Brookhaven National Laboratory

A magnifying glass on a sunny day brings light to a point hot enough to scorch paper. This furnace does the same thing with a different light and a different optic. The light is a 750 W infrared halogen bulb. In place of a lens there is a mirror shaped like an egg, and an egg-shaped mirror has a useful property: light leaving one point inside it all lands on one other point. Put the bulb at the first point and everything it emits arrives at the second. That second point is the hot spot, and the lamp is rated to reach 1200 °C there, with no heating element and no contact. The formal name is an ellipsoidal reflector furnace, or optical oven.

My mentor, Dr. Kenneth Evans-Lutterodt, works at NSLS-II, Brookhaven's National Synchrotron Light Source II, which produces x-ray beams used to work out what a material is made of and how its atoms are arranged. He wants to run those measurements while the sample is hot, to see how a material changes as its temperature goes up.

A conventional furnace is the wrong tool for that. Its walls would block the beam, and it would heat everything around the sample along with the sample. A spot of focused light heats the sample and leaves the x-ray path clear. On the poster the light converges from above and the x-rays cross sideways.

The bulb is a commercially available part, run from a 1000 W DC supply. Cooling air moves through the light source while the lamp is on, to keep the structural materials around it from overheating.

• The probe. A type-K thermocouple, which is two different metals joined at a tip. The junction puts out a small voltage that depends on its temperature. A Moxa reading device turns that voltage into a number the computer can store.
• The stage. Three motors, one per axis, driven by a Newport ESP 302 motion controller. Give the controller a position and it goes there, so the probe sits where the code says rather than where a hand left it.
• The code. Python. Move to a position, read the thermocouple, write down both, move to the next one, and keep going until the list of positions runs out. After a full scan it draws every reading as a color-coded map.
• The cameras. Two Prosilica GT cameras, each looking through a microscope and connected to a computer, pointed at the sample from different angles. The setup stays covered while the lamp is running, so the sample gets watched on a screen from outside the light's line of sight as the power goes up.

The map is the reason for scanning rather than taking one reading. A single number tells you the probe was hot. A grid of numbers tells you where the hot region sits and how quickly the reading falls off toward the edges. On the poster the hot region shows up as one bright patch off to one side of the scanned square rather than at the center, and finding that patch is how the focus gets located.

That map covers a 20 mm square, 5 to 25 mm on each axis, sampled as a 26 by 21 grid, so the steps are 0.8 mm along one side and 1 mm along the other. The same grid then repeats at 13 heights, so the beam can be followed up and down as well as across.

I ran the scan at about 50 W, far under the lamp's 750 W, to protect the probe while still bringing out the shape. The probe read about 100 °C at the hottest point. So the map fixes where the beam lands and what shape it is. It does not give the temperatures the lamp reaches once the power goes up.

The probe is what the motors carry. Once the map is made the probe comes off and the sample goes on in its place, so the sample ends up sitting where the readings were taken. That swap is also why the probe and a sample are never in the light at the same time.

A lamp this bright and this hot makes safety protocol a large part of the setup work. We moved the experiment to a spot with easier access to airflow and submitted Safety Approval Forms.

One thing I did not expect: past its melting point the indium held its shape. Surface tension keeps a small bead of liquid metal roughly spherical, the way a drop of water stays round on a leaf, so the sample looked much the same after melting as before, only smoother. Taking the power above 500 W did not change that. There was no moment to point at, which is why the evidence is a change in texture rather than a puddle.

That leaves the 250 W figure resting on my judgment of when the surface changed, and two people watching could pick different moments. Three things would replace the judgment with a measurement: step the power in fixed increments and hold each step; use a shape whose change cannot be missed, such as a wire bridge that sags or a cube whose corners round off; or let the camera catch the glare that comes off a liquid surface.

Move the bulb, swap the mirror or knock the stage, and the map changes. It describes this furnace on the days I scanned it and nothing else, so the map itself does not carry over to anyone else's bench. The method carries over: motors, a probe, the scan code, and a melting-point check. So do three things that turned up along the way. A melt can announce itself as a change in surface texture rather than a puddle. A shiny probe reads low. And scanning at low power gives shape and place without putting the probe at risk. Any lamp like this one has to be characterized before it is useful, and the method is a repeatable way to do it.

The rig and the code stayed with the group at Brookhaven and are in use there. The code is not public at this time.

What would come next is a calibration. Melt several metals with spaced melting points, tin, zinc and aluminum, at the same spot, and each one gives a true temperature at one lamp power. Together they make a ladder: real temperatures against power, which is what the map is missing, and a measurement of how far the probe reads low at each rung. The probe's error belongs to the probe, not the sample, so it stays the same whichever metal is on the stage. The samples should be alike in size and finish, with shiny ones roughened first, because a bright surface reflects light away the way the probe does and sits cooler at the same power. A probe coated dull, or a pyrometer reading the sample directly, would remove the reflection at its source. And feeding a sample sensor back to the power supply would let someone set a temperature instead of a wattage.