Barnett
Barnett
Barnett
by H.A.R. BARNETT
1. INTRODUCTION
1.1 This exercise all started out of a harmless puzzle in the July 1993 issue of the Staple Inn Actuarial
Society's publication "The Actuary". The puzzle was suggested by P. R. Watson and required the
solver to discover on which day of the week the 13th of the month was most likely to fall on the
assumption that the Gregorian calendar continued indeffitely. The cycle of 400 years includes 97 leap
years and thus consists of an integral number of weeks; or put another way, the average length of the
Gregorian year, 365.2425 days is 7 times 52.1775, giving exactly 20,871 weeks in 400 years. The
solution appeared in the August 1993 issue, together with a subsidialy puzzle which gave rise to the
1.2 This led to the appreciation of the fact that the number of days in the true solar or tropical year to four
decimal places is 365.2422, or to eight decimal places 365.24219878, and a new puzzle was set to
devise a modification to the Gregorian system to reduce the error of about 3 days in every 10,000 years.
2.1 The Gregorian calendar is a modification of the Julian calendar, which itself set out to correct the
accumulated errors in the Roman calendar. Full details of the changes in that calendar before the time
of Julius Caesar are hard to come by, but much information can be obtained from Ovid '2' and
Sandys "'. Initially the Romans adopted a lunar system (hence months) and were apparently inhibited
by a superstition against even numbers, so the months were alternately of 29 and 31 days, and the
Calends (Ist), the Nones (5th or 7th) and the Ides (13th or 15th) all fell on odd days of the month.
Sometimes additional months were inserted (intercalation) but by the time Julius Caesar amended the
calendar in 46 B.C. a year of 450 days was required to bring the se'mns back to the right place in the
year. But the pre-Julian months, as amended by Julius, form the basis of the months we still use; hi
plan was for months to be alternately of 31 and 30 days (the superstition about even numbers having
apparently been forgotten or ignored) and Julius was careful to ensure that the month of Quintilis (which
was renamed after the Emperor himself) should be one of the greatest months (31 days). Julius
Caesar's astronomers calculated the length of the year as 365.25 days, so Februaty (only) was lefi with
29 days, with a 301h day in every fourth year, giving the pre-Gregorian system of 365, 365, 365 and
366 days in the year. It was not until 7 B.C. that the month of Sextilis was renamed after Augustus
Caesar (although he had become emperor in 43 B.C.), and to make August a 31 day month a further
day was taken from February. The counting of the A.D. years from 31st December in the year 1 B.C.
The months in which the Nones fell on the 7th and the Ides on the 15th (March, May, July, October)
are all now 31 day months, but I have found no reference to the fact that if Julius had alternate 31 and
30 day months right through from March to January, October would only then have been a 30 day
month. Unfortunately Ovid (2', who analysed each imponant day of each month, stopped when he
reached the end of June; but the lengths of the months are not really relevant to the current exercise.
When Pope Gregory XI11 realised that the seasons were no longer exactly where they should be, he took
his astronomer's advice that the length of the year was really 365.2425 days, so he introduced the
system still in force today, of making the century years ordinary (non-leap) years unless they were
multiples of 400. The astronomer also advised that there should then (in the 16th century) be an
adjustment of 10 days to correct the accumulated error. Why it was 10 days and not 12 (for the A.D.
years 100, 200,300,500, 600,700, 900, 1000, 1100, 1300, 1400 and 1500) is not clear. Although
Ovid " put the summer solstice at June 26th. Frazer's " notes say it was June 24th. whereas now it
falls on June 21st or 22nd. We shall probably never know whether Pope Gregory (or his astronomer)
deliberately altered the seasons from the way they were in the days of Julius Caesar, whether the 10 day
adjustment was a mistake, or whether it was an intentional device to ensure that the seventh day A.D.
followed throughout the centuries (some not until the current 20th century) with the adjustments for the
surplus days increased from Gregory's 10 days according to the century of adoption (1 1 days for Britain
A MATHEMATICAL EXERCISE
My first solution to the puzzle described in 1.2 was to convert to non-leap years those which are
divisible by 3,200, but to reinstate as leap years those divisible by 80,000, to reconvert to non-leap years
those divisible by 800,000, and to re-instate as leap years those divisible by 4 million. This wwld give
an average of 365.242199 days in the year (ie: accurate to 6 decimals). I subsequently realised that if
this system of exceptions stopped at 800,000, with no change at 4 million the result would be more
accurate at 365.24219875 (accurate to 7 decimals), while to obtain accuracy to 8 decimals would require
A.B. Pepper came up with an elegant alternative, correct to 8 decimals. After converting to non-leap
years those divisible by 3,200, the reinstatement as leap years.would not occur until those divisible by
G.P. Mann suggested a sequence with multiples of 2,800 as non-leap years, but multiples of 14,000 as
leap years, but multiples of 70,000 as non-leap years, and multiples of 980,000 as negative leap years
A further suggestion by both Pepper and Mann was to dispense with the 400 year cycle in favour of a
500 year cycle. so that after every 4 leap but every ICQnon-leap the sequence continued: every 500 leap
but every 5000 double leap (367 days), but every 1 million leap.
3.5 1 suggested a modification to this. Consider a watch which gains 1 minute per week (as does mine);
when it has gained half a minute I move the large hand back one whole minute (thus keeping it
synchronised with the supplementary hand which makes a revolution every minute), and by this means
my watch is never more than half a minute out plus or minus. So I felt that the system in 3.4, rather
than going out in the same direction for 5,000 years, might have an intermediate stage at 2500 years,
thus: every 500 leap, but every 2,500 double leap, but every 5,000 leap, but every million non-leap. (A
similar half way modification to the other suggestions would not be appropriate as they would require a
3.6 However, none of these suggestions is really practical, which is why I have labelled this section as a
mathematical exercise. With an average year of between 365 and 366 days there is never likely to be
acceptance of a system requiring any year to be 364 or 367 days; and that only leaves the systems
described in 3.1 and 3.2 which can involve years which are somewhat awkward numbers (albeit
4.1 It was at this stage that I acquired a copy of ~veni"'and I can do no better than quote from the book:-
Minor r e f o m have taken place since the time of Pope Gregory. By agreeing to convert A.D. 4000,
8000 and 12000 to common years we have reduced the difference to 1 day in 20,000 years. Finally, at
an Eastern Orthodox congress held in Conrtantinople in 1923, yet another rule was adopted in the
parade of legiskzion engendered by the drive for accuracy. It stated that century years divisible by 900
will be leap years only if the remainder is 200 or 600. The resulring calendar is accurate to I a@ in
44,000 years.
I had not seen in any book of reference that years which are multiples of 4,000 should be non-leap years
(more of this later). And the Eastern Orthodox rule seems to me to be wrinen in mathematical
gobbledygook; century years divisible by 900 are not leap years anyway, unless they are also divisible
by 3,600, while (in my understanding of the word "remainder") if they are divisible by 900 there is no
remainder. It has been suggested that the Eastem Orthodox rule may have been mistranslated, and that
it intended to change to an o r d i i year any fourth century year which, when divided by 900, left a
remainder of 200 or 600. But this would mean, in the first ten millennia A.D., reducing to non-leap
years the years 2000, 2400, 5600, 6000, 9200 and 9600,which would be too many reductions; and far
too niany if the intention was to treat 4000 and 8000 A.D. in the same way.
4.2 One of the interested actuaries had already communicat+ with a representative of the Royal Greenwich
Observatoly, Cambridge, who at first had seemed encouraging, but later lost his enthusiasm; he said the
matter had already been "done to death", civilisation will have changed markedly in the long term, and
any adjustments will be made as and when necessary. So far as being "done to death" is concerned, he
offered no book of reference to confirm this. Civilisation will no doubt have changed somewhat in
1,000,000 years or even in 100,000 years, but can we be so sure it will have changed by 20000 A.D?
And when it changes, are we right to assume there will be no homo sapienr (or perhaps some being
even more sapienr) who will be interested in counting time in solar years? And as for adjustments "as
and when necessary", what will happen if the astronomers (or governments) of different nations cannot
agree when they are necessary? Surely it is better while a United Nations Organisation is in existence to
try to get some agreement to correct the error which is already accumulating?
4.3 "'
1 was not to be put off by the R.G.O8s lack of enthusiasm, and after reading A V ~ N 1 wrote to the
same official. In his reply he said that so far as he knew there had been no official adoption of the
4000, 8M)O, I2000 rule. Also he agreed that the Eastern Orthodox rule, as quoted, was gobbledygook.
'
He commented that any possibility of d e f ~ ag Universal Calendar receives short shrift from all
participants at major conferences. It sounds as though the use of astronomical knowledge is too
4.4 I would not dream of advocating a Universal Calendar! I believe both the Jewish and the Moslem
calendars deal with leap years by the intercalation of additional months from time to time, with lengths
of year varying by tmsiderably more than one day. But 1 would advocate endeavouring to obtain
agreement by governments of all countries which have adopted the Gregorian calendar. Indeed, the fact
that one modification has apparently been adopted in the U.S.A. (see Aveni's "royal we") and another
by many countries of Eastern Europe and of Asia, indicates how necessary it is for such agreement to be
obtained.
GENERAL CONSIDERATIONS
The foregoing is merely a preamble to interested readers in the subject! It seems desirable that (weather
permitting) the seasons should come round the same time each year. Why the solstices and equinoxes
should ever have been fixed by the preJulian, Julian and Gregorian calendars in the months of March,
June, September and December, and not at the beginning or end of these months - nor even in the
middle - I have been unable to discover, but now users of the Gregorian calendar have become used to
The precession of the equinoxes causes variations which, however, "return to base" every 25,800 years
but, as these movements are slow and, ultimately, self-correcting. I propose to ignore them.
The solar year is gradually shortening. In the London "Sunday Telegraph" of January 29th, 1995 it was
reported that Dr George Williams, a geologist at the University of Adelaide, had proved that 620 million
years ago the length of the solar year was between 393 and 407 days. This represents a reduction, on
the average, of between ,00009 and ,00014 of a day (or between 8 and 12 seconds) every 2,000 years,
due to the spin of the earth k i n g slowed by the drag of the ocean tides, the earth's orbit around the sun
having remained essentially unchanged . And Whitrow '4' states that the solar year decreases by about
,00006 days every 1,000 years, having been approximately 365.24232 days when the Julian calendar
was introduced in 45 B.C.; this coincides with Dr Williams' conclusion. So, as the number of days in
the solar year is decreasing, the tendency will be for more and more leap years to need to be
"suppressed" or converted to non leap years, thus increasing the existing calendar errors. This seems to
make it imperative to provide for the errors of which we are already aware.
5.4 The slowing of the spin of the earth causes another complication. From time to time an additional
second (a "leap second") is needed; one such was added to midnight on 31st December, 1987. As the
slowing continues, more and more leap seconds will be required.. In due course one will be needed
every year, and when this occurs it would be desirable for a "new second" to be used, of such length of
time that a year of new seconds will be equal to a year of old seconds plus one old second! Aveni "'
states that the variable length of the day is irregular, but the calculations of Whitrow @'and of Dr
Williams seem to indicate that the second needs redefining about once in every 200 years. This is
6. RECOMMENDATION
6.1 Had it not been for the information in 5.3, my recommendation would have been to avoid a complicated
system by keeping to simple and easily remembered numbers; this could have been achieved by
suppressing the leap year (i.e.: to substitute a non-leap or ordinary year) whenever the year A.D. is a
multiple either of 4,000 or of 1.0.000 (or of both). This would give an average calendar year of
365.2422 days and a remaining error of about one day in 800,000 years.
6.2 However,.the continual reduction in the length of the solar year (see 5.3). although apparently minute,
has quite a sizeable effect even in what I would regard as the medium term (say up to the year 20000
A.D.). If y is (year - 2,000) + 2,000 then 10y is the number of seconds reduction after the year 2000
(ignoring fractions) and the accumulated reduction (in days) by any year after 2000 is
of 2,000 remaining in 6.1 apart from 2000 itself) there will still be an error of over 6 days by the year
20000 A.D. (This paragraph assumes the rate of reduction in the solar year will remain unchanged; and
it is desirable for posterity to b o w the sixth decimal figure in the rate of ,00006per 1,000 years; the
figure of 9.4 in the above table may correctly be any number of integral days from 8 to 11 depending
6.3 So it will bi necessary not only to agree that all years qfler 2000 A.D. which are multiples of 2,000
should be ordinary (or non-leap) years, but also a note should be left for posterity in the medium t e ~
that one further leap year should be suppressed in the I lth, 13th. 15th. 16th (or 17th for the sake of
smoothness), 19th and 20th millennia, subject to the caveat in brackets at the end of 6.2. (The
additioh information in 5.3 does, of course, make nonsense of the modifications to the Gregorian
number of days in the solar year remains the same, then some time between the 36th and 42nd millennia
the solar year will be 365.2400 days, and by then all century years should be non-leap years. While
after about 4 million years the 366 day year should disappear altogether, and distant posterity will need
to decide which years should have only 364 days! So it will be. a very long time before actuaries
constructing mortality tables will need to ask themselves "when is a year not a year?"
6.5 1 would suggest that the IACA, through its membership in many countries, should pass the
recommendations made in 6.3 above to the U.N., to be adopted by all member countries using the
Gregorian calendar.
7. ACKNOWLEDGEMENTS
7.1 My thanks to Katie Bedford of Cheltenham Ladies' College for sending me extracts from Sandys "' and
Ovid "'; to my daughter, Fiona Wild, for following this up by presenting me with a copy of
Ovid '2'; and to Nick Andenon, Brian Clark. Rod Marshall and Tony Pepper for suggesting
References:
Empires of Time : Calendars, clocks and cultures (I.B. Tauris & Co Ltd)
"' Ovid
Time in History - views of timefrom prehistory to the present day (Oxford University Press)