Why Time Needs Measuring

Time: Clocks, Calendars and How We Measure It ๐Ÿ†

Course 1 ยท Chapter 1 ยท The first problem of timekeeping

Imagine you are a hunter-gatherer living some forty thousand years ago, and you want to meet your cousins at the river when the salmon run starts. You cannot send a text, there is no calendar on the wall, and nobody has invented a number for "next month". What you have is the sky. The Sun comes up and goes down. The Moon grows fat, thins away and returns. The seasons swing round and the cold comes back. These three rhythms are the oldest clocks there are, and for almost the whole of human history they were the only ones. This course follows the long story of how people learned to count them, argue about them, build machines to imitate them, and finally replace them with the vibrations of atoms.

What was run, and what to trust
Every Python snippet in this chapter and in the three exercise solutions was executed, and the output in the comments is what Python printed. The astronomical constants (day, month and year lengths) are the standard published values, written here from memory without looking them up again during writing. The descriptions of the ancient bones are summaries of what archaeologists have proposed, and the lunar readings of them are interpretations, not settled facts.

Three Natural Clocks

Each natural clock comes from a different motion of the Earth or the Moon, and that is the root of every calendar headache in the chapters to come.

CycleWhat is really movingWhat a person seesLength
Daythe Earth spins on its axissunrise to sunrise24 hours (mean solar day)
Monththe Moon orbits the Earthnew Moon to new Moonabout 29.53 days (synodic month)
Yearthe Earth orbits the Sunone full round of the seasonsabout 365.2422 days (tropical year)

The Earth spinning, the Moon circling and the Earth circling the Sun are three separate motions. Nothing in physics makes one of them a tidy multiple of another. As the next section shows, the numbers do not fit.

Why They Do Not Divide Evenly

How many months are in a year? If the answer were a whole number, calendar-making would be easy. Divide the year by the month:

YEAR = 365.2422 # tropical year, in days MONTH = 29.530589 # synodic month (new Moon to new Moon), in days print(round(YEAR / MONTH, 3)) print(round(12 * MONTH, 3), round(YEAR - 12 * MONTH, 3)) print(round(13 * MONTH, 3), round(13 * MONTH - YEAR, 3)) # 12.368 # 354.367 10.875 # 383.898 18.655

A year contains 12.368 Moon cycles: twelve and a bit. Twelve months add up to 354.367 days, which is 10.875 days short of a year. Thirteen months overshoot by 18.655 days. So a calendar built purely on the Moon slides backwards through the seasons by nearly eleven days a year, and a calendar built purely on the Sun ignores the Moon completely. Every calendar in history is a compromise between these two facts, and the compromises fill chapters 3 and 4.

The day has the same trouble with the year. The year is 365.2422 days, so if you count days you must carry that extra 0.2422 of a day (nearly six hours) somewhere. A calendar of exactly 365 days loses about a quarter of a day each year. After four years it is nearly a whole day (0.97 of a day) behind the seasons.

Year 365.2422 days 12 months 354.367 days (short by 10.875) 13 months 383.898 days (over by 18.655) 365 days short by 0.2422 per year (about 5 h 49 min)

A Closer Look at the Day

Even "a day" is less simple than it sounds. There are two sensible ways to measure one, and they give different answers.

Solar daySidereal day
Measuredfrom noon to noon (the Sun back at its highest)from one passage of a fixed star to the next
Length24 h 0 min 0 s on averageabout 23 h 56 min 4.09 s
Used fordaily life, civil timeastronomy, pointing telescopes

Why the difference? While the Earth spins once, it also moves a little along its orbit around the Sun. After one full turn relative to the stars the Earth has to turn a little further for the Sun to be back at its highest. Over a whole year, this adds up to exactly one extra turn of the Earth, so a year contains one more sidereal day than solar days: 366.2422 sidereal days, against 365.2422 solar days. The sidereal day is the solar day scaled by that ratio:

YEAR = 365.2422 solar_day = 86400.0 # seconds sidereal_days_per_year = YEAR + 1 # one extra turn per orbit sidereal_day = YEAR * solar_day / sidereal_days_per_year h, rem = divmod(sidereal_day, 3600) m, s = divmod(rem, 60) print(round(sidereal_day, 2)) print(int(h), int(m), round(s, 2)) print(round(solar_day - sidereal_day, 2)) # 86164.09 # 23 56 4.09 # 235.91

The sidereal day is 86,164.09 seconds, or 23 h 56 min 4.09 s, which is 235.91 seconds (about 3 min 56 s) shorter than the solar day. That is why a given star rises about four minutes earlier each night, and why different constellations dominate the sky in different seasons.

Neither day is perfectly constant. The solar day varies slightly in length through the year (chapter 2 explains the equation of time), and tides slowly drag on the Earth's spin so that the day is getting longer by roughly a couple of milliseconds per century. For now, take the "24 hours" day as an average.

The Oldest Records: Notches on Bone

How early did people count these cycles? We cannot know what they said or thought, but some objects look very much like tallies. The most famous are two bones from Africa:

  • The Lebombo bone, from a cave in the Lebombo mountains on the border of South Africa and Eswatini, is a piece of baboon leg bone about 43,000 years old, with about 29 clear notches cut in it. The bone is broken at one end, so there may once have been more. A count near 29 is close to a lunar month, which is why it is sometimes called a lunar tally.
  • The Ishango bone, found in what is now the Democratic Republic of the Congo, is roughly 20,000 years old and carries groups of notches in three columns. It has been read as a tally, as a lunar record and even as a simple arithmetic aid.

Be careful with these claims. Marks on bone could be a hunting tally, a trade record or decoration, and nobody can prove which. What is safe to say is this: tens of thousands of years ago, people made patterned marks in sets that look like counting, and counting the Moon is one reasonable explanation. A tally keeps one thing the memory cannot: a number you can check tomorrow.

Tally Drift: Why Counting Is Not Enough

Suppose a bone-carver cuts one notch per day and starts a new column at each new Moon. They must decide how long a month is. Real months last 29 or 30 days, never both at once. Counting every month as 29 days makes the tally lag the Moon. Counting 30 makes it run ahead. How quickly?

MONTH = 29.530589 def drift_days(months, tally_length): """Days by which a fixed-length tally has drifted from the Moon.""" return months * (MONTH - tally_length) for length in (29, 30, 29.5): print("tally", length, "-> drift after 12 months:", round(drift_days(12, length), 3), "days") # tally 29 -> drift after 12 months: 6.367 days # tally 30 -> drift after 12 months: -5.633 days # tally 29.5 -> drift after 12 months: 0.367 days

A fixed 29-day month falls 6.4 days behind the Moon in a year. A fixed 30-day month runs 5.6 days ahead. Even the best fixed guess, 29.5 days (alternating 29 and 30), is still 0.367 days out after twelve months and a full day out after about 33 months. A tiny error in a repeated count does not stay tiny: it adds up. This is the key idea of the whole course, and we will meet it again with calendars, with pendulum clocks, with ships at sea and with computers. A timekeeper is judged less by how well it starts than by how its error grows.

What Makes a Good Clock?

We can now ask what we mean by a "good" clock. People use three words loosely, but they mean different things. Definitions vary a little between fields. These are the working meanings used in this course.

WordQuestion it answersExample of failing
AccuracyHow close is the clock to the true time, on average?a clock that runs 10 minutes fast
PrecisionHow finely can it be read, and how repeatable is the reading?a clock with only an hour hand
StabilityHow steady does its rate stay from one day to the next?a clock gaining 1 s one day and losing 3 s the next

A very stable clock is the most useful, even if it is inaccurate, because a steady error can be measured once and corrected for ever. A stable clock that gains exactly 2 seconds a day can be set right by subtracting 2 seconds every day. An unstable clock cannot be corrected in advance. This is why later chapters spend so much effort on making the rate steady (pendulums, balance springs, quartz, atoms) rather than on setting the starting time.

Errors are often quoted as a fraction. A clock that gains 1 second in a day is wrong by 1 part in 86,400. Engineers write this as parts per million (ppm):

SECONDS_PER_DAY = 86400 for err in (1, 2, 10): ppm = err / SECONDS_PER_DAY * 1_000_000 per_year = err * 365 print(err, "s/day =", round(ppm, 2), "ppm =", per_year, "s/yr =", round(per_year / 60, 1), "min/yr") # 1 s/day = 11.57 ppm = 365 s/yr = 6.1 min/yr # 2 s/day = 23.15 ppm = 730 s/yr = 12.2 min/yr # 10 s/day = 115.74 ppm = 3650 s/yr = 60.8 min/yr

A clock only 1 second a day out looks wonderful, yet it is 6 minutes a year out, and 10 seconds a day means an hour a year. For navigation, science and the internet, these small errors are not small at all. By the end of the course you will see clocks that are wrong by less than one second in many millions of years.

Common mistakes
Treating "a year" as 365 days exactly (it is 365.2422, which is the reason leap years exist); assuming 12 months make a year (12 Moon cycles are 10.875 days short); confusing the solar day with the sidereal day (they differ by about 4 minutes); and mixing up accuracy with stability (a clock can be badly wrong yet perfectly steady, or right on average yet jumpy). Another common slip: reading a tally bone as proof of a specific ancient calendar. Treat those readings as plausible interpretations.
Exercise 1

Using a year of 365.2422 days and a month of 29.530589 days, find how many months are in a year, and how far 12 and 13 months miss a year. Then search for the whole number of years (up to 39) whose length is closest to a whole number of months, printing each new best fit. Which one wins, and by how many days is it off?

๐Ÿ“„ View solution
Exercise 2

Three clocks were compared with the true time at midnight on 8 days, giving the offsets listed in the solution file. For each clock, find the mean daily rate (a measure of accuracy) and the spread of the daily changes (a measure of stability) using Python's statistics module. Which clock is the easiest to improve, and why?

๐Ÿ“„ View solution
Exercise 3

A tally maker cuts one notch per day. Using a Moon month of 29.530589 days, find how far the tally has drifted after 12 months if every month is counted as 29, 30 or 29.5 days, and after how many months each version is a whole day out. Then simulate alternating 29- and 30-day months and print the running error for 12 months.

๐Ÿ“„ View solution

Quick Reference

  • Three natural clocks: day (Earth spins), month (Moon orbits), year (Earth orbits the Sun)
  • Lengths: mean solar day 24 h; synodic month 29.530589 days; tropical year 365.2422 days
  • They do not fit: a year is 12.368 months, so 12 months fall 10.875 days short
  • Sidereal day: 86,164.09 s (23 h 56 min 4.09 s), about 236 s shorter than the solar day
  • Drift: a small error repeated many times adds up (1 s/day is 6 min/yr)
  • Accuracy / precision / stability: closeness to true time / fineness of reading / steadiness of rate
Coming next
Time: Clocks, Calendars and How We Measure It 2 is about shadows and water: how a stick in the ground became a sundial, why the Sun is not a perfect clock, and how water clocks kept time when the Sun was hidden.