Differential Equations › Section I · Partial differential equations › Topic I.9

PDEs in science and engineering

The heat, wave and Laplace equations of this section turn up far beyond bars and strings, from microchips and drums to atoms, stock options and medical scanners. This topic surveys where they appear and how engineers solve the many problems that no series formula can reach.

The same equations, everywhere

This section solved three PDEs on simple shapes: heat in a bar, waves on a string, steady states on rectangles and disks. Their importance comes from how often the same mathematics reappears. In simple linear models, a quantity that spreads from high concentration to low obeys a heat-type equation, a disturbance that travels at a definite speed obeys a wave-type equation, and a steady balance with no sources obeys Laplace's equation. Real problems usually involve three dimensions, awkward shapes or nonlinear terms, so they are mostly solved numerically.

Diffusion in three dimensions

In a solid, the heat equation becomes \(u_t = k\,\nabla^2 u\), where the Laplacian now includes \(u_{zz}\). Engineers solve it to design heat sinks for processors, insulation for buildings and cooling for battery packs, and they solve its steady form, \(\nabla^2 u = 0\), when temperatures have settled. The same equation, with the diffusion coefficient \(D\) in place of \(k\), describes a dissolved substance spreading through a fluid or tissue. The flux law there, Fick's law, has the same shape as Fourier's law: flow runs down the concentration gradient at a rate proportional to it.

Worked example: how far can oxygen diffuse?

Oxygen diffuses through tissue with \(D\) of about \(2 \times 10^{-9}\) m²/s. Use the time scale \(L^2/(\pi^2 D)\) from Solving the heat equation to compare a distance of 0.1 mm with one of 1 cm. 1. For \(L = 10^{-4}\) m:

\frac{L^2}{\pi^2 D} = \frac{10^{-8}}{9.87 \times 2 \times 10^{-9}} \approx 0.5\ \mathrm{s}

2. For \(L = 10^{-2}\) m the distance is 100 times larger, so the time is \(100^2 = 10{,}000\) times longer, about 5000 s, or 1.4 hours. 3. Interpretation: diffusion is fast over a fraction of a millimetre and hopelessly slow over centimetres. 4. Check: this is why large animals need blood vessels reaching close to every cell.

Waves in two and three dimensions

The wave equation generalises to \(u_{tt} = c^2\nabla^2 u\). It governs sound, seismic waves, ripples on a pond and, component by component, light in empty space. A circular drumhead is a membrane with a fixed rim. Separating variables in polar coordinates produces Bessel's equation for the radial part, linking back to Special functions in applications. The drum's frequencies are set by zeros of Bessel functions; they are not whole-number multiples of the lowest one, with the next few at about 1.59, 2.14 and 2.30 times it, which is why a drum has a less definite pitch than a string.

Schrödinger's equation

Quantum mechanics is built on a PDE for the wavefunction \(\psi\) of a particle of mass \(m\) in a potential \(V\):

i\hbar\,\psi_t = -\frac{\hbar^2}{2m}\nabla^2\psi + V\psi

Separation of variables turns it into an eigenvalue problem whose eigenvalues are the allowed energies. For a particle confined to \(0 \le x \le L\) with \(V = 0\) inside, the spatial part solves \(X'' + \lambda X = 0\) with \(X(0) = X(L) = 0\), the very problem met for the heat equation, with \(\lambda = 2mE/\hbar^2\). So \(\lambda_n = (n\pi/L)^2\) gives

E_n = \frac{n^2\pi^2\hbar^2}{2mL^2}

Energy is quantised because only these eigenvalues fit.

Biology, finance and fluids

Adding reaction terms to diffusion gives reaction-diffusion equations; Alan Turing proposed in 1952 that such systems could produce spots and stripes like those on animal coats. In finance, the Black-Scholes equation of 1973 for pricing options can be transformed by a change of variables into the heat equation, a standard route to its well-known pricing formula. Fluid flow is governed by the Navier-Stokes equations, nonlinear PDEs for velocity and pressure that model air over wings and blood in arteries. Whether smooth solutions always exist in three dimensions is one of the Clay Mathematics Institute's Millennium Prize Problems, with a $1 million reward.

Finite differences

Most real PDEs are solved on a computer. The simplest approach places grid points \(x_j\) a distance \(\Delta x\) apart and steps forward in time by \(\Delta t\). Write \(u_j^m\) for the temperature at point \(j\) after \(m\) steps. Replacing \(u_t\) by a forward difference in time, and \(u_{xx}\) by a central difference of neighbouring values, gives the explicit scheme

u_j^{m+1} = u_j^m + r\big(u_{j+1}^m - 2u_j^m + u_{j-1}^m\big), \qquad r = \frac{k\,\Delta t}{\Delta x^2}

Each new value is the old one nudged towards the average of its neighbours, a discrete version of the concavity rule. The scheme is stable only if \(r \le \frac{1}{2}\).

Worked example: one step, stable and unstable

A bar has grid values 0, 40, 40, 40, 0 °C, with both ends held at 0. Take one explicit step with \(r = 0.25\), then with \(r = 1\). 1. With \(r = 0.25\), the second point becomes \(40 + 0.25(-40) = 30\); the middle stays 40, since its neighbours average 40. 2. Result: 0, 30, 40, 30, 0, a sensible cooling from the ends. 3. With \(r = 1\), one step gives 0, 0, 40, 0, 0, and a second gives 0, 40, −40, 40, 0. 4. Check: a bar between 0 °C and 40 °C can never drop below 0 °C, so the negative value shows the instability that \(r > \frac{1}{2}\) causes.

Worked example: choosing a time step

An aluminium bar has \(k\) of about \(9.9 \times 10^{-5}\) m²/s. With grid spacing 1.0 cm, what is the largest stable time step, and what happens if the spacing is halved? 1. Stability needs \(r \le \frac{1}{2}\), so \(\Delta t \le \Delta x^2/(2k)\). 2. Substitute:

\Delta t \le \frac{(0.010)^2}{2(9.9 \times 10^{-5})} \approx 0.51\ \mathrm{s}

3. Halving \(\Delta x\) to 0.50 cm divides the limit by 4, to about 0.13 s. 4. Check the cost: twice as many points and four times as many steps means about 8 times the work, which is why engineers often switch to implicit schemes that have no such limit.

Finite elements and medical imaging

Finite element methods divide a region of any shape into small pieces, such as triangles or tetrahedra, and approximate the solution by simple functions on each; they are standard in engineering design. Fourier analysis also sits inside hospital scanners. An MRI machine measures signals that are essentially samples of the two- or three-dimensional Fourier transform of the tissue, and a computer inverts the transform to build the image. CT reconstruction recovers a slice from many X-ray projections, and the projection-slice theorem ties that problem to Fourier transforms too.

Common mistakes to avoid

First, choosing a time step that breaks the stability limit, so a simulation grows wild oscillations and impossible values. Second, refining the grid spacing without shrinking the time step by the square of the factor. Third, assuming a linear model when the physics is nonlinear, as in turbulent flow governed by Navier-Stokes. Fourth, forgetting that a three-dimensional problem needs the full Laplacian, not just \(u_{xx}\). Fifth, expecting a drum's overtones to be whole-number multiples of its fundamental, as a string's are.

Key takeaways

The heat, wave and Laplace equations generalise with the Laplacian \(\nabla^2\) and model diffusion of heat and chemicals, sound, seismic and electromagnetic waves, drums, quantum energy levels, option prices and fluid flow. Diffusion times grow as the square of the distance. The particle in a box reuses the eigenvalue problem from the heat equation, giving \(E_n = n^2\pi^2\hbar^2/(2mL^2)\). Most real PDEs are solved numerically: the explicit finite-difference heat scheme is stable only when \(k\Delta t/\Delta x^2\) is at most one half. Finite elements handle complex shapes, and Fourier methods drive MRI.

Flashcards

Flashcard

∇²u

The Laplacian, uₓₓ + uᵧᵧ + u_zz

Replaces uₓₓ in higher dimensions.

Test your understanding

Question 1 of 5

In the explicit heat scheme with r = 0.5, grid values are 0, 0, 60, 0, 0 with both ends held at 0. What are the values after one step?

Question 2 of 5

For a material with k = 2.0 × 10⁻⁵ m²/s and grid spacing 2.0 mm, what is the largest stable time step for the explicit heat scheme?

Question 3 of 5

A substance diffuses across a 1 mm layer in a time T. Roughly how long does it take to diffuse across a 3 mm layer of the same material?

Question 4 of 5

For a particle in a box, how does the energy E₃ compare with E₁?

Question 5 of 5

Which statement about the Navier-Stokes equations is correct?

← Previous topic