How to read a light curve

A thin line beside a passage says how far its claim is proven: measured, implemented, published, illustrative. Colored numbers carry the same mark. A passage with no line is stated. How claims are marked.

The two axes

Across is time. Within one night it is usually hours; across many nights, days. Astronomers often count time in Julian Days, one running day count that saves the calendar arithmetic when nights fall in different months. On my variable-stars page, a single night is drawn in hours, and each point is a 10-minute average (implemented).

Up and down is magnitude, the astronomer's brightness scale, and it runs backwards: a bigger number is a fainter star, so magnitude 10 is brighter than magnitude 11. Light curves put brighter at the top, so the numbers on the side get smaller as you go up. My variable-stars page labels that axis "magnitude, brighter up" (implemented).

The scale is a ratio, not a ruler. Five magnitudes is exactly a factor of 100 in brightness, so one magnitude is a factor of about 2.5. A dip of 0.1 magnitude is about 9% less light, and 0.01 magnitude is about 1%.

Why every curve is relative: comparison stars

Through a night the air changes. A star sinks toward the horizon and shines through more air, haze drifts in, a thin cloud passes. All of that dims every star on the frame at once.

So a light curve is measured differentially: the star's brightness is compared with steady stars on the same frame at the same moment. What dims them all cancels out; what changes only in your star is left.

On my variable-stars page, brightness comes from differential photometry against an ensemble (a group) of steady Gaia DR3 stars on the same frames, put on an approximate V scale from Gaia colors. V is the standard green-yellow filter band most variable-star work is reported in. The page's values are untransformed, meaning they were not converted to the standard system using standard stars, so levels can sit a few hundredths of a magnitude off a standard V (implemented).

That page also draws a steady comparison star alongside each night's curve, for scale (implemented). Read the two together: if the comparison star wiggles as much as your star, the wiggle probably came from the night, not the star. My AAVSO guide gives the same advice before you submit: leave out a curve whose comparison star wanders as much as the variable.

Relative also means the level can shift between observers. Two people using different comparison stars or filters can get curves with the same shape, offset by a few hundredths of a magnitude. Compare shapes and timings first, levels second.

Error bars and noise

Every point is a measurement with an uncertainty. An error bar usually spans one standard deviation, called one sigma, either side of the point. If the noise behaves, about two points in three land within one bar of the true curve, and about one in twenty lands more than two bars away.

Fainter stars get bigger bars. A faint star sends fewer photons, and the sky's own glow and the camera's noise weigh more against them. That is what the faint-star button above shows (illustrative).

Averaging helps, but slowly: noise shrinks with the square root of the number of points, so averaging four points halves it. My variable-stars page plots 10-minute averages rather than single subs (a sub is one single exposure) (implemented).

Sigma is also a yardstick for "is it real?". A swing of 5 sigma is five times the noise; with well-behaved noise, chance alone makes a swing that big less than once in a million tries. When points scatter much more than their bars, something else is adding noise, such as clouds, a close neighbor star or a frame edge, and the bars understate it.

Period and phase folding

Many variable stars repeat. An eclipsing binary is two stars orbiting each other, lined up so that each one passes in front of the other. The light dips deeper when the hotter star is hidden (the primary eclipse) and less when the other one is (the secondary).

A night is often shorter than one cycle, so each night catches only a piece. Here are six made-up nights of the same made-up star, each in its own color:

If you know the period, you can fold the curve. For each point, work out where it falls in the cycle: its phase, the fractional part of (time − a reference time) ÷ period, a number from 0 to 1. Plot brightness against phase instead of time, and every night's piece lands on the same curve.

Try the presets. At the true period, all six nights trace one curve. At 1% too long, each night slides sideways a little more than the last and the dips smear. At half the period, the deep and shallow eclipses land on top of each other, so phase 0 shows two layers of points. At twice the period the curve looks clean but shows two deep eclipses per cycle, which tells you the trial is too long; its spread number still rises, because each slice of phase now covers twice as much of the real cycle (illustrative). Half and double the true period are the classic traps in period searches, because they still line up much of the data.

My variable-stars page does this in its "full cycle" view: it stacks every session of a star by where it falls in its cycle, so many partial sessions make one full cycle (implemented). The periods it folds on come from the Variable Star Index (VSX), the AAVSO's catalog of known variable stars, not from my own period search (published).

The labels on my variable-stars page

Variable stars in Seestar Collective images defines four labels and has a filter button for each (implemented). Here they are with that page's own definitions, word for word:

Clearly varies

on at least one night the star's swing is at least 5 sigma and at least twice that of its comparison stars, and those comparison stars stay flat.

Likely varies

nearly as strong, or the comparison stars weren't quite flat that night.

Held up across sessions

I stacked every session of the star by where it falls in its cycle (the "full cycle" view), separately for two groups of telescopes chosen before anyone looked at the data. Both groups showed the same variation on their own.

Has a timing

a time of minimum or maximum that passed a 5-sigma test.

In plain words, using the sections above:

• Clearly varies is a one-night test. The swing beat the noise five times over, beat the comparison stars' own swing twice over, and the comparison stars stayed steady. It says the star changed that night; on its own it doesn't say the change repeats (implemented).
• Likely varies is the same test, just short of that bar, or passed on a night whose comparison stars weren't quite flat (implemented).
• Held up across sessions is the stronger check. It is a fold, done twice, on two groups of telescopes split before anyone looked. A fluke in one telescope or on one night would show up in one group, not in both.
• Has a timing means at least one measured time of minimum or maximum; for an eclipsing binary, a minimum is the middle of an eclipse. Timings collected over years show whether a period is slowly changing.
• No label doesn't mean the star is constant. In the page's words, a star that doesn't vary there "may simply have been caught at a quiet phase" (implemented).

How many stars carry each label, from the page's own totals (as of 2026-10-13; measured in October 2026, from sessions between 2024-06-08 and 2026-10-11 UTC; data: the Seestar Collective):

The same page lists 385 times of minimum or maximum (measured). It calls all of its measurements preliminary, and I have submitted none of them to the AAVSO; the people who took the images can, under their own observer code (implemented). A few of its stars share their light with a close neighbor, or depend on which aperture size is used, and their cards say so (implemented).

Next steps

• Look up your own telescope on the variable-stars page and read its curves with the axes, comparison star and labels above in mind.
• If your Seestar caught a variable, submit your light curves to the AAVSO under your own observer code, after checking each curve the way the guide describes.
• Light curves need a time for every sub, so make sure your subs are saved: what your Seestar saves, and which files to keep.
• The lines beside passages mark how far each claim is proven; see how claims are marked.