The Casimir Effect

Two uncharged metal plates. Parallel, in a vacuum chamber, a few micrometres apart.

Two uncharged parallel metal plates in vacuum, with no visible field or matter between them, yet arrows point inward to show the unexplained attraction.attraction?

They pull on each other.

Not strongly. The force is small enough that you need a torsion balance to see it. But it’s there, and it’s measurable, and it’s reproducible, and there’s nothing in the gap doing the pulling. No charge. No air. No detectable field. The plates are inert metal, and the space between them is about as empty as space ever gets.

This is the Casimir effect, predicted by Hendrik Casimir in 1948 and pinned down experimentally by Steve Lamoreaux in 1997. It isn’t exotic physics. It shows up inside the chips in your phone, in the behaviour of thin films, and in the failure modes of microscopic moving parts. Nobody serious doubts that it happens.

Whether anyone has a comfortable feeling about why it happens is a separate question. The mechanism is understood. The mechanism is also strange, and the only way through it is to figure out what “vacuum” actually means.

The vacuum isn’t empty

Ask anyone what’s inside a vacuum. They’ll say nothing. That’s what the word means. Quantum field theory says otherwise, and has said so for nearly a century, and what it says is there has been measured in enough independent ways that the question is settled.

What QFT puts in place of nothing is fields. The electromagnetic field exists everywhere; it has a value at every point in space, all the time, light or no light. Most of the time you walk through it without noticing. In a perfectly empty universe, you might guess the field’s value should be zero everywhere. No light, no waves, nothing happening. The uncertainty principle has rules about this.

Position and momentum can’t both be pinned down with perfect precision. You’ve probably heard that part. The same rule applies to a field: it can’t hold exactly zero value with exactly zero rate of change. So it doesn’t. It wiggles. Everywhere. Always. There is no off switch. The wiggle has a name in physics: virtual photons. They are flickers of electromagnetic field at every wavelength, blinking in and out of existence faster than any instrument can resolve a single one of them.

Their fingerprints sit all over modern physics. They shift hydrogen’s energy levels by a measurable amount, the Lamb shift. They bend the electron’s magnetic moment a hair off its naive prediction. Without them, an excited atom would never emit a photon; it would sit in its excited state forever. And they’re the reason two metal plates in a vacuum pull on each other, which is why we’re here.

Here is the picture. Empty space is filled with every electromagnetic wave at once, each at the lowest energy the uncertainty principle will allow it, which is not zero. You don’t see any of this happening. The waves go in every direction, at every wavelength, and on any scale bigger than an atom they wash out into a uniform nothing.

Below: ten waves, each at its own wavelength, each cycling through its phase at its own rate. Above: their sum — what the field at this slice of space actually looks like. The sum never settles, because the ingredients are always drifting in and out of alignment.

In empty space, every wavelength fits. There’s nothing in the way.

But if you put two metal plates in the way…

Only some waves fit

Metal reflects light. Inside a good conductor, the electric field has to be zero. Free electrons rearrange themselves at the surface to cancel anything trying to penetrate, so a wave hitting metal can’t go through. It bounces.

Now put two pieces of metal facing each other, parallel, separated by a distance dd. Anything electromagnetic that lives in the gap has to drop to zero on both surfaces, simultaneously, the entire time. This is the same boundary condition a guitar string lives under, and it has the same kind of solution. Only the standing-wave modes that fit cleanly between the boundaries are allowed:

λn=2dn,n=1,2,3,…\lambda_n = \frac{2d}{n}, \qquad n = 1, 2, 3, \ldots

The longest wavelength that fits is λ1=2d\lambda_1 = 2d. Anything longer has no home in the gap — not faint, not damped, just absent. The geometry doesn’t host it.

Interactive Casimir explainer showing continuous vacuum modes outside the plates, discrete standing modes between them, and the resulting inward pressure imbalance.d = 1.00 μmarrows show the vacuum pushing on each side of each plate
1 nm10 nm100 nm1 μm10 μm
longest allowed wavelength in the gap
λ₁ = 2.00 μm
force on 1 cm²
1.30 × 10⁻⁷ N
radiation pressure across the scene

Outside the gap, the vacuum keeps its full spectrum. Inside, only standing-wave modes that fit between the plates are allowed. Below, the same scene redrawn as a pressure map: the thinner spectrum inside the gap means less pressure.

Drag the slider. As the gap shrinks, the longest allowed wavelength shrinks with it, and longer-wavelength modes drop out of the allowed set. They don’t squeeze through. They don’t tunnel. The geometry simply has no room for them. Meanwhile, the vacuum outside the plates is unconstrained and keeps its full spectrum.

You now have two vacuums sitting next to each other with different mode content. Outside, every wavelength. Inside, every wavelength shorter than 2d2d, and nothing longer. Each missing mode inside is a tiny bit of radiation pressure not pushing outward on the plate. The outside pressure pushes at full strength. The inside pressure is short by exactly the modes that don’t fit. So the plates feel a net push inward.

There’s another way to read the same picture, and you’ll see it phrased both ways in textbooks. The vacuum between the plates carries less zero-point energy than the vacuum outside, because it has fewer modes available to carry it. So when the gap shrinks, the total energy of the configuration drops. The plates fall together for the same reason a ball rolls downhill: there is somewhere lower to be.

The mode-counting argument

The simplest version of the calculation is to add up the zero-point energy of every allowed mode in the gap, do the same for an equal volume of unconstrained vacuum, and take the difference. The Casimir energy is the answer to that subtraction. Squeeze the plates, watch the answer change, and you have the force.

Naively, this is hopeless. The sum doesn’t converge. There are arbitrarily short wavelengths in the spectrum, each carrying a sliver of energy, and the total goes to infinity. The same thing happens outside the plates. Two infinities, one inside and one outside, and ordinary subtraction won’t get you anywhere.

What rescues you is that both infinities come from exactly the same place: the very-short-wavelength end of the spectrum. At those wavelengths the plates barely matter. A 1-nanometre wave can’t tell a boundary one micrometre away from a boundary infinitely far away. Whatever short-wavelength contribution you have inside the gap, you have an essentially identical contribution outside it. Subtract one from the other carefully and the bad parts cancel. The mathematical name for “carefully” is zeta-function regularisation. What survives the cancellation is the long-wavelength part of the spectrum, the part you watched dropping out of the widget. That part is finite. And it’s negative. The clean answer for the force per unit area is:

FA=−π2ℏc240d4\frac{F}{A} = -\frac{\pi^2 \hbar c}{240 d^4}

The minus sign tells you the force is attractive. The 1/d41/d^4 tells you it gets extreme fast: halve the gap and the force goes up by a factor of sixteen.

Interactive log-log plot of Casimir force per area versus plate separation, with a draggable marker.force per area F/A10⁹10³10⁻³10⁻⁹10⁻¹⁵1 nm10 nm100 nm1 μm10 μm100 μm1 mmplate separation d (log scale)
1 nm10 nm100 nm1 μm10 μm100 μm1 mm
1 mm²10 mm²1 cm²10 cm²
selected distance
d = 1.00 μm
force per area
1.30 × 10⁻³ N/m²
force on 1 cm²
1.30 × 10⁻⁷ N
≈ weight of 13.3 μg on Earth

Drag the orange marker or use the slider. The green line is the ideal parallel-plate result on a log-log scale.

At a 1 micrometre gap across a 1 cm² plate, the force is about 1.3×10−71.3 \times 10^{-7} newtons, roughly the weight of a grain of pollen. Detectable, but only with apparatus built specifically to detect it.

Close the gap and the numbers get serious. At 10 nm the force per unit area is comparable to atmospheric pressure. At 1 nm it’s about a thousand times that. MEMS devices — the accelerometers and micromirrors inside your phone, things a few hundred nanometres across — operate squarely in this regime, and the Casimir force is one of the dominant forces acting on their moving parts. It can pull two microscopic components together and cold-weld them into a single chunk of useless metal. Engineers call the failure mode stiction, and they spend real effort designing around it.

The plates attract

Casimir wrote down the prediction in 1948. Marcus Sparnaay tried to measure it ten years later and got a result consistent with the formula, but his error bars were too wide to rule out alternatives. The clean experiment came in 1997, when Steve Lamoreaux at the University of Washington built a torsion balance fine enough to see the force directly and confirmed Casimir’s prediction to within 5%. The experiment has since been repeated at the 1% level across dozens of geometries: sphere-plate, plate-plate, gold, aluminium, in fluids, in gases. The 1/d41/d^4 scaling shows up cleanly in the data. As results in quantum field theory go, the Casimir effect is about as well-tested as anything we have.

So: two parallel plates in a vacuum chamber, no charge, no current, no material between them, no field detectable by any normal instrument. They pull on each other. The pull comes from the vacuum, which is not empty and which has a structure the plates change by being there. Closing the gap lowers the configuration’s energy, and in physics, lower is where things go.

Empty space had been doing all this the whole time.

—

Key references: