Experiments 18

How Cold Is Space

In 1990 a NASA satellite held a heated blackbody up to the sky and turned it until the two matched. That is how we know the temperature of the universe: 2.725 degrees above absolute zero. Here you turn the heater, with COBE’s real data and its flight log.

Rust · WASM · Rerun · three.js · COBE FIRASView source 

Match the sky.

Turn the heater until the needle over the yellow ring goes still.

Heater5.00 K−268.15 °C
What the detector sees: sky and heater light, mixed. Matched, the ripples cancel.

Infrared sky: COBE DIRBE at 100 microns

The most perfect glow ever measured

In January 1990, John Mather showed the first FIRAS result to a meeting of the American Astronomical Society. It came from nine minutes of data, and every point sat on a blackbody curve. The room stood up and applauded.

This is the final version, from the whole mission. The squares are what FIRAS measured; the line is a perfect glow at 2.725 degrees above absolute zero. On the top chart the error bars are thinner than the line.

01002003004005101520Brightness (MJy/sr)-500-25002505005101520Gap from the curve (kJy/sr)
Frequency, in waves per centimeter. Bottom: the gap between each point and the curve, zoomed in 187 times, with each point’s error bar. Every point lands within its error bar. The bars grow at high frequencies, where the glow is faint; near the peak each point is pinned down to a few parts in 100,000.

Data: Fixsen and colleagues, 1996, from NASA’s LAMBDA archive.

Notes on how it works

Notes

How FIRAS measured the sky

COBE’s Far Infrared Absolute Spectrophotometer never looked at the sky on its own. Light from the sky and light from a small blackbody inside the instrument, the internal calibrator or ICAL, went into the same interferometer, and the detectors read only the difference. Setting the ICAL close to the sky’s temperature cut the signal by about 99 percent. That near-null is what made the result absolute instead of relative. The heater on this page is the ICAL, simplified: one dial, one needle. A second blackbody, the external calibrator, swung into the sky horn to check the whole chain; the log shows it there about one-eighth of the time.

The ripple trace is that difference as the interferometer would see it. It is computed from the same numbers that drive the needle, summed over FIRAS’s 43 low-frequency channels (2.3 to 21.3 waves per centimeter) with an idealized instrument, not the real one’s response.

What the needle compares

The sky side of the needle is FIRAS’s own temperature map: 6,067 sky patches, averaged over those within 4° of where you point, about the size of the instrument’s 7° beam. That map is on the 1996 calibration, where the sky averages 2.7277 K. It is rescaled by one factor, 0.99917, to the final value of 2.72548 K (Fixsen, 2009). The match counts when the heater is within 0.08 thousandths of a degree.

Dust is why the Milky Way never goes still. FIRAS’s dust map is fit with one dust spectrum, then weighed with the frequencies that matter most for temperature. No heater setting can cancel what is left: about 0.1 thousandths of a degree far from the plane, and 1 to 5 on it.

The speed

Moving through the glow makes it a little warmer ahead and colder behind: ΔT/T = v/c × cos θ. From two matches, the page solves for v using the direction FIRAS’s own map gives, galactic longitude 265.1° and latitude 47.9°. That map alone says 3.34 thousandths of a degree, or about 368 km/s. Planck later measured 369.8 km/s. The direction lies in Crater, next to Leo; the opposite side is in Pisces. It is the solar system’s motion, a mix of the Sun’s orbit around the galaxy and the galaxy’s own fall through space.

The recording

NASA’s LAMBDA archive now publishes FIRAS’s time-ordered data, with a guide dated February 2026. The COBE window uses one of its four detector channels: 567,325 interferograms from 20 November 1989 to 21 September 1990, each with its time, sky pixel, pointing, and COBE’s latitude, longitude and altitude. Positions are turned to space coordinates with Earth’s rotation angle; the Sun’s direction comes from a short ephemeris. The heater temperatures are the instrument’s own housekeeping log: during sky observations the ICAL read between 2.749 and 2.767 K. The window jumps to a real moment when FIRAS looked where you point.

The recording is built with the Rerun Python SDK and plays in the Rerun web viewer (0.38.1). The viewer’s 14 MB of WebAssembly comes from jsDelivr, only on large screens or when you open the recorder. Coastlines are from Natural Earth.

The picture of the sky

The sky is COBE’s other infrared camera, DIRBE, at 100 microns, with zodiacal light removed. It shows dust. The colors only grade its brightness, from faint to bright; they are not temperature. The red and blue halves that appear after you find our speed are the dipole from FIRAS’s map.

The sound is an analogy. Two tones a little apart beat, like an out-of-tune guitar string, and go quiet together when they match. FIRAS made no sound.

Sources