Twenty-Three Where Fifty-Three Appeared

The number came back wrong. Not slightly off—wildly, impossibly wrong. Thermal expansion coefficient for the aluminum test bar: 53 × 10⁻⁶ K⁻¹, more than twice the published value of 23 × 10⁻⁶ K⁻¹. Either every material science textbook since 1950 is lying about aluminum, or my interferometer is broken.

Started this because the grating ruling engine fails when temperature drifts 2°C—thermal expansion changes groove spacing by 34%, which destroys diffraction quality. The obvious solution: measure thermal expansion directly using optical interference. Same physics as coating thickness monitoring, but aimed at dimensional stability instead. One wavelength of sodium light equals 589 nm, which means one fringe shift represents 294.5 nm of physical movement (light travels down and back). Count fringes, multiply by λ/2, divide by temperature change, and you have the coefficient of thermal expansion measured to sub-micron precision.

Built a Fizeau setup Wednesday night. Simpler than a Michelson because you don’t need a beam splitter—just place an optical flat directly onto the test specimen with a tiny air wedge between them. Interference fringes appear where the gap exists. Heat the specimen, watch the fringes shift as it expands, count how many pass by a reference point, and the math falls out.

Used the 75 mm optical flat from the estate sale bin, the one I’d been cleaning with distilled water after learning acetone destroys AR coatings. Test specimen: 100 mm aluminum bar, 25 mm square cross-section, from the metalworking scrap pile. Clamped the bar to the bench, placed the flat on top, illuminated with a sodium street lamp positioned 2 meters away through a slit.

Fringes appeared immediately—parallel yellow stripes across the flat’s surface, spaced about 8 mm apart. Beautiful. Textbook example of equal-thickness interference. The pattern looked like a topographic map where each contour line represents 294.5 nanometres of elevation change in the air gap.

Heating the bar with a heat gun (variable temperature, set to 100°C), I watched the fringes begin to move. Slowly at first, then faster as the aluminum warmed. They marched across the surface like waves rolling in. Each fringe that passed the reference point (a scratch marked on the flat with a Sharpie) represented 294.5 nm of expansion.

Five minutes of heating. Counted 23 fringes. Bar temperature measured with a type-K thermocouple taped to the far end: started at 21.3°C, stabilized at 43.7°C after the heat gun shut off. Temperature rise: 22.4°C.

Total expansion: 23 fringes × 294.5 nm = 6.77 µm.

Coefficient of thermal expansion: α = ΔL / (L₀ × ΔT) = 6.77 µm / (100 mm × 22.4°C) = 3.02 × 10⁻⁵ K⁻¹ = 30.2 × 10⁻⁶ K⁻¹.

Aluminum’s published value: 23 × 10⁻⁶ K⁻¹.

Difference: 31% high.

Repeated the measurement with longer heating time to reach 60°C. Got 38 fringes over 38.7°C temperature rise. Same calculation: 52.7 × 10⁻⁶ K⁻¹. Even worse—now I’m reading 129% high.

What Else Moved

Spent an hour checking for errors. The fringe count was solid—recorded video through a loupe and played it back frame-by-frame. Definitely 38 fringes. Sodium wavelength: 589 nm by definition. Temperature measurements looked reasonable. The math is four operations long and none of them involve logarithms or trig functions where I might have messed up a calculator input.

Started questioning whether the aluminum bar was actually aluminum. Maybe it’s an alloy with magnesium or silicon that changes the expansion coefficient? Checked with a file and a magnet—soft, non-magnetic, oxidizes to white powder. Aluminum. Probably 6061 or similar structural alloy, which has α = 23.6 × 10⁻⁶ K⁻¹, essentially identical to pure aluminum.

The thermocouple could be wrong. Checked it in boiling water: 99.1°C at 710m elevation (predicted boiling point: 98.9°C). Within error. Checked in ice water: 0.3°C. Fine.

Recalibrated the heat gun temperature setting by measuring output with the thermocouple. The dial says 100°C but it’s delivering 110°C at the nozzle. Doesn’t matter—I’m not using the heat gun temperature for the calculation, only the thermocouple reading on the bar itself.

Then I touched the optical flat. Still warm.

Of course it’s warm. The flat is sitting directly on the aluminum bar, which is at 60°C. Glass has thermal expansion coefficient around 9 × 10⁻⁶ K⁻¹. Lower than aluminum, but not zero. The flat is expanding too.

Worse: the wooden bench under the clamped bar is also warming up. The clamp is steel. The bench surface is plywood over two-by-four framing. Wood expands anisotropically—different rates along and across the grain—with coefficients ranging from 3 to 10 × 10⁻⁶ K⁻¹ depending on species and moisture content.

The entire assembly is expanding together. The bar, the flat, the bench, the clamp. The fringes I counted represent relative motion between the bottom surface of the optical flat and the top surface of the aluminum bar. But the flat is rising as the bench expands underneath it, and the bar is rising as it expands upwards from its clamped base. I’m measuring a combination of aluminum expansion, glass expansion, wood expansion, and whatever strain gets distributed through the mechanical coupling.

No reference surface is actually stable. There’s no fixed point.

The Measurement I Didn’t Build

Proper thermal expansion interferometry requires isolating the reference path from the test path. In a Fizeau setup, that means the optical flat must stay at constant temperature while the specimen changes. You’d need active cooling—maybe a water-jacketed flat holder circulating 20°C water from a temperature-controlled bath, or an air gap with forced convection, or a massive heat sink big enough that the flat’s temperature doesn’t drift during the ten-minute measurement cycle.

Or you’d switch to a Michelson interferometer, where the reference arm and test arm are physically separated. The test specimen sits in one arm, isolated inside an insulated enclosure with a heating element and a viewport. The reference mirror sits in the other arm, mounted on a thermally stable platform—granite block, Invar base, something with near-zero expansion. Beam splitter combines the paths, detector counts fringes.

LIGO does this at the 10⁻¹⁸ meter scale. Four-kilometre arms in vacuum tubes with seismic isolation and active stabilization. Same fringe-counting physics, better implementation. They’re measuring spacetime compression from gravitational waves; I’m trying to measure aluminum warming up by 40°C. The precision requirements are different by a factor of 10¹⁵ but the failure mode is the same: environmental coupling ruins everything.

I could build a thermally isolated Michelson, but it’s not a Wednesday-night project. Need viewports, vacuum fittings or at least sealed enclosures, circulating coolant, temperature sensors, and probably three iterations of prototype before it works. More importantly, I’d be solving mechanical engineering problems (how to isolate heat flow) rather than learning optical physics (how interference measures displacement). The hobby becomes building a stabilization platform, not using interferometry.

Measurement Boundaries

The journal entry from May came back: “Measurement technique interferes with the phenomenon.” Modal analysis changed mushroom resonances by adding transducer mass. Optical pyrometry measures emissivity and temperature simultaneously without separating them. Thermal expansion interferometry measures everything expanding at once unless you isolate the reference.

The tool shapes what you can know. Sometimes by coupling to the process. Sometimes by requiring infrastructure you don’t have.

Stopped heating the aluminum bar, let everything cool back to room temperature, and watched the fringes march backwards as thermal contraction reversed the expansion. Same phenomenon, opposite sign. At least the physics works symmetrically.

The ruling engine thermal drift question remains unquantified. I know 2°C variation causes 34% groove spacing error because that’s calculable from α = 23 × 10⁻⁶ K⁻¹ and 3.33 µm nominal spacing. But I can’t measure my own shop’s thermal stability without first solving the reference stability problem that makes thermal expansion interferometry difficult in the first place.

Circular dependency. Measurement requires the thing you’re trying to validate.

The optical flat is back in its case. Heat gun unplugged. Thermocouple coiled in the drawer. Aluminum bar still clamped to the bench, waiting for a use that doesn’t involve counting fringes that measure nothing useful.