You are on page 1of 29

MONTHLY NOTICES

OF THE

ROYAL ASTRONOMICAL SOCIETY


GEOPHYSICAL SUPPLEMENT
Vol. 5

No. 7

1948 July

THE FIGURES OF THE EARTH AND MOON (Third Paper)+


Harold Jeffreys, F.R. S .
(Received 1948 February 7)
Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

Sumniary
The paper consists of a general adjustment of the data relevant to the radius and ellipticity of the Earth, the lunar parallax and the differences of the moments of inertia of the 'Moon. The anomalies of gravity recently found to persist over distances of IOO and 30' imply similar anomalies in the deflexion of the vertical and provide a means of estimating the additional uncertainty that they introduce. When this is taken into account it is found that the data are satisfactorily consistent. Attempts are made to fmd additional checks on the existence of the anomalies of gravity represented by harmonics of degrees z and 3 other than the main ellipticity term, but they are indecisive, and the final solution is a compromise between solutions with and without them. T h e chief changes from current values are a reduction of the Earth's equatorial radius from 6371'388 km. to 6371*099fo-116km.(s.e.) and the value of g at Potsdam from 981.274 gals. to 9812606fo.oo10 gal. ; owing to recent work by the Anglo-Iranian Oil Company the British stations are now strongly connected with the new determination of the absolute value of g at Teddington. The change in the equatorial radius is due to the use of free-air instead of isostatic: reductions ; reasons are given why the latter may have introduced a systematic error.
1 . 0 . The usual standard of reference in geodesy is the Hayford spheroid, with which is associated the International Gravity Formula of Cassinis. The ellipticity of the Earth produces measurable perturbations of the Moon, two of which also involve the Moon's figure, and give useful information about the latter. There are several previous discussions of the data, the chief being those of de Sitter t,who has largely relied on the hydrostatic theory initiated by Clairaut and carried to the second order of small quantities by Callandreau, Darwin and himself. If the hypotheses of this theory are granted, the Earth's ellipticity can be calculated from the precessional constant (dynamical ellipticity) with greater accuracy than any geodetic or astronomical observations can give, and de Sitter's final adjustment (completed by Dirk Brouwer) depends on this calculation.

Earlier papers were M.N., Geophys. Suppl., 6 1-13, 1937; M.N., 101, 3 4 , 1941. f Bull. Astr. Z n r t .N e t h . , 1,97-108, I924 ; 4.5741,1927 ; 8,213-230,1938 (with D .Bmtawa).
G

18

220

Harold yegreys

Recent work has shown that all the results obtained hitherto are in need of revision. The hydrostatic theory has long been known to be false for the Moon and accordingly was dubious for the Earth. The value of the precessional constant H depends on the mass of the Moon, which depends on the solar and lunar parallaxes and the lunar inequality. The solar parallax and the lunar inequality had been taken from the work of Hinks on the Eros opposition of 1900, but Spencer Jones has shown from the observations at the opposition of 1931 that they need changes of several times their apparent uncertainties. The lunar parallax has long been noted for a slight discordance between the measured value and that calculated from dynamical principles, but the comparison depends on the adopted values of four quantities, all of which need readjustment. Accordingly, whether the hydrostatic theory is right or not, the value of H needs to be redetermined and the ellipticity based on it to be recalculated. The mean value of gravity over the Earth has also been shown by new measurements of the absolute value of g and re-examination of the old ones to need a change between 10 and 20 parts in a million. De Sitter gives formulae for making small corrections to his adopted values, but in all cases he has eliminated the ellipticity in favour of the precessional constant and thereby made his results inapplicable if the hydrostatic hypothesis is rejected. He remarks that if isostasy is true the changes would be inappreciable, but evidence from gravity has made it clear that there are widespread variations of gravity from any formula consistent with the hydrostatic state. Isostasy is not true, and the interesting question now is to find out how far it is wrong. These variations of gravity provide a means of finding out. Even the adopted equatorial radius given by Hayford is under suspicion because it depends on the use of isostatic reductions and therefore assumes that isostasy is true. If it was true, gravity over the land would be systematically greater than over the oceans by amounts that could be detected by analysis of the observations. Actual analysis, however, shows that the difference, if there is any, is in the opposite direction, and it is possible that the isostatic reductions have introduced a systematic error. For all these reasons the data relevant to the figures of the Earth and Moon need rediscussion. 2.0. Geodetic data. General discussion.-The equatorial radius of the Earth, a, enters into several of the relations connecting astronomical and geophysical constants, usually in combination with some function of the ellipticity, e. Ultimately all estimates of it rest on trigonometric survey, but unfortunately there are difficulties of interpretation. A survey concerns a limited region, necessarily on land, and the practice is to choose the values of a and e of the spheroid that most nearly represents the form of the level surfaces in that region. If the Earth was a spheroid, the surface being exactly a level surface, each survey would give independent estimatesof a and e for that surface, and these could be combined by the usual methods to give more accurate values for that surface. The actual level surface that includes the ocean surface is not an exact spheroid, gravity showing a complicated system of small departures from it, and accordingly the a and e determined from a survey are representative only of the region and not necessarily estimates applicable to the Earth as a whole. For astronomical purposes, however, we do need values for the Earth as a whole, on account of the relation of a to the lunar and solar parallaxes and of e to the part of the departure of the external field from

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

The Figures o f the Earth and Moon

221

symmetry that is expressed by the zonal harmonic of degree 2. For purposes of the external field theory we take the gravitational potential as U =f M / r + U', where 7 is measured from the centre of mass and U ' therefore contains no spherical harmonics of degree less than 2. Then the level surfaces are surfaces of constant Y,where Y = u+&fPf cos%js. Part of one of these level surfaces coincides with the ocean surface. Within the land we can extrapolate this surface by simply taking points at distances below the outer surface equal to the measured heights. (The usual definition of the geoid, I think, introduces unnecessary complications, on account of the attraction of matter above it, but the difference is of the second order, and in practice the geoid used is that just defined.) Then the astronomical spheroid can be taken unambiguously to be the one with minimum mean square departure from this surface. The definition remains unambiguous if U' contains such harmonics of degrees 2 and more that Y is not constant over an exact spheroid. The process of inferring a and e for this standard spheroid from the values given by separate surveys involves complications similar to those already discussed in relation to gravity." The derivation of a gravity formula from observations of gravity, like the astronomical effects, depends only on external field theory, and the most direct treatment rests directly on the free-air anomalies. But these show irregularities, which can be classified roughly into (I) very local disturbances, say up to ' I in horizontal extent, (2)disturbances persisting over 1 0 ' squares but not correlated with adjacent 1 0 ' squares, (3)disturbances persisting over 30" squares but not correlated with adjacent squares, (4) disturbances expressible by spherical harmonics of degrees 2 and 3. Each type is associated with disturbances of Y and therefore of the relation between distances measured astronomically and by geodesy. The first type correspond to local deflexions of the vertical, which may be treated as random variations in the reduction of a geodetic arc. The second persist over a considerable fraction of the lengths of the arcs, the third possibly over the whole length of an arc. The fourth will overlap the longest arcs surveyed. Now what is done in the survey of an arc is essentially to determine the relation between the distance travelled over the standard level surface to each point and the direction of the normal to the level surface. According as the survey is along a meridian or a parallel, it therefore determines the curvature of the meridian or the parallel. It might appear that for sufficiently long arcs of meridian it would determine the variation of the curvature along the meridian ; but in attempts to do this, the ellipticity has always been found to have a standard error of about 10in its reciprocal, which is so large in relation to that given by other methods of estimating the ellipticity that the information about the ellipticity given by a single arc can be treated as negligible. The practical method is therefore to adopt a reasonable trial value of the ellipticity and find, subject to it, what value of Q fits the data best. I n the United States and India, geodetic arcs have been surveyed in both directions, and an estimate of the ellipticity can be made, essentially by comparjng the curvatures of arcs of meridian and parallel. But in any case all the disturbances of types (z), (3), (4) introduce correlations between adjacent residuals when an ellipsoid is

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

* M.N., Grophy~. Suppl.,5, 1-22, 1941 ; 55-66,


G

18*

222

Harold Jeflreys

fitted to the data, and the uncertainty of a tends to be underestimated. For type (2) this can be detected in the study of a long arc and allowed for, but the estimate of the additional uncertainty is based on a small number of degrees of freedom. But for types (3)and (4) the disturbance can be almost wholly absorbed by adjusting the estimate of a, and consequently the estimztes of a made for different regions will differ. This additional variation could be detected by comparing different s an unknown. If an inconsistency is found it can regions, now reintroducing e a be allowed for by a corresponding increase of the standard errors. The essential point is that a as given by a survey is the a for the spheroid that best fits the level surface near the arc ; and this is not necessarily the spheroid that best fits the level surface for the Earth as a whole. Consequently the determination of the latter involves additional corrections and its accuracy depends not only on the intrinsic uncertainties of the separate surveys but on our ability to determine and eliminate their differences from the general standard. The importance of the 10 variation can be seen in the results for the United States, India and South Africa. Hayfords solution for the United States was a = 6378388 & 18 metres ; I/e = 297.0 +- 0.5 (p.e.). Hayford remarked that these probable errors are found on the hypothesis that all the residuals are accidental in character, but that as there is some evidence of systematic errors, these probable errors should be somewhat increased to represent the actual uncertainties in the values. He gave a map showing in separate enclosures blocks of consecutive residuals with the same sign, often including five and sometimes as many as ten. The solution also takes the depth of compensation as fixed. These complications have been discussed by Helmert *, who increases the uncertainties to 53 metres and 1.2 (standard errors). De G r a d Hunters map of the Indian geoid shows a change of elevation above Hayfords spheroid from -20 feet in the south to + 35 feet in the centre, falling again to below -20 feet in the north; the two minima are separated by about 12 in arc. The South African survey was fully discussed by W. Bahn.1 The solution for the whole arc, with 11298.3 for e, was$ a = 6378307 & 179 metres. But separate analyses were also made for three somewhat overlapping sections and gave corrections to Clarkes 1880 value of -2645 k 450m., + 1031 f 438 m. and +213 & 385 m. Similar correlations are noticeable in the west European arc of meridian; they are less marked in the Russo-Scandinavian arc, but this may be because on an average the stations are farther apart. I think that the use of isostatic reductions, and still more of Bouguer reductions, is unsuitable for this problem. Both have the property that they make the residuals vary more smoothly than the free-air ones, whether intensity or direction of gravity is being considered. But they estimate and remove the attraction of the outer parts of the Earth on specific hypotheses, and do not restore it ; consequently they make non-random changes in the external field, which may contribute to the estimates of a and e. I t is indeed plain that some of these reductions (including Hayfords) would introduce first harmonics into the field, because the indirect

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

* Sit2. Bms. Akad. Wiss., 1911,1-19. t M.N., Geophys. Suppl., 3,42-51, 1932.

3 Gerlandc Bktr. z. Geophys., 10, 5 1 p j j 1 , 1910. 3 All uncertainties are standard errors except when the contrary is stated.

The Figures of the Earth and Moon

223

effect (that is, the disturbance of the position of the level surfaces with regard to the surface of reference) was not taken into account when they were made, and the continents are not symmetrically placed over the Earths surface ; but the existence of such harmonics in gravity or in the distance of the geoid from the centre of mass is contrary to the law of gravitation. But even if the indirect effect is taken into account it is not clear that systematic errors will not be introduced. On both hypotheses the geoid within the continents will lie above the spheroid that best fits the oceans, curving down to it near the coasts, and therefore, if either is true, the radius found from land observations will be systematically less than that of the Earth as a whole ; then the reductions might be expected to remove a systematic error. The hypothesis that the internal stresses are such as to require the least strength consistent with surface elevation, and zero strength below a certain depth, while giving results qualitatively similar to the usual forms of isostasy, gives rather more disturbance of the external field and arguments concerning isostasy would apply more strongly to it ; but as no reductions have been made according to this hypothesis it does not arise here. The data on the distribution of gravity can be used to test the existence of such a systematic departure, and conceivably to allow for it if it exists. From my table * it appears that for harmonics up to degree 5, if compensated, the disturbance of gravity due to an increase of level of 5 km. would range from 6 to 20 mgal., and such harmonics would represent the main features of the distribution of land and sea. Without compensation the difference would, of course, be greater. But in my analysis of gravity I found four longitude terms, three of which have the opposite sign to those in surface elevation. There is plenty of evidence that within 10 squares free-air gravity is positively correlated with height, but this correlation was estimated and the data for each square were reduced to the mean height of the square. The question remains whether the 10 means are correlated with the mean height. I took from my table (pp. 18-22) the data for the land squares where the mean free-air anomaly had a standard error of 15 mgal. or less, and fitted an expression a + b h to them by least squares, treating all as of equal weight. The result was b = ( + 3-6f 10.0) mgal./km.

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

T h e result is in the expected sense but not significant. It is of no use to compare the whole of the land squares with the whole of the sea squares, because the mean over the Pacific must be practically the same as over the opposite hemisphere. For Europe and Africa I got a mean of + 8.4 f 3-6mgal. ; Atlantic, 6-4k 5.7 ; N. and S. America, -1-6 f 5.7. Harmonics of about degree 4 would represent the Atlantic and we should expect the Atlantic value to be 6 to 8 mgal. less than the continental ones. The American value may need an increase of 6 mgal. to correct a possible error at the base station at Washington, but in any case it is clear that the data do not support the systematic difference indicated on the hypothesis of exact isostasy. The difference between land and sea gravity on the hypothesis of exact compensation would be large enough to be observed, but it is not. Consequently isostatic corrections will not remove a systematic difference between land and sea ; on the contrary, they will create one. It may be remarked in passing that the observed difference between land and sea observations would be expected on a quite plausible hypothesis. Suppose

* M.N., Geophys. Suppl., 3,16568, 193%

224

Harold Jeflrqs

that the fractures in the last period of mountain formation left the crust in a state of minimum stress and that the fractures sealed. Later denudation would lower the land and transfer sediment to the sea floor. The result would be to reduce gravity on land and increase it over the sea until the new stresses became enough to produce a new- set of fractures and general readjustment. I n fact the sign of the differenceis what we should expect if adjustment does not quite keep pace with denudation, and nobody supposes that it does. Accordingly I think that the only systematic correction connected with height that needs attention is the geometrical reduction of measured distance to distance over the geoid, and as this has been applied in all reductions it needs no further comment. A theoretically satisfactory method would be to use observed gravity itself to determine the form of the geoid (but not its size). If we use the departures of gravity from a formula appropriate to an ellipsoid of given form, we can determine the e elevations of the geoid above that ellipsoid ; but survey gives the actual form and scale of the geoid in the region covered, and comparison would give an estimate of the size, from which errors due to the fact that the external field is not that associated with an ellipsoidal geoid would have been automatically eliminated. The practical difficulty of the method is of course the incompleteness of the gravity survey. Stokes's integral formula supposes gravity known everywhere ; to apply it to existing data would require interpolations over vast ranges. We could not, for instance, apply it to the United States without full gravity data for the North Pacific. The following practically possible methods need attention. In all cases it is supposed that free-air reductions have been made. (I) T o use the existing measures with uncertainties as given directly by the surveys. This would lead to an underestimate of uncertainty, which would be detected by computing x2. (2) Where there is noticeable correlation between adjacent residuals in the survey, this can be allowed for by computing a new uncertainty from means of several consecutive residuals. The drawback of this method is that if, as in South Africa, the correlation persists over a large fraction of the length of the arc, the estimated uncertainty may be on only I or 2 degrees of freedom, and the arc may be seriously overweighted or undenveighted in consequence. (3) We can use the variations between I O O squares and 30" squares found from gravity to estimate the scatter likely to occur in measures of the radius, and combine this with the uncertainty directly given. This would be substantially better than method (2) because these variations are based on some hundreds of degrees of freedom. The individual values are not sufficiently accurately known to be of much use, but the averages are well determined. (4) As gravity itself appears to have disclosed some harmonics besides the constant and the main ellipticity term, it should be possible to calculate their effects on the measures, and, if these terms are genuine, allowance for them would remove a systematic error. It is clear that method (3) should lead to the best estimate of uncertainty; if the scatter of the separate determinations is reduced when corrections are made according to (4) it will serve as a check on the reality of these longitude terms in gravity. T h e latter rest on the basic assumption that the higher harmonics (which i n any case are needed to describe the 30" variation) are such that their contributions

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

The Figures of the Earth and Moon

225

to the estimates of the lower ones can be regarded as random ; they would make no contribution at all if the survey was complete, but with the actual distribution of gravity observations it is not certain that a large part of the estimates of the coefficients of the lower harmonics is not due to higher ones that have been inadequately eliminated in the solution. 2.1. Correction o f afor known deformations o f thegeoid.-Position on the Earth's surface is specified by the astronomical latitude and longitude A, that is, by the direction of the plumb-line. Geodesy measures the arc between two values of q5 We have therefore to express the for constant A, or two values of h for constant arc over the actual geoid in terms of the changes of 4, A along it. We use a suffix notation ; the direction cosines of the normal are

+.

g ax, * We take as surface of reference a non-rotating sphere of radius a and the same mass and use suffix o for quantities measured on it. Then

Ia Y zi= - -

(1)

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

Yo=f M / r .

(4

+'

Corresponding points on the two surfaces are those where the 1, have the same values. We have, if U"= U'++w2r2 COS2+' =g(, (3) being the geocentric latitude, and if we neglect squares of small quantities,

Thus for given li a aUn


to the first order in U". If x,, x,, x3 are the usual Cartesian coordinates,

(7)

Hence

226

Harold Jeflreys

Derivatives with respect to Y no longer appear, as we should expect, since the only property of Y that is needed to establish correspondence between normals is that it is constant over the respective level surfaces. If Uis a harmonic of the first degree the two surfaces only differ by a constant displacement, and the quantities on the right should be constants. This is checked by taking U= cos $ cos A, cos $ sin A, sin $ in turn. For an arc of meridian take A=o; then to the first order in 5 cos$.(a++)

whence

-ds = a + + + - - - ,a2+ .

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

a4

For a parallel 4 is constant ; then we find similarly

It may be verified that these reduce to a if is a first harmonic. The estimates of a are adjusted to various trial values of e, and e2a may not be negligible. If we retain it we find directly for a meridian
ds = a I - ~ e ( I + 3 c o s 2 # ) + e 2 - - e2 2sin2$cos~~ , (19) d$ and for a parallel ds = a(r + e sin2$ + e2 sin4$ - ie2sin2$ cos2$}. (20) cos$A The terms of order e are given by McCaw. Where a has been estimated as a, by reference to a trial value e,, the terms of order t 2may be supposed to have been eliminated with sufficient accuracy in the reduction of the observations. Let be the departure of an ellipsoid (a, e) from a sphere, and let

c1

e -eo=e, 5 The correction equations are, with sufficient accuracy,


a -ao=a,

(21)

a-a,.

ge(1+3cos2$)+[+

a--,+ -= o a$2
aha

for a meridian survey ; and for a parallel survey a + aoesin2$+ 5 -tan

4-a+ + S C L ~ $a - 5 = 0.

a+

(23)

Here 5 represents the departure of the geoid from the nearest spheroid and therefore may include the terms representing the ellipticity of the equator and higher harmonics. We denote the coefficients of e by w , so that if +=o the correction does not alter a + we.

* Geog. Journ., 64, 120-129, 1924 ; 6g, 572, 1927(discussion on paper by A. R. Hinks).

The Figures of the Earth and Moon

227

If [= Pi(cossh, sinsh) the terms in 5 in (17), (18) are as follows, for n = 2 and 3, each multiplied by cos sh or sin sh :
Meridian Parallel

p: pi p:

pi

3(3sin24 -I) -30 sin 34 I5 cos 34 -30 COS 3+

--y(cOs+ -cos3+) -30(4 -sin2+)

- $(5 + cos 24) -%( 5 sin + sin 34)

2.11. The correction o f sntall systematic errors o f slope.-The problem of correcting an estimate of the radius of the Earth made by a survey for systematic departures of the level surface from a spheroid is one of a general type. In the ideal case the problem is to fit a set of measures at given intervals of (or A) by a linear form A +a+, using the method of least squares ; thus A and a are found ' from

Z ( A + a+) = 2 s ; x$(A + a+) = q s . (1) Suppose now that s is affected by a systematic error q, so that the corrected measures are s -q ; we want the corrected values A + 6A, a + 6a that would be obtained if s - q was used instead of s. We have

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

x(6A + @ a ) = - x q ;

Z+(6A + +6a) = - q q .

(2)

If the observations are numerous and uniformly distributed from 4, to & we have nearly
($2

-+,PA + W f -+:Fa=
-+:Fa
=

-jqd+,

(3)

4<+f -+:PA +
whence

- +qdh

(4)
(5)

~ ( -+1)36a 4 ~=-

I(+ -$4 -++2)q d+.


1

In the survey problem A is an error at the base station and is not of general interest. qis small and is not given directly, only through its derivative. Since q is small and usually varies slowly, it is worth while to derive an approximate formula based on two representative values of dq/d+. We replace +1, +2 by & k, and take

Suppose that at

+= k kl, dq/d+=pl,p2.

Then

j:,+qd+ = i(pl +p2)k3+ 5yk5 -WW


Thus the terms in y will be zero if k, is chosen so that

228

Harold Jeflreys

I n cases where the systematic error is small, so that Sa is comparable with the standard error found for a, it will be enough to use this approximate formula ; and then returning to (5) and taking
T

we have

The correction is therefore reduced to the calculation of the given derivative at two values of 4. 2.12. Treatment of the 10"and 30" variations.-I found that the variations of gravity between 1 0 ' squares could be represented by a variation whose mean value over a 10" square is taken as random with a standard deviation T~ of 2 0 5 mgal., and a variation whose mean value over a 30' square is taken as random with a standard deviation T~ of 12.5 mgal. Now for high harmonics the mean square deviation of the radial component of gravity is nearly equal to the sum of those of the two transverse components ; hence in either meridian or prime vertical we should expect mean square deviations of the vertical of T/g1/2 in circular measure, that is, 3"-0 and 1"-8respectively. These are comparable with the residuals in surveys, so that the 10" and 30" variations of gravity are consistent with the amount of correlation between consecutive residuals that actually exists. For a survey line IOO long the former would put a wrong by & 1.2 parts in 10,000, or f800 metres, roughly. With this standard error the differences between the parts of the South African arc would not be remarkable for a selected case. For an arc 30" long the T~ variation would give a standard error of a of order I in 40,000, or f 150metres. This is comparable with the intrinsic uncertainty of most of the measures of a and must be taken into account when the measures are combined. The method of 2.1 supposes q given as a function of $,A. I n the T~ and T~ variations only the .mean square variation of q is supposed given. Suppose that errors of deflexion f a occur at random and that observations are spaced at interval I ; then if there are observations at $ = 0,I . ..nl, and the individual errors are E ~ we , shall have

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

(CC$~ -(n
and therefore

+ I ) q 2 ) S a= a x ( $ -$)er,
{&n(n

+ I ) ( n +2)}fua=au/l.

The actual situation is complicated by two facts. q is a continuous function and is not to be taken constant over a square ; we have only the scatter of the mean deviations in squares. So long as the sampling points are identical with the centres of the squares this gives no trouble, but actually the surveys use many s filled in by interpolating intermediate points. The variations are to be regarded a a continuous function, and it is known that if this is carried out completely the mean square deviation of the function is equal to that of the original values.* This complication therefore does not affect the result.

* R. A. Fisher and J. Wishart, Proc. Camb.Phil. Soc., 23,912-921,1927.

The Figures o f the Earth and Moon

229

In addition the length of the survey is not an exact multiple of 1; suppose that it is (n +K)l, where o <k <I. The starting-point of the survey may be considered to be chosen at random with respect to the variation of q. Then there is a chance K that n + z squares will be overlapped, I - K that n+ I will, and in working out expectations we can interpolate &(n + I)(n + z ) linearly. Clearly for each variation u = T / g d Z , in circular measure, and ao/l is equal t o 540 metres for T~ and 107 metres for T ~ . The following table gives the standard errors of estimates of a arising from them for arcs of different lengths. They have been taken to 1 0 0 ' to indicate the importance of connecting the RussoScandinavian, Egyptian and South African arcs. The contribution from T~ is. the larger for all lengths of arc, the ratio of the contributions to oafalling from 8 a t 1 0 ' to 1-5 at 1 0 0 ' . TABLE I Length (") u (metres) Length (") u (metres) Length (") u (metres)
I0 I2

I4 16 18
20
22

807 649 560


501

24 26

459 424 378 346


321

28 30 32 34 36 38 40 4 5

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

quotes eight determinations of a. 2 . 2 . Application to survey data.-McCaw Some of ;hem had had Boiguer or isostatic adjustments applied. I take the following free-air solutions. Each value a, is subject to a trial value of e, which I denote by e ,. A4 denotes that the arc is along a meridian, P along a parallel. a is in metres. I give the standard error, a, on the assumption of complete randomness of the residuals. I n addition I give a adjusted t o the Helmert value of the ellipticity, 1/298-3. For the parallels I give both the longitude difference of the ends and the actual lengths of the arcs.
Arc
I.
2.

Mean
lat.

Length
299.15 299.15 299.15 299.15 298-3 2 9 8 ' 3

a 0

West European*

Russo-Scandinavian* 3 . Europe 5zo* 4 . Europe 4 7 ' 5 * 5. Indiat

M M

P
P

P
M

49"-72 7 ' 58O-o 25" 52" 6 0 " ( 3 7 ' ) 47"'s 20'(13~-5) 2 4 ' ~~"(20')
-25" 25"

6 . S. Africaf

6378000+ - 65 +455 57 -650 +358 +307

I55 127
10s

650 182 I79

(aadjusted) 6378-f - 50 +455 19 -683 +358 +307

Helmert considers the use of the Indian meridian too unreliable, and I agree. I n partly mountainous country the free-air residuals tend to be larger in magnitude in mountains than in plains, and allowance for variation of the standard error with position would be worth making, as for gravity. Its effect would be to reduce the importance of the mountainous regions in determining the general form of the Earth ; but as this allowance has not been made the result for the Indian meridian may depend greatly on the arbitrary position taken for the northern end of the arc.

t F. R. Helmert, Z S . ~ EtdRunde, . 17-34, 1913. 1 W.Bahn, Gerkands Beitr., 10, SI!J-~JO. 1910.

' F . R. Helmert, Sitz.

Reuzr. A M . Wiss., 525-537, I&.

230

Harold Jejfreys

Helmerts discussion shows little sign of correlation of consecutive residuals in the European arcs, but for the 52 parallel his final solution has rejected the station at ,each end. Bahns results show strong correlation in the South African arc, but a -re-estimate of the standard error on two degrees of freedom is not worth making. For the United States, Hayford made a free-air solution, but did not give uncertainties. I find that his equations * lead to the equations, with independent uncertainties, a + o.678e = -1.925 0.33, (1) e= -0.72 f 0.23, (2) where Iooa = a -6378206, e = 20000(e -1/2950). These are equivalent to
a + 1356000 e - - = 6378062 f 33,

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

2 9 4

These uncertainties depend on the hypothesis that the errors are random. This is likely to be at least as far wrong for the free-air residuals as for isostatic ones, and from Helmerts results for the latter it appears that the standard errors might be doubled. The unsatisfactory feature of this procedure would be that the failure of randomness is the sort of variation represented by the T~ variation, and if we did it and then corrected by Table I we should be correcting twice for the same thing. Now with a change of notation we write
a = a -6378000, e = 107(e -1/298.3) = 107(e -0.0033523). T h e equations of condition are, with the uncorrected uncertainties ul,
I.
2.

(3) (4)

3. 4.
5.

6. 7.

8.

a-0*163e= - 5 0 f I 5 5 u+o.rore= +455frz7 19f 105 a+o*396e= a+o*347e= -683fQ0 a+o.roge= +358fr82 U-0934&= +307f179 a+o.156e=+ 64% 33 el=+ 1 7 f 1 1 5

+ 107
+ 91 + 73
+

Calc.

0-c
- I57

+364
- 54 -759 +267 $-I53 25

+ + 89
-

76

+ I 91 54

61

+ 78

T h e solution is
e = -61 f 91, a= + g 7 + 3 1 ; but gives x* = 153 on 6 d.f. This is far too large. The only notable individual contribution is that for the Russo- Scandinavian arc, which is not under any special suspicion. The South African arc might have been expected to make a large contribution, as it is the clearest case of great underestimation of the uncertainty ; but it gives only 0.7. Now the data come from four continents, four from Europe,

* Supplementary Investigation of

1909, p. 39.

23= one each from Africa and India and two from North America. It is quite possible that there may be variations in Europe that affect extensive regions and have not been revealed by inspection of residuals in the separate arcs. Accordingly we proceed to make an alternative solution with allowance for the 10"and 30"variations. We meet a serious difficulty because the United States survey does not consist of two arcs but of a complicated pattern. There is a long parallel from -124" to -75", in latitude 39", crossed by three arcs of unequal lengths in longitudes of about -IZZ", -98" and -80". The method adopted was to replace the equations of condition given for the United States by a pair of determinations for a meridian and a parallel that would be together equivalent to them. We have in fact (1000/33)2=g18; (I000/115)2=75. We find g18(a' + 0.136e')~ +75et2=785(a' +0.253e')~ + 133(a' -0360e')~. (6) The coefficient 0.253 is chosen because it is the value of w for a parallel of 39"; the rest of the coefficients have been determined algebraically. The coefficient -0.560 is found to be the value of w for a meridian in mean latitude 37'7, which is a plausible value. Accordingly we can take for the United States an alternative pair of equations of condition a' + 0.2536' = + 66 f 36, (1') a' -o.56oe'= +52 +87. (2') The former would correspond to the actual parallel of 39"; the range of longitude is 49" and the actual length 38". The latter corresponds to a summary of the western and central meridians and the eastern oblique arc. The ranges of latitude are 33" to 49", 26" to 49" and 25" to 43". The adjustment to allow for the T~ and T~ variations will be the mean of the values from Table I. Then the revised statement of the data will be as follows :
I. 2.

The Figures of the Earth and Moon

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

3. 4.
5.

6. 7. 8.

~'-0*163e'=- 50f348 a'+o-IoIe'= + 4 5 5 f 3 5 6 19f243 u'+o*396e'= a'+o*347e'= - 6 8 3 f 8 7 0 a'+o*Io5e'= + 3 5 8 f 4 6 2 a'-o.g34e'= + 3 0 7 f 3 8 0 a'+o.q3e'= 66f216 52f453 a'-0*560e'=

- 191
+351

0-c

X'

9'3
I '0
0'0 0 '7

42

-751 255 55 - 16 - 146

+ +

0.3
0 '0
0'0

0'1

The solution is

+118frrg; e'= -143fzgo. (7) Equations with independent uncertainties are e'= -r43+zgo. (8) a'+o-o47e'= +rr1+118; The resulting value of x2 is 2.4 on 6 d.f., which is on the small side but not sufficiently so to suggest that we have overestimated the allowance for the 10"and 30" variations. Had we got x2=o*5 it might have been a very different matter. On the face of it the result appears quite satisfactory. Then we have a =6378.118 f o-rrgkm. ; e = 0-0033360f o.oooozgo ; I/e= 299.76 2-63. (9) The value of a is so insensitive to that of e that it is already clear that Hayford's value is too large.
a'=

232

Harold Jeflreys

2.21. The Longitude ternls.-My discussion of gravity led to the following terms depending on the longitude in the elevation of the geoid above the mean spheroid, in metres.
~6~1P~cos~h+~zP~coszh+1~ 1.50P~singh. ~ 8 P ~ c o s h + (10) This being taken as given, the corrections to the equations for a, e can be worked out. I give them as corrections to the right sides of the equations of condition, the contributions from the harmonics being given separately.
P B '
West European Russo-Scandinavian Europe 5 2 ' Europe 4 7 ' 5 India .S. Africa U.S.A.

p:

p:

U.S.A.

0 0'0 M - 48 152 73 M - 70 - 8 I74 - 4 1 1'3 P + I 0 4 f 78 229 0.9 69 -t- 72 342 +113 p 42 -4- 30 + 58 + 93 -460 +113 -573 0' 4 P -198 - 84 + 6 - 126 - 4 4 +116 - 160 0'1 r V 1 $. 21 - 54 - 68 223 127 17 96 0'1 - I1 p -177 - 1 1 0 90 - 142 + 1 1 4 -256 I '4 123 - 181 0 ' 2 M + 5 - 88 - 20 - 7 - 48

+
+ +

+ +

Total Calc. 0-C obs. 126 +I19 + 7 526 +116 +4IO

XS

+ +

+ +

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

4' 4

T h e solution is
a'= +117&1rg, e'= -11+290, (11) and gives x2=4-4. This is quite normal, but it is rather surprising that correction of an apparent systematic error should have increased x2. The changes in a and e are of no importance in relation to the standard errors. It was hoped that this test would give an additional check on the existence of the longitude terms, but it is indecisive. They had appeared clearly genuine according to the method used in analysing gravity, but this is subject to the basic assumption that the places of observation were so situated that harmonics of degrees of 4 and more would contribute at random to the estimates of those of degrees 2 and 3, which were themselves treated as orthogonal, except when they clearly were not. This was the only practicable method of procedure, but the hypothesis needs further examination. There is no doubt that widespread departures of gravity from a simple ellipticity formula exist, but the incompleteness of gravity survey is still such that we cannot say how much of them is due to low harmonics and how much to those of degrees from 4 to about 12. The values of a and e corresponding to this solution are a=6378-117 k o - ~ ~ g k m . ; e=o.o03351z &o.oooo~go; I/e =298.40 k 2-61.

f grdy.-Most observations of gravity are made by 3.1. The absolute value o comparison of the periods of the same pendulum in the field and at a base station, the base station itself having been more or less directly compared with Potsdam, where an absolute determination g p =981.274 & 0.003 gal. was made by Kuhnen and Furtwangler.* This corresponded well with the mean, reduced to Potsdam, of results found by earlier observers, but the separate determinations ranged from 981.253 (Bessel, Schumacher and Peters) to 981.282 {Defforges). Two recent determinations by J. S. Clark t at the National Physical

* Ve~off. Preuss. Geodb't. Inst., 1906. t Phil. Trans. A, 238, 65-123, 1940.

The Figures of the Earth and Moon

233

Laboratory, Teddington, and by P. R. Heyl and G. S. Cook" at the Bureau of Standards, Washington, have indicated that the Potsdam determination is between 10 and 20 mgal. too high. H. L. Dryden t points out that the final value was greater than any separate determination. Kuhnen and Furtwiingler had in fact noticed a possibly systematic variation of the estimated g p with the mass of the pendulum and attributed it to bending of the knife-edges. They therefore made a least squares solution to extrapolate to zero mass. N O such effect was noticed in the work of Heyl and Cook.(with, however, a smaller range of pendulum mass); and according to Dryden, W. R. Osgood has investigated the matter theoretically and concluded that the effect of bending of the knife-edges would not be more than one part in a million. Accordingly Dryden considers the extrapolation unwarranted and prefers to combine the data without allowing for a systematic variation with mass. His results differ appreciably according to the method of weighting used, but agree that the Potsdam observations themselves indicate a value of g , appreciably less than that adopted. None of Dryden's methods of weighting seems to me quite satisfactory, and he does not compute an uncertainty. I shall adopt one similar to one that I have found satisfactory in seismologicalwork. The chief difliculty is that four of the pendulums used beat approximate seconds, the other half-seconds, and it is not obvious how to weight the latter in comparison with the others. In the original paper the observationsare combined into groups (second averages) which are treated as of equal weight. Dryden considers it best to weight each combination of pendulum with the method of support equally, irrespective of the number of observations. I prefer an intermediate procedure. The second averages are as follows :
Pendulum Support
A

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

us

OL
P

Length of seconds pendulum (1o-Icm.) 9942330 2373 2346

Pendulum Support C

Length of seconds pendulum


( I 0 4 m.)

OL

us
P

2304
2222

2216

0s
2265

:z]
2130
2350
2215

2196

us

D
2264 2334
2250

OL

UL OL P

us

2232

2323}2z88 2252

0s

2333

2290 2358) 2326


2322

0s
2283
2250

us

2262
2002

2156
2242

i"!:]
1890 I974
2062

1948

*J. Res. Nut. Bur. Stand., 1 7 , 8054339, 1936. t J. Res. Nut. Bur. Stand., 29, 303-3 14,1942.

234

Harold Jeffreys

E is the half-seconds pendulum. For each combination we form residuals against the mean for that combination and derive a standard deviation for ABCD together, subject to the pendulum and support being kept the same ; the result is 41 on 13d.f. For E we get 67 on g d.f. To test for differences between pendulums and between supports, we form a general mean for ABCD and compare with the means for the separate pendulums and for the separate supports by forming sums of squares of the residuals. The results are :
Pendulums supports

Sum of squares 6618 2814

Expectation

Excess

1876
2100

4742 714

There is clearly a significant difference between pendulums but not between methods of support. We therefore take means for each pendulum but combine with the uncertainty a contribution 5 40 x 10-5cm. for the part of the variation that persists for each pendulum. The weighted mean is then (9942277& 22) x I O - ~ cm., and g , =981.2634& 0.0022gal. This is virtually on 3 d.f. For E it is plain that the variation between supports i s significant, and we may as well determine a mean and uncertainty directly from the support means. We get for the length (9942179& 117) x I O - ~ cm., g , = 981.2537& 0.0115 gal., and combining the two sets g, = 981-2633 & 0.0022gal. This agrees closely with Dryden's final recommendation. Clark's value for Teddington is given as g , = 981.1815f 0.0015gal. The uncertainty represents an average residual without regard to sign derived from the means of eighteen series of ten measurements each, and the scatter corresponds closely to known sources of uncertainty. The standard error found in the usual way would be 0.0005 gal. There appears to be a mistake in Clark's calculation of the correction for flexure, besides the one noted by him in the errata. His equation ( 2 ) , p. 109, would be correct only if the elastic pendulum had the same angular velocity at all points, and this is false. The theory of flexure given by Kuhnen and Furtwangler, pp. 6974, is correct and suited to a pendulum of any section. But Clark's pendulum was the best yet designed for making flexure small and I shall suppose that his value of the correction for it (I mgal.) is correct within about & I mgal., and therefore adopt a standard error of 1.0mgal. Hey1 and Cook give the value for Washington as g , =980.080 & 0.003gal. The stated uncertainty here is the mean of the average residuals in seven series of observations out of eight made. A mean of the seven series, with a standard error found in the usual way, is g , =980~0802 & 0.0015gal. 3.2. We proceed to discuss the most direct of the various comparisons that have been made between these stations. A direct comparison of Potsdam with

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

The Figures of the Earth and Moon


the Bureau of Standards was made by Lt E. J. Brown* in 1933. with no stated uncertainty, gw -gp= -1.174.

235

His result is,

In deriving this result he rejected one series of observations out of 36 ; this series gave a residual about twice the next largest and the rejection seems justified. T h e effect is that in Table 1 1 1the entry B4 for knife-edge A11 should be changed from 980.0984 to 980-1010. The entry B8 for knife-edge A I is wrongly computed ; it should be 980-0943 instead of 980.1006. This can be verified by comparison with the entry for Bg on knife-edge A 11, the difference in the periods of the pendulums being 3014x 10-7sec. in both cases. Six pendulums were of bronze, three of invar. On applying checks for consistency I found no abnormalities for different pendulums of the same material or for different knife-edges. The means computated from internal consistency were Bronze pendulums Invar pendulums 980~1013 f 0.0004 (11 d.f.) 980-0969 k 0-0010 (5 d.f.)
Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

The differenceis four times the standard error. I t cannot be regarded as definitely significant if we have regard to the small number of degrees of freedom of the second estimate, but it seems undesirable to combine the two until we have further checks. Brown remarks that the difference may be partly due to a slight error in the temperature coefficients of the pendulums ; this would support the estimate from invar pendulums. Magnetization is a known source of disturbance for invar, but great care has been taken to minimize both these sources of error, and on the data by themselves we cannot say which is the more reliable. As the various means are on the Potsdam standard we shall take

g , -gp

-1.1727f0-0004 (bronze) -1.1771k 0-0010 (invar).

The connection between Teddington and Potsdam is indirect. Teddington has been compared with Greenwich (both stations) by H. L. P. Jolly and G. T. McCaw, and Greenwich is well connected with Potsdam directly and through other British stations and De Bilt. I have previously given a least squares solution for the principal British stations t, but now prefer a slightly different treatment, as Dr Bullard suggested to me that on account of the closeness of the two Greenwich stations the difference of g between them could be calculated far more accurately than it could be measured. He gives g , -g, = + O*OOOI gal. for the difference between the National Gravity Station and the Record Room at Greenwich. With this value we can eliminate g , from all equations of condition. Using the values quoted by Bullard and Jollyf, I get

gtp -g1= + 0.0067 ; gT -g, = + 0.0064, whence we have an equation of condition, combining both equations, gT -g, = + 0.0066. This determination was under some suspicion as the observations were not checked by repetition at the original base. Recently, however, the Anglo-Iranian Oil Company has carried out comparisons of Greenwich (National Gravity Station,

* U.S.C. and G.S., Spec. Publ., a04, 1933. t Theory of Probability, 126-9, 1939.
1 M.N., Geophys. Suppl., 6 450,
1936.

a '9

236

Harold yegreys

g2), Kew,g, and Teddington with their base at Chipping Sodbury." the value at the last by g 6 we have

Denoting

g, -g6 = -6.71 mgal., g, -g6 = + 5-82 mgal., g T 3 6 = -0.88 mgal.


These determinations were made with a gravimeter and are believed to be trustworthy to 0 . 1 mgal. Consequently they may be taken as definite in comparison with pendulum data and used to eliminateg,, g, from all the equations of condition (the residuals of which, however, continue to give useful information about the accuracy of the pendulum observations). Then we find

g, =g, g, =gT -5-9 mgal. ; and we find s = 1.5 mgal., and on the g o = 981.26773f0.00094 g , = 981.18820 & 0.00079 g, = 981.18830& 0.00079 g, =981.20080& 0.00079 g , = 981.26419 & 0~00103 g , = 981.12308f 0~00123 g , = 981.19410 f 0*00079 g, -gp = -0.07990 f 0~00079

g, =g, + 6-7 mgal., -5.8 rngal. ; Potsdam standard De Bilt Greenwich, Record Room Greenwich, National Gravity Station Kew Observatory Cambridge, Pendulum House Southampton, Ordnance Survey Teddington, N.P.L.

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

Bullard and B. C. Browne t have compared Teddington with Washington directly, using six invar pendulums. Their result is g, -g, = -1.0969& 0~000gg gal. The standard error of the comparison is appreciably less than in the comparisons of English stations but not unduly so, since more pendulums were used. E. J. Brown's uncertainty for the bronze pendulums, also six in number, does look incredible, and is under some suspicion on account of the possible inaccuracy of the temperature correction. Provisionally I replace it by I milligal. We then have the following equations of condition :
0-?1

x:
2.7
1.0

gp=981.2633&0.0022 gT= 981-18I 5 f0.0010 gm= 980.0802 fo.0015 g,-gT= - 1*0969&0*~009 gw-gp= - I'I727&0'0OIO g,-gp= - 1.1771f o ~ o o I o gT-gp= -0'0799 f o-0008

+3'7
+re0

0-c,
+2'7
+om8 -2.9 +0.3
+0.4
0'0

-3.6
-0.2

5.7
0.0

Xt 1'5 0.6 3-8


0-1

+3-1 -1.3 -0.8

9.6 1.7

...

...

0.2

21-7

1'0

6.2

0'0

The solution is

gT= 981-1805; gw=980-0838 gals. gp= 981-2596 ; This gives the residuals 0-C, in milligals. x2= 21.7 on 4 d.f. is far too large, and nearly half of it comes from E. J. Brown's comparison with the bronze pendulums,

t Proc. Roy. Soc., A, 175,110-117,

Letter from M r L. H. Tarrant to D r E. C. Bullard.


1940.

The Figures o f the Earth and Moon

237

already under some suspicion, even though we have already increased the estimate of uncertainty for them. Rejecting this determination we get gp =981-2606& 0.0010, gT=981*1807& 0*0008, g, =980.0831 f0-0009, and the residuals 0-C,. x2=6.2 on 3 d.f. is not remarkable, and no separate contribution reaches 4 . Though it is slightly on the large side it may be held that it is about time we did get a x2 larger than its expectation. The largest residual is less than the uncertainty actually claimed by Hey1 and Cook. On the other hand, some of the estimates combined are on few degrees of freedom ; gp on 3, gw -gT on 5, gw-gp on 2. The tendency is for the use of uncertainties estimated on few degrees of freedom themselves to lead to large x2,and the final estimates of uncertainty should be regarded as estimated on not more than 5 or 6d.f.; the posterior probability of an error more than twice the estimated standard error will be larger than if the posterior probability followed the normal law. Taking the values ofg, andg, just given as standard and readjusting the other British stations and De Bilt we find : g o =981.2543 & 0.0009 0*7@gp 0*27sg~, g, = 981.1748 +%T, g 2 = 981-1749 + %T, g, = 981-1874 + %T, g, = 981.2508 k 0.0008 + 0-zoSg, + 0-806gT, g, = 981.1097 2 0.0011 -tO*136gp +0'876gT, g, = 981.1816 +sgp For general purposes the main conclusion is that the adopted value ofg,, on which observed values ofg are based, needs a reduction of 13-4i 1.0mgal. The British values are now firmly connected with Teddington. 4.1. The distri6ution o f graoity.-My solution is, in milligals, Potsdam standard, g =979772'5 k 1.9 (3439'9 5'o)p2+ 5'3p4 +c4-0 1.4)Pgcos zA +(1.30 & 0.68)Pi cos 2A + (4.2 k 2.4)Pi cos A + (0.46 f0.26)P: sin3X, P, being taken as a function of geographic latitude. The comparatively large uncertainties represent chiefly the scarcity of observations in many large regions, including most of the southern hemisphere. The mean square of the longitude terms is 80 mgal.,. They are somewhat affected by errors of the base stations ; or instance, it is now probable that there is an error of about -6 mgal. at Washington, which has been transmitted to the whole of the United States, and the comparison of India with Potsdam is still somewhat uncertain. Nevertheless it seems highly improbable that the whole of the terms could be explained in this way. It would be more likely that such errors, if they could have produced the longitude terms, would also have produced some of the five impossible harmonics ; but the method of analysis gave these an opportunity of showing themselves if they were in the data, 2nd no evidence for them was found. However, the failure to find confirmation of these terms in the comparison of surveys suggests that another test is needed. I took from the table on p. 58 * those

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

M.N., Geophys. Suppl., 5, 55-66, 1943.


G

19'

238

Harold Jeffreys

30" squares where a mean anomaly against the International Formula has a standard error <13 mgal. (this limit being chosen so as to include several of the squares 30" S.) ; from each mean I subtracted the corrections to the constant and Pzterms ; and then for each I took the mean of its neighbours. I then formed the correlation coefficient for the squares and the means of the neighbours. The result was : sum of squares, 8354 ; sum of squares of means of neighbours, 2857 ; product sum, -273 ; number of comparisons, 26 ; correlation coefficient, -0.055. Now even a third harmonic keeps the same sign over distances of order 60, and the correlation coefficient should have been positive. Actually the square centred on IZO'E., o"N. contributed -594 to the product sum. But if the low harmonics gave a mean square variation as much as half the T~ variation we should expect the correlation to be strongly positive. A casual inspection of the table would suggest that there is an appreciable positive correlation; we notice for instance a block of positive signs from 30" to 90" N., -30" to +30" E., five negative out of six in the block to the west, and five negative out of six in the block to the east. It is simply that the 30" square centred on ooN., 120"E. contains two 10" squares with enormous positive anomalies, apparently reasonably well determined. But if two 10"squares can produce such an effect on the total the only conclusion is that the harmonics of degrees 2 and 3 cannot be determined satisfactorily without a much more general survey. The tendency has been to make gravity surveys primarily in exceptional areas,with the result that we know quite a lot about gravity near mountains and ocean deeps; but we still know nothing like enough about gravity in apparently ordinary regions, such as Siberia, most of the Pacific, Australia, Africa and Brazil. If in the normal equations (p. 62) * we put all the longitude terms zero, we get
am=

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

+1-5+1.8;

a2,= -1-4f4-6,
a , , = -6-1 f 5.0.

compared with the solution including the longitude terms


am=

+2.5 k 1.9 ;

The former solution somewhat underestimates the standard errors, because the variation contributed by the longitude terms has been removed; if it is treated as random it must be redistributed among the 30" squares, and the second pair of standard errors will be more representative. Then a gravity formula, without longitude terms, will be

k 5*0)p2+5'3p4g=979771'5 f 1'9 +(3444*4


Both these formulae are based on the Potsdam standard, and the constant terms, on the combined standard, will become 979759.0 f 2 ' 1 ; 979758'0 k 2 . 1 .

De Sitter has pointed out that in the theory of the figure of the Earth the second order terms are smaller if we use mean gravity and the mean radius as fundamental constants instead of equatorial gravity and radius. At first he also suggested that the estimated mean radius was less sensitive to changes of ellipticity than the equatorial radius. From 2.2 (14) this appears to be untrue, and de Sitter did not, apparently, maintain it in later papers. The outstanding advantage is in the

Lor.cit.

The Figures of the Earth and Moon


definition of the rotation parameter m (p, in de Sitter's notation). radius and gravity where sinaq$ = 6, the errors in the equations r;=(I -e)d,

239 If r, andg, are

are of the second order in e, m with respect to the main terms. Consequently the accuracy of the determination of Y, and g, practically permits a definitive calculation of m. The international formulae depend on e = 1/297.0 =0.0033670, and we take eff= e -0.0033670. Then if a=6378111(1 +u)metres,
Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

g,= 979*7569(I +w) gals.,


we find

r, = 6370944(1+ u -6e"),

nt=3.497g x 10-3[1 +o.gg77(u - v ) -0-332ge"I. We make two solutions for eNaccording as we reject or keep the longitude terms. Since the uncertainty of e does not affect the second-order terms to the accuracy required, we get
= $m -e

+o.oo00305 =

(ooa52863} 0.0000075, 00052701


f

00033687

0.0000075,

I/e = (296.85, 296.22) f0.66. In terms of the notation of 2.2, e' = ( 164, 236) f75. The uncertainties of u, w are of the order of I O - ~ and I O - ~ and will not make an error of more than I in the last figure. Similarly I e"I is not likely to exceed 0-OOOOIOO and may give an error in m of the same order of magnitude. Hence the determination of nt may be regarded as definitive in equations when its uncertainty is combined with that of e. 5.1. The muss o f the Moon.-This requires a knowledge of the monthly inequality in the Sun's longitude and of the solar and lunar parallaxes. For the former Spencer Jones * gave L = 6".4390f o.oo15(p.e.) ; (1) from a rediscussion of the data ti1 get L =6"-4378f 0.0017 (s.e.) (2) Spencer Jones gets Pa790 A 0.00'1( p . ~ ) (3) for the solar parallax. As the p. e. given might be interpreted by values differing

* Mm., R.A.S., 66,11-66, 1942; M.N., 101, 356-366, 1941.


f

M.N., 102, 194-204, 1942.

240

Harold Jejfreys

by a factor of 3 I take a weighted mean of his solutions (b), (c) (M.N., 101, p. 361)~namely 8".7888 & 0.0011 (s.e.), (4) which gives x2=2.1 on I d.f. The formula given by Spencer Jones for the mass of the Moon is

L=- I*
I

+p 3422.526sin I"'

and from his values he derived


1 / p= 81.271 & 0.021 (p.e.).

The denominator in the last factor is, however, the calculated value of the lunar parallax recommended in de Sitter's 1938 paper (equation (56), p. 224), and is subject to some corrections, one of which depends on the mass of the Moon itself. The full expression is (3422.526 f O * O O ~ ) [ I + 0 ~ 0 0 4 0 + 4~ 033277(u -v) + O * O O O ~ I W +o.ooorrx +o~ooog1t,b], (6) where p-1=81-53(1 +z). (7) The terms in w , x and t,b depend on the hydrostatic theory of the figure of the Earth and replace a direct expression in terms of the ellipticity. But even with allowance for the inaccuracy of this theory it appears to be the most accurate way of estimating the ellipticity. w and z are closely related ; w = 0.67472. From the data given so far u -w is about -0~000040& 0~000016 at most. x and t,b are of order IO-.. . Substituting for L and TO in (5) from (z),(4) we get

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

z = -0-oo31g & 0~00031, p-l= 81.269 & 0.025. The uncertainty comes almost wholly from L and no. The close agreement
with Spencer Jones's value arises from the approximate cancelling of the larger corrections. 6.1. The Lunar Parallax.-This has been derived visually from measures made at Greenwich and the Cape of Good Hope. I t can also be calculated dynamically from the values of gravity at the Earth's surface and the Moon's mean motion. The two determinations have been stated to be in disagreement, but de Sitter's comparison * gives the difference as

V -D

= -0".12

& 0"-06(p.e. of visual value).

(1)

On the face of it a residual of twice the probable error, that is, 1-35 times the standard error, is not very convincing evidence of disagreement. Nevertheless, it is of some interest to see whether the residual provides any useful information with regard to the figure of the Earth. There are several previous discussionst, but revisions of the data make further examination desirable.
. t'C W.D. Lambert, Astr.Journ., 38, 181-5, 1928.
*Bull. Astr. Inst. Neth., 129~61, 1927.

The Figures of the Earth and Moon

241

The observed value gives V'( =3422"*40 & 0.09 -0".0578i?-' -0"*422h1-o"-301h2 - 0 ~ 0 1 0 4 +~0~ . 0 1 3 8 +o"-o168h. ~~ (2) The terms in h,, h,, el, E , represent corrections for departure from the spheroid ; the elevations of Greenwich and the Cape above the spheroid are h,, h, kilometres, el and e2 are corrections to the spheroidal value of -4 for given 4, and h is the correction to the adopted selenocentric distance of the crater observed, Mosting A. de Sitter's standard of comparison is rl =6371.238 km., g, =979.770, 1/e =296.96, p-l= 81-53. (3) With the same standard of comparison the calculated value is 7r'a = 3422".519& 0-ocg+0"-188r1 -f"x68g1 -0"~0138~-~+0"~1708p-~. (4) (The sign of the coefficient of 8p-1 is given wrongly in the next equation of de Sitter's paper.) On examination of the uncertainties, we see that those arising from the four correction terms are of the order of 0.022, 0-002,0.01 and 0-0004 respectively. Hence we gain nothing by trying to adjust g, and p to make ( 2 ) and (4)consistent ; but comparison of ( 2 ) and (4) may yield some useful information about rl and e. Allowance for the correction of -0.0136 gal. tog, and -0261 to p-l decreases the constant term in (4)to 3422".4go,so that the discrepancy, such as it was, is brought down to the standard error by the correction to the mass of the Moon. Adapting to our standard form a' = Ioa(a-6378.000), e ' = 107(e-0-0033523), we have for the visual value 7r'a =3422".324+0~0005070e' & o"-og, (5) and for the calculated value T'C=3422".406+0~00018a' +o-ooo0777e'& o"*oog. (6) We must also allow for the deflexion of the vertical at each observatory. This is directly related to the irregularities of gravity ; from the discussion of survey data and the variations of gravity we may take the deflexion to be 3"-5 for local variations, & 3".0for 1 0 ' variations, and f 1".8for 30" variations, making k 4"-9in all. The contribution to the visual value is +0"-085, so that the uncertainty arising from lack of detailed knowledge of the deflexions of the vertical is comparable with that of the visual value itself. Then we have the equation of condition 0.18~' -o-43e'= -82 f 124. (7) The contributions from the longitude terms are h1= + 0.075 ; ha = + 0.043 ; 1= -3"'O ' 9 +0".4, (8) and the terms in question are -0"-008. I t is surprising that they cancel so closely ; the contributions from the separate harmonics are appreciable. The right of (7) then becomes -90 & 124. 7.I . Derivation of the ellipticity front the precessional constant. If H = ( C -A)/C, where A and C are the Earth's principal moments of inertia, and the interior of the Earth is in a hydrostatic state, it is known that the ellipticity can be calculated from

+'

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

242

Harold Jeffreys

H with higher accuracy than is given by any other method. De Sitter in particular
laid great emphasis on this determination. His last estimate of e-l was 296.753 ? 0.086 (p.e.), and therefore apparently about five times as accurate as any other. Most of the uncertainty comes from the mass of the Moon. Bullard has made a revised calculation using Spencer Jones's determination of the m a s s of the Moon and Bullen's results for the distribution of density within the Earth, and gets 297.338 f 0050 (p.e.).* Having regard to the apparent accuracy, the change is serious. But the Earth is certainly not in a hydrostatic state, and we must consider how far departures from this state may introduce further errors into the estimate. A detailed revision of the Clairaut theory would be difficult because this theory takes as dependent variable the ellipticity of a layer of constant density. If we use a model with hydrostatic conditions below a certain depth and non-hydrostatic conditions above it, the interface will not in general be a surface of constant density in either region and the boundary conditions will be difficult to state. I n addition it is not certain what mechanical equation should replace the hydrostatic equation in the outer region. Accordingly it seems likely that an extremely simplified model may be as good as any that can be suggested at present. We suppose that over a surface of radius aa there is a surface density UP,. The corresponding-disturbance of external potential is

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

u '= irrf up,, 9.3


and the disturbance of gravity over a level surface is

a45.4

g' = $~fa%P,. (2) The elevation of the level surface above a sphere is UlJgaand the ellipticity induced
is

T h e contribution to C -A is given by A(C-A) u'= -f r8 p2, whence

(4)

A( C - A )

- ~7ra4a40,

If k is the coefficient of P, in g '

k = 4 rrfa4u,
and

(7)

3k. Ae= - - ,
28

a4k AHzF.

But f M + g a 2 ; hence

Ma3k A H = --. c g

(9)

* M.N., Geaphys. Suppl., 5,

186-192, 1948.

243 We regard e, Has composed of portions el, Hl related according to the hydrostatic theory, and e,, H, related by (8). Then for small changes it is known that

The Figures o f the Earth and Moon

3 =0*1874Ant +0.8138- AH1


el

nt

Hl '

while

; , H , 2we have, nearly, nearly ; hence neglecting e

Ae =0*1874Ani +o-8138 lA H +0-500 H 2 , e ni H, e Am AH e = 0.1874 - + 0.8138 - - IOOH,. nt H


The discussion of the low harmonics in gravity has led to an estimate of K as +5*9mgal.* Then the corresponding uncertainty in e, as .found from H, i s
f 1OOe Ma2k -- f 6.0 x
Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

IO-~.

cg

T h i s estimate assumes the low harmonics validly estimated. If we reject them


we shall expect smaller values of It, but the stresses indicated by the T~ and T~ variations of gravity reach about 150 x los dynes/cm.', which is not much less than would suffice to maintain the low harmonics even if strength is zero for the inner 90 per cent of the radius. We therefore have no reason to infer that these harmonics are absent. Provisionally we can take the additional uncertainty of e as half that given by (14). De Sitter and Brouwer give the following equation for e on the hydrostatic theory : c1= 296-753[1 -0-1874(~- v ) -0.8138~ +0.1696x -0.8098$]. (1.5) Here u -v = -I m/o*oo344993 = -0~000041 f :o.oooo34, (16) I w = H/0*003279423, (17) a0 is effectively determined by their x, which, with my revised estimate of the mass of the Moon, is -0.00319 0~00031; w = 0.67472 = -0.00215 f 0~00021, (18) H= 0.00327237 f 0~00000069, x = IO~(K -0~00000050) = 0~00018, (19) $ =X -0~00040= -0-00024 f :0~00018, (20) the last two equations being according to Bullard's calculation. Then we find e ' = 297.341 f 0.065, (21)

e 0.00336314 o.oooooo74. (22) According as we reject or keep the longitude terms in gravity we must increase the standard error of e to 0~0000031 or 0~0000061, and we have the respective equations for e ' i e'= +108+31 or e'= +108+61. (23)

* M.N., Geophys. Suppl., 5, 88, 1943.

244

Harold Jefreys
Longitude terms omitted a+o.o47e= I I I & I 18 Survey e= - I43 fZgo Gravity e = + 1 6 4 f 75 Parallax 0.18~-0.43e= - 82f124 Precession e=+108f 31

8.1. Combination o f data.

Calc.
+I12

0-C
I

k
00

+115

+I15

- 32
+I15

+ 49 - 50
- 7

-258

0.8
0 4

02
00

I 4

The solution is
a=

+97f116;

e= +115+29,

whence
a =6378.097f 0.116 km.,

e = 0.0033638& 0~0000029,
r1= 297.28f0.25,

a-0-022e = + 95 f 116 irrespective of e.


Longitude terms retained survey ut+0-047e= I 16f I 18 el=- 11f2w Gravity e = + 2 3 6 f 75 Parallax 0*18~-0.43e= - 9of 124 Precession ot=+108f 61

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

Calc.
+IIO

0-C 6

X
0.0

+I57 +I57 - 49 $-I57

+ 79 - 41
- 47

-168

0.3
11
0.1

0.6

21

The solution is
a=

+99+116;

e= +157&46,

whence
a = 6378.103& 0.116km.,

e = 0.0033680& 0~0000046,
e-l =296.91f0.40,
u -0.022e = +gg 116 irrespective of e. The general conclusion is that the agreement is quite satisfactory whether we keep the longitude terms in gravity and allow for their further consequences or not, In the lunar parallax question we have not allowed for the 8h term, and there is n o reason to suspect systematic error in the astronomical observations. It has been convenient to take together these four types of information, chiefly because in all of them a main source of uncertainty is the irregularity of gravity or the associated deflexions of the vertical. I n the lunar parallax equation the uncertainty arises nearly equally from the possible defiexions of the vertical and the uncertainty of the astronomical observations. It is already known that the data from the Moon will not determine e with a standard error less than about go x I O . Hence there seems to be no point in carrying out further calculations with both sets of values. We have really two superposed probability distributions for a and e, which, in view of the search for a test between them and its indecisive result, we may take to be equally likely., Then the total probability of e follows a law of the form

245 A compact summary is found by choosing a normal law with the same first and second moments as this distribution. The conditions are easily found to b e
e = +(el +e,) f o ;
08

The Figures o f the Earth and Moon

= iu;

+ &4 +a(el -e,)*

and with the actual values found e = 0.0033659 k o.ooooo43, a = 6378.100 & 0.116 km. The uncertainties are nearly independent. 9.1. Data from the M o o n ' s motions.-These have been considered previously.* It is convenient to use the parameter J , which is related to e by the equation
J =e -am -0~0000040. (0 ~, The data up to the present give e =0.0033659 f O - O O O O O ~ whence J =0-0016370f 0*0000043. (21 The lunar data have been discussed also by Spencer Jones t, who has made Borne corrections to the coefficients given by de Sitter. If

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

J'++K'=L', the contributions to the motions of the perigee and node should be

(3) (4)

3897J +380L' -1192K' ;


We put

-3648J -460L'.

(5) (6) K ' = I04k, (7) and the equations become (probable errors being converted to standard errors when necessary) Lunar perigee 0~390j+0~038X -0-11gk- -0.110f 0.056, (8) Lunar node -0.3651' -0.046X = -0.031 f0.052, (9). Inclination of Moon's axis X -0-005k = + 0-ozg f 0.016, (10) Libration in longitude k = +0.60 k0-44, (11) Terrestrial data j = -0.030 f 0.043. (12) The equation from the Moon's latitude term has been dropped, as it is affected by the obliquity of the ecliptic, estimates of which may have systematic errors, according to Spencer Jones, and when the uncertainty of this is taken into account the amplitude becomes too uncertain to add much information. I n the equation from the libration in longitude I have replaced the derivation from f by the direct determination of K' given by the observations. The solution is J = 0~0016370 5 0~0000041, j = -0.030 f 0.041 ; (131 = + 0.034 f0.016 ; L '=0.0003734 f 0~0000016, (14) k = + 0.070 f 0.34 ; K ' = 0~000070 & O - O O O O ~ ~ , (15) xa=o.8'on 2 d.f.

L' =0.00037+ IO-'A,

J =0~0016400 + 10-y,

* M.N., Geophys. Suppl., 4, 1-13, t Cape A m & , 13, pt. 3, 1932.

1937 ; M.N., 101, 34-36, 1941.

246
Then

Harold Jeflreys
e = 0.0033659& 0~0000041, e-l= 297.10 & 0.36 ; e' = 136f 41,

a = 6378.099f 0.116, the uncertainties of a, e being nearly independent. 10.1. General discussion.-The outstanding departure from generally adopted values of the terrestrial and lunar constants is the reduction of Hayford's value of the Earth's equatorial radius by 289 & 116 metres. The final value of e is practically identical with Hayford's. The explanation lies mostly in the abandonment of isostatic reductions, the survey arcs having been reduced to the geoid by the simple geometrical formula. The object of isostatic reductions in estimating the dimensions of the Earth is chiefly to eliminate a possible systematic difference of the curvature of the geoid between continents and oceans and thereby to make land surveys more representative of the Earth as a whole. But if there was such a difference it wculd be associat :d with a systematic differex: of gravity between continents and oceans, which is not shown by the observations; in fact observations suggest a slight difference in the opposite direction. Accordingly it is probable that isostatic reductions have introduced a systematic error. The estimates of the variation of gravity over ranges of 1 0 ' and 30' have permitted a uniform method of combination of the survey data, since they provide estimates of the probable deflexions of the vertical that persist over arcs comparable in length with survey arcs. In combination with the reduction of the radius and the increase in the estimated mass of the Moon they remove the +discrepancy between the Moon's visual and dynamical parallaxes. The equation provided by these is a useful supplement to those given by the surveys. The data do not decide on the reality of the additional terms in gravity represented by harmonics of degrees 2 and 3 . They can be equally well reconciled whether these are present or not, and the final values of a, e represent a compromise between the two hypotheses. De Sitter's equation determining the ellipticity from the precessional constant has been corrected for the probable departures of the Earth from a hydrostatic state. The increase of its apparent uncertainty is considerable, but it remains the most accurate single equation for the ellipticity. The various comparisons of data, taken together, give x2 =6.3 or 8.7 on 14 d.f., according as the longitude terms in gravity are omitted or retained. Both values are on the small side but not seriously so. Some of the estimates of uncertainty of the equations of condition, however, are not based on least square solutions but o n personal impression, and it is possible that they are a little too high. This refers in particular to the correction to de Sitter's equation, and to the calculated values of the motions of the Moon's perigee and node. The lunar nutation has not been considered in the present paper. Though the observed value departs appreciably from that calculated from the lunar inequality and the rate of precession, its apparent precision is not high enough to give important supplementary information about the Moon's mass and the precessional constant H. Lamb * remarks that the correction to the calculated nutation for fluidity of the core would be appreciable, and I have confirmed this 7, but the actual value af the allowance has not )-et been calculated.

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

*Hydrodynamics, p. 727, 1932. M.N., 108, No. 2, 1948. For summary see: M.N., Geophys. Suppl., 5,

282,

1948.

The Figures of the Earth and Moon

247

1 1 . 1 . General adjustnzent.-Having arrived at a pair of values of a, e that reconcile the available evidence we now proceed to derive values of the other constants compatible with them. We find rl/a= 0.9988771 & 0~0000014, Also r, = 6370.937 & 0.116 km. co =7.~921152x ~ o - ~ / s e c . w has been corrected for precession by de Sitter. We use the theory of the second-order terms that makes the geoid an exact ellipsoid. Then when sin24 = 8, P,(sin 4) = -& ;

g , = 979'7720 -0.0021 -0.0134 = 979-7565f0-0021 gal., nt = 0.003449787 & 0~000000063. If


g =go(I /3 sin24 y sin2z+), = zm -e + 0~0000305 = o~oo52891& 0~0000041,
y=

+
+

Downloaded from http://gsmnras.oxfordjournals.org/ by guest on July 22, 2013

-0~0000059,

g, =gO(I 0~0017630 & 14 -0~0000052), go= 978.0373(1 f :0~0000024).

In de Sitter's formula in terms of e for the dynamical lunar parallax


Sr, = -0.301 & 0.116 ; Se-l= + 1-14f 0.36 ;
and
T'(

&-I=

6gl = -0.014 f :o.002 ; -0.261 f :0.025


-O".O#

=3422".519 & O".OO9 -0".054


= 3422"*422 & O"*OZ4.

+ O " * I I G -O"*OI5

fO".OZI & o"-ooz f0".005 &.0"-004

The observed value is


T ' ( = 3422".40 & O"'I2

-0".06 & O"'O2

=3422"'34 & O".I2.

The difference is well within the standard error. T h e changes in a and p have decreased it, that in e increased it. The weighted mean is T ' ( = 3422"-419& 0.024. Substituting in the equation for the mass of the Moon, we get
w = -0~00208 0-00021, H =0.00327260 f :0~00000069. To this value of H corresponds on the hydrostatic theory c1 = 257.324 f :0.065 ;
p-l= 81.278 & 0.025, x = -o.oogog 5 0~00031;

but as already stated, I prefer not to adopt this value. L' -0.oo5K in the final solution is practically identical with the value given by the inclination of the Moon's axis alone, and the observed value of the latter, with its uncertainty, stands unaltered. Also
f = "-- K' - 0.81 f 0.09.

L'

The greater part of the uncertainties of the motions of the perigee and node comes from the calculation, and the analysis can do nothing to correct the observed values.
160Huntingdon Road,
Cambridge : 1948 Febrwry 6.

You might also like