Astra For Me THE SKY, THIS MONTH
ECLIPSES  ·  THE SAROS

Three clocks that agree once every 18 years

The Moon keeps three separate periods and none of them divides into the others. Every 223 of the first, all three come back together within a few hours, and the same eclipse happens again. We measured all three over a century, then walked a real eclipse series forward to watch it hold.

Astra For Me 13 Sep 2026 12 MINUTE READ
MEASURED

Anything in a violet panel we computed from DE421 ephemeris data. You can check it.

INTERPRETED

Anything on a gold rule was said by people, at a particular time. We say who, and when.

Why there is not an eclipse every month

START WITH THE ONE THAT IS OBVIOUS

There is a new moon roughly every twenty-nine and a half days, and at every one of them the Moon stands between us and the Sun. If everything moved in one plane that would be an eclipse each time.

The Moon's orbit is tilted about five degrees to the plane the Earth goes round the Sun in, so at most new moons the Moon passes above or below and its shadow misses entirely. Here is every new moon of this year, with how far the Moon sat off that plane at the moment of conjunction.

EVERY NEW MOON OF 2026
New moonOff the plane
18 Jan 20263.37°
17 Feb 20260.93°
19 Mar 20261.75°
17 Apr 20263.89°
16 May 20264.94°
15 Jun 20264.68°
14 Jul 20263.21°
12 Aug 20260.90°
11 Sep 20261.68°
10 Oct 20263.83°
9 Nov 20264.93°
9 Dec 20264.60°

2 of 12 fall within 1.5°, which is roughly where a shadow can still reach the Earth. The others range out to 4.94°, and nothing happens at all.

So eclipses need two things at once, a new moon and the Moon near that plane, and the two have different periods. The rest of this article is about what happens when you ask how often they line up again.

Three months, none of them the same length

AND ALL THREE ARE NEEDED

The Moon comes back to the same place three different ways, and each answer is a different number of days.

The synodic month is new moon to new moon. The draconic month is one crossing of the Earth's orbital plane to the next, which is what decides whether the shadow lands. The anomalistic month is closest approach to closest approach, which decides whether the Moon looks big enough to cover the Sun completely or leaves a ring of it showing.

We measured all three from a century of real positions rather than quoting them.

MEASURED OVER A CENTURY
Synodic month29.53059 days, from 1236 of them
Draconic month27.21224 days, from 1341
Anomalistic month27.55455 days, from 1325
Span18 Jan 1950 to 24 Dec 2049

Each period is the slope of a line through every event of the century, so a single perturbed month cannot pull the answer. The instants agree with an independent check to within 179.2 seconds.

Now multiply. Take 223 synodic months, and see what the other two do over the same stretch of time.

WHERE THE THREE MEET
MonthDaysCountTotal
synodic29.530592236585.32
draconic27.212242426585.36
anomalistic27.554552396585.54

The draconic column lands 0.95 hours from the synodic one and the anomalistic column 5.18 hours from it, across an interval of more than eighteen years. That is the saros.

Nothing requires this. Three periods set by unrelated parts of the Moon's motion happen to come back into step after eighteen years, to within a few hours out of a hundred and fifty thousand. It is the sort of coincidence that makes a technique possible.

Three periods with nothing to do with each other, back in step to within a few hours out of a hundred and fifty thousand.

What that interval looks like on a calendar

AND THE EIGHT HOURS THAT MATTER
ONE SAROS
Length6585.32 days
Which is18 years, 10 days, 7.7 hours
Leap days in that span5
Earth turns further by116°

The familiar figure is eighteen years and eleven days. Ours says 10, and the difference is bookkeeping: this particular interval happens to contain 5 leap days rather than four. The hours are the part that never changes.

Those extra hours are the interesting part. The Earth does not care that an eclipse is due; it keeps turning, and by the time the geometry repeats it has carried you 116° further round. So the next eclipse of a series happens over a different part of the world, roughly a third of the way west.

Three saros periods put that right, because three lots of eight hours is a day. After 54.09 years the shortfall is down to 0.8 hours, about 12° of longitude, and the eclipse comes back to roughly where it started. That period has its own Greek name, the exeligmos, and we will come back to why that matters.

Walking a real series

SIX ECLIPSES, ONE FAMILY

The test of all this is whether it works on an actual eclipse. We took the total eclipse of 11 August 1999, 11:03 UTC, the one that crossed Europe, and stepped forward and back by exactly one measured saros at a time.

ONE SAROS SERIES
DateSun inSun to MoonOff the planeDistance, km
20 Jul 1963Cancer 27°22′-0.59°0.58°375661
31 Jul 1981Leo 7°50′-0.29°0.54°374482
11 Aug 1999Leo 18°21′-0.05°0.49°373313
21 Aug 2017Leo 28°54′0.15°0.44°372155
2 Sep 2035Virgo 9°29′0.27°0.40°371015
12 Sep 2053Virgo 20°08′0.34°0.34°369892

The Moon never gets more than 0.59° from the Sun in longitude across 6 steps, and stays within about a degree of the plane throughout. The distance varies by 1.6%, which is what decides whether an eclipse in the family is total or leaves a ring.

The dates are worth reading. The step lands on the eclipse that crossed the United States in 2017 and on the one due over the western Mediterranean in 2035, both of which are real eclipses of the same family, and neither was put in by hand. The only input was the length of a synodic month.

Look at the last column but one, though. The Moon is not returning to exactly the same place: it is walking across the plane at about 0.05° per step, in one direction, and it does not stop.

That is why a series ends. Extrapolating the drift until it walks out the other side of the window gives about 62 eclipses over 1127 years. Published counts run a little higher, sixty-nine to eighty-seven eclipses over twelve to fifteen centuries, and the gap is honest: our estimate assumes the drift stays constant, and it does not.

Who worked this out, and what they called it

THE INTERPRETATION, AND ITS HISTORY

From here down we are reporting rather than measuring.

BABYLON, BY THE MID 6TH CENTURY BC AT THE LATEST

Babylonian astronomers had the 223-month recurrence long before anyone could explain it. They had no model of the Moon's orbit and no notion of an orbital plane; what they had was several centuries of dated eclipse records, and a lunar eclipse is visible from half the Earth at once, so one city's archive was enough to see the pattern.

The scholarship on this is specific about what can and cannot be said. Brack-Bernsen and Steele, working from the cuneiform tablets themselves, place the cycle in use by the middle of the sixth century before Christ. No individual is credited with finding it, and none should be: it emerged from record-keeping rather than from a discovery.

LONDON, 1691

The word saros is a mistake, and a well documented one. Edmond Halley found the term in the Suda, a tenth-century Byzantine lexicon, which reported it as a measure among the Chaldeans, information that had reached it from Berossus by way of Eusebius. He attached it to the eclipse cycle in a paper for the Royal Society.

The Babylonian word behind it, šāru, meant 3600, a unit for very long stretches of legendary time. It had nothing to do with eighteen years. Guillaume Le Gentil pointed the error out in 1756 and the name stuck anyway.

So it is wrong to say the Babylonians called it the saros. They found the cycle; the name was given to it by an Englishman two thousand years later, out of the wrong book.

THE AEGEAN, ABOUT 100 BC

The most striking evidence that the cycle was worked rather than merely known is the Antikythera mechanism, recovered from a wreck off the island in 1901 and read properly only in the last twenty years.

On its back is a spiral dial of 223 months, four turns of it, carrying engraved glyphs that mark which months a lunar or solar eclipse was predicted in and at what time of day. Nested inside it is a much smaller dial divided into three, marked with corrections of nought, eight and sixteen hours. That is the exeligmos: the three-saros period, and the eight hours this article measured, cut in bronze.

The readings are Freeth and colleagues, published in Nature in 2006 and 2008. Be careful what you claim for the machine: it predicted that an eclipse was possible and roughly when, in the Babylonian manner. It did not compute what you would see from a given town.

NINEVEH, 7TH CENTURY BC

The interpretation attached to all this was not about individuals. In the Babylonian and Assyrian omen tradition an eclipse was a message about the king and the state, and the omens were compiled in the series Enuma Anu Enlil, whose lunar eclipse tablets Francesca Rochberg edited in 1988.

It was taken seriously enough to be acted on. The Neo-Assyrian royal correspondence, edited by Simo Parpola, documents the substitute king ritual actually being performed against specific eclipses: another man was installed on the throne to absorb what the omen threatened while the real king withdrew, and was disposed of afterwards. Five performances are attested between 679 and 666 BC.

Whatever one makes of the practice, it is a reminder that these people were not being decorative. They believed the sky had told them something about a named person, and they acted on it at the cost of a life.

AND HOW IT BECAME PERSONAL

The move from omens about the state to charts about a private individual is usually told as a clean break at the Hellenistic period. Specialists no longer tell it that way. Rochberg, who has done more than anyone to edit the primary material, argues for continuity: individual birth horoscopes appear in Babylonia from the fifth century before Christ, and the shift in emphasis was gradual rather than a moment anyone can date.

We would rather leave that unresolved than give you a tidy date that the people who read the tablets do not accept.

What we do and do not claim

OUR OWN POSITION, DECLARED

The arithmetic above is the app's own arithmetic, and none of it depends on astrology being true. It is a statement about three periods of the Moon's motion, and an astronomer would sign every line of it.

What the sky does not tell you is what an eclipse means. That part came from Babylonian scribes reading a message about their king, and it has been rewritten many times since. We compute the first part and we tell you where the second came from, which is a different thing from telling you it is true.

One practical note. The saros predicts that an eclipse will happen and roughly where; it does not give the path across the ground, and modern predictions do not use it for that. Its real job now is bookkeeping, sorting eclipses into numbered families so that one can be compared with its relatives.

YOUR OWN SKY

Your chart, and the transits over it, computed from the same ephemeris data as everything above. Free.

Compute your chart

Sources

SO YOU CAN CHECK THEM YOURSELF