Partial Differential Equations Intro

Equations for whole fields at once: temperature everywhere, waves everywhere, with boundary rules deciding which evolution is allowed.

A sine heat hump of height 1 on a rod flattens to 0.37 by time 1 as diffusion drains its curvature

Why Does This Exist?

Heat spreads, waves travel, membranes settle: quantities defined at every point evolving under local rules plus edge constraints. ODEs track finitely many numbers; PDEs track fields, and the boundary conditions (held ends, insulated walls, radiating surfaces) select among infinitely many evolutions that all satisfy the interior rule. Diffusion models, physics-informed networks, and image smoothing all discretize PDEs, so knowing which equation smooths, which transports, and which equilibrates prevents modeling nonsense. Multivariable ground is in multivariable calculus and the derivative atoms in partial derivatives.

The one idea you need first

$u(x, t)$ is the whole profile at time $t$ . The PDE ties time change $u_t$ to space shape ( $u_x$ , $u_{xx}$ ): heat says $u_t$ follows curvature $u_{xx}$ (peaks fall, valleys fill), waves say acceleration $u_{tt}$ follows curvature, Laplace says $u_{xx} + u_{yy} = 0$ at rest.

Think of It Like This

A crowded room settling down

Shouting in one corner of a hall (a heat spike) does not stay put: neighbours hear and repeat, the spike spreads and fades, and the room tends toward one murmur level fixed by the doors (boundaries). A clap (a wave pulse) instead travels intact to the far wall and echoes. A settled seating plan nobody wants to leave (a Laplace solution) has everyone as comfortable as their neighbours' average: no interior reason to move.

Heat diffuses, waves transport, Laplace rests. The doors decide the final murmur, the echo timing, and the seating: same hall, different doors, different fate.

Where the analogy stops: real PDEs quantify with exact decay rates and speeds from eigenvalues and characteristics, not vibes about murmur.

How It Actually Works

The three archetypes. Heat $u_t = \alpha u_{xx}$ smooths: high modes decay as $e^{-\alpha k^2 t}$ , so fine detail dies first. Wave $u_{tt} = c^2 u_{xx}$ transports at speed $c$ along characteristics, preserving shape. Laplace $\Delta u = 0$ equilibrates: solutions average their neighbours (mean-value property), attaining extremes only on boundaries.

Boundaries and order. Dirichlet pins values, Neumann pins fluxes, Robin mixes them. Order counts the highest space derivative; linearity decides whether superpositions hold. Separation $u(x,t) = X(x)T(t)$ splits clean geometries into ODEs; transforms and Green's functions handle the rest.

Numerics. Grids turn $u_{xx}$ into stencil differences with stability limits (explicit heat needs $r = \alpha \Delta t/\Delta x^2 \le 1/2$ ); finite elements and spectral bases handle complex shapes.

Worked example

Rod of length $\pi$ with ends pinned at $0$ , diffusivity $\alpha = 1$ , initial $u(x, 0) = \sin x$ . Mode $\sin(kx)$ on length $\pi$ decays as $e^{-k^2 t}$ , and here $k = 1$ :

u(x, t) = e^{-t}\sin x

Check: $u_t = -e^{-t}\sin x$ and $u_{xx} = -e^{-t}\sin x$ agree. At center $x = \pi/2$ and $t = 1$ : $u = e^{-1}(1) \approx 0.368$ . The hump keeps its sine shape while its height drains to 37 percent: pure decay with no travel, the signature of heat versus waves.

Watch Out For

Solving the interior while ignoring the boundaries

Same PDE with pinned versus insulated ends gives decay versus conservation: the interior rule never selects alone. Dropping boundary terms (or imposing the convenient ones) produces a beautiful solution to the wrong problem.

Symptom: simulations that "almost match" but leak or accumulate at edges, or analytic modes that violate the stated end conditions. Fix: write boundary conditions before solving, verify each mode satisfies them, and test conservation behaviour (Neumann heat preserves total mass, Dirichlet drains it).

Meshing a high-dimensional field like a 1-D rod

Grid points multiply across dimensions: $100$ per axis is $10^4$ in 2-D but $10^6$ in 3-D and hopeless beyond. Naive explicit stepping then dies twice: no memory for the grid, no stable step for the stiffness.

Symptom: 2-D prototypes that cannot extend to 3-D, or step limits shrinking with refinement. Fix: use implicit or spectral time-stepping for stiffness, sparse or adaptive grids, and low-rank or neural-field ansatzes when dimensions grow.

The Quick Version

PDEs evolve fields: heat smooths, waves transport, Laplace rests.
Boundaries select the evolution: pin values, fluxes, or mixtures deliberately.
Sine heat on a pinned rod of length $\pi$ decays as $e^{-t}$ , reaching 0.37 by $t = 1$ .
High modes die first under heat: detail fades before shape.
Verify modes against boundaries; respect stability limits and dimensionality when discretizing.