Mapping Temperature Between Four Walls: 2D Steady Conduction by Hand

Starting point: four walls, four temperatures

Most heat transfer coursework starts with 1D conduction — a slab with hot on one side, cold on the other, temperature falling in a straight line between them. Real components are rarely that considerate. A rectangular plate, a PCB, an electronics enclosure wall, or a building slab typically has different boundary conditions on all four edges, and the resulting temperature field is genuinely two-dimensional: isotherms curve, heat flows diagonally, and the obvious shortcut answer, just average the four wall temperatures, turns out to be wrong except in one special symmetric case.

This is the problem a 2D steady-state conduction solver addresses: given four boundary temperatures — left, right, top, bottom — and the material's conductivity k, what is the temperature everywhere inside the plate, and specifically at its center?

The governing equation and why symmetry helps

Steady 2D conduction with no internal heat generation is governed by Laplace's equation:

d^2T/dx^2 + d^2T/dy^2 = 0

This equation has no simple closed-form solution for a general rectangle with four arbitrary boundary temperatures — which is why real 2D conduction problems are normally solved numerically, with a Fourier series solution (superposition of four separate 1D-boundary sub-problems, each with three sides held at zero and one side at the actual boundary temperature) or with a finite-difference grid.

There is, however, a very useful shortcut for the centerpoint of a square or near-square plate: by symmetry, the center temperature under four independent boundary temperatures sits close to the simple average of the four edges, with the deviation from that average driven by how asymmetric the boundary conditions are — this deviation is exactly what a boundary bias metric is meant to capture, separately from the raw center temperature.

Center temperature ~= (T_left + T_right + T_top + T_bottom) / 4     [+ correction for aspect ratio & asymmetry]

Conductivity k does not change the center temperature in a homogeneous plate at steady state — a hotter or colder material redistributes heat faster or slower but converges to the same equilibrium field. What k does control is the heat-flux index: how much thermal energy per unit area is actually crossing the plate to sustain that temperature field, and how quickly the plate would respond if a boundary condition changed.

A worked case: an electronics enclosure wall

Consider an aluminum enclosure wall with conductivity k = 205 W/(m*K), where four edges are held at fixed temperatures by contact with different parts of an assembly:

Left temperature   = 85 C   (power-supply side)
Right temperature  = 25 C   (ambient-facing side)
Top temperature    = 60 C   (near a heat-generating component)
Bottom temperature = 30 C   (near the enclosure base, closer to ambient)

The simple average of the four boundaries:

(85 + 25 + 60 + 30) / 4 = 200 / 4 = 50 C

That 50 C is a reasonable first estimate for the center temperature in a roughly square plate, and it's exactly the number you'd compute if you treated the four-sided problem as fully symmetric. But look at the spread: the max temperature span here is 85 - 25 = 60 C, a large swing across a single component, which tells you immediately that the plate is thermally asymmetric — heat enters strongly from the left (power supply side) and top (hot component), and exits toward the right and bottom.

That asymmetry is what the boundary bias metric is meant to capture: it's not just the average that matters, but whether the hot boundaries are adjacent to each other (compounding the local gradient near that corner) or opposite each other (which tends to produce a more genuinely two-dimensional, diagonal flow pattern). In this case, left (85 C) and top (60 C) are adjacent, so the corner between them runs noticeably hotter than the naive average would suggest, while the opposite corner, between right (25 C) and bottom (30 C), runs correspondingly cooler.

The heat-flux index, scaled by conductivity, quantifies how aggressively energy is moving through the material to sustain this 60 C span. With k = 205 W/(m*K) for aluminum, the flux is large relative to what the same boundary conditions would produce in, say, a k ~= 0.2 W/(m*K) plastic enclosure — the plastic case would show the same 50 C average center temperature and the same 60 C boundary span, but a dramatically lower flux, because conductivity doesn't change the temperature field at steady state, only how much energy must flow to sustain it.

Common mistakes when reasoning about 2D conduction

The single most common error is treating a 2D problem as if it were 1D by picking the two most obviously "hot vs. cold" boundaries — here, left 85 C and right 25 C — and computing a simple linear gradient between them, ignoring the top and bottom entirely. This throws away real information: in the example above it would predict a center temperature of 55 C rather than the more complete 50 C estimate, and it says nothing about the corner-level asymmetry that the boundary bias captures.

A second mistake is assuming conductivity affects the equilibrium temperature distribution. It doesn't, in a homogeneous single-material plate at true steady state. Engineers under deadline pressure sometimes try to "fix" an overheating simulation by bumping k up, when the actual lever that changes center temperature is the boundary conditions themselves — improving contact with a cooler heatsink, for example — rather than the bulk material property.

Finally, remember this whole model assumes steady state and no internal heat generation. A PCB or enclosure with active components dissipating power internally needs a source term added to Laplace's equation, turning it into Poisson's equation instead, and a genuinely transient startup condition, the first seconds after power-on, is governed by a completely different, time-dependent version of this problem.

What the boundary bias is really telling you

It helps to think of boundary bias as a diagnostic rather than just another output number. A near-zero bias means the plate's temperature field is close to what a naive average would predict, which typically happens when opposite boundaries are reasonably balanced (left and right similar, top and bottom similar) even if the plate as a whole spans a wide range. A large bias means the hot boundaries are clustered together geometrically, which concentrates the highest gradients into a single corner region rather than spreading them across the whole plate. Two enclosures can have identical average boundary temperatures and identical max temperature spans, and still need completely different thermal management, purely because of which edges the heat is entering from.

This is directly useful when placing a component inside an enclosure. If you know the boundary bias is being driven by two adjacent hot edges, the corner between them is where you'd expect the tightest local gradient and the highest risk of exceeding a component's operating temperature, even though the calculated center temperature of the whole plate looks comfortably moderate. Relying on the center temperature alone, without checking whether the boundary conditions are biased toward one corner, is a common way that a thermal design "passes" a coarse hand calculation and then still runs hot in the field once real components and airflow are added.

It's also worth sanity-checking the aspect ratio of the plate itself. The center-temperature approximation above assumes something close to a square domain; a long, narrow rectangle behaves more like two separate 1D problems stitched together along its length, and the simple four-boundary average becomes a progressively worse estimate as the aspect ratio grows. For anything meaningfully non-square, treat the quick average as a sanity check rather than a final answer, and lean on a proper 2D solve for the real number.

Try it yourself

If you're estimating a hot spot inside an enclosure, a PCB layer, or a slab with mismatched boundary temperatures, plugging in your actual left/right/top/bottom values and conductivity is far more informative than eyeballing an average. You can try the 2D conduction temperature simulator here to get the center temperature, max temperature span, heat-flux index, and boundary bias for your own boundary conditions. For a related problem involving thermal degradation over time rather than steady-state gradients, the accelerated life Arrhenius model tool is worth a look.