]\gt ; .
On the Eqnilibrinm of Rotating Liquid Cytinders ' . .
H. JEANS , B.A. , Newton of Collegc , bridyc .
by Professor G. .
DARWIN , F.R. Received Iarch 20 , 190 S1 .
As a preliminary to cking the problem determining the urations of a rotating mass of liquid , I was led to consider whether some method could not be devised for calculating the potential of a homogeneous mass in a manner more simple than that usually adopted .
What was obviously was a calculus enablin us to write down the potential of such a mass by an algebraical transformation of the equation of its boundary , instead of by an integration extending throughout its volume .
There was found to be no difficulty in reducing the calculation to a of algebraical transformation , but in three-dimensional the transformations required were , in general , as impracticable as the integrations which they were intended to eplace .
This was because the transformations depended upon continued application of the formula which expresses the or powers of spherical harmonics as the sum of a series of rmonics .
As soon , however , as we pass to the consideration of two-dimensional problems , spherical hal.monicsllay be replaced by circular functions of a single ) The transforn ] tion vow geable , and this reason the 1 ) resent 1 ) ) deals only with two-dimensional problems .
The first part of the ) contains a short sketch of a of two-dimensional potentials .
I have , however , confined myself stl.ictly to such problems as are required for the solution of the main problem under discussion , namely , that of the rotatin liquid ; the method does not to be one of general applicability .
POTENTIALS OF ENEOUS S 2 .
We shall suppose the cross-section of the cylinder of which the potential is required , to be bounded by a single continuous curve enclosing the origin .
Let ( 322 .
) .
J. H. ON THE ] V the potential of this cylinder , supposed to be of homogeneous matter of density The value of V must be finite and continuous at points except -ity , and its erent i coefficients must ) finite and continuous at ) points .
Also V tisfy all points outside at all points inside .
At infinity V must vanish , except for a tern ] proportional to These conditions suffice to determine V uniquely .
For if there were two distinct V and , the function ( V \mdash ; V ' ) would satisfy at all points of space , would be finite and continuous , together with its differential coefficients , at all points of space , and would vanish ab infinity , except for a term proportional to only solution satisfying these conditions is known to be , hence function satisfying the conditions laid down above must be the potential of which we are in S3 .
Let us use polar co-ordinates in conjunction with orthogonal co-ordinates , ?
/ , let us also introduce variabIes , defined by us to the curve itten in the form . . . . . . . . .
( 1 ) , and let us imaoineo that this equation is solved explicitly in the . . . . . . . .
( 2 ) .
In general will be a function of , and the equation ( 1 ) may , and probably will , be tisfied for .
values of and than those which occur on the surface Let us , however , supl ) that we one of such that this value is a -valued function of ) ) of , and is equal to .
Let us suppose that we have succeeded in expanding this of 1n a series of and descending powel .
S of these each supposed COllvergent at every point of S. Let us . . . . . . .
( 3 ) , . . .
( 4 ) , . . .
( 5 ) .
We shall consider only the case in which the surface has the plane as a plane of met r. In this the equation .
] nains the equation to the curye aftel the sig of is changed , and we therefore have as a second form of this equation , ( 6 ) .
There is efore a solution of this equat i as a function of in the form . . .
( 7 ) where by , 4 and 5 .
LIQUII ) Since we have } ) posed equation ( 2 ) to epresent the complete solution of equation ( 1 ) , the equation . . . . . . . .
( 8 ) must represent the complete solutio of equatio1l ( 6 ) , if the of remains unaltered .
Let us suppose that yalues of the multiple-valued function what same ) are exhibited on the appro- priate RIEMANN'S surfirce .
The locus of points at which this function is equal to ( also a definite and unique of position upon RIEMANN'S surface ) will be the complete system of points satisfying equation ( 6 ) , and will therefore include the curve S. This curve will not , howeyer , be drawn upon a } ) lane , but upon a surface .
Now so long as the cnrye @ does not ) ossess a cusp branch point we have at oint of hence it follows thatJlO ) Rllch 1 ) of the lie the curve .
Let the be a single closed curye surrounding the origin , and it , that we can always so arrange the surface that no branch line intersect the curve S. In this case it is for a solution of the folYn of equation ( 6 ) to satisfy at every poin ' the curve , the curve now being regarded closed curve lyinoo on one sheet of a RIEMANN'S face .
By simply the axes of and we call imagine the function epresented on the sanJe RIEMANN'S surface , we see that at every point on the old curye we shall have .
We have therefore proved that if a pal.ticular solution of the of equation ( 3 ) be found to esent the curve , then solution ( 7 ) will epresent the same curve .
We have also seen that if a family of curves is descl.ibed by continuous deformation starting from some curve , the general solution can ) eal.rived at continuous variation starting from the solution for the curve , so of the 'possesses cnsp or .
For present purpose it is sufficient to have proved that if we have , at every point of the surface . . . . .
( 9 ) , then we have also , at every point , ( 10 ) .
Let us noly introduce a new function defined by ( 11 ) , where is a constant .
We have by differentiation , ( 12 ) , ( 13 ) , and therefore at the surface , by equations ( 9 ) and It follows that has constant value the surface , and this , by a suitable choice of the constant , may be taken to be zero .
Also it follows that at all points on the surface , where denotes differentiation with respect to the normal .
We therefore have at every point of the surface . . . . .
( 14 ) , . . . . .
( 15 ) .
.64 .
Let us denote the potential at a point inside the surface by , arid that at point outside by .
Let us examine , as a trial solution for V , ( 16 ) , ( 17 ) .
Since , the greatest value of at any point on is equal to the greatest radius , say , which can be drawn from the origin to S. Since the series is , by hypothesis , convergent for all points on the boundary , it follows that the radius convergence of the power series must be greater than , and hence that is convergent at all points inside surface S. Hence also must be convergent at all points inside S. The same is obviously true if is written instead of Hence it follows that if is defined by equation ( 16 ) , then and its first differential coefficients will be finite and continuous at all points inside S. In a precisely similar manner it can be shown that if is defined by equation ( 17 ) , then will be finite and continuous at ] points outside except at , and bhat the first differential coefficients of will be finite at all points outside S. From what has been said it follows that and finite at the boundary .
The value of at the boundary is , and this vanishes by equation ( 14 ) .
Hence V is finite and continuous adl points of space ( except infinity ) .
Since the series and have been supposed to be convergent on the boundary , it follows that the first differential coefficients of must be convergent on the boundary , hence that theSe ( coefficients finite at all points space .
At the boundary , by equation ( 15 ) .
Hence it follows that the first coefficients of V are finite ( ' at all ) oints of space .
At a point inside and similarly at a point outside OF ROTATING IQUID C Lastly , by actual tegration , we find as the value of from equation ( 17 ) , ( 18 ) .
The tell lllay be replaced by ; hence V vanishes at infinity , except a to We have now seen that V satisfies the conditions which must be satisfied by a potential ; ] the potential will be given equations ( 16 ) .
S 5 .
Calculated ation , the value of is where the integration extends throughout the -section of the cylinder , and is the distance of the point at which the potential is being evaluated , from the oiut of the cylindel .
We have .
Now if , we have .
, and hence Upon integl.ation we obtain .
The cylinder being symmetrical about the plane , we have and hence equation ( 19 ) becomes .
Let the area of the cross-section be , and its centre of gravity be at so that ; then equation ( 15 ) becomes . .
Comparing this with equation ( 1s ) , we find that . . . . .
( 21 ) , . . . .
, ( 22 ) .
These last equations enable us to find , by a process of mation , the ctioD and the avity of the is J. H. JEANS ON QUILIBRIUM given by equation .
I have made no attempt to discuss these from the point of view pure mathematics ; we thel.efore pass at once to some simple trations of theoly .
Cylin der . .
Let us attempt to find the potential , cross-section , and centre of gravity of the cylinder , of which the cross-section is the circle . . . . . . .
, or , in polar co-ordinates , . . . . . . . .
The equation tc this curve ( 25 ) , or , solving explicitly for .
Now the minimum value of which can occur on the curve is , and this is than if .
In this we have and therefore the solution for is We accol.dingly have . . . . .
( / 26 ) . .
If we dete1mine from the condition that and shall become c-qual at the boundary , we have the known values for .
The equations for A and reduce to , and these , , are obviously in agreement with the known results .
Elliptic : 7 .
us next find the potential of the elliptic cylinder . . . . . . . . .
( 28 ) .
The equation to the surface is or , explicitly , . . . .
OF LIQUID ) At points the } ) the min nun value of is ; hence , gent expansion for is .
expansion we obtain at once , ( 31 ) .
' We call o in once ; if we equire its ) in finite we may roceed as follows:\mdash ; equation ( 30 ) , function of function of same function of . . . .
The results obtained agree the known results if aken of the same .
There is a second solution , obtained the of one of these roots , and this corresponds to a mathematical ) of which one is tive .
The full nificance of will ) later .
ansion in of a S 8 .
When the equation to the ce is of a degree hioher b second , it will not , in general , be possil ) to a solution in finite terms of the form ofequation ( 29 ) .
Suppose , howeyer , that the surface forms one of a ) of , the lnily being described by the variation of a parameter , and let this family be chosen so that } surface is one for which the solution is we the value of when , we shall assume that it will be ) OhHible to find the value of in a series of ascending powers of , and the equations the various coefficients of will ] lave a unique solution .
The solution for will be a function of and .
The maximum values for and will therefore be functions of , and will be expressible as series of ascending powers of .
Substituting these values respectively in and , we obtain two series in ascending powers of .
Let the radii of convergence of these two series be and espectively .
Then is convergent inside and over the boundary provided is convergent outside and over the boundary provided .
The solution for found by expansion in powers of will satisfy the conditions which it is assumed to satisfy in S3 so as is less than the smaller of the two quantities and S9 .
Let us suppose our solution in of to have been obtained .
The values of and can be as series of the yariables or , and as such will have cles of gence i slJace , having the origin as centre .
Let .
be convergent inside circle of adius p , and outside a circle of radius .
When , R. ' ; the circle being wholly inside , and the circle wholly tside the S. When ches the value , the circle ; for values of than , the circle either intersects , or lies outside , the surface S. Suppose that : then for values of such that the circle lies wholly outside , while ] does not lie ] olly inside S. Thus if .
and are defined by the same power series which have been found for them for values of less than will be convergent at all points inside , but will not be convergent at points outside S. We have now to inquire whether it is possible for to give the true value of the internal potential , even when the sel.ies found for fails to represent the external potential .
Let the equation to the sul.face be written in the ( 33 ) , and let the solution be written in the form . . . . . . .
( 34 ) .
We shall only consider case in which , so that former furlction is , by hypothesis , convergent at the boundary and at all points inside it .
We take to meall the same thing as our former , and suppose defined by equation ( 34 ) .
Now the value of cannot become infinite at any point on the boundal.y , since we suppose the bounding surface not to extend to infinity for any value of .
Hence it follows our function will be finite at all points of the boundary so long as .
Moreover , in the in which our formel is convergent region , must become identical with , and will therefore be finite .
The function will therefore be finite at the surface , and will vanish at infinity to an .
at OF ROTATING LIQUID least equal to 1 This function is real at eyery point of space , and is a solution of LAPLACE'S equation .
Hence the function is finite at every point of space .
It is therefore clear that if , in equations ( 16 ) and ( 17 ) , we replace by , , and make the corresponding changes in , , we shall have a tion which satisfies all the conditions of the ) , subject to the single condition that is less than .
At infinity will be capable of expansion in the form of equation ( 5 ) , , and we now see that equations ( 21 ) can be obtained in the same manner as before .
lt is clear that the values obtained for V , and will be the true , even if they have been obtained by the use of divergent series , provided only that ) series remains convergent up to boundary .
In the case in which , a similar proposition is .
values of such that 10 .
llow considel the case which is greater than either .
Let the values of which been found for values of less than either or be extended , by a process o ' continuation to points outside their circles of convergence , and let values so obtained define the functions nd V These functions will tain infinities , the position of these infinities depending upon the value of When , all the infinities of lie outside ; all the infinities of lie inside S. Let us suppose that up to some value of , say , no infinity crosses , but that ( if possible ) at one of these ities is found on the boundary .
The values of and are functions of satisfies the requisite equations from until is equal to the smaller of the values Hence it mnst continue to satisfy for all values of , until the condition found in 3 is violated , i. e. , until of is such that the curye possesses a cusp or branch point .
Also and satisfy the requisite conditions of finiteness , uniqueness , and uity until reaches the value .
Now as approximates to from the direction in , the value of V at same point of the boundary ( viz. , the point at the infinity occurs when ) will increase indefinitely , becoming ultimately infinite when .
This value of will , howevcr , giye the true solution for all values of less than , and there will be a superior finite limit to the value of V. It therefore follows that there can be value infinity the , and the of and found by " " colltinuation\ldquo ; of the power series will give the true values of and until the whole solution is invalidated by the occurrence of cusp branch point .
Summing up , it ) ears that we may neglect the question of collvergelcy of series altogether : so long as the values obtained either or are possible values , they must be true values .
But care must be taken not to pass through values of the pal.allletel .
such that possesses a cusp or branch point for these ] .
J. II .
THI S 11 .
Illustrations of these remarks are afforded by the examples of SS 6 and 7 .
In tions ( , ( 24 ) , let us egard c as a variable parameter , so that , as , the equations represent the different members of a family of circular cylinders .
seen the sel.ies obtained for only remains convergent so long as Equations ( 26 ) and ( 27 ) represent the values of and expanded in powers of Now the of is convergent , no matter how great may be , hence we know represents the true value of for all values of .
The value of has as circle of ence the circle , and this intersects the boundary as soon as attains the value .
Hence value obtained for will fail to give a true solution as soon as exceeds the value , although it will always giye the value of at points outside the circle .
Again , in , let us as a variable parameter .
The value found for is convergent for all values of , and therefore resents the true solution for all values of , provided that we start from a true solution , and do not pass through a value of at which a cusp occurs .
Under this condition the series for will give the true value of at all points at which it is , and the expression given in equation ( 32 ) will give the true value at all points , this the expression for which would be found by " " continuation\ldquo ; of the series , or ( what is the same thing ) by the methods of S 9 .
The does not possess a cusp except in the critical cases of In the former the ellipse reduces to a pair of parallel lines , and the points at infinity rank as cusps .
In the latter the ellipse reduces to a doubled straight line joining the points , and , , these points rank as cusps .
Hence a solution will remain the true solution so long as does not pass through either of the values or , i.c , , so as does not change sign .
This is the meaning of the condition found in S 7 , that and must be taken with the same sign .
We can see this from another point of view , as follows .
The solution ( 29 ) can be ibited on a RIEMANN'S ce of two sheets , the branch points being given by When ( the ellipse reduces to a circle ) these points coincide in the origin , and destroy one another .
As l ) increases , the points move along the ; of , and nately meet the curve when at the points , i.e. , meet the curve at its cusI ) as soon as cusps occur .
Similarly , as decreases from the value oints move along the axis of , and meet the curye when at the points Circular Cylinder .
S 12 .
As example of expansion in a series of powers of a parameter , let us considel ' the nder of ] the equation is .
( 35 ) , 1 , in .
( 36 ) .
OF ROTATING LIQUID ) We are in search of a solution for expanded in powers of in the form . . . . . . .
( 37 ) , in which , . .
are functions of Substitute this assumed solution in tion ( , and we obtain terms of and higher in the coefficients of the various powers of , we obtain , \amp ; c. Solving these equations in succession , we obtain , terms oflower degree in , \amp ; c. Hence we obtain at once ( 38 ) , and , except in the special case of ?
In this way we can , when is , write down the to any required degree of accuracy .
Dcformed Elliptic Cylindcr .
S 13 .
As a final illustration , we shall find the potential produced by a small deformation of the surface of the elliptic cylinder . . . . . . .
( 39 ) .
Let the deformed surface be ( 40 ) , the solution be , .
as first powers of , here ction of is the solution when all the 's vanish .
Substitute solution ( 41 ) in equation ( 40 ) , equate coefficients of , and we obtain \ldquo ; solving for . . . . . . . . .
We can express in the form ( 43 ) .
Hence tion ( becomes . . .
Hence } find as the yalue of terms in , \amp ; c. .
( 44 ) .
The value of given by equation ( 43 ) is a solution of equation ( 39 ) .
Substituting this value , the coefficients of the two highest powers of , we find as tions determining and , equations will be required later .
ROTATING LIQUID CYLINDER .
Theory .
S 14 .
We now pass to the main problem before us , and consider the equilibrium of a cylinder rotating with angular velocity The equation to the cylinder for a rotation equal to zero is . . . . . . . .
( 45 ) .
When the rotation is different from zero , we shall suppose the equation to the face }eferred to its axis of rotation as origin to become OF ROTATING LIQUID or , in polar co-ordinates , .
This equation is not sufficiently general to repl.esent all cylinders which are netrical about the initial line .
The value of known at the we shall always be to find a function such that is finite and continuous , together with its first differential coefficients , at all points inside the boundary , and sucl ] that ) side the boundary , nd at the ) value of near the origin can be nded in the form . . . . .
( 48 ) , and this series will a circle of ence , B. lt ; only curve lies wholly inside this circle that the can be by equatio of the of ( 47 ) .
Let us , however , ttach a ) tion ( which the right hand becomes divergent at , as follows .
n , s supl ) that the value of the function 7 ) given by tion ( is culated from its known values inside the circle , the values utside this circle obtained by a process of " " ] tlation.\ldquo ; we shall } ) equation ( 47 ) to epresent the locus of at which Obviously , with this convention , equation ( 47 ) is sufficiently general to ) resent any surface .
If this surface is to give an equilibriunl configuration uIlder a rotation , we must have stant . . . . .
at the surface .
Now .
is a spherical at all points inside the surface , and equation ( 49 ) can be writtell in the form constant , or , what is the same thing , ( \mdash ; constallt ( 50 ) .
This equation is satisfied at the surf ce , and each is a solution of LArLACE'S equation at eyery point inside ; hence the equation must be satisfied point inside S. Now can be calculated by the methods eady explained , we obtain equation of the form , which gives the value of at all points inside a certain circle of convergence .
The value of inside its circle of convergence is , from equation ( 48 ) , ; SO , H. JEANS ON EQUILIBRIUM hence equation ( 50 ) , at points inside both circles , becomes constant .
) .
We luust therefore have , for all positive integral values of ? ?
\mdash ; . . . . . .
( 52 ) .
The condition that equation ( 4G ) or ( 47 ) may represent an equilibrium configuration is therefore . . . . . .
( 5S ) , mnd the of that each fraction is to 1 .
Since for configuration the axis of rotation must coincide with the centre of gravity , we see that if the coefficients of ( 46 ) satisfy ( 53 ) , the curve must be referred to its centre of gravity as origin .
The points of bifurcation are given by the Hessian of this system of equations .
For our purpose this may be most conveniently written in the form , .
( 54 ) .
Our method will , reasons already explained , break down as soon as a cusp or branch point occurs on any linear series .
Any solution will be a true solution provided we can pass from this solution to another , known to be a true solution , over a path through a system of linear series , without passing through a point at which a cusp or branch point occurs .
Now the occurrence of a cusp or branch point indicates , in the physical problem , the division of the mass of fluid into two separate and when this occurs the solution breaks down for a second reason also : for equation ( 49 ) is only true when the surface of the fluid is continuous ; when the surface consists of two distinct parts , the constant on the right-hand side has different values for the two parts of the surface .
This limitation will not cause trouble in the present stigation .
For we are only desirous of tracing the changes in the guratio of the fluid up to the separation into two parts , and even if method had enal ) us to proceed beyond this point , it would hayp ) ) ifless to OF ROTATING LIQUII ) CYLINDERS .
Series of linders .
15 .
If it were possible to calculate the 's and solve equations ( 53 ) in most general case , we should arrive at a complete knowledge of the system of linear series of ations .
This impossible , we shall start from known series , and calculate successive series by mining tious points cation .
Now we know ( S 12 ) that for the small values the is of the form Hence there is solution of the system of tions ( given by This is the series of circular configurations , esponds to the sel.ies of in th three-dimensional problem .
When the soluLiou breaks down hysically , since the pressure at point of the liquid becomes negative .
In fact , when reaches the the sel.ies gives place to series of annular each of which has the value .
We can adjust the radius of the annulus so as to give any desired amount of angular momentum greater than the critical value which occul .
S in the circular configuration when Points of on S16 .
To seal.ch for points of bifurcation on this linear ) by in equation ( 54 ) .
Eyery term in the determinant on the left hand now nishcs , except the terms of the leading diagonal , the equation reduces to The different roots correslJond the ) arc } givell by , 2 , 3 , 4 , 5 , .
, 5 , 666 , 75 , 8 , . .
1 .
The first of bifurcation 1nay be rejected once , the critical ' : vibration\ldquo ; being merely a ) cement of the cylindel}id A displacement in which only occurs the will alter potential energy only by a ter1n to the square of .
Hence the principal vibrations correslj to the diffel.ent values of to , and that only occurs for a value of in each .
When , all these vibrations are stable .
When eaches the -point of bifurcation of order the vibration of ordel becomes unstable , and , since there is only one polnt of bifurcation of order , this vibration remains unstable for values of greatel than the value at point of bifurcation .
We therefore see that by the time that reach e the limiting vibration is able .
J. H. ON THE Series of Elliptic Cylindc ) stable is of is the point of bifurcation occurs for the value The to the surface is . . . . . . .
( 5 5 ) .
In we found tion ( the corres value of , in is .
( 57 ) , this is howevel great ltlay be .
The values of can be written down once .
We have and all the othel functions vanish .
Hence there is a general solution of tions ( in ) the vanish except and . . . . .
( 58 ) , where is given by equation ( 57 ) .
This is the lineal series of which we are in search .
It is obviously a series of elliptic ders , and corl.esponds to the of soids in the threedimensional problem .
equations ( 57 ) and ( 58 ) we have . . . . . . . .
( 59 ) .
We see that as we moye along this the vadue of contin decreases to .
The Jnomentnnl , howevel , increases finite to an infinite yalue .
Uns .
Before ' points of tion on this sel.ies , let us iefly examine the the other points of on the circular series , these sel.ies known to .
Near the point of bifurcation the of the series of is . . . . .
( 60 ) .
S 1 we calculated the values of as far as .
If we , it appeal .
S that all these functions vanish excel ) and that is given ) Tha series is given by equation ( 60 ) , until become ciable .
{ ROTATING LIQUID The of is given by It therefore appears that the angular velocity decreases ( we recede from the ) oint of bifurcation .
the series the bounding curve is ) eriodic i with a period \ldquo ; and it is easily seel ) that at the remote end of the series is a curve consisting of equal and sylhlnetrically arranged , these extending the limit to infinity , being of zero breadth at ] points except in the neighbourhood of the In fig. 1 I the the of error in the value of tion ( caused the of will ) seen to be of ) order of Since the value of curve to be lcgitimate to neglect Points of reation oSeries .
S 19 .
Let us now return to the series of elliptic cylinders , and search points of bifurcation on this series .
We have found in S 13 that when is finite , ) when the squares and products of the ining 's may } ) neglected , we have ( eqnatio1l ( 44 ) ) for values different from 2 , terms linear in \amp ; c. . .
( 61 ) .
1\mdash ; 2 Let us now examine the form assumed equation ( 54 ) at npon this series .
We have , in the first place , [ R. J. H. ON THE QUILIBRIUM We have also ) whenever is greater than Hence we see that the determinant ) the left hand of ( 04 ) reduces to the products of the terms in its leading diagonal , and that the equation itselfis equivalent to the separate equations . . . . .
( 62 ) taken for values of from 1 to to a root of ( 62 ) there is a point of bifurcation , and the linear series starting the point must be found from equations ( 52 ) .
From these equations it ears that linear series corresponding to a root of ( 62 ) will be such that , as far as the first of small quantities , exists only for the values Of these series the series be rejected , as corresponding only to a step of elliptic cylinders , and not to a new series at all , and the series be rejected , as corresponding merely to a change of origin .
We are left with the values , 4 , 5 , .
, and for any one of these we lave , equation ) , points of tion a accordingly given by the equations . . . .
( 6.3 ) , where has values 3 , 4 , 5 , .
These points of bifurcation are points on the series of elliptic cylinders , hence , and are connected by equations ( 57 ) and ( 58 ) .
If we eliminate and from the three equations ( 57 ) , ( 58 ) , and ( 63 ) , we find , as the equation giving points of bifurcation of rder n ( 64 ) .
equation must be solved by aphical methods .
In fig. 2 the curve which is concave to the axis of is the parabola . . . . .
( 65 ) .
Fig. 2 .
OF ROTATING LTQTIID CYLINDF , 8.5 The remaining curves are the graphs of 6 ) for the values , 4 , 5 , .
The curve ( 66 ) cuts at the point , and .
The value of is .
It is }refore o } ) vious that the cnrves are convex to the axis of , and since for any value of the of is greatest for that curve for which is least , it is obvions that the curves can never intersect .
We therefore see that the will meet each of the curves ( 66 ) once , once only , for values of between ctnd 1 .
Moreover the smaller is , the smaller the value of at the intersection .
As we move along the series of elliptic cylindel .
S , the of increases to 1 .
Hence there will be an infinite number of points of bifurcation on this series , of orders 3 , 4 , 5 , . . .
The point ( we arrive first is th of order ; those of orders 4 , 5 , . .
follow in succession .
As before , we find that the configuration at the end of series of elliptic infinitely long and thin ellipse ) is unstable for every vibration .
The .
series which we expect to ) is that of order .
To find the point of bifurcation of fieries we to solve the equation and the solution is found by spection to be From equations ( 52 ) and ( we find that ) this point of bifurcation and .
The elliptic cylinder the point of bifurcation is therefore ) cylinder . . . . . .
( 67 ) or , in Cartesian co-ordinates , . . . . .
( 68 ) .
If we reduce the linear scale of this the area is , we fnd its equation and for its angular momentum , times greatest angular momentum which the circular form is stable .
Series of -shaped S20 .
The configuration of the new linear series of order .
the ) of bifurcation , of the form . . . . .
This new series is seen to be the series correS } ) onding to ' series of shaped figures .
Instead of making a separate problem out of the of he eonstants and , we shall , in order to avoid ] ) a later * H. , ' ' vol. 7 , p. 347 once to the equations determining the general configuration of this series .
We therefore replace equation by . . .
( 70 ) .
will be assumed to be the general form of snrface in the linear sel.ies under discussion , the quantity bein a ) al.ameter which vanishes at the point of bifurcation .
equatio expressing explicitly the solution of ( 70 ) ) be supposed to } ) . . . .
( 71 ) .
is the value of when , and therefore satisfies . . . . .
, is ( of and of . .
3 ) .
If calculate the yalue of , find ( 74 ) , yhere , if is -en the value of is . . . . . .
the of given by ( 74 ) , the to be the , surface const .
or , dividing tbroughout by \mdash ; lls independent of . . . . . .
( 76 ) .
uation ( Jnust be identical with ( 70 ) , the right-hand members ) ) arlnonics , and bence we must have , therefore , equation , for positive values of ? .
Instead of ) eing given by equation ( 73 ) , the value of may now be supposed to be . .
( 77 ) .
If we the limitation that the curve is to of , we must .
If replace .
by new unknowns } .
, equation ( 77 ) in the synlJnetrica ] ( 78 ) , in which we know that must ultimately } ) equal to , in that centre of gravity Dlay coincide with the ( cf. equation 22 ) .
OF flOTATING LIQUID CYLINDERS .
Now will be a function of , clnd the elatio between and our this relation lly the definition of .
We have , however , already assumed that the point of bifnrcation .
We thel.efol'e take as the tion between . . . . .
( 7 !
in which , itud , . .
undetermined .
The value of givell by equation ( 71 ) is now . .
( 80 ) .
If we substitute this in equation ( 70 ) , of we ) posing ) it to be a solution , we ( 81 ) , and if we coeHicients ) successive ) } of in obtain . . . . . . . .
( 82 ) , ( 83 ) , ( S4 ) , ( 85 ) .
Equation ( 82 ) is , as it ought to be , identical with ( 72 ) .
to determine the constants hese ]aving been found , equation ( 84 ) bless us to determine , and so on in succession .
S21 .
As as first powers of , we know the is of by equation ( 69 ) .
We fore assume once the \mdash ; . . . . .
, in which i Since , and the ] takes the form . . . . .
( 87 ) .
Since , we .
( 88 ) , nding in the ) form , and taking ( 89 ) .
Fronn equations ( 82 ) we have ( 90 ) .
By we .
( 91 ) .
tion ( now into the .
ting ) the coetlicients of the ious ] ) of we obtain The tion is , as it ought to be , all identity .
We assigll to any value , and thel.efore take , this bein equivalent to fixing the linear scale of measurement of .
Solving the remaining equations in succession , we obtain the following scheme of values :\mdash ; vanishin of shows that the celltre of gravity of the curve is , as it ought to be , the igin .
We now have as the value of , equation ( 86 ) , .
( 93 ) . .
We no ceed to the determination of .
Equation ( 84 ) takes the ( 94 ) .
The value of is of the ( equation ( 78 ) ) , ( 95 ) .
8 !
Suppose this substituted for in tiou ( , in the obtained the coefficients of ious powers of ) owel .
of eater than ) fourth it } seen equation ) ) ) ) by The equation obtained by equating to zero the coefficient of any power of than the fourth will be of the form a linear homogeneous function of , . .
and there is an equation of this form for every value of greater .
This system of equations can only be satisfied by taking We may therefore assume for ( equation ( 95 ) ) ) ansion of the ' ( 9 ) , which is writtell for From equation ( 82 ) we have and from equation ( 87 ) , . . . . . .
From these last two we obtain .
( 98 ) With the help of equations ( 97 ) and ) ) we can write equation ) in the form .
It is clear upon examination of this equation that the equations found upon equatin the coefficients of , \amp ; c. , will contain only terms ) , \amp ; c. , without constant terms .
We therefore take Equation ( 99 ) now contains only eyen powers of we can calculate coefFicients of these powers must series for equation ( 93 ) we get . . . . . . .
( 100 ) .
Next we have from equation ( 82 ) .
Squaring t ' and subtracting fromN each which been obtained for and ( equations ( 89 ) and . .
now that part of the -hand side of equation ( 99 ) which doehl We .34375 By addition the sum of in question is found to be , we have Collecting ) various series , we find the assumed by equation ( 99 ) , .
the of the various powers of , we obtain Solving , we obtain in succession , OF ROTATING LIQUID CYLINI ) We therefore as the value of ( 101 ) .
S23 .
We now to the determination of .
Equation ( 85 ) takes ] form ( 102 ) .
All the on , except in the line , in and of degree 5 at most .
is true of the on the left-hand which are multiplied by multiplied by are of even , two at most .
It is therefore clear that we may at once take ; and assume for an expansion of the \mdash ; . .
( 103 ) .
We now calculate the various series which .
We and from equation ( 97 ) , This last bracket can be at calculated the series of the last page ; we have ( 104 ) .
Next From equations(93)and(90 ) we have ( 105 ) .
From equations and we J. THE 8 ' @ hence , multiplication , . . . . .
tioJlS ( and ( I 00 ) hence , tion , erefore ( From equations and ( 89 ) we have equation ( 93 ) , tion , of these last series , We also @ . . .
By multiplic of series ( 91 ) and , find . .
lf we collect the various series which have been obtained and substitute them in tion ( lye find , the equivalent of this equaion , OF LIQUID .
, the fficients of the ious powers of ) tain . . . . . . . .
( 10 - \mdash ; . . . .
( 107 ) , - \mdash ; \mdash ; 16 ( 108 ) , - , \mdash ; \mdash ; \mdash ; \mdash ; ( 109 ) .
lving the first two of these equations , we obtain . . .
( 110 ) , . . .
( 111 ) .
Equations ( 108 ) and ( 109 ) may writtell - \mdash ; . . . . .
( 112 ) , . . .
Multiply ( 112 ) and ( L13 ) by 20 and 9 respectively , and subtract , and we obtain . . . . .
( 114 ) .
From equations ( 110 ) and ( 111 ) we obtain and hence , by comparison with ( 114 ) , This vanishing of supplies a test of the accuracy of the Equation ( t12 ) becomes , after tion , . . . . .
Our equations do llot us to determine separately ; they however , all satisfied by taking . .
( 11 J. H. EQUILIBRIUM where is unknown .
We therefore have tion 1 terms in , \amp ; c. ( 118 ) .
the value of given by equation ( 111 ) , we find as the values of .
1 , S24 .
Collecting the values of the various constants , we find as the equation to the , terms in , \amp ; c. ( 119 ) .
The occurl .
the } ljtity can easily be accounted for .
For we a solution .
( 120 ) , corresponding to a parameter is ] lected w the otation by the relation . . . . .
( 121 ) , then we can obtain this same solution in another form -by replacing the parameter by a new parameter .
As as this leaves the relation ( 121 ) between and unaltered , whilst the equation to the surfaces as far as becomes . . . . .
( 122 ) .
It accordingly ) that in ( 119 ) the value of is entirely at our disposal .
We shall therefore take yntion of .
There is a laarge priori that the linear series we are now considering will be stable for small values of , but it will be well to rigorously examine the question .
It appears from esearches that the whole investigation reduces to determining be the the angular momentum is a maximum or a minimum at We therefore to calculate the angular momentum as far as , and the to our }epd upon the sigIl of the containing As far the equation to the Slll.face in polar co-ordinates is ( equation . .
lnent of inertia is * H POINCARE , ( Sir la Stabilites de l'Equilibe des Figures .
'Phil .
Trans vol. 198 , p. :333 .
OF LIQUID CYLINDERS .
taken the , and therefore by ' ( 124 ) , where is iven by equation ( 123 ) .
Let us assume a solution as far as of the . . . . .
, where are functions of only , given by . . . . . .
( 126 ) .
The value of corresponding to solution ) is , and if we substitute this value for the similar yalues and in equation ( 123 ) , and equate the coefficients of . . . . .
( 127 ) , whence , by elimination of We can eliminate frcm this equation by the help of equation ( 126 ) .
The resulting value for contains only even powers of ; if simplify this , and transform the numerical coefficients to decimals , we find Now we require to find the coefficient of in I ( equation ( 124 ) ) , this is .
We therefore require to know the value of for , 3 , 4 , 5 .
This integral can easily be eyaluated for all positive integral of ; the values which we require at present are as follows:\mdash ; , Substituting these values , we find at once that is a positive quantity .
J. ON QUILI the inertia the .
velocity both increase as increases from moment of is a } for the value of Pear , Cnrves .
equation to the curves ) series has already been calculated as .
In order to obtain a still better idea of the shape of the curves I have carried calculation two degrees .
The calculation of these last two degrees is , and I omitted all details in to save space .
The method , precisely similar to that was followed in the calculations of SS20-23 .
It was found that the coefficients multiplying terms in and were inconveniently to this , the parameter has been changed to .
After making this change we find , as far as , for the equation to the surface expressed in polar co-ordinates , and for the tion determining ? . . . .
( 128 ) . . . . . .
( 129 ) . .
We must next consider within what limits the calculated terlns of equation ( 128 ) will give good to the complete equation .
It is clear that for given values of and the worst approximation may be expected when Let us consider the function , by .
( 130 ) .
of is expressed doubly infinite series , of which only a few terms known .
, the value of is kllown to be accurately smadl values equation will give with considerixble accuracy , but for largel values of .
and , the terms calculated will be inade to good to the value of .
What then , we inquire , are the values of and oyer which this approximation may be regarded as good ?
The coefficient of each power of .
is an series of powers of , of which terms up to have been calculated .
A glance at these will show that the cbpproximation is tolerably good long as , but begins to break down as soon as exceeds this unit Supposing that we ssigned to solle defi value less , the value of will be ( by an series of ) owers of , of only the first seven arc known .
small values these first few terms will give a sufficiently good approximation , for approxunation be and for still greater values the series will become divergent , so that first few terms give no at all .
It will be seen from inspection of equation ( 128 ) that the approximation will be lerably good so long The conditions under which the calculated terms will give good approximation may be supposed to be that , and .
In fig. 3 is represented the plane of .
The part this plane which the approxi m is good is that bounded by the four curves This is the portion which is shaded in the VOL. CC .
\mdash ; A. J. ON THE this same the thick curves represent the locus , calculated upon the supposition that the calculated terms of give a snfficiently ] to the whole .
For the greater part of the curve fhis lptio is not justifiablo , so the curve adjustment , the amount of adjustment increasing as we recede from the shaded portions of the .
The most important points on the curve are those at which These ) with sufficient accuracy for our present ) be taken to be S28 .
points -hicb axis meets the curve of which the equation is are iven 1 accordingly enables us to trace the motion of these points as we move along the linear , as increases from zero upwards .
At we have , of course , equal opposite ro the positiye root increases , Fig. 4 .
while the root numerically decreases .
Remembering that the centre of gravity of the curve must remain at the origin , we see that this indicates a general of the half of the curve which ] , with a diminution in the thickness of the forward half , and consequent gthening of this half hese , become more marked as increases , until we reach the value , at which a new ture presents itself For here there are two new roots occurring at the point This indicates that the fluid separates into two portions when the is reached , the point of separation being .
We at on ce struck by the great inequality in size between the primary and satellite : the extending approximately from to , and latter only from to .
The ratio of the linear dimensions will therefore ) something like 8 to 1 , but it must be emembered that our results require considerable correction on account of the imperfections in our approximations .
S 29 .
In fig. 4 the thick curye is the elliptic cylinder corresponding to , and the dotted curve is the adjacent curve corresponding to a small value of .
OF LIQUIL ) ) In figs. 5 and 6 the curves are those correspondin to and .
A glance at figr .
3 will show that there are difficulties in the way of drawing these latter curves with much accuracy .
I have given in detail some of the calculations used in drawing the curve 1 , in order that the may udge fimself as to the closeness or otherwise of the .
The curve is of course much easier .
Before passing on to the calculations , two points ought to be noticed .
It will be noticed that in the various -series ( the coefficients of powers of / in equation ( 128 ) ) the terms last calculated are without exception of the same sign as those previously calculated .
There is therefore some justiiication for hoping that the remainders in these series will be of the same sign as these last .
If this is so , the error introduced by the neglect of these remainders could , to an appreciable extent , be reduced by adjustment in the value of .
Thus we shall be to calculate the curve for ( say ) , and shall obtain a curve which is much more like the curve some smaller value of say than it is like the curve .
Regarded as an } ) at tracing a surface of equilibrium the ) will be much smaller than if regarded an at tracing the curve Fig. 5 .
( ii .
) It will be noticed that the sign of the leading terms in each of the sel.ies multiplyin , is positive .
An examination of the method by which these leading terms are calculated will show that this is a general law : all the leading terms after are of positive sign .
Thus the error will be reduced by supposing the series ( 128 ) continued to higher powers of by a suitably chosen sel.ies of terms .
I have accordingly done this in the calculations , and the conjectural throughout , enclosed in square S tracing this curve I started from , and calculated a series of points on the curve for decreasing values of .
The approximation at was found to * The effect of these corrections must , of course , ) small ; but , at the same time , it seems well to make use of any definite knowledge that we possess .
1 oo .
J. H. ON THL be good , the itnd the .
occurring only the decimal place .
As the ) ) tion ioves , at the error occurs only in fifth decimal place .
At the error again in the third place , and fter this the ) tion is bad .
My plan was to for smaller values of well I could , taking to keep the values of ?
in defect excess of their true values .
The curve was then ) lotted out on ) ruled with of 1 millin ] .
, the unit oflength taken to be 50 millims. * The of the elliptic cylinder of fig. 4 is known to be sq .
millims. , and this would also have been blJe c of the curve it been accurately .
The uf the ( obtained by ting ) ' ) , howevel , found to be sq .
millims. molnents about the of two parts of the ye and Fig. 6 .
ought of coul S to be equal : these found to be respectively 298,290 cub. millims. and 302,850 cub. millims. It efol e obvious that the curve been too mucb tened in the region in which eadjusting the curVe in this region so as to divide the as equally as possible , I with rors in moment of sq .
millims. nillims .
is the in fig. 5 .
It be seen thab the is one of about of Curve .
( Fig. 6 .
) S31 .
equation of the curye is found to be the in being those mentioued at the end of S29 .
* This been photographically reduced to 25 millims. before ting .
OF IQUII ) We calculate in succession the value of the radius vector when , \amp ; c. The value of is a root of lnethod will be to find a value of which is small , calculate at .
The root is then given by Wheu When \mdash ; .
therefore When When ) therefore Fv When When ; ; efore R When .
When \mdash ; Wheu ] ' . .
Whell \mdash ; When \mdash ; .
J. H. JEANS ON QUILIBRIUM is efore i at about , but it is clear that we are already in the region which the approximation ceases to be satisfactory , and that we are close the in which the series become actually.divergent .
of the lleighbourhood of the point calculated we come to a point which is a pair of equal roots for therefore " " minimum\ldquo ; in the yalue of .
This is the " " lleck\ldquo ; of which the first signs pparent in fig. 5 .
Let us refer to all the matter to the left of this " " neck\ldquo ; the " " \ldquo ; to all that to the right as the " " satellite Let the exact line of division be a vertical line at a distance 2 from the origin .
The imary h been drawn with accuracy ; the satellite must be drawn in the dopted in the difficult region of the former curve .
The of the found to be 11778 .
millims. , this leaves 1312 .
millims. to be accounted for by the satellite and the errol in drawing .
The centre of ravity of the primary was found to be millims. ) the left of the the error equally as possible , I rived at the curve of fig. 6 .
The area of satellite are respectively 11778 and 881 sq .
millims. , the error in area is a of 431 sq .
lnillins .
, and that in the moment about is that ) excess of 431 .
millims. at the point millims. , .
The error in the whole curve is therefore about 3 per cent. ; that in the satellite is tely of the same .
of magnitude as the satellite itself .
S 32 .
The following table sums up the results have been obtained , and also contains some new esults .
The monlents of momentum of the last curves were obtained by a ) ocess of counting on ) are not to great OF Ull ) 10,3 In this table the quantities of which the numerical values are doubtful are marked with a query .
A single query indicates a probable error of 1 or 2 per cent. ; a rlouble query indicates that the may be compalable with the quantity itself Let us examine the state of after separation taken place .
The satellite is describing an orbit about the primary , both bodies with the same angular velocity .
This angular velocity is , to within a few per cent. , given by . . . . . .
If satellite exerted no attraction upon the primary , the figure of the primary would be a figure of equilibrium under the influence of a rotation given by ( 131 ) .
The force exerted by the satellite may be divided into two parts , a uniform force in the direction .
and a tide-generating force of the usual kind , If the former of these existed alone , the the still be one of under of by ( I31 ) .
We see that the actual tioI 1 of the be under otation given by ( 131 ) , rbed by the reneratin potential is caused by the satellite .
Since this tide-generating potential -is small , except in the immediate neighbourhood of the satellite , it ought to be possible to emove the tides from the surface of the primary , and form a pretty good idea of the configuration which would be configuration of the primary except for tidal disturbance .
If this is dolle with the primary of fig. 6 , it will be found that the remaining curve is very good ellipse .
We may therefore conjecture that curve ( 9 ) is an ellipse deformed by the tidal influence of its satellite .
Now the ellipse corresponding to the amount of rotation given by ( 131 ) is curve 4 of the preceding table .
We see that the axes are in the ratio , ftnd ; is in good agreement with the ellipse )( the tides in ) .
The momentum of the ellipse of unit area of which the axes , are in the atio : ] .
( curye 4 ) is 44 .
If we reduce this so as to to an ellipse of area Jstend of to one of unit area , we find an angular momen tum of 38 .
Since this ellipse must be supl)osed to rotate not about its centre , but about the centre of gravity of itself a satellite about one-fifteentb of its mass , situated at the axis , this angular momentum must be increased to about ) .
small ) ancy between this and the value 40 obtained for curve 9 may be accounted for partly by of pproximation , and partly by the increase of caused by tidal deformation of the ellipse .
We can check our result in another .
The equation the ellipse . . . . .
, the force at , a point near the extrcmity the axis and outside 1 .
04 ON EQUILIBRIUM OF LIQUII ) CYLINDERS .
may , to a good approximation , be taken to be , and if is the centre of gravity of the satellite , this must be equal to the factor introduced to allow for the displacement of the centre of gravity of the primary .
, we find the ' and , this yives the Now is the sum of the semi-axes of primary and satellite divided by the semi-axes of the primary .
The equation just found is therefore about as true as could be expected , the linear diameters of primary and satellite being approximately in the ratio of 4 to 1 .
The ellipse as curve 4 is stable , and , since the mass of the satellite is small compared with that of the primary , we may suppose the combination of primary and satellite to be stable .
Thus , if ' conjecture as to the interpretation of curve 8 is sound , it appears that the linear series which comlnences with curve 5 remains stable until the mass separates into two lnasses .
The motion of a -cooling mass will therefore be through the following cycle of changes .
Firstly , increase of the ratio [ angular momentum area ) ] until we reach curye 5 .
Then motion along POINCARL\ldquo ; S Iinear series until we reach curve 9 .
At this point separation takes place , and the primary is left as a tidallydistorted form of curve 4 .
As the satellite recedes the tidal distortion decreases , and as the value of the ratio [ angular momentum area ) again increases , the confiul'ation moves along the Jacobian series of elliptic cylinders until curve 5 is again ched .
This completes the cycle , and the continual repetition of the cycle can only be ended by solidification , or some similar cause which is outside our present considerations .
I have not attempted to give any discussion of results from the point of view of dynamical astronomy .
The complications introduced by the heterogeneity and compressibility of natural substaJlces , as well as by the difference between the twodimensional and three-dimensional ) , are so great that any discussion with reference to the actual conditions of astronomy would } ) impossible in the present paper .
I have had the advantage of frequent tions with Professor DARWIN on the subject of this my thanks are also due to Professor FOBSYTIJ for advice in connection with the earlier sections .

