A definition for the term "time scale":
- 16th General Assembly of the International Astronomical Union
Grenoble, France (1976)
Resolution No. 4 by Commissions 4 (Ephemerides) and 31 (Time)
- a useful time scale is generated by any process which enables dates to be assigned to events
The concept of time has been refined throughout history, and new understanding usually produces a new time scale. The list in this web page aims mostly at describing when and why new time scales were developed. At the end of this web page are links to other web pages which better explain the definitions of various time scales.
Many visits to this page are prompted by APIs used in computer languages that want to handle elapsed time. Terser than the many details here I point to the fundamental problem which confounds POSIX time_t and other computer implementations of time.
I have some plots of some of these time scales.
The rotation of the earth under the moon and sun produces tides. Reckoning time by tides is sufficiently simple that rocks can do it. Sedimentary rocks called tidal rhythmites contain records of earth paleorotation dating back three billion years.
Apparent Solar Time has been used since prehistory. It is reckoned at any location by observation of obvious phenomenon such as sunrise, sunset, or passage through the meridian (noon). Reckoning time by sunrise and sunset is sufficiently simple that plants can do it.
Reckoning time by local apparent noon is sufficiently simple that anyone with a stick in the ground can do it. The ellipticity of the earth's orbit and the obliquity of the earth's equator produce a variation in the duration of a day reckoned by apparent noon. The variation is known as the "equation of time", and it is large enough that a pendulum clock is stable enough to measure it. The "equation of time" is best visualized in combination with the annual variation in the latitude of the sun in order to produce the analemma.
The earliest almanacs tabulated the "equation of time". Until 1930 in the British Nautical Almanac and 1935 in the American Ephemeris and Nautical Almanac the position of the sun was tabulated at apparent noon as well as at mean noon.
The IAU has had a love/hate/love relationship with MJD.
It should also be noted that the use of JD or MJD for the UTC time scale is problematic and ambiguous at the precision of one second. JD and MJD express the elapsed count of some form of "day" as real numbers along a presumably unsegmented, continuous number line. The UTC time scale (and, historically, GMT as used in practical situations before the advent of UTC) contains changes in rate and discontinuities. In particular, there is no obvious way to represent a leap second of UTC (or the smaller leaps present in the available forms of GMT and UTC before 1972) using JD or MJD notation.
The needs of railroad schedules produced the adoption of standard time zones within which every railway station clock shared the same mean solar time. The UK railways began to adopt Standard Time in 1840, and the US railways adopted Standard Time in 1883.
The mean solar time of the Greenwich meridian. Sadler (1978) wrote a monograph on the sordid history of GMT which should be read by everyone. Beyond that are a few other details below.
Prior to the long sequence of usages below it is
relevant to point out one thing:
GMT does not now have, and never has had, leap seconds.
(Except when what was called GMT did have leap seconds, which was for the first few years starting 1972-01-01 when the radio broadcast time signals switched from the old rubber/elastic seconds version of coordinated time to the new leap second version of coordinated time. The US NBS broadcasts of WWV and WWVB and the UK broadcasts did this. In the case of the US this can be attributed as a misnomer associated with the fact that there had been no official documents giving the name UTC. In the case of the UK the BBC newsreaders still employ the term GMT. For the sordid history of the confusion between GMT and UTC in other nations see preprint 11-662 at the 2011 colloquium "Decoupling Civil Timekeeping from Earth Rotation" and preprint 13-505 at the 2013 colloquium "Requirements for UTC and Civil Timekeeping on Earth")
For the only time scale which really has leap seconds see UTC.
Nevertheless, the practical and historical reality is that the available forms of GMT have always had leaps, just not leap seconds. The leaps happened every time any official clock was reset to agree with calendar days of earth rotation.
Some agencies currently assert that GMT is currently used as a synonym for UTC. Below is a history of meanings which have been applied to GMT. There is no current definition of GMT which is authoritative in all contexts.
References to GMT before this time cannot have had any contemporary meaning. They are probably better interpreted as indicating what is now known as UT.
The mean solar time of Greenwich attained a unique distinction at the International Meridian Conference held in Washington during October. By international vote it became the one time, the cosmic time, the universal time which most nations agreed to use.
The exact wording of the protocols had a curious effect in England because it indicated that the transit instrument then in use at the Greenwich Observatory, the Airy Transit, should define the prime meridian. All of the maps of England, however, used the meridian occupied by the previous transit instrument at Greenwich, the Bradley Transit. By strict interpretation of the International Meridian Conference (IMC) results, all maps of England suddenly had incorrect values of longitude. This change of longitude origin went unrecognized until 1956.
Within the pages of the British Nautical Almanac and Astronomical Ephemeris the time 00:00:00 GMT had previously meant noon, but starting this year 00:00:00 GMT now meant midnight. Sadler (1978) indicated that the use of the same name "GMT" for a quantity entirely opposite in meaning to the original term was by order of the Admiralty.
Because the British Nautical Almanac had declared the value of what had once been noon now to be midnight the IAU recognized that the term "GMT" had been rendered useless as a precise term in most archival and practical contexts. Anyone encountering a document containing the term "GMT" could only be certain of its meaning if the document were verified to have originated before 1925. For documents after 1925 there would be no way to be certain whether the author had meant "old GMT" or "new GMT".
UT as determined by transit observations from the observatory affiliated with the radio station. This was before the seasonal variations of UT2 had been codified, so depending on the observatory the values would initially have been UT0 and later something approaching UT1.
Through this era it was typical to reset the master clocks controlling the radio broadcasts by a fraction of a second whenever it was indicated by the transit observations. These clock resets were, in effect, "leap milliseconds", and they were consistent with the way that mechanical clocks had always been reset in order to keep track of mean solar time. Tabulating the differences between various radio broadcasts had been a principal mission of the BIH from its inception, and only through those records can the meaning of old radio broadcasts be interpreted.
These many usages for GMT recall a bit of childhood nonsense:
"When I use a word," Humpty Dumpty said, in a rather scornful tone, "it means just what I choose it to mean - neither more nor less."
"The question is," said Alice, "whether you can make words mean so many different things."
"The question is," said Humpty Dumpty, "which is to be master - that's all."-- Lewis Carroll, Through the Looking Glass
In 1928 the IAU approved the use of this term as a synonym for Universal Time. The US almanac used this term from 1925 until 1952 to indicate that days were being reckoned from midnight. The British almanac could not be convinced to use the term GCT, and continued to use the term GMT with the new astronomical meaning.
Part of the argument against GCT was that the legal civil time of Greenwich was set forward one hour in the summer. In 1935 the IAU recognized that the ambiguity caused by the disagreement between astronomical usage and civilian usage would never permit complete adoption of GCT. The term GCT was discontinued by resolution 2 of Commission 3 (Notations) and Commission 4 (Ephemerides) at the IAU 5th General Assembly in 1935. Starting in 1939 the US almanac used "GCT or UT" until 1952 when it switched to UT alone.
The congressional act that authorized the 1884 International Meridian Conference (IMC) called for a "standard of time-reckoning throughout the globe". The body of the proceedings contains 72 instances of the phrase "universal day" and 57 instances of the phrase "universal time". The title of the resulting proceedings called this "a universal day." Resolution 5 of the IMC stated "That this universal day is to be a mean solar day", and Resolution 4 stated that the initial meridian for measuring longitude would be Greenwich. Thus the resolutions of the IMC specified the use of the mean solar time of Greenwich, and Universal Time became synonymous with GMT reckoned from midnight.
It is important to note that Universal Time has always been a conventional construct based on measurements of sidereal time. It is not easy to measure the position of the sun with great precision. Even if it were easy the solar measure would still have to be converted to UT by application of some conventional formulae including the "equation of time".
The proceedings of the 2nd General Assembly of the IAU in 1925 in Cambridge contain discussion and consternation over the change of terminology introduced in the UK Nautical Almanac at the beginning of the year. They were much concerned about the inability to unambiguously interpret whether a document using the term GMT had the old or new meaning.
The General Assembly created a "Temporary Commission on Time" (see page 15) which adopted this conclusion:
Astronomers have not yet reached sufficient agreement on the unique terminology that would be desirable to use for time, and are not prepared to formulate a rule about the subject. It is therefore desirable that astronomers exactly define the time they provide.
The terms Greenwich Civil Time (G.C.T.), Weltzeit (W.Z.) and Universal Time (U.T.) denote time measured from Greenwich Mean Midnight, and are not ambiguous.Unfortunately the use of the term "Greenwich Civil Time (G.C.T.)" proved to be politically unacceptable.
Resolution 1 by IAU Commission 4 (Ephemerides) at the 7th General Assembly in 1948 recommended (see page 4)
that the designation 'Universal Time' (Temps Universel; Weltzeit) only be used by astronomers to designate mean solar time, reckoned from midnight of the Greenwich meridian.Again we see the IAU making UT synonymous with GMT.
Through the 1950s UT was used as the independent variable of the ephemerides.
UT2 is constructed from UT1 by adding an empirical formula to remove the effect of the annual seasonal variations in the rotation of the earth. In that sense UT2 is more smooth than UT1, and it provided a better gauge of "mean" for mean solar time. The correction from UT1 to UT2 is at most about 0.035 s. The UT2 time scale has no official name, but early mentions of it in publications by the BIH used the terms "Temps Universel uniforme provisoire" and "provisional Uniform Universal Time".
The CCIR elevated the status of UT2 in 1963 when Recommendation 374 specified that radio broadcast time signals should be in close agreement with UT2. In 1966 CCIR Recommendation 374-1 repeated that specification.
Before the year 1970 measurements of the random variations in the rotation of the earth using the precision made possible by atomic chronometers had revealed UT2 as a dubious exercise, for there are other variations in the rate of UT1 -- some of which are predictable, and some which are not. Nevertheless, because of the the first two versions of CCIR Recommendation 374 it is arguable that UT2 remains the legal time scale for those countries whose legal documents assert that time is based on GMT or the Greenwich Meridian. Sadler (1978) points out that UT2 was of little relevance by 1978.
The distinction between time epoch and time interval is not clear to everyone.Commission 31 produced an explanatory note (see page 16) drafted by W. Markowitz, H.M. Smith, and L. Essen in the hope of clarifying the situation which included this quote
The epoch of U.T. is determined by the angular position of the Earth around its axis; it is required for various scientific and technical purposes and for civilian use, sometimes without delay.
Sadler (1978) provided a detailed review of the history of mean solar time, GMT, and all forms of UT before 1984.
During the 1970s new techniques for observing earth orientation were proving to be far more precise and accurate than previous means. These included satellite observations (laser and radio), lunar laser ranging (LLR), and very long baseline interferometry (VLBI). Traditional optical transit measurements of stars had always been limited to observations at night, and certain systematic effects had never been seen.
Aoki et al. (1982) gave the expression for UT1 which came into use in 1984 and they explain why it is no longer based on the sun.
The current version of UT1 is exactly what is needed for geophysical investigations. It permits evaluation of the length of day (LOD) with a precision that reveals changes due to storm systems and changes in ocean currents. But in continuation of a trend that began over a century ago, there is really no publicly available measured quantity which can indefinitely be used as an authoritative value of mean solar time. Indeed, Fukushima (2001) demonstrated that in about 360000 years the value of UT1 given by the IAU 2000 resolutions will differ by a full day from the count of calendar days experienced by people living on the earth.
The goal behind the new expressions defining UT1 has been to ensure continuity of the value and rate of UT1 at the instants of change. These new expressions very closely track the hour angle of the mean sun because they were designed to match a quantity which once did that, but there is no longer a quantity specifically designed to indicate the hour angle of the mean sun. Continuity and uniformity of UT1 has been deemed more important.
Historically no attempt has been made to create expressions for UT which will be valid for more than a few centuries. Demanding a better expression for mean solar time is not immediately reasonable because, in a Clintonesque sense, it depends on what the meaning of the word "mean" means. In particular,
Within the next millennium the IAU will have to consider defining and naming a new version of UT to serve as the quantity which is measured by analemmatic sundials. A name for that time scale might perhaps be Analemmatic Universal Time (UTA).
From 1895 through 1984 the Vernal Equinox was the location on the non-relativistic celestial sphere defined by Newcomb's expressions. UT1 was determined from GMST using Newcomb's formula. Nevertheless there were several instances of potential discontinuity in UT1 when new fundamental catalogs of stars were adopted.
In accordance with IAU resolution from 1976, 1979, and 1982, in 1984 a new set of astronomical constants came into use along with a switch from the FK4 star catalog to the FK5. From 1984 into 1997 the Vernal Equinox was defined by the expressions in Lieske et al. (1977). UT1 was determined from the expressions in Aoki et al. (1982), who also point out that the new expressions and catalog resulted in shifts of the longitudes of some observing stations.
From 1997-02-27 through the end of 2002 there was an additional correction to the equation of the equinoxes based on IAU 1994 resolution C7 recommendation 3. This correction was required by the vastly increased precision that VLBI had contributed to the measurement of earth orientation.
Based on IAU recommendations in 2000, starting in 2003 the Vernal Equinox was abandoned as the coordinate origin for measurement. A point with some of the original properties has been resurrected by Capitaine et al. (2003). They also point out that the effort to maintain continuity of UT1 had invalidated the classical interpretation of GMST and the equation of the equinoxes.
At the IAU 26th General Assembly in 2006 the IAU approved Resolution B1 which explicitly contain a new definition for the ecliptic, and thus for the equinox, to come into use starting in 2009.
Since 1984 GMST has not been the Greenwich hour angle of the mean equinox, and the equation of the equinoxes has not been the true right ascension of the mean equinox. Beginning in 2003 UT1 is no longer determined from GMST, but from the Earth Rotation Angle (ERA).
Defining the Vernal Equinox to an accuracy of one arcsecond is easy, but the rapid motions of the rotation pole of the earth and the non-planarity of its orbital motion hinder precise definition of the equator and ecliptic. The Vernal Equinox is not well defined at a precision of one milliarcsecond. This ambiguity became evident during the 1970s as VLBI began to contribute to earth orientation measurements, and it became problematic during the 1980s as VLBI and other new techniques replaced traditional optical astrometric measurements. By the year 2000 there were at least three different definitions for the location of the equinox being used for various purposes. VLBI precision is approaching microarcseconds, and it is hard to measure anything when the origin of measurement is not known as well as the measurement itself can be made.
In 1979 Guinot proposed the non-rotating origin (NRO) as a new and unambiguous scheme for defining earth rotation. Capitaine (1986) gave an early description of the differences between the traditional and NRO-based schemes, and the implications of the NRO were studied through the 1990s. In the literature through the year 2000 the ERA was known as the "stellar angle" (not to be confused with a "sidereal angle", which would refer to an equinox).
At the IAU 24th General Assembly in 2000 Resolution B1.8 directed that the equinox-based scheme for determining UT1 should be replaced by a NRO-based scheme beginning in 2003. This change was sufficiently fundamental that in 2002 the IERS held a workshop entirely devoted to exploring the implications of the change. The terminology and pedagogy for these new elements of fundamental astronomy are very new even after considerations by the IAU Working Group on Nomenclature for Fundamental Astronomy.
ERA is the first officially sanctioned quantity whose name explicitly disavows that earth rotation is time. This reflects a change in the purposes of IAU Commissions 19 and 31 over the course of history. Initially Comm. 31 (Time) was solely concerned with earth rotation, and Comm. 19 was solely concerned with polar motion. Now earth rotation is handled by Comm. 19, Comm. 31 primarily considers atomic time, and Comm. 4 continues its long history with dynamical time.
Fukushima (2001) pointed out that the non-rotating origin (NRO) introduced by the IAU 2000 resolutions for the measurement of ERA and UT1 actually rotates significantly over long time intervals. In particular Fukushima showed that precession of the equinoxes will cause the non-rotating origin (previously referred to as the CEO and now to be called the CIO) to circle entirely around the sky over an interval of approximately 360000 years. (Other papers on the implications of the IAU 2000 resolutions regarding time scales and reference frames are mentioned below at the end of the section on Fully Relativistic Time.)
The above forms of earth rotation time are consistent with the nomenclature that has developed throughout history. In conjunction with its orbital motion, earth rotation naturally produces the days, weeks, months, and years which are counted by calendars. Earth rotation time traditionally and evenly subdivides into 24 hours of 60 minutes of 60 seconds. In the sexagesimal notation used for these subdivisions the time tag 12:00:00 has long been understood to indicate that the sun is nearly overhead.
Earth rotation time is robust across an interruption in civilization because it is derived directly from observation. It is a straightforward matter to recover earth rotation time to a precision of microseconds even after an interruption of a century, and it will always be easy to recover to better than a second.
Although rocks, trees, fishes, bugs, little furry creatures, and most of humanity reckon time by the rotation of the earth, physicists reserve the term for a more uniform concept. The irregularities in the rotation of the earth mean that no form of earth rotation time measures time in the sense used by physicists. All forms of earth rotation time are now better described as "time-of-day". In the vocabularies of most of the languages of the world the long history of the words for "time" has been to mean "time-of-day". In order for "time" to have a simple relationship with a calendar, that "time" must be a form of "time-of-day". In the general population there is not yet a clear understanding of the distinction between the old traditional meaning of time and the new physical meaning.
Dynamical time means time determined by comparing observations of the motions of objects with physical models that describe that motion. The celestial bodies in the solar system move like hands of a clock. Using a theory of motion for those bodies it is possible to read the hands of the clock and determine what time they are telling. But that dynamical time has no relation to the rotation of the earth.
At the 5th IAU General Assembly in 1935 Commission 4 (Ephemerides) resolved that all national ephemerides should use the same value for the Gaussian gravitational constant (k) when calculating the motions of the planets. The president of Comm. 4 was E.W. Brown who was renowned for having developed a workable theory of the motion of the moon. Brown was tasked with ascertaining a value of k which was acceptable to all parties and reporting that value at the 1938 General Assembly. Brown died just prior to the 1938 General Assembly, but the agreeable value of k had been determined. At the IAU 6th General Assembly in 1938 resolution 2 of Comm. 4 adopted the value k = 0.017202098950000; this was the original value used by Gauss and later by Simon Newcomb for his tables. The value of k fixes the scale of the solar system in astronomical units, but in conjunction with Newcomb's tables it also fixes the duration of the Ephemeris day. Thus, the rate of what would eventually be called Ephemeris Time was irrevocably fixed in 1938.
Newcomb had suspected that the variations in the motion of the moon were directly correlated with variations in the motions of the inner planets, and Spencer Jones (1939) demonstrated the correlation unequivocally. This was direct evidence that the rotation of the earth varied on a time scale of decades. It meant that Universal Time was not uniform and therefore not suitable for use when calculating ephemerides.
Because Newcomb's tables had been based on astronomical observations from the 18th and 19th centuries, the length of the ephemeris day happened to match the length of the mean solar day from around the year 1820.
The difficulty with Ephemeris Time is that it can only be measured in retrospect after reducing observations of the orbital motions of bodies in the solar system. The motion of the sun with respect to background stars is extremely difficult to measure. The motion of the planets, except perhaps Mercury, is too slow to provide a precise measure. As a result, the primary means for determination of Ephemeris Time were observations of the motion of the moon as compared with the theory of Brown.
During the 1960s it became evident that there were deficiencies in Brown's analytical theory of the moon. Several corrections to the analytical theory were named by IAU Comm. 4 during the 1960s, and each correction resulted in a new form of Ephemeris Time.
During the 1960s it also became evident that the ever increasing numerical abilities of computers provided a workable alternative to analytical theories. Ephemerides of the motions of bodies in the solar system began to be calculated directly by numerical integration. The numerical integrations incorporated relativistic effects which Newcomb could not have known to include in his expressions. The passage of time on the surface of the earth was being measured much more practically by atomic chronometers. Atomic time agreed well with the time indicated by observations of the solar system compared to numerical integrations. Also, it was not clear whether ET was proper time or coordinate time. This inevitably led to the demise of Ephemeris Time. At the IAU 14th General Assembly in 1970 Guinot suggested that Ephemeris Time could be replaced altogether by using atomic time alone, but by 1973 it was understood that dynamical time and atomic time might differ despite the best of intentions.
The almanacs and ephemerides continued to be based on Ephemeris Time and on the analytic theories rooted in Newcomb's expressions through the end of year 1983.
The definition of Ephemeris Time set a precedent for the nomenclature of time which has created some confusion. The expression used for the mean longitude of the sun in ET was the same expression from Newcomb which was used for the calculation of UT. ET was a strictly uniform time scale not related to the rotation of the earth, and UT was a non-uniform time scale directly related to the rotation of the earth. But the fact that both used the same formulae meant that both were reported using the traditional nomenclature for time which had always been based on rotation of the earth. Values of UT and ET were given using multiples and divisions of days; i.e., calendar years, calendar months, hours, minutes, and seconds.
In particular, recommendation 5 from Comm. 4 called for the creation of new time scales which were intended to serve as relativistic replacements for Ephemeris Time. In order to give time for study and implementation of these changes it was resolved that they would not be used in the ephemerides until 1984. Coincidentally this was 100 years after the International Meridian Conference.
Despite resolutions that TCB should be used for all solar system calculations, the ephemerides and many other papers on solar system dynamics continued to use the name TDB for their independent variable.
The IAU 26th General Assembly in Prague approved several resolutions. Resolution 3 produced a new and rigorous definition for TDB.
By so doing TDB has effectively been converted from a Platonic or theoretical time scale into a practical one which is tied directly to JPL DE405.
Each different ephemeris has a different instance of Teph. Typically the values of Teph closely match the values of TT over the interval of the observations which were used to construct the particular ephemeris. Any instance of Teph is a barycentric coordinate time scale which has 0.01720209895 as the value of the Gaussian gravitational constant.
In early discussions of the IAU Working Group on Nomenclature for Fundamental Astronomy it has been suggested that the various instances of Teph might someday be given an official name something like Barycentric Ephemeris Time.
In 2005 the IAUwgNfA has decided that Teph is what TDB was always meant to be, and they are recommending that the two be considered synonymous.
Dynamical time is robust across an interruption in civilization because it is derived directly from observation. As long as the historical record of astronomical observations is preserved it is a straightforward matter to recover dynamical time to a precision of milliseconds even after an interruption of centuries.
Although there had been previous efforts to build atomic resonators, Atomic Time as we now know it came into being in 1955 when Essen and Parry at the UK NPL created the first workable chronometer using the cesium resonance. By 1958 the cesium resonance had been compared with astronomical observations of the ephemeris second, and the atomic second of 9192631770 Hz was born. Thus 86400 atomic seconds match the length of one ephemeris day, but the mean solar day has been longer than that for quite a while. So atomic time has no relation to the rotation of the earth, which in the long run means it has no relation to a calendar.
A form of time scale used in long-wave radio broadcast time signals of the US and Germany during the 1960s. The carrier frequencies and most second intervals were atomic seconds, but there were frequent steps of 100 ms to 200 ms inserted in order to match UT2. These steps were the precursor to leap seconds in the later form of UTC. In 1966 broadcasts of this time scale were classified as experimental by CCIR Recommendation 374-1.
In particular, SAT was used by the long-wave transmissions of WWVB in the US and the PTB in Germany. Some devices manufactured by Hewlett-Packard for automatically synchronizing with broadcast time used WWVB, and were therefore synchronized with SAT, not UTC. Complete records of all the steps applied to SAT are not readily available. As a result it can be difficult to determine a relationship with sub-second precision between historical timestamps expressed in SAT and in TAI.
TAI is the continuation of time scales which began with the first cesium atomic chronometer in 1955.
TAI has always been a statistical combination produced after the fact based on the available ensemble of atomic chronometers. The differences between TAI and various other atomic time scales are published monthly by the Time Department of the BIPM in Circular T.
TAI has not ticked uniformly throughout its history, but once a TAI has been assigned to an event its value is never revised. The variations in the rate of TAI can be traced via the annual publications of TT(BIPMxy).
As seen below in the 2007 entry for TAI, the BIPM disavows that TAI can be available for use in operational systems. This means that no device can provide TAI in real time, and any such claim cannot be based on any authority other than its manufacturer.
At this point the various laboratories keeping atomic time scales had only begun to recognize their significance and utility. These atomic time scales were unrelated to each other and only comparable after the fact through laborious computations. Subsequently the various time service bureaus which had then been operating atomic chronometers agreed to set their various atomic time scales to agree with their values for UT2 on this date.
This date was before high-precision international coordination of time had begun, and before the various time bureaus agreed on globally self-consistent values for the longitudes of their observatories. Guinot (2000) points out that because in 1958 the values of UT1 used by various observatories differed by several hundredths of a second of time, so also did their values of UT2. Because these atomic time scales were later incorporated into the atomic time scale of the BIH it can be said, in an approximate sense, that TAI was set equal to UT on this date.
At the 14th meeting of the Consultative Committee on Time and Frequency (CCTF) Dr. Dennis McCarthy of USNO presented report CCTF99-18 on The Future of Leap Seconds.
The resulting discussion produced a letter encouraging and recommending that applications which need a time scale without leap seconds use TAI.
An additional output was a circular letter from the Director, BIPM, emphasising the utility of TAI, as opposed to UTC, for systems requiring uniform time, i.e. free from the discontinuities arising from the application of leap seconds.
At a meeting of the Consultative Committee for Time and frequency (CCTF) held here in April 1999, attention was drawn to potential problems that might arise if a number of new, independent, uniform atomic time scales were developed for satellite navigation and electronic communication systems. There is apparently a perceived need for such uniform time scales to avoid problems in navigation systems resulting from discontinuities in UTC produced by the periodic introduction of leap seconds.
There was no consensus within the CCTF for any proposal to change the definition of UTC. Instead, I was asked as Director of the BIPM to draw your attention and that of agencies developing satellite navigation systems, to the option of using TAI which is, of course an international uniform time scale. I remind you of the ITU Recommendation ITU-R 485-2 (1974-1982-1990) in which it is recommended that "time data should be issued wherever possible either with reference to Coordinated Universal Time (UTC) or to International Atomic Time (TAI)". It is clear that if the leap seconds of UTC cause problems in any particular application, the preferred alternative is TAI.
The CCTF recommends, therefore, that in conformity with this ITU Recommendation developers of future satellite navigation systems and electronic communication systems should link their time scales to TAI as the only alternative to UTC and that, insofar as it is feasible, existing systems take steps to align their time scales with TAI. This is in conformity with the CCDS Recommendation S4 (1996) on the "coordination of satellite systems providing timing", in which it was recommended that "the reference times (modulo 1 second) of satellite navigation systems with global coverage by synchronized as closely as possible to UTC". To facilitate the direct use of TAI for satellite navigation systems, the time community is willing to take any steps that are necessary to make TAI easily accessible to users. UTC remains the basis for worldwide timekeeping, but TAI is recommended for those applications requiring uniform time. I urge you to take the necessary steps to inform your constituents of the characteristics of both UTC and TAI so that appropriate use may be made if these international scales. I enclose a few documents that may be of help in this respect.
After more than a year of planning, meeting, and writing the Precision Time Protocol (PTP, officially IEEE 1588-2002) was approved as a standard.
PTP time is defined as TAI - 10 s for dates subsequent to 1972-01-01T00:00:00. This means that a POSIX system can set its system clock value of time_t equal to PTP time and use the "right" zoneinfo files to produce human-readable timestamps in UTC.
CAVEAT: A close reading of the 2002 PTP spec indicates that the contents Appendix B may be inconsistent with the rest of the specification. The above paragraph is based on Appendix B and looks like it may be wrong. It may be that PTP is intended to be elapsed seconds of TAI starting on 1970-01-01T00:00:00 TAI. This text will be updated when clarity is certain.
The BIPM contributed document 7A/51-E to the ITU-R WP7A meeting regarding Question ITU-R 236/7. (The letter was reproduced as Document CCTF/09-27 in the Report of the 18th meeting of the CCTF.) The letter says that the CCTF "stresses that TAI is the uniform time scale underlying UTC, and that it should not be considered as an alternative time reference." This appears to indicate that the CCTF and BIPM are not comfortable with the notion that operational systems might be employing TAI as their time scale.
The final sentence of the BIPM document reads:
In the case of a redefinition of UTC without leap seconds, the CCTF would consider discussing the possibility of suppressing TAI, as it would remain parallel to the continuous UTC.This makes it unclear whether TAI will continue to exist.
After another year of effort a revised standard for PTP (officially IEEE 1588-2008) was approved.
PTP time remains defined as TAI - 10 s for dates subsequent to 1972-01-01T00:00:00, but the standard refers to "1 January 1970 TAI" which is a date before TAI had been defined or approved.
CAVEAT: The above paragraph may be wrong. See the warning in the 2002 PTP section above.
The GPS satellites adopted a time scale which was synchronized with UTC on 1980-01-06, and it has been steered in close synchrony with TAI since then. Therefore the difference TAI - GPS has been 19 s to within a microsecond. For practical purposes this means that GPS time strongly resembles the variants of TA(k).
See the current revision of IS-GPS-200 for full technical details. Note that getting precise time from GPS requires that the receiver must apply the polynomial corrections for the SV clocks, the GPS time scale itself, and the ionosphere as detailed in sections 126.96.36.199 and 188.8.131.52 of IS-GPS-200.
The satellites for the initial and experimental Chinese GNSS were known as BeiDou-1. They used a time scale which was synchronized with UTC during the year 2000, so the difference between it and TAI was 32 seconds.
The new and operational Chinese GNSS is known as BeiDou-2 or Compass. Its time scale was synchronized with UTC on 2006-01-01, and it has been steered in close synchrony with TAI since then. Therefore the difference TAI - BST has been 33 s to within a few nanoseconds. In effect, this means that BST is another variant of TA(k).
India is launching satellites known as the IRNSS (Indian Regional Navigation Satellite System). Their time scale is IRNSS Network Time (IRNWT). IRNWT was synchronized with UTC during the year 2006, so the difference between it and TAI is 32 seconds.
The name for the radio broadcast time scale which was suggested by Sadler (1978) in his monograph on the sordid history of people and agencies changing the name or characteristics of time scales which were already in use.
The name of the time scale in one of the presentations from E. F. Arias at the ITU-R SRG 7A colloquium in Torino on 2003-05-30 as a likely candidate for radio broadcast time signals at some point in the future. TI would be a purely atomic time scale offset from TAI by a fixed integer number of seconds. In order to avoid discontinuities for systems using radio broadcast time signals the offset would be equal to the offset of UTC at the instant of switching from UTC to TI. The nominal year at which the switch would occur is 2022.
Atomic time is not robust across an interruption in civilization because it depends on the continuous operation of complex equipment. If civilization were disrupted the existing atomic time scales would be lost. The precision to which a new civilization could match new atomic time scales to the current ones would depend upon astronomical observations. If dynamical time were the only available scheme, then the match would not likely be as good as a millisecond.
TT can be realized by various methods, but the most practically available form is derived directly from TAI. Irwin and Fukushima (1999) have provided the best relationship between TT and TCB. Soffel (2002) and Petit (2002) provided further details about the IAU 2000 resolutions which clarified TT.
This is TT calculated simply by presuming that TAI has been free of defects since it read 1977-01-01T00:00:00. The value of 32.184 s is the best available estimate of the difference between TDT and TAI on that date.
Terrestrial Time is not currently robust across an interruption in civilization because its principal realization depends on atomic time. If civilization were disrupted then TT could be reconstructed to a millisecond precision using dynamical time. TT(pulsars) might provide a way to achieve microsecond precision, but this possibility will not be clear until after that new time scale has been in operation for many years.
The forms of dynamical, atomic, and coordinate time above are not based on earth rotation. They have no connection with days in the traditional sense; thus they have no simple relationship with the concept of a calendar. It is important to remember that the 24-hour cycle of tags like 12:00:00 really only makes sense for earth rotation time.
To their credit, the BIPM do report TAI and TT using modified Julian date (MJD) notation. Nevertheless, it remains commonplace to use the traditional calendrical and sexagesimal notations when counting the seconds which are defined by these time scales. (Indeed, for the sake of human cognition this notational convenience is very nearly a requirement.) Unfortunately, that usage leads to even greater confusion about the meanings of time scales.
Denoting TAI with a tag in the form 1958-01-01T00:00:00, and denoting TT, TCG, or TCB with a tag in the form 1977-01-01T00:00:00 should be considered as a convenience only. They are merely counts of elapsed time where one anonymous and indistinguishable second follows another. It is very appropriate to count these forms of time using decimal notation from the epochs represented by those tags. But there is no observable event that happens cyclically at 12:00:00 for dynamical, atomic, or coordinate time -- the entire notion of such a cyclical process is contrary to their uniformly-incrementing conceptual definitions.
This category covers two separate concepts: "radio broadcast time signals" and "UTC". As Sadler (1978) pointed out, in recent usage the two concepts have tended not to be distinguished carefully. The history given here attempts to be careful in discerning which has been which.
UTC has always been two things
Starting in 1972 UTC has counted two things
Nothing resembling the name UTC was used prior to 1960, so any claim that UTC can be used before then is inappropriate. The name UTC did not appear in any official context until 1974, so any claim that UTC was used prior to 1974 is almost certainly a reinterpretation of history which does not correspond to anything in contemporary documents.
Radio broadcast time signals and UTC do not fit into any one of the above categories, but they are dependent on them. Since the name "UTC" came into use it has always been both Universal Time and Atomic Time as provided by radio broadcast time signals. As such, UTC has been a practical time scale. The nature of the goals that radio broadcast time signals have been intended to meet and the procedures for meeting those goals have changed over the years.
Prior to the advent of atomic chronometers it had not been possible to keep the broadcast time scales of widespread systems synchronized to within a millisecond. The original goal of what was informally called UTC was for radio broadcasts around the world to be synchronized with each other while providing a time which matched the expected value of UT2 as closely as could be predicted in advance. The length of the second in radio broadcast time signals was adjusted ("elastic seconds" or "rubber seconds") at the beginning of each year, and any accumulated error caused by mis-prediction of the rate of UT2 was corrected by applying occasional steps of 50 to 100 milliseconds to the value of the broadcast time.
One last note before going into the detailed list of the evolution of UTC. The last time that everyone involved in the definition of UTC agreed on the concepts was 1964, the same year that the term UTC first appeared in print. Already by that date the different international organizations involved in regulating radio broadcast time signals were holding as many as three different meetings during a calendar year to discuss what to broadcast. Explaining the concepts of the time scales in the broadcasts, and explaining how it was safe to use those time scales, were less important than who had control over the broadcasts. The result has been that, since the term UTC first appeared in print, no two international specifications have agreed on how to handle the sequence of timestamps in radio broadcast time signals. A flame war erupts almost every time any organization starts a discussion about the technical details of UTC.
In the UK the agencies of the Royal Greenwich Observatory, the National Physical Laboratory, and the General Post Office, and in the US the agencies of the US Naval Observatory, the Naval Research Laboratory, and the National Bureau of Standards agreed "to co-ordinate the time and frequency transmissions" beginning in 1960 where "the co-ordination plan operates within the framework of the IAU and the CCIR."
The word "co-ordinate" was the same word used to describe the existing collaboration between the US and UK to share the work of producing their national ephemerides. Beginning with the 1960 edition this effort had already resulted in the two countries issuing publications which contained identical sections of astronomical and navigational tables.
The report from Commission 31 (Time) to the 11th General Assembly of the IAU mentioned the issues of atomic time vs. universal time, interval vs. epoch, astronomy vs. navigation vs. physics, and the roles of the IAU, CCIR, and URSI. Starting from this point the tensions become visible over the nature of what radio broadcast time signals should be trying to provide.
During the 11th IAU General Assembly in Berkeley the meeting
of Commission 31 (Time) referred to the US/UK broadcast time
The process of broadcasting synchronized signals was called
co-ordination, the resulting signals were
called co-ordinated, and the stated goal was
to achieve agreement to within 1 ms.
Commission 31 resolved that the BIH should determine the annual changes to the frequency offset from cesium to be used to co-ordinate the radio broadcasts. At this meeting Commission 31 decided to address the longstanding issue of the conventional longitudes of the observatories around the world. The IAU directed all of the time service bureaus to adopt new, globally self-consistent longitudes for their observatories starting in 1962.
The responsibility for coordinating the frequency offsets and steps was transferred to the BIH. A table of the frequency offsets and steps is available from the Paris Bureau of the IERS, which is the organizational descendant of the earth rotation portions of the BIH.
During this same period some radio broadcasts were providing SAT (see above). Also, at least until 1967 the Soviet Union and China were broadcasting another form of time coordinated in the USSR independently of the BIH.
Before assigning absolute meaning, any time-stamp from this era intended to be used with sub-second precision should be subjected to historical scrutiny about the mechanisms of its provenance. This includes the origin of the POSIX epoch at 1970-01-01T00:00:00.
The 10th Plenary Assembly of the CCIR approved Recommendation 374 to supersede Recommendation 319. Recommendation 374 does not contain any form of the words "coordination", "coordinated", nor the letters "UTC". The text states that the broadcasts of time signals should "be offset to keep the time pulses in close agreement with UT2" and "maintained within approximately 100 ms of universal time UT2" with steps of exactly 50 ms.
Recommendation 374 contained the details of what is now informally referred to as the original form of UTC even though the CCIR did not use that name. In particular, 374 described the Best Current Practice of broadcast time signals using a frequency offset from the cesium resonance and monthly small steps, both coordinated internationally.
The US NBS continued to announce their broadcast time signals using the standard time at the radio transmitter, as offset from the mean solar time of Greenwich.
Editorship of the Bulletin Horaire from the BIH passed from Anna Stoyko into the hands of Bernard Guinot. The first issue produced by Guinot was Series J, Number 1 giving the values of time for 1964 January and February. With the issues from this year the bulletin started to use the name UTC for its time scale constructed from the monitoring of worldwide radio broadcasts. Note, however, that at that time the BIH was more than a year behind in the processing of the time values. Despite being the tabulations for time during 1964-01/02, the content of Bulletin Horaire, Series J, No. 1 contains information about events as late as 1965-05-01, and it was not received by subscribers until late in 1965 June.
On page 2 of that issue Guinot wrote a definition in boldface print
L'échelle intermédiaire TUC (UTC en anglais) sera appelée Temps Coordonné.This is the earliest known use of the term UTC in any publication.
The bulletin did not make a clear choice of language for
There was not yet a clear precedent for using CUT
"Coordinated Universal Time" (in English) nor
TUC "Temps Universel Coordonné" (in French),
nor UTC (following the pattern that Markowitz of the IAU
had worked out with the time service bureaus for UT0,
UT1, and UT2).
The canonical ordering UTC is consistent with neither French nor English, but it is consistent with the UT0/UT1/UT2 nomenclature scheme. Despite the canonical ordering, it remains common even today to see usage where the letters are permuted (and worse misnomers such as Consolidated Universal Time, Universal Correlated Time, Universal Coordinating Time, Universal Time Calibrated, Universal Time Code, etc.).
The report of Commission 31 to the 12th General Assembly of the IAU noted that most time services had switched to using quartz crystal clocks instead of pendulum clocks.
At the 12th General Assembly of the IAU Commission 31 again resolved that radio broadcasts should use the frequency offset from cesium chosen by the BIH to match UT2. They noted that high frequency radio broadcasts of time signals permitted worldwide synchronization to within about 2 ms, but that transporting cesium chronometers and using satellite telecommunications could attain synchronization to about 1 microsecond. They also produced this regrettably prophetic quote
The distinction between time epoch and time interval is not clear to everyone.along with an explanatory note written by none other than W. Markowitz, H.M. Smith, and L. Essen (see page 16). That note says
The epoch of U.T. is determined by the angular position of the Earth around its axis; it is required for various scientific and technical purposes and for civilian use, sometimes without delay.
For the next few years the term UTC was plainly informal and it appeared in various publications with various different abbreviations. Some authors wrote in English using U.T.C. alongside U.T.0, U.T.1, and U.T.2; others wrote in French with T.U.C, etc. Publications from the USNO used UT2C.
Despite being the first person to use the term in any publication, on page 62 of The Measurement of Time: Time, Frequency and the Atomic Clock Guinot declines to take credit for the name of the time scale
which had been spontaneously christened Coordinated Universal Time (UTC)
In context it is fair to say that the name "Coordinated
Universal Time" and its various abbreviations had
roughly this meaning:
The kind of Universal Time which we are able to broadcast in radio signals that are internationally coordinated to be synchronized to 1 ms or better.
to maintain the epoch of the time-signals within about 100 ms of Universal Time UT2The significant change in 374-1 was to describe (and denote as experimental) the broadcasts of Stepped Atomic Time (with no frequency offset from cesium and many more steps) as a new Best Current Practice.
IEEE proceedings V55 #6 published "Some Characteristics of Commonly Used Time Scales" by George E. Hudson. One contemporary sentiment of that era is plainly visible where he wrote
the deplorable problem raised by the existence of a great variety of auxiliary time scales, some denoted by a random nomenclature.and he concluded
the proliferation of artificial derived scales is to be deplored.
He listed several atomic time (AT) scales including A at US NBS, A.1 at USNO, TA1 at Swiss LSRH, and one at UK NPL. He listed five "slightly different universal time scales" including "UT0, UT1, UT2, universal atomic (UA), and stepped atomic (SA)." He listed the two different schemes approved for radio broadcasts by CCIR Rec. 374-1 using the terms "UTC" (which, as noted above, is not used in 374-1) and "SAT" (for Stepped Atomic Time, and he also notes that WWVB called this NBS(SA) but had previously called it SAB). He further notes that WWV and WWVB actually use NBS(UA) (see 1968-12 below for when NBS switched to following the BIH version of UTC).
The IAU noted the nomenclature "UTC" at its 13th General Assembly in Prague when referring to CCIR Recommendation 374-1 even though the term "UTC" does not appear in the CCIR recommendation. The IAU discussions show that while the radio regulators thought of the broadcasts as being UT2, the astronomers recognized that could not be the case, and they used the internal term of art "UTC" for which there existed no formal recommendation nor resolution.
In the transcripts of the joint meeting of Comm. 31 (Time) and Comm. 4 (Ephemerides) are discussions and resolutions on subjects including
In particular, the IAU approved Resolution 6 from Comms. 31 and 4 ( English, French) in which the IAU clearly indicated that "Universal Time" is a proper name for time based on rotation of the earth. In combination with Resolution 5 and the discussions which produced both resolutions it is very clear that the members of of these IAU Commissions and the General Assembly considered purely atomic time scales to be distinctly different than Universal Time.
US NBS Special Publication 236 first used the name Coordinated Universal Time with the abbreviation UTC as the time scale coordinated by the BIH. It used the terms UT2 and GMT interchangeably, and the terms UT, UTC, and GMT were equated to each other.
By this point it is evident that everyone familiar with the time broadcasts knew them as UTC, but the relationships between UTC, UT2, UT and GMT were expressed in different ways by different people. This is not surprising given that the CCIR still specified the use of UT2 and had never used the term UTC.
In anticipation of a coordinated coordinate rate for USNO and NBS, on August 24, 1967, the Coordinated Universal Time clock of the Bureau, UTC(NBS), and all UTC transmissions of NBS were advanced by 200 microseconds. This left NBS about 35 microseconds early relative to USNO.(As seen above, they also provided a unique use of forms of the word "coordinate".) On 1968-10-01 the two scales agreed, and both agencies adjusted the frequencies of their time scales. After that date they steered the frequencies of their time scales to keep the time difference less than 3 microseconds.
CCIR Interim Working Party 7/1 had been created to resolve issues with broadcast time signals. Despite resolutions from scientific unions calling for an international and interdisciplinary committee to study options for broadcast time signals, in 1969 CCIR Interim Working Party 7/1 unilaterally deemed that radio time signals should broadcast atomic seconds with occasional leaps of full seconds.
IWP 7/1 produced CCIR Plenary Assembly Document VII/1008 which was the draft version that became CCIR Rec. 460. Section 2 of the draft originally included the phrase "or integral multiples thereof". This is the closest that the world ever got to having a "double leap second".
The report of Commission 31 (Time) to the 14th General Assembly of the IAU covered events of the three years since the previous IAU GA. It appears to have been written in late 1969 or in 1970 before the CCIR meeting. The title of the section on radio broadcast time signals was "Coordinated Time", and it begins with the words "The universal coordinated time scale (U.T.C.)". These are indications that the members of IAU Comm. 31 who wrote the report did not then share a unanimous nomenclature for the time scale in radio broadcasts. It also looks as though some members did not agree that the atomically-regulated broadcasts should be associated with the name Universal Time.
The report of Commission 31 included an unofficial synopsis which is the only publicly-available record of the CCIR efforts to redefine broadcast time signals. The report indicated that changing time signals to have leap seconds might cause difficulties for aircraft collision avoidance systems.
The report also discussed the alternatives for changing radio broadcast time signals in a way that very much prefigured the situation that has been happening in the ITU-R starting in 2000 and continuing through 2015. Alternative (b) in the report matches the draft revision of ITU-R TF.460 submitted to the Radiocommunications Assembly in 2012 by calling it "complete disuse of U.T.C. system replacing it with a coordinated uniform time scale without offsets and steps and therefore not approaching U.T." Thus the IAU is on record opining that UTC without leap seconds is not a form of Universal Time.
Disregarding the requests from various scientific unions the delegates at the 13th plenary session of 12th Plenary Assembly of the CCIR unilaterally decreed that on 1972-01-01 (in less than two years) broadcasts of time signals would begin using the new scheme of leap seconds. This was codified in the text of the new CCIR Recommendation 460. (Most CCIR, and ITU-R, texts were not freely available, but this one was republished by the US National Bureau of Standards on page 31 of Monograph 140, Time and frequency: theory and fundamentals (Byron E. Blair, 1974).)
It was during the Plenary Assembly that the words "or integral multiples thereof" were removed in order to limit the havoc that the change might cause to navigators. The maximum difference DUT1 = (UT1 - UTC) was specified as 0.7 s, which prompted the creation of signals encoding DUT1 which can only indicate values up to 0.7 s.
It is relevant to note that the term "UTC" does not occur in the text of CCIR Recommendation 460. The focus of the discussions was on the technical characteristics of the radio broadcast time signals. The proceedings give no indication that a name for the time scale used in the radio broadcasts was important.
Rec. 460 prescribed that the broadcasts
"maintain approximate agreement with Universal
Time", but it gave no instructions on how to
implement such a scheme.
Rec. 460 prescribed that those instructions were to be specified later.
Rec. 460 also prescribed that the director of the CCIR should transmit its text to the scientific unions, including the IAU.
Rec. 460 was a significant change of practice because it prescribed a new and untested form of broadcast, as opposed to previous CCIR documents which had described existing signals by broadcasters who had a long history of interacting with their users. Whereas the national laboratories had an ongoing means of interacting with their users, the CCIR was not constituted for giving prompt answers to questions about the usage of the signals.
Under Rec. 460 the duration of one second became unrelated to the duration of one day. This satisfied the two demands outlined in 1964 where the physicists required uniform seconds as a measure of elapsed time and the astronomers required that the value in the broadcasts would conform to a measure of the calendar days of Universal Time. Rec. 460 was thus an implicit abrogation of a fact that everyone knew to be true; a day would no longer be the same as 86400 seconds. This abrogation was not clearly communicated.
At the 14th General Assembly of the IAU in Manchester the action of the CCIR to put leap seconds into radio broadcast time signals was discussed at the meetings of Commission 4 (Ephemerides), Commission 31 (Time), and at a joint meeting of 4 and 31. The IAU members used the term "UTC" when discussing the time scale used for radio broadcasts.
At the meeting of Commission 31 it was pointed out that the CCIR had failed to send a letter informing the IAU of the change, so the IAU was unable to respond officially at its 14th General Assembly in 1970. That made it impossible for the IAU to produce an official response before the IAU General Assembly in 1973, which was after the change would be implemented.
The resolutions from the 14th General Assembly of the IAU include
CCIR Recommendation 460 went into effect. At the end of 1971 a special offset of -0.1077580 seconds was applied to step UTC so that ( TAI - UTC ) was exactly 10 seconds. Henceforth the length of the UTC second has matched the length of the TAI second, and the value of UTC has been adjusted via one second leaps to keep it within a second of UT1.
From this day onward the duration of one second has been unrelated to the duration of one day. In radio broadcast time signals the duration of a second has been determined by measuring cesium atoms, and the duration of a calendar day has been determined by measuring the rotation of the earth.
Without explicitly saying so the effect of CCIR Rec. 460 was to abrogate the connection between the concepts of time and date which had been understood by everyone since antiquity. The clock was disconnected from the calendar. Starting in 1972 the leap second is the only way that the concepts of date and time have been kept in sync.
J. Terrien (the director of the BIPM who had been at the IAU meetings on radio broadcast time signals since 1964) noted that "the BIPM does no experimental work on the measurement of time and frequency" and apologized that "regrettable misunderstandings, especially between astronomers and physicists, have crept into discussions on time and frequency."
The CCDS declared that UTC can provide a firm basis for an internationally acceptable time system.
The 13th Plenary Assembly of the CCIR approved Recommendation 460-1, the first revision of the document specifying that radio broadcasts of time signals have leap seconds. It incorporated advice from the IAU 15th General Assembly in 1973 and raised the maximum allowable difference of (UT1 - UTC) from 0.7 s to 0.9 s. This was the first instance when the CCIR documents described the implementation rules of leap seconds as used in the time scale for radio broadcasts.
This was also the first CCIR document to be approved by the plenary assembly and the first instance of any CCIR recommendation about radio broadcast time signals which used (in French) the word "coordonné" and the term "TUC" or (in English) the word "coordinated" and the term "UTC". (It was another 4 years until the CCIR issued Rec. 536 to specify a single acronym for UTC.)
The report of Study Group 7 to the 14th Plenary Assembly of the CCIR contained a number of statements that are even more interesting in retrospect. Its section on UTC started with the words "UTC was introduced by Recommendation 460 in 1972". Those words give no indication that the CCIR recognized any form of UTC previous to 1972, and they omit any mention that the CCIR did not apply any name to the radio broadcast time scale until 1974. The SG7 report asserted that the mean solar time of the Greenwich meridian is Universal Time (UT). The SG7 report acknowledged the 1975 recommendation from the CGPM as the basis for civil time (but did not explicitly reiterate that the CGPM had recognized UTC as a form of mean solar time). The SG7 report also pointed out that some countries had already redefined their legal time to be based on UTC and said "the UTC time scale is the general reference for civil time". (This is the first instance in which the notion of a reference time scale appears. In the aftermath of the 2012 RA and WRC the phrase "reference time scale" had become jargon.) So SG7 recognized in report to the CCIR plenary that the use of UTC had already accumulated political implications without clearly realizing that those would eventually compete with technical considerations. Finally it indicated that the implementation of UTC was complete and there was no further need for Interim Working Party 7/1. Thus, just as the world of computing systems was beginning to recognize a need for what would become POSIX, the CCIR decided to abolish the working party which was best suited to answering questions about operational details.
The 14th Plenary Assembly of the CCIR approved working document 7/1007 as the second revision of the document specifying that radio broadcasts of time signals have leap seconds in CCIR Recommendation 460-2. (Once again, the revision to Rec. 460 was mostly to incorporate the words and actions of other agencies, in this case CCDS Recommendation S1 (1974)). The Plenary Assembly requested the views of the Observer for the International Astronomical Union regarding the sentence "GMT may be regarded as the general equivalent of UT." (That is in Annex I section A about the UT1 variant of the time scale Universal Time.) The IAU observer "said that the use of GMT was neither authorized nor approved, but merely noted, by his organization." The Chairman of Study Group 7 (G. Becker of FRG) proposed putting parentheses around the GMT sentence, and the assembly approved Rec 460-2 with that amendment.
The Plenary Assembly also approved working document 7/1008 as CCIR Recommendation 535. The aim was to recommend the usage of the term UTC to replace GMT in all international telecommunications activities and documents. This recommendation was explicitly based on consideration of the 1975 action of the 15th CGPM and the 1973 action of the 15th General Assembly of the IAU, both of which had acknowledged UTC as a form of mean solar time. In particular this recommended that UTC replace GMT in the Radio Regulations of the CCIR itself, and the resulting wording made it clear that this usage applied not merely to time, but to calendar dates as well. In combination with the words in the SG7 report that distinguish UTC from the mean solar time of them Greenwich meridian this resulted in a longstanding confusion because Radio Regulation 2.5 defines the date using the concepts of longitude and earth rotation where UTC itself is only connected to those concepts because there are leap seconds.
The Plenary Assembly also approved working document 7/1009 as CCIR Recommendation 536. The aim was to introduce "language independent time-scale notations". Explicitly based upon consideration that in 1971-10 the 14th CGPM had defined International Atomic Time "using the designation TAI" and that in 1975-05 the 15th CGPM had recommended the use of Coordinated Universal Time "using the designation UTC" this recommended the use of those abbreviations. Thus the CCIR did not take credit for creating the abbreviations TAI and UTC, but rather accorded that to the CGPM. Finally, Recommendation 536 explicitly recognized UTC as a form of Universal Time, and the CCIR recognized UT as mean solar time.
The introduction by the chair of SG7 starts off with a paragraph that is, in retrospect, astonishing for the way it crows about how UTC with leap seconds has been acepted by many agencies as solving all problems:
The year 1982 will see the tenth anniversary of the system of Coordinated Universal Time (UTC) which was introduced in essentially its present form on 1 January 1972 by Recommendation 460. In the past ten years the UTC system has gained widespread acceptance as the reference time scale for timekeeping purposes throughout the world, not only for technical and scientific applications but also for time in everyday affairs. It is particularly gratifying to the Study Group that the World Administrative Radio Conference in 1979 (WARC-79) accepted the terms of Recommendation 535 and adopted UTC as the reference time scale for all radiocommunication activities. Likewise, UTC was also accepted in 1980 by the VIIth Plenary Assembly of the CCITT as the time scale for all other telecommunications activities. Together, these two actions constitute a powerful endorsement of the careful work and the intensive discussions extending over more than one plenary period which led to the present UTC system, as specified in Recommendation 460.This introduction appears to be an early instance of using the phrase "reference time scale" when referring to UTC.
In contrast to the introduction in 1978, this introduction refers to the existence of "the UTC system [...] prior to 1972" even though no document approved by the Plenary Assembly had used that name until 1974. So again we see a CCIR document asserting facts that are inconsistent with the recorded history of the CCIR.
The 15th Plenary Assembly of the CCIR approved the third revision of the document specifying that radio broadcasts of time signals have leap seconds as CCIR Recommendation 460-3.
In retrospect, the absence of publications containing clear and consistent specifications makes it no surprise that external agencies (e.g., IEEE, ANSI, ISO, POSIX) later produced standards for programming interfaces describing something that they called UTC as a time scale with properties that differed from the UTC in radio broadcasts. It is amazing to see the extent to which SG7 failed to foresee the the problems with leap seconds that started to appear within the next decade.
This is the publication date of IEEE Standard Portable Operating System Interface for Computer Environments (IEEE Std 1003.1-1988) also known as POSIX.
Chapter 2 (Definitions and General Requirements) section 2.3 (General Terms) defines
Appendix B (Rationale and Notes) section B.2.3 (General Terms) explains
For POSIX the choice of goals was that simplicity and calendar days were more important than precise time. The result of this standard has been that a machine cannot handle precise time with leap seconds and also be a Conforming Implementation of POSIX.
The C programming language was standardized as ANSI X3.159-1989 and then as ISO/IEC 9899:1990. Some time between the drafts in 1988 and the final standard the section on time.h with the definition of struct tm changed the allowed values of element tm_sec to the range [0-61] with the explanation that this was to handle a "double leap second". There has never been such a thing as a "double leap second". In 2001 Mark Brader claimed that this was his fault.
This erroneous notion of "double leap second" highlights that ITU-R recommendations are expected to be implemented by everyone, yet they are not freely distributable so their content is not widely known. As seen in this history of UTC, ITU-R recommendations may be approved in the absence of any consensus, rationale, implementation details, or interoperability tests.
At the 16th meeting of the CCTF Ron Beard reported on the 2003 UTC colloquium at IEN in Torino.
The conclusion of the SRG was that the creation of a new time scale, to be known as "International Time", was not recommendedFrom this it was clear that the SRG of ITU-R WP7A had already decided to disregard the advice they had received from the international experts they had gathered to discuss the future of UTC.
Other members of the CCTF discussed in response to Beard.
Dr Boulanger expressed the opinion that it was wrong to change the definition of the (UTC) timescale without changing its name as it could lead to a diminution of trust. Dr McCarthy pointed out that there was a precedent for this. The frequency steps which were once part of UTC has been dispensed with. Dr Arias said that the word "universal" would no longer be appropriate for UTC. Dr McCarthy recalled that the word "coordinated" was chosen to reflect the coordinated change in the different timescales, then in use in the UK and the USA, to the (new) UTC. Dr Arias clarified her earlier point; she believed that the word "universal" was appropriate only for a timescale that was "linked" to the rotation of the Earth.
As of 2004-09 the index of contributions to WP7A on the ITU website contained a new document from the United States whose title said it was a proposed revision of ITU-R TF.460-6 (the document requiring leap seconds in radio broadcasts of time signals). Although it was not possible to see the content of that document, it seems likely that it was a revision of document nc1893wp7a on the FCC website from the United States Working Party 7A that holds the federal charter to interact with the ITU-R.
Document nc1893wpa from USWP7A proposed that UTC should switch to having leap hours on 2007-12-21. It is relevant to note that on this date the legal time of the US was mean solar time, so a document which did not include some kind of leap would have been contrary to the will of congress.
At the meeting of ITU-R WP7A on 2004-09-28/10-01 the other delegates scoffed at the notion of leap hours.
On 2005-09-19 the USWP 7A released the 2005 version of its
Proposed Revised Recommendation ITU-R TF.460-6.
The document was not approved by the US State Department,
but its text would have ceased adding leap seconds as of
The fact that this date would have been the same as the end
of the Mayan calendar long count was a source of
At the meeting of ITU-R WP7a on 2005-11-08/11 there was no consensus for change.
UTC has always been a hybrid of the needs for atomic frequency and time interval and the needs for the counting of calendar days. This is because the intended purpose for UTC is to satisfy two separate goals: seconds of uniform length, and days that match the calendar progression of earth rotation. As a result
On the other hand, the full second leaps which have been introduced in UTC since 1972 (and the millisecond leaps in radio broadcast time signals before 1972) are nothing new for civil time. Prior to atomic chronometers there were no practically available clocks which were as stable as earth rotation. Even the best temperature-controlled quartz crystal clocks needed to be reset -- i.e., leaped -- regularly to agree with the astronomical observations measuring calendar days. Throughout the long history of developing clocks for timekeeping it had always been that way. Prior to 1960 the leaps had been applied individually, by each local timekeeping agency, for its own set of time signals, as deemed necessary by the astronomers running the transit instrument at the local observatory. The only real change that happened in 1972 was to agree that the leaps should be full atomic seconds, coordinated internationally, and that there should be a nomenclature scheme for referring to those leap seconds.
What happened in 1972 was that the clock was disconnected from the sun. Each individual second became disconnected from earth rotation, but the leap seconds allow the individual days of the calendar to remain connected with the sun.
The possibility of changing UTC to omit leap seconds means ignoring astronomy and disconnecting the calendar from the rotation of the earth. It would make the progression of civil time predictable, but time tags such as 12:00:00 would be completely unrelated to having the sun overhead at noon. Over the passage of centuries the difference between 12:00:00 and noon would become increasingly obvious.
Reversing what Julius Caesar did 2000 years ago, it would remove the calendar from the hands of astronomers and put it back into the hands of politicians. It would change the meaning of all contracts based on days. Bank interest would effectively be accumulated per second rather than per day. Midnight debut showings of blockbuster movies would be based on cesium atoms and political decrees with no reference to sunrise or sunset.
Note also that if UTC is redefined without being renamed, the result will be the same sort of archival chaos as was created by the British Admiralty when it redefined GMT in 1925. Because the definition of UTC has always been a hybrid, anyone finding the term UTC in a document will not be sure of its intended meaning even if the document originated while UTC had leap seconds. After such a change it will be unclear whether authors intended the use of "old UTC" or "new UTC", or whether the difference is important at all. This question will have to be asked:
Should an instance of the term "UTC" in a document be interpreted simply as specification of a conventionally-available atomically-regulated civil time, or was it originally intended to specify a form of mean solar time and related to calendar days?Persons reviewing computer source code and documentation will have to ascertain whether the original intent of authors and system operations involving any quantity called UTC require a time scale with properties that match mean solar time or not. Legal systems which have relied on the fact that UTC is a form of mean solar time may require revision. Legal and operational systems which seem to work at first may fail later as the difference between UTC and UT1 grows. Changing the characteristics of UTC without changing the name may seem easy at first, but it has widespread consequences of potentially considerable cost which must eventually be paid.
UTC without leap seconds would be inconsistent with the nomenclature established by resolution V of the 1884 International Meridian Conference and adopted by the IAU, and also inconsistent with the 1975 CGPM resolution on UTC.
The robustness of UTC across an interruption in civilization is schizophrenic specifically because UTC is a combination of two incommensurate concepts. Inasmuch as UTC is a vehicle for communicating mean solar time, UTC is robust because by its own definition it admits that the conventional value of mean solar time need not be accurate to much better than one second. Inasmuch as UTC is a vehicle for atomic time, UTC cannot be robust because atomic time is not.
In 2003 Landon Curt Noll submitted some historical views on how POSIX got the way it is to the LEAPSECS mailing list.
The standard defines seconds since the epoch ( 2004, Base Defs sec 4.14, 2008, Base Defs sec 4.15) as ignoring the existence of leap seconds. As a result, the rationale (2004, Rationale A.4.14, 2008, Rationale A.4.15) admits that not all POSIX seconds have the same length, and it is also fuzzy about the definition of the epoch ( 2004, Rationale sec A.3, 2008, Rationale sec A.3).
To put it simply, the POSIX standard is self-inconsistent. A rigorously defined time scale should either be a count of mean solar seconds (which are subdivisions of calendar days), or atomic seconds (which are unrelated to calendar days). POSIX tries to be both and as a result it cannot address leap seconds meaningfully.
POSIX time shares most of the caveats of Julian Date (above). There are two main distinctions between POSIX time and the forms of Julian Date. First, POSIX time is an integer count of seconds instead of an arbitrary-precision real number of days. Second, the quality of chronometers and the representations of integers in computing hardware place severe restrictions on both the precision of instants of time and the useful span of dates which can be represented by POSIX time.
See the related calendar and bibliography page for more ruminations on the POSIX time scale.
There is no way to go back and make POSIX time retrospectively self-consistent. This was tried with the "right" zoneinfo files, and the result was confusing and somewhat chaotic.
There is a way to make POSIX time self-consistent starting at some future date -- even if UTC retains leap seconds. POSIX time (as well as the NTP which usually underlies it) can become self consistent if the available broadcast time scale stops having leap seconds. This could be accomplished by changing the name of the broadcast time scale, omitting leap seconds from the newly-named broadcast time scale, and letting UTC become a time zone in the zoneinfo files. This is effectively the strategy espoused in the conclusion of the 2003 Colloquium on the future of UTC held in Torino. See the prescription on this page for details.
The year 1601 is prior to the development of reasonably accurate escapements for clocks. The year 1601 is prior to the development of the telescope. The year 1601 is 75 years prior to the establishment of the Royal Greenwich Observatory, so not even GMT makes sense.
As noted above, nothing that could possibly be called UTC existed prior to 1960. UTC has always been an atomically-regulated time scale. Prior to atomic chronometers all practical time keeping was performed in subdivisions of the day. Subsequent to atomic chronometers it becomes necessary to decide whether a time scale is counting calendar days or SI seconds.
For values subsequent to 1972 trying to do both means that Microsoft Windows file time faces the same problem as POSIX time. For values prior to 1956 there is no possible way to attribute an unambiguous meaning to a Microsoft Windows file time. The whole notion is utter bilge.
See the related calendar and bibliography page for more ruminations on the Microsoft Windows file time.
Also note that Microsoft has rejected the notion that Windows is capable of noticing leap seconds or their absence because
The W32Time service cannot reliably maintain sync time to the range of 1 to 2 seconds. Such tolerances are outside the design specification of the W32Time service.(It is likely the case that any other OS running on motherboards with piece-of-crap hardware clock implementations would experience the same issue.)
The .NET framework from Microsoft takes the above notion even farther into fantasyland.
According to the docs on msdn the DateTime structure "represents dates and times with values ranging from 12:00:00 midnight, January 1, 0001 Anno Domini (Common Era)" using "100-nanosecond units called ticks, and a particular date is the number of ticks since 12:00 midnight, January 1, 0001 A.D. (C.E.) in the GregorianCalendar calendar".
According to calendar historians, at that date the Roman empire under Augustus was probably still engaged in the process of having no leap years at all (in order to correct for having one every three years starting in the year Julius died). This is not to mention the fact that the Common Era would not be defined for another millennium, the Gregorian Calendar for another half millennium after that, and the atomic chronometer another four centuries after that.
Dr. David Mills at the University of Delaware began implementing the Network Time Protocol (NTP) in the early 1980s. The current versions of NTP are able to keep computers all around the Internet synchronized to better than one millisecond. Dr. Mills has written this book on NTP as a definitive reference.
Principal documentation for NTP can be found at the web site of Dr. David Mills and at the NTP project web site. The IETF has an active working group on NTP and provides status of the effort.
It is almost certainly a mistake for me to write anything further about NTP, but its characteristics place it on this page.
The NTP timescale is kept and exchanged via an unsigned 64-bit fixed point integer where the upper 32 bits represent integer seconds and the lower 32 bits represent fractions of seconds to a resolution of around 200 picoseconds. The origin of the current NTP era is 1900-01-01T00:00:00 UT, and the NTP counter will wrap around in the year 2036. Given that UTC with leap seconds originated in 1972, and that atomic time did not exist before 1955, it is not clear that any meaning dare be attributed to the fractional bits of the NTP chronometer during most of the first half of the present NTP era.
The NTP counter ignores leap seconds. As such, its practical properties are very similar to POSIX time. NTP ticks in SI seconds, but its counter accumulates mean solar seconds. At the sub-second level NTP time corresponds directly with TAI or UTC since 1972. At the resolution of one second NTP corresponds to mean solar time. Differences NTP over long spans of time correspond to the historical tradition where "time" means calendar days of earth rotation.
The properties that are suitable for LCT remain an open question. Some kind of solar time has been the basis for most calendrical schemes since the dawn of history. Any form of solar time is guaranteed to be non-uniform, and hiding that non-uniformity has been a goal for centuries. The non-uniformity is too subtle for humans to notice, but the number of human-created systems which can detect that non-uniformity continues to increase.
Atomic time is necessary to many systems used by civilization. Current applications for navigation require that everyone be able to agree on its value to about 10 nanoseconds. Mean solar time is traditional as civil time and necessary in order to construct a calendar that counts days. Current applications for civil time require that everyone be able to agree on its value to about 10 milliseconds. There is little dispute that it is far more important for everyone to be able to agree on the value of civil time to within 100 nanoseconds than it is for civil time to have a value that tracks mean solar time to within 100 milliseconds.
Humanity could adopt perfectly uniform atomic time now at the expense of
letting our descendants figure out how to return to mean solar time if
they so desire.
We could admit that both mean solar time and atomic time are important at the expense of retrofitting all timekeeping systems, legal systems, and the general populace with the understanding that there are two different kinds of time.
The relative merits of these and other options are not yet clear.
Perhaps by forcing the question might we trigger civil timekeeping chaos where different jurisdictions choose different answers to the questions?
For those who are designing operational systems that rely on stable inputs of time scales the many changes in the definitions, meanings, names and uses of time scales evoke another pop-culture quote:
I am altering the deal. Pray I don't alter it any further.-- Darth Vader, The Empire Strikes Back