]\gt ; PHILOSOPHICAL TRANSACTIONS .
I. Further Consideration of the Stabitity of the Pear-shaped of a Mass of Liquid .
By .
G. H. DARWIN , K. C.B. , F.R.S. , ofessor of , and Felon ) of Trinity College , Receiyed October 29 , \mdash ; Read December 12 , 1907 .
INTRODUCTION .
IN vol. 17 , No. 3 ( 1905 ) , of the ' Memoirs of the Imperial Academy of St. Petersburg , ' M. LIAPOUNOFF has published an abstract of his work on figures of equilibrium of rotating liquid the title " " Sir un Problem de Tchebychef\ldquo ; In this paper he explains how he has obtained a rigorous solution for the figure and stability of the pear-shaped figure , and he pronounces it to be unstable .
In my paper in the ' Philosophical Transaction had arrived at an opposite conclusion .
The stability or instability depends , in fact , on the sign of a certain function which M. LIAPOUNOFF calls , and which I denote , where is equal to % M. LIAPOUNOFF tells us that , after having seen my conclusion he repeated all his computations and confirmed his former result .
He attributes the disagreement between us to the fact that I have only computed portion of an infinite series , and only used approximate forms for the elliptic rals in the several terms .
He believes that the inclusion of the neglected residue of the infinite series ] lead to an opposite conclusion .
In my computation the function is decisively negative , and being numerically greater than 2 which is positive , of the two is negative .
The inclusion of the neglected residue undoubtedly tends to make this whole function positive , but after making the revision , explained in the present paper , it remains incredible , to me at least , that the neglected residue can amount to the total needed to invert the It may be worth mentioning that in revising my work I notice that owes its negative sign to the term .
This term arises from the energy of the double layer , called .
It comes from the portion of the term , which gives rise to a in with a negative sign .
This * Series , vol. 200 , pp. VOL. CCVIII .
term involves under the integral sign the factor , all the other factors being positive .
If we attribute to and to various values between and zero , we see that in part of the range the factor is positive and in other parts negative .
A general inspection does not suffice to determine whether the positive portion outweighs the negative , as in fact it does .
Therefore , in order to feel abundantly sure that no gross mistake had been made , I computed by quadratures the eight constituent integrals involved in the final result , and confirmed the correctness of the value found by the rigorous evaluation .
The analysis of the investigation has been carefully examined throughout , and I lave , besides , applied the same method to the investigation of MACLAIJRIN'S spheroid , where the solution can be verified by the known exact resul As a further check , the formulae of the paper on the Pear-shaped Figure have been examined on the hypothesis that ellipsoid of reference reduces to a sphere .
The several terms correctly reproduce the analogous terms in the paper on MACLAURIN'S spheroid , but in effecting the comparison it is necessary to note that the variable of the Pear-shaped reduces to , whereas in the paper on MACLAIyRIN'S spheroid the esponding variable denotes , where is radius vector and the radius of the sphere .
Dissent from so distinguished a mathematician as M. LIAPOUNOFF is not to be undertaken lightly , and I have , as explained , taken especial pains to ensure correctness .
Having made my revision , and completed the computations as set forth hereafter , I feel a conviction that the source of our reement will be found in some matter of principle , and not in the neglected residue of this series .
I can now only express a hope that some one else will take up the question .
In the revision of the computations , the methods now used are much better than the old ones .
In as far as this paper is a mere repetition of the former work with improved methods , the results will only be stated in outline , but I now show how any of harmonic functions may be computed without approximation , and how the functions of the second kin may be found rigorously .
The Cambridge University Press is now bringing out a collection of my mathe , and when we come to the paper on the stability of the pear-shaped , the new of computation will be substituted for the old .
This } ) apel .
is to the former one on Stability , and it will only be intelligible in connection therewith .
As before , I refer to the papers in the ' Philosophical Transactions ' " " Harmonics\ldquo ; and the " " Pear-shaped Figure 'Amer . .
Soc. ' 1903 , vol. 4 , p. 113 , on " " The Approximate Detcrminatiou of the Form of MACLAITRIN'S Sphcroid ( further notc on the same )iect , recently sent to the same Society .
'Phil .
Trans , vol. 200 , p. 251 .
, p. 461 , and vol. 198 , , p. 301 .
STABILITY OF TH1 OF { 3 S1 .
Tloe Rigorous Expression forthe Functions .
Rigorous forms have been found in previous papers for all the hal.monics of orders up to the third incIusive .
For harmonics of the fourth order rigorous algebraic forms may be obtained in all cases except when , 2 , 4 , but these are exactly the cases to be considered in this investigation .
We have , then , to show how these functions of higher orders may be evaluated rigorously for an ellipsoid of known ellipticity .
The only case required is that in which both and are even , and although all the forms might be evaluated by processes similar to those indicated below , I shall confine myself to this case .
We have seen in " " Harmonics\ldquo ; that if denotes of this paper , and denotes .
It is well known that and Hence we may clearly write in the form Since when is not zero , and when 6 is zero , it follows that .
For the zonal ha1monics this gives .
The determination of the other 's depends on that of the , which we shall considel later .
Another form of will be useful , .
: It is obvious that \amp ; c. , \amp ; c. , \amp ; c. Thus , when the 's are computed it is easy to obtain the \amp ; c. We know that ( the cosine function of ) is the same function of that is of , except as regards a constant factor .
Hence it follows that where is a constant factor .
Now I desire to define and exactly as in " " Harmonics This definition has already been adopted as regards , but it remains to adjust the constant so as to attain the same end as regards .
When and even , was defined thus : cog Since it follows that the term independent of in is The term in in is The term in in is so forth .
In accordance with the definition to be adopted , these terms in the three cases respectively are : 1 , .
Hence must be chosen so as to fulfil that condition .
Pursuing only the case of in detail , we have STABILITY OF THE PEAR-SHAPED FIGURE OF EQUILIBRIUM .
If , then , we must have , \amp ; c. Thus when , \amp ; c. , are found , it is easy to compute , \amp ; c. , and , \amp ; c. Our formulae tend to the differences between large numbers , and this defect becomes more pronounced as the order of harmonics increases .
The fault is mitigated by using the forms the case of a lower harmonic , however , such as the fourth , we may just as well use the form for , \amp ; c. , and powers of We must now show how to co mplete the evaluation of 's for the zonal harmonics .
It appears , from p. 486 of " " Harmonics that , when is even , we have to solve the equation ' ending with , where .
We are to take that root which vanishes when vanishes .
Although the equation for is of order , yet at least for such an ellipsoid as we have to deal with , it is very easy to solve it by successive rapid approximations .
It is clear that we may write the equation in the form An analytical approximation is found by neglecting the continued fraction in the second term on the left , and we then obtain If this value of is used in computing the first term of the continued fraction , and if the quadratic is solved again , we obtain a closer approximation .
We then use the second approximation and include one more term in the continued fraction , and proceed until no longer changes .
It is shown on pp. 486 , 487 of\ldquo ; HarmoIlics\ldquo ; that It be remarked that the ctor 2 in each of these equations on the left , excepting in the first one ; also we are to take In the course of the successive approximations for the determination of , each of these fractions is naturally .
evaluated .
Therefore it is only necessary to extract values found in the course of solving the equation for As a verification , which shows whether the equation has been correctly solved , we have It is now obvious that we are able to find all the 's in terms of , which is unity .
We then multiply each by its appropriate power of or that is to say , we ' form for , 2 , , , and introduce the results into the formula for .
A closely analogous method enables us to find all the other types of function for an ellipsoid of known ellipticities , but , except for certain harmonics of the fourth order , it is possible to obtain rigorous analytical solutions .
Approximate analytical forms are given in " " Harmonics and the approximation may be carried further if desired .
The following tables give the coefficients in the several functions for the critical Jacobian ellipsoid with which we are dealing:\mdash ; STABILITY OF THE PEAR-SHAPLD FI UR OF QUILIBRIUM .
* The second ) co ) with } , } the npper bign the lower for In the cas of , I had in the inadx ntly ch:thgod the ) of ?
of course , introducing any error , since they as a product .
These functions are only givcn in their pproxjmate forms .
As it is desirable to use the other in the higher zonal harmonics , I the coefficients , \amp ; c. , in these cases .
It will be noticed how much smaller the numbers involved .
COEFFICIENTS of Terms in when expressed in Sines and Cosines .
S2 .
The Rigorous Expression for the rmomcs of the Second The integral denotes , and denotes .
Thus is , in fact , the harmonic function of the second kind .
is clearly determinable from In the original paper was found by quadrature , and this defect in my procedure is referred to by M. LIAPOUNOFF as a cause possibly contributory to the discrepancy between our results .
Quadrature was not , perhaps , a very satisfactory method , and the defect will now be made good by finding these integrals in terms of the and elliptic integrals .
It will appear that my former results were sufficiently near to the truth for practical purposes .
The functions or are of eight types , determined by the oddness or evenness of and , and the association with a cosine or sine function of .
In " " Harmonics\ldquo ; the types are indicated by combinations in groups of three of the four letters , S\mdash ; denoting Even , Odd , Cosine , Sine ; for example , OES means odd , even , associated with a sine function .
All the roots of the equation or are real , and when the form of the function has been determined by the method of S1 the equation may be solved .
Hence these functions are expressible as the products of a number of factors ; and it is to be noted that it is not necessary to adopt the same definition as in " " Harmonics because the function may be multiplied by any constant factor , without affecting the result .
For brevity , let , where ' is of " " Harmonics The parameters , and , as elsewhere , define the form of the ellipsoid .
An alternative notation will be needed , in which we write STABILITY OF THI PEAR-SHAPED FIGUPffi OF At the surface of the ellipsoid , and we shall , as before , write .
At the surface of the ellipsoid we have then In this notation A consideration of the eight types of harmonics shows that they may be written as tollows : Type .
EEC , ?
, , OOC , , , OEC , ?
, OES , , , .
Using and generically for any one of these and .
the corresponding function of the other kin , we have $ Or , changing the variable of integration to To effect the ration the reciprocal of the square of $ must be expressed in partial fractions .
Inspection of the eight forms of functions shows that with appropriate values of to be given hereafter .
In every case but that of OE8 some or all of are infinite .
VOL. CCVIII .
\mdash ; A. In terms of This has to be divided by and integrated , and the result will be expressible in terms of the elliptic integrals Accordingly we require certain integrals , which are given on p. 313 of the " " Pearshaped Figure but in somewhat different fo Here and elsewhere denotes , and denotes The als needed are as follows:\mdash ; ( 1 ) The last two of these admit of considerable simplification , as will now be shown .
It is proved in the " " Pear-shaped Figure\ldquo ; that the elliptic integral disappears from the expression for $ for all cases up to the third harmonics inclusive .
I have also proved numerically that for the Jacobian ellipsoid the like is true for all the even zonal harmonics up to the tenth inclusive , and for the tesseral harmonics and .
It is thus certainly true in all cases used by me , and I do not care to spend perhaps much time in proving algebraically the general truth of the law .
The last two of our integrals will occur in the form and I that the coefficient of in this expression always vanishes , as is to be true in all cases actually computed .
STABILITY OF THL -SHAPLD FIGUlffl QUILIBRIUM .
Hence Now the coefficient of in this same combination of is In this we may substitute for its value , and thus find .
This expression together with ( 1 ) give all the required integrals , and it only remains to tabulate for the several types of function .
For the sake of brevity I write , the factor which would vanish being in each case omitted .
When there is one is to be interpreted as unity .
TABI of Values of ' EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS co EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS armonic .
EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS EES OOC OOS OEC OES EOC EOS In all the types with the omission of the term which would be infinite .
We have generally .
Hence by logarithmic differentiation of the expressions for the several types of we find results given in the following table :\mdash ; TABLE of the B Integrals .
In the case of the zonal harmonics , is always less than unity ; for harmonics of rank 2 one of the is greater than unity and the rest are less ; for rank 4 two of them are greater than unity and the rest less .
For the zonal harmonics there is some gain in simplicity by putting We then take the equation and find all the roots , say If we solve the corresponding equation for the tesseral harmonic of rank 2 , we find one root for to be negative .
If this root corresponds to STABILITY OF TH PEAR-SHAPED IGURE OF LQUILIBRIUM .
, we must put , so that Similarly for the harmonics of rank 4 , two roots correspond with imaginary angles , and so forth .
Subject to this explanation we may now .
the roots as defined by Since , we have , and If we write ( factors ) , our former may be written in the form , and the several coefficients in the expression for be expressed as trigonometrical functions\mdash ; some of which may , however , be hyperbolic .
We thus have This formula agrees with the result given for in S4 of " " The Pearshaped Figure although the formula is there expressed in terms of , and is replaced by its equivalent In the case of the even zonal harmonics of order , all the 's are real angles , and it facilitates the solution of the equation for to note that , with rough approximation ( improving as the order of harmonic increases ) , The following numerical values apply to the critical Jacobian ellipsoid:\mdash ; For the fourth harmonic ; the rough approximation gives and For the sixth ; the ough a ) proximation being For the eighth ; the approximation being For the tenth ; the approximation being The values of the several a found by quadratures were in every case too small ; the correct values are given in the table below .
I find that for quadratures gave too small a value by a part ; for % by a part ; for by an part .
The method which I have given above fails for the tenth zonal harmonic , unless we use logarithms of more than seven places ; and it is not worth while to undertake so heavy a piece of computation .
I conclude by extrapolation that for quadrature ( carried out on exactly the same plan as in all the other cases ) gives too small a result by a part of itself I therefore augment in this case the result of the quadratures and find ; this enables us also to compute The following table gives the results of the whole TABLE of Logarithms of and Integrals .
S3 .
Note on S 15 .
" " The of \ldquo ; It has been found best to make some changes in this part of the work .
The integrals to be evaluated are denoted The integral may be made to depend on ( which is the same as of the original paper ) , and therefore I only evaluate the latter .
These integrals were originally found as the erences of certain other functions , but it is not hard to give formulae for finding them directly .
I have done this and recomputed the whole series of values .
o1lly errol ' of any mome1lt which I ) found in my previous vork is that in some of the cases I had forgotten to introdnce the factor in some of the after effecting the dratures .
The error in my results from this oversight was fortunately not scrious .
There is a misprint on p. 286 .
The function should be scc4 see2 STABILITY OF THE PEAR-SHAPED FIGURE OF EQUILIBRIUM .
The series of values were also computed independently from the series given on p. 537 of ' ' Harmonics A little consideration will show that the differences of the series of functions with the signs of the odd differences changed , gives the series of functions Stated analytically A table of the natural numbers ) is given , but the table of differences is not reproduced .
The following is the result of the recomputation:\mdash ; of the Functions ( Natural Numbers ) .
TABLE of Logarithms of the Functions .
The results in the original paper were not so accurate as I had thought they were .
There was a mistake in the differences which give the series , affecting the values from onwards , but as no use was made of the series as published , the mistake did not affect the final result .
S4 .
Note on SS16 , 17 .
Th.e An improvement has been made in the method of computing all these .
The functions to be integrated were written in every case with a common factor ; now this is equal to .
In consequence of the substitution of this value for the common factor we are able to obtain the result as the sum of , instead of the difference between , two numbers .
Another consequence is that we can dispense with the series of functions denoted and A single example of the way in which this change is applied will suffice .
If we write , and denote by , , corresponding functions with , S ' for , S and in place of it is easy to show that The computations as revised gave Taking , I found This differs by 4 in the seventh place of decimals from the old value .
In evaluating when we use the form of involving , \amp ; c. , we have to put and are easily found from , S. I then write , \amp ; o. , and , when denotes the order of the harmonic concerned , write , These functions are then combined to give the required integrals .
STABILITY OF THE PEAR-SHAPED FIGURE OF EQUILIBRIUM .
A similar notation enables us to evaluate , which is given by so that appropriate forms being attributed to , .
The result of the procedure sketched is the following series of values , in which only the cases , 4 , are derived from approximate forms .
For the sake of comparison I add the approximate yalues of as computed from the formulae in " " Harmonics In the cases , 2 , the approximate results , derived from that paper , are multiplied by such factors as to make the approximate formulae for and agree with the exact one when ) ; and for and to make the coefficients of and agree with the exact formulae .
TABLE of Logarithms of Note that is negative while remains positive .
The calculation of the integrals for and was very laborious , and as the results tend to present themselves as the differences between large numbers , it is difficult to obtain accuracy with logarithms of only seven places of decimals .
The integrals are much the most troublesome ; indeed I do not claim close accuracy for ; and as it appeared to be impossible to compute to nearer than 10 per cent. from the formula , I computed the several constituent integrals for the tenth harmonic by quadratures and combined them to find .
The results derived from the approximate formulae of " " Harmonics\ldquo ; are given for the sake of comparison .
They clearly give somewhat too large a value for the higher harmonics .
I believe and to be nearly correct .
VOL. CCVIII .
\mdash ; A. If allowance be made for the difference of definition adopted in this paper from that used in " " Harmonics\ldquo ; as regards the second zonal harmonic , it will be found that , when set out graphically , fall into an evenly flowing curve .
The corresponding test for the 's is not quite so convincing , but there is nothing which implies a mistake .
The values of fall well into line , and so do but there is a gentle elevation in the neighbourhood of .
In consequence of this slight wavin of the curve I recomputed the whole again independently , after it had been recomputed and verified once , and special attention was paid to and S5 .
Final Synthesis of Nurnerical Results , Conclusion .
The several numerical values are combined just as in the original paper , but the numbers , of course , differ a little from those obtained before .
The following table gives the final stage , inclusive of the additional terms now computed:\mdash ; I then find .
From these we find ; whence 0553908 0008037 0316007 Denominator 0229864 STABILITY OF THE PEAR-SHAPED FIGURE OF EQUILIBRIUM .
The Numerator divided by the Denominator is , whence With , from S7 of the " " Pear-shaped Figure , \ldquo ; we have Thence we find These values differ sensibly from the old ones .
The moment of inertia with is given by The moment of momentum is As before , we find the pear-shaped figure to be stable , because the moment of momentum is greater than that of the critical Jacobian , provided that the infinite series does not amount to too great a sum .
If be the uncomputed residue of 2 , I then find , as before , that the moment of momentum is The coefficient of will be positive and the pear stable , provided that or The eighth zonal harmonic gave a contribution of 0000196 , and the tenth of 0000063 .
These are respectively a ninth and a twenty-seventh of the critical total .
The pear is then stable unless the residue of the apparently highly convergent series shall amount to more than 27 times the value of the last term computed .
M. LrAPOUNOFF claims in effect to prove that this is the case , , but to me it seems incredible .
I look for the discrepancy between our conclusions some other direction ,

