_1 h_2 h_3}{h_i} A_i \right)$ Here, we must not only differentiate the vector components $A_i$, but also the scale factors, because the geometry itself is changing.

• ​​Curl (of a vector $\mathbf{A}$):​​ The measure of the "rotation" or "vorticity" of a vector field. $\nabla \times \mathbf{A} = \frac{1}{h_1 h_2 h_3} \det \begin{pmatrix} h_1 \hat{\mathbf{e}}_1 & h_2 \hat{\mathbf{e}}_2 & h_3 \hat{\mathbf{e}}_3 \\ \frac{\partial}{\partial u_1} & \frac{\partial}{\partial u_2} & \frac{\partial}{\partial u_3} \\ h_1 A_1 & h_2 A_2 & h_3 A_3 \end{pmatrix}$

These formulas look complicated, and they are. But they are also completely general for any orthogonal coordinate system. Just plug in the appropriate scale factors—for instance, $h_{\rho}=1, h_{\phi}=\rho, h_{z}=1$ for cylindrical coordinates—and you have the operators ready for action.

At this point, you might be thinking this is a terrible bargain. We traded the simple elegance of Cartesian calculus for a thicket of scale factors and gnarly derivatives. But here is the payoff, and it is profound.

Physical laws are not a matter of opinion or convenience; they are truths about the universe. They cannot depend on the coordinate system we choose to describe them. A key identity in electromagnetism (and fluid dynamics) is that the divergence of a curl is always zero: $\nabla \cdot (\nabla \times \mathbf{A}) = 0$. This implies, among other things, that there are no "magnetic monopoles." This fundamental law of nature must hold in any valid coordinate system.

Let's see if it does. If we take the general formula for the curl, plug its resulting components into the general formula for the divergence, and embark on a perilous journey of partial differentiation using the product rule on all the components and scale factors... something miraculous happens. Terms expand, jostle, and rearrange. And then, a symphony of cancellation begins. A term here cancels a term there. After the algebraic dust settles, every single term pairs up with an equal and opposite partner, and we are left with... zero. Utterly, beautifully zero.

This is not a mathematical accident. It is proof of the self-consistency and power of our framework. It is the universe reassuring us that even though our descriptive language has become more complex, the underlying physical truth remains simple and invariant. The machinery of curvilinear coordinates, from basis vectors to metric tensors, is the language we needed to see this unchanging truth, no matter what geometric lens we choose to look through. This is the inherent beauty, and the profound unity, that modern physics is built upon.

Now that we have acquainted ourselves with the machinery of curvilinear coordinates—the scale factors, the metric tensor, the new forms for gradient, divergence, and curl—you might be tempted to ask, "Why go through all this trouble?" Why abandon the simple, rectilinear comfort of our old friend, the Cartesian grid? The answer, and this is the wonderful part, is that the universe simply doesn't care about our graph paper. The laws of nature are written in a language that is independent of any particular coordinate system. By learning the language of curvilinear coordinates, we are not just learning a clever computational trick; we are gaining the freedom to choose our own point of view, to align our mathematical description with the natural symmetries of the problem at hand. This freedom transforms problems from intractable messes into things of beautiful simplicity.

Let's take a journey through a few different realms of science and see how this powerful idea plays out.

The great laws of nineteenth-century physics, like those of electromagnetism and fluid dynamics, are statements about fields. An electric field $\mathbf{E}$ or a velocity field $\mathbf{v}$ pervades space, and its behavior is governed by differential equations. A central concept in this description is divergence, which, as we've seen, measures the "sourceness" of a field at a point. For instance, Gauss's law in electrostatics relates the divergence of the electric field to the density of electric charge $\rho$: $\nabla \cdot \mathbf{E} = \rho/\varepsilon_0$.

Suppose you are given an electric field, perhaps a complicated-looking one, expressed in some peculiar, non-Cartesian coordinate system. How would you find the distribution of charges that creates it? You would have to calculate the divergence. You would find, as demonstrated in calculations like the one in, that the formula for divergence involves the scale factors $h_i$ of your chosen coordinates. The formula looks different in polar, cylindrical, or any other system, yet it always measures the same intrinsic physical quantity: the flux of the field out of an infinitesimal volume. The scale factors are the geometric "price" we pay—or rather, the correction we must apply—to ensure our calculation is physically meaningful in our chosen curved grid. The physical law itself, the connection between divergence and charge, remains steadfast and true.

This same story repeats itself throughout mechanics. Imagine mapping the flow of a river. In some places, water wells up from underground springs; in others, it drains into sinks. If the fluid, like water, is incompressible, then for any region with no springs or sinks, the amount of fluid flowing in must exactly equal the amount flowing out. The net "outflow"—the divergence of the velocity field—must be zero. This simple, powerful statement, $\nabla \cdot \mathbf{v} = 0$, is the continuity equation for an incompressible fluid. It is a fundamental law of nature. Now, whether you describe a swirling vortex using polar coordinates or the flow around a cylinder using cylindrical coordinates, this law holds. The specific mathematical form of $\nabla \cdot \mathbf{v}$ will change, adapting itself to the geometry of your coordinates, but the physical content is invariant.

The pinnacle of this coordinate-free thinking is found in the theory of continuum mechanics, which describes the behavior of deformable materials like a steel beam under load or a flowing glacier. Here, we must go beyond orthogonal coordinates and embrace the full power of tensor calculus. The state of stress inside a material is described by the Cauchy stress tensor, $\boldsymbol{\sigma}$. Newton's second law, $F=ma$, when applied to a continuous body, takes on a new, elegant form: the divergence of the stress tensor, plus any body forces, equals the density times acceleration. In the language of general coordinates, this is written as $\sigma^{ij}{}_{;j} + \rho b^i = \rho a^i$. Notice that little semicolon! It denotes a covariant derivative, a generalization of the partial derivative that correctly accounts for the fact that in a general curvilinear system, the basis vectors themselves change from point to point. The terms needed to make this correction are the Christoffel symbols, $\Gamma^k_{ij}$, which are derived directly from the metric tensor. It's a beautiful thing: the geometry of our coordinate system, encoded in the metric, tells us exactly how we must formulate our physical laws to get the right answer. The physics and the geometry are inextricably linked.

This connection between geometry and motion is revealed in another profound way. If you track a small blob of fluid as it moves and deforms, its volume will change. The rate at which its volume changes is directly related to the divergence of the velocity field. Specifically, the rate of change of the volume-scaling factor, the Jacobian determinant $J$, is given by $\dot{J} = J (\nabla \cdot \mathbf{v})$. This is not an approximation; it is an exact, geometric result that is a cornerstone of continuum mechanics.

Nowhere is the freedom to choose coordinates more crucial than in quantum mechanics. The central equation is the time-independent Schrödinger equation, an equation for the wavefunction $\psi$ of a particle, whose square $|\psi|^2$ gives the probability of finding the particle at a certain location. For a particle of mass $m$ in a potential $V$, it reads $(-\frac{\hbar^2}{2m} \nabla^2 + V)\psi = E \psi$. The $\nabla^2$ is the Laplacian operator, and it is the bane of many a physics student.

Consider the simplest atom: hydrogen, an electron orbiting a proton. The electron is attracted to the proton by a potential that depends only on the distance between them, $V(r) = -k/r$. If you try to write this in Cartesian coordinates, $V(x,y,z) = -k/\sqrt{x^2+y^2+z^2}$, and then try to solve the Schrödinger equation, you will be met with a truly horrendous mess. The variables $x$, $y$, and $z$ are all mixed up. But, if we recognize the spherical symmetry of the problem and switch to spherical polar coordinates $(r, \theta, \phi)$, the potential becomes wonderfully simple. The real magic, however, is what happens to the Laplacian. While its form in spherical coordinates looks complicated at first glance, it has a structure that allows the entire Schrödinger equation to be separated into three a ordinary differential equations—one for $r$, one for $\theta$, and one for $\phi$. This separation is what allows us to solve the hydrogen atom problem exactly and discover the quantum numbers that govern atomic structure. Without spherical coordinates, there would be no analytical chemistry as we know it.

This is a general principle. The ability to solve the Schrödinger equation by a "separation of variables" depends on a deep harmony between the shape of the potential $V$ and the geometry of the coordinate system. For a given coordinate system, only potentials of a specific mathematical form will allow for separation. In polar coordinates, for instance, a potential of the form $V(r,\theta)=V_r(r) + V_\theta(\theta)/r^2$ is separable, but an additively separable one $V(r,\theta)=V_r(r) + V_\theta(\theta)$ is generally not. The geometry, through the scale factors that appear in the Laplacian, dictates the rules of the game.

Moreover, the very foundation of quantum mechanics relies on the proper use of these geometric concepts. The probability of finding a particle in a small volume must be the same regardless of the coordinate system used. This means the integral of $|\psi|^2$ over a volume must be invariant. The infinitesimal volume element, which is $dx\,dy\,dz$ in Cartesian coordinates, becomes $\sqrt{g}\,dq^1\,dq^2\,dq^3$ in general coordinates, where $\sqrt{g}$ is the square root of the determinant of the metric tensor, also known as the Jacobian of the coordinate transformation. This $\sqrt{g}$ factor is essential for conserving probability and ensuring that physical observables are represented by Hermitian operators, a cornerstone of the theory.

The power of these ideas is not limited to physics. They represent a universal language for describing geometric structures. In the field of complex analysis, for example, the Cauchy-Riemann equations are the very definition of an analytic (smoothly differentiable) function. In Cartesian coordinates, they are a simple pair of equations relating the partial derivatives of the real and imaginary parts of the function. If one transforms these equations into any orthogonal curvilinear coordinate system, they take on a new form. This new form looks different, but it must express the very same intrinsic property of the function. And what determines the new form? Once again, it is the geometry of the coordinate system, captured by the scale factors, which dictate the stretching of the grid lines.

Mathematicians, in their explorations of geometry, have invented a stunning variety of coordinate systems, each tailored to a particular family of shapes. Ellipsoidal coordinates, for instance, are defined implicitly as the roots of a cubic equation, creating a beautiful orthogonal grid of confocal ellipsoids, hyperboloids of one sheet, and hyperboloids of two sheets. These are not mere curiosities; they are precisely the right "viewpoint" from which to solve Laplace's equation for the gravitational or electrostatic potential of an ellipsoidal body.

In all the examples so far, we started with a coordinate system defined by an analytical formula. But what about the really complex shapes we encounter in engineering and science—an airplane wing, a turbine blade, a human heart? No simple formula will create a coordinate system that neatly fits such convoluted boundaries. Here, the idea of curvilinear coordinates takes a modern, computational turn.

Instead of defining the coordinates by a formula, we can compute them. One of the most elegant methods is to define the new coordinate functions, let's call them $u(x,y)$ and $v(x,y)$, as solutions to Laplace's equation, $\nabla^2 u = 0$ and $\nabla^2 v = 0$. We "pin down" the coordinate lines by setting their values on the boundaries of our complex domain. For instance, we can map the four sides of a simple square to the complex boundary of an airfoil. Then, we solve Laplace's equation numerically, typically using finite difference or finite element methods, to find the values of $u$ and $v$ throughout the interior. The resulting level curves of $u$ and $v$ form a beautiful, smooth, custom-built curvilinear coordinate system known as "harmonic coordinates." Because they are born from Laplace's equation—the smoothest of all things—these coordinate systems are guaranteed not to fold over on themselves (their Jacobian determinant remains positive), providing a well-behaved grid on which we can then solve much more complex equations, like the Navier-Stokes equations for fluid flow. It is a spectacular inversion of logic: we use a physical law to create the very mathematical grid we need to solve other physical laws.

So we see, the study of curvilinear coordinates is a journey that takes us to the heart of what it means to describe the world. It is about the principle of invariance, the freedom of perspective, and the deep, beautiful unity between the geometry of space and the laws of nature. It is a tool that not only enables us to solve problems on the chalkboard but also empowers the most advanced computer simulations that shape our modern world.