]\gt ; On the bility of a Random of the Stars in By , University , London .
Received Read April 2 ( 1 ) The )ject of the present paper is to discuss certain statistical distributions of the stars , and to consider how far they are in accordance with our experience so as that extends .
For direct tests we require .
more numerous deternlinations of parallax an are at present available .
The largest sample of parallaxes hitherto published is that due to Kapteyn , * but , for reasons where stated , I do not feel able to include this in my survey .
Small series collected by Newcomb and determined by the observers will be considered below .
The main point of my investigation will , , turn on certain parallax .constants , which frequently arise in stellar statistical ations , and which -are capable of determination ithout a knowledge of absolute paralla I shall use the following notation : stellar Ilitude , bsolute stellar distance , stellar parallax , a constanlt , depending on the unit of measurement and the diameter of the earth 's orbit .
, is the logarithm of the .
A rule over any of these quantities will denote their mean values .
will represent their standard deviations , or roots of mean square deviations from the mean will denote what is termed a coefficient of variation , or the ratio of a .standard tion to the corresponding meau value .
, and are the parallax constants which naturally a in the course of stellar statistics , and which to some extent can be measured without the aid of absolute parallax determinations .
The letter will be used throughout for a correlation coefficient , the attached subscripts the characteristic whose association is under discussion .
Let represent the light of a ; then this depends : ( a ) on the size of the star ; ( b ) its intrinsic htness ; its distance from us ; " " The Palallaxes of 3650 Stars of difl.erent Galactic Latitudes J. C. pteyn W. de Sitter , ' Publications of the Astronomical Laboratory at Groningen , ' No. 20 , 1908 .
I have retained for parallax in accordance with much ] , although it introduces some confusion when occurs in the same equations , and still more if it appears in the form and are needed for many other purposes .
In manuscript I invariably write II , but this never meets with the p1inter 's approval .
Prof. K. Pearson .
On the Improbability [ Apr. 5 , possibly ( d ) , on the existence of interstellar absorption .
Or , if represent an absorption coefficient , we shall have or , , where depends solely on the constitution of the star .
But if we measure in terms of the stellace , i.e. , the light , , of a star of tenth-magnitude as unit , and use 's value of the constant , we have and transferring to parallax , or Here is entirely independent of the star 's parallax ; it contains a constant , 5 , which disappears when we consider deviations from mean values , and a quantity , , which depends on the intrinsic brightness and size of tlJe star .
With our present knowledge there is no reason whatever for any association between the distance of a star from the sun and its size or ntness .
We shall therefore suppose .
Further , if there be an absorption term , we llay suppose it so small that for our approximation we can put .
Thus we write : whence it follows that for any series of stars It is clear that for system of stars , the easily calculable must be eater than the value 5 from any ested theoretical distribution .
We thus a means of testing the possible nonapplicability theoretical systems .
Again , let thele be a second magnitude , , arising from a light phenomenon at the same or the same distance from us as can arise in the cases of variable stars and of double stars .
Then we have Whence , if be the number of stars dealt with , 1910 .
] Random bution of the Stars in Thus , if there be no association between the actual emitted , we should have the left-hand side of which is determinable for any series by observation and calculation .
This provides another method of whether a theoretical distnbution is applicable in the case of phenomena at the same distance which are assumed to be independent .
In eneral which enables us to ascertain the actual relationship between capacities , if we caIl 1pothesise a reasonable distribution for the stars umder discussion .
Thus , for example , we might , pel'haps , assume to a first approximation that double stars are scattered at random in space , and might inquire whether any relationship exists between the brilliancies of the primary and secondary .
( 3 ) My proposed discussion will naturally break np into two parts .
First , the consideration of what values these parallax constants take for theoretical stellar distributions ; and , how far any system of stars obsel.vable .
us , and for that very reason a selected system , is likely to be a random sample of such theoretical systems .
I col-sider first some sample theoretical distributions .
( 4 ) Both Tewcomb and have investigated in the first place the effect on rallax frequencies of the uniform distribution of stars space .
This appears an obviously reasonable suggestion , ] , as xveighty reasons will arise for this standpoint .
I take lirst:\mdash ; Th Values )for a of St a Spher of Let be the density of the stars per unit ] , then , or S , or Hence Again , or ' or VOL. LXXXIy.\mdash ; A. Prof. K. Pearson .
On the Improbability of [ Apr. 5 , Hence Lastly , if be the modulus and , integrating by parts , or Again , where .
Thus we have Hence , integrating twice by parts and evaluating the indeterminate terms , we find or The special quantity generally needed , , is given by Now the results reached are , supposing the stars dealt with to be uniformly distributed throughout space , of some interest .
We see that they are quite independent of the radius over which we extend the sphere of our observations .
Further , we remark that , and ought in all cases where we may reasonably assume a uniform distribution to come out greater than ( 5 ) I will now proceed to modify these constants by supposing the stellar universe in two spherical layers about the sun , each layer uniform but individual density .
Let be the density up to radius , and the density from to .
Then it follows that and It is clear that to raise the values of , and 25 , we must make less than .
In other words , in a general way these parallax constants will 1910 .
] Random Distribution of the Stars in be if the stellar universe is rarer at greater distances from us .
But the changes are not very rapid .
If we make , then for we find only , and for , a very improbable supposition , 20- has only risen to .
For , the values for the same densities are and A dense nucleus round the sun appears the only in which substantial change could be made in these parallax constants .
( 6 ) I do not stay to discuss these points , because another method of investigating the problem will throw on other matters also .
If we supposed our steHar universe to be spheroidal in form , or that we dealt with a spheroidal portion of it , we should clearly be the same result , after a certain limit , as supposing the density to decrease with the distance from the sun .
We need not reject such a system with the sun in the centre as necessarily idle , for we are only considering the nature of the changes which very approximations to the actual state of affairs will make in the parallax constants .
In particular , the limit to the oblate spheroid be looked upon as a flat stellar universe , and this " " bunlike\ldquo ; universe is worth consideration .
Again , the limit to a prolate spheroid is a columnar universe , and , the constants of such a columnar universe being determined , we can superpose two such colunmar universes , and obtain the parallax constants for a double drift system .
The polar axis of ouu : spheroid being and the equatorial axis , I find for the mean distance of the population from its centre where , for the oblate spheroid ; and where , for the prolate spheroid .
Further for both .
Hence , can readily be found .
In the case of the parallax we have for the oblate spheroid where and Prof. K. Pearson .
On the Improbability of [ Apr. 5 , For the prolate spheroid where and These equations have been found by htforward but somewhat laborious integration .
The determinations of the mean , , and standard deviation , , of the parallax logarithm are not so simple .
We have Let , then for the oblate spheroid , and .
Change the ratiou to , and we have on integrating by parts , where , and vanishes when .
Next , where .
Whence .
Put again , and change to an integration with regard to .
We find 1910 .
] Random of the Stars in Integrating by parts , the first term will be found to vallish between the limits , have or where I is the .
It will be found that or To find the last integral we put The latter integral is finite from to 1 .
The curve given by was plotted for the range to 1 and mechanically rated .
For the prolate spheroid by a similar process we have , where and or 25 dr .
To find the , I take The subject of the latter integral vays remains finite .
The curve was plotted and mechanically rated up to the values required .
Prof. K. Pearson .
On the bility of [ Apr. 5 , The following table gives the chief stages in the determination of :\mdash ; Oblate Spheroid .
Ratio of axes .
Prolate Spheroid .
in the above equations , we obtain , after some arithmetic , the values of , and The limiting value of is , and again of The limiting value of is 1 and of is 1 , again of 1910 .
] Random Distribution of the Space .
55 I shall speak of as the " " normal\ldquo ; value of the constant of the stellar system , , since it is the value it would take if the stars were distributed at random through space .
It is clear that to suppose the distribution spheroidal and not spherical will allow of some increase from the normal value , but far for the bunlike\ldquo ; than the columnar universe .
It is desirable at this .
to compare these ers with actual data , however slender .
( 7 ) Kapteyn 's parallax determinations .
excluded , I can only give results deduced from the 72 stars cited by Newcomb* ; having the best determined parallaxes , and from 163 stars observed at Yale .
The zero and negative parailaxes must of necessity be excluded , because they lead to infinite distance and infinite logarithm of parallax P. This reduced for and the number of stars to 69 Newcomb and 122 admittedly most slender and imperfect material .
We have , after some considerable arithmetic , For I have , to see what effect it would produce , included the zeroparallax stars .
The result is to ) arate more markedly the Newcomb and series , and this was to be expected .
For Newcomb excluded all but the best determined parallaxes , and so , many zero cases which would certainly be included by the Yale observers .
For the Newcomb and stars boive ularly , accordant results , but wholly out of accord with the " " normal\ldquo ; system .
Nor is this to be wondered at .
Parallax is subject to a very large error of observation , and our apparent and have not their true values , but equal to and , where and measure the variability in observational elror in distance and parallax respectively .
Now let us take and , i.e. assume the riabilily in error to be as great as the quantity to be determined .
This cannot be true , because to admit it is to assert that parallax determinations are wholly unreliable , but it will serve to indicate that the maximunl limit to error does not suffice to reduce our observations to the " " nornal\ldquo ; values .
our apparent values by we find and * ' The Stars : a Study of the Univel'se , ' 1902 .
Chase , Smith , and Elkins , ( Parallax Observations on 163 Stals 'Traus .
Yale Univ. Observe vol. 2 , pp. 1\mdash ; 207 .
Prof K. Pearson .
On the Irnprobability of [ Apr. 5 , and for Newcomb and Yale stars respectively .
The reduction is considerable , but it will not bring the observations into ' accord with the extreme values of for our spheloidal universe , still less with the theoretical\ldquo ; no1mal\ldquo ; values of , and .
The theoretical and observed values for are still further out of agreement .
It must be at once admitted that the argument against a uniform distribution of the stars , based on these data , is not of much weight unless as confirming other results .
In the first place we must suppose that we take a random sample of all the stars within a given distance ( b ) of the sun , i.e. with reater than a given parallax .
In both cases the stars have been selected because they had big proper motions .
Newcomb 's stars have been largely chosen because they had a parallax greater than , i.e. parallaxes only sensible in the third pJace of figures have been rejected , but we cannot assert that they are a random sample of all stars within a given sphere .
Still the general agreement in between the two independent series when we cut off all the zero parallaxes is remarkable , and carries some if not very great now to our fundamental equation we find for the 69 Newcomb stars , which is reater than the maximum for , i.e. , and leaves a minimum value of .
But if we take the Yale stars we find , which is less than the apparent value , , of , so that if the latter were correct we should have a negative value for .
This can only be satisfactorily accounted for on the assumption that and are not independent in our selection .
In other words , although there is bly no organic relationship between intrinsic brightness and distance , still whenever we take a series of stars by any form of selection , we introduce a spurious correlation , which will generally of course be negative , between intrinsic brightness and parallax ; that is to say , of the stars at a great .
stars with small parallax\mdash ; only those of great intrinsic brightness are admitted to our series .
The For both series together ( 245 stars ) , or with maximum correction quite incompatible with This question of a spurious correlation arising from the manner in which the observer selects parallax stars is not without considerable interest , and worth following up in the case of the Yale stars .
Our general relation is , which leads us at once to : and .
Now we can directly find , and .
omitting Algol we have for the 121 Yale stars with positive parallax : , whence we deduce and .
Now ( see p. 48 ) , we have : stant - ' 1910 .
] Random Distribution of the Stars in Space .
modification which this idea requires in our theory will be developed in the following sections .
( 8 ) In dealing with star series it is not unusual to include all stals up to a given nitude , and counts have been made which are fairly reliable of all stars up to the 75 magnitude ; indeed , we may bs said to have practically complete lists of all the lucid stars .
It will be clear , then , that if we can modify our theory , so that we do not include all stars up to a iven distance from the sun , but all stars up to magnitude , we shadl be able to compare it much more sntisfactorily with observation .
Now , this process of cutting off from consideration all stars above a iven m tude introduces a spurious correlation between magnitude and parallax , or between logarithm of parallax and the emission .
We reduce the " " density\ldquo ; of the more distant layers of stars , because the fainter stars in these layers do not reach the admissible magnitude .
In order to reach any conclusion , it is , I think , needful to assume a law of distribution of the size and intrinsic brightness factor of the stars .
Returning to our equation for the magnitude , we have ( see p. 48 ) , any absorption factor where is the modulus , and is the distance .
I shall put , and instead of the of the previous work write briefly for .
F. Thus and be a factor the variation of which depends solely on the size and intrinsic htness of the individual star .
* Now , we need some law of frequency to distribute the character / , and it is not unreasonable to assume it in the first place to follow the same \mdash ; the Gaussian\mdash ; which has been adopted by many astronomers for other stellar characters .
I shall consider this point ; at present , it will be at least a guide to theory , and indicate the possibility of where measures the amount of emitted at the star .
Hence if and are positively oorrelated , and will be legatively correlated , or we find in the data that the light emitted and the parallax of the star are substantially correlated .
This , of course , is an indirect result of the further stars being only included in the list when they have relatively greater absolute brilliancy than the near stars .
and are used for the and of the earlier parts of this paper as more convenient for the following rather lengthy algebraic analysis .
Prof. K. Pearson .
On the Improbability of [ Apr. 5 , extensions .
We have now carefully to distinguish between the actual stars in the universe and those which will be found in our list , which we will suppose includes all , or at any rate a perfectly random sample of all , the stars up to nitude m Let be the mean value of for all the actual stars , and suppose the frequency of given by const.const .
if we wril , for brevity .
Assuming a uniform distribution of stars of density , we have between and stars Hence , since the hypothesis is that the intrinsic brightness of the stars and their distance from the sun are independent , it follows that the number of stars characters lying between and is of the form const .
, say .
Now we are to take all possible values of and subject to the condition that does not exceed .
Clearly our limit is to be given by , or , where is written for for brevity .
Hence , if be the total number of the selected stars , This determines and the constant of our frequency distribution .
We can now proceed to find the mean and the standard deviation of the emitted of the selected population of stars .
avel 9Random Distnbution of the .
Stars in Space .
Therefore , , and Similarly or , and accordingly Thus , while the effect of cutting off the series at a given nitude is to lower Che mean light emitted , it makes no change whatever in the variability of emitted .
Neither the mean nor the standard deviation of tho light emitted depends at all on the nitude at which we close our series .
I now turn to the mean and variability of the logarithm of the distance .
Hence or .
60 Prof K. Pearson .
On the lmprobabihty of [ Apr. 5 , This should be compared with the result on p. 50 .
It gives the modified form taken by the mean arithm of the parallax .
Further , since Or , we have the remarkable eneral proposition that : The mean of any selection of sta , not any over the ought to be of gnitude below this value , if the .
are distributed uniformly through space .
I next pass to the standard deviation of .
We have Whence , after some reduction , we find But it is easy to see that is identical with , or of our previous notation .
Thus We see accordingly that the quantity when we cut off at a given magnitude will for every magnitude be still constant , but it will exceed by the quantity the " " normal\ldquo ; value on the assumption that a random sample has been taken of all stars .
I now proceed to find the product moment of and ?
with a view to determining their correlation . .
Whence , after some reductions , and substitut.ing for and , we find : 1910 .
] Random Distribution of the in Space .
Reverting to our equation , : we have ; 61 .
Thus the effect of off our sample at a given magnitude produces a most important result ; we replace the statement that must always be greater than by much more striking one : The standard viatio b of of stars off at dc is whatever that be , to the } of parallax constant .
We have here a very simple test of the uniform distribution of the stars space .
I return to the result But There is thus correlation between and , which means a positive correlation between the logarithm of the parallax and , or , finally , ative correlation between the arithm of the paralla-x ( or the parallax ) and actually emitted .
In other words , cutting off at a iYen mnitude has introduced a spurious correlation between parallax and emitted , between a star 's distance from the sun and its size and insic bhtness .
It is not probable that such a relationship actually exists in the totality of the stellar universe , but it should be apparently demonstrable in all actual collections of stars .
For example , suppose the maximum to be emitted by to certain spectral ( or colour ) classes , then we should expect to find a correlation between parallax and spectral class of the character .
I think this " " cutting off at a given nitude \ldquo ; is most } ) robably the source of the fairly correlation already noted as existing for the Yale observations between spectral class and ] ) 'Monthly otices , R.A.S. , ' vol. 68 , p. 428 , 1908 .
62 Prof K. Pearson .
On the bility of [ Apr. 5 ; Further , this cutting off at magnitude will also modify the natural relationship between magnitude and parallax .
Taking and by , summing and dividing by the number of pairs in the usual way , we have Thus , , and is a c.onstant .
Further , since , we have Finally , the correlation of the logarithm of the parallax with the magnitude will be given by From the value previously give for we find : or we are able from observable quantities to determine the correlation of parallax with the actual light emitted .
* It may be noted that when there is a correlation between parallax and magnitude , a low correlation will be found between parallax and actual light emitted .
( 9 ) Before we proceed to compare the results of the preceding section with obseryation , it is fitting to consider how far they really involve the assumption made that the lights actually emitted by the stars obey a Gaussian frequency distribution .
It is probably quite as legitimate to assume this as to suppose that stellar proper-motions follow the Gaussian law , but I wish to show that it is by no means needful to assume this in order to obtain the chief results reached in the previous section .
I have taken the Gaussian curve in the first place because it has a certain sanction from astronomical usage , but our results are really much more general .
It is open to anyone to argue that the universe is a dying system , that there are a majority of dark stars , or stars of small intrinsic brilliancy .
If this be true , not only will the distriIn the case of the Yale stars and , and we find and , against the observed values of and This is , of course , further evidence that the Yale series is not a random sample " " up to a given magnitude\ldquo ; of stars distributed uniformly through space .
bution ontrinsic bliancies gskew frequency ctlrve , butl 9 ' the bution of the logarithms of the factors which measure actually emitted will do the same thing .
Accordingy we need , to cover this case , a frequency distribution far more genel'al than the Gaussian , and one that can represent all degrees of skewness , from a type like A to one like B. This is met by using the curve ' which ranges from to , and which can take the form if be negative .
I have discussed this curve at length in a memoir in the 'Philosophical Transactions , ' vol. 186 , , p. 373 .
By supposing or to have this type of distribution , we shall at any rate determine how far our results are peculiar to the form or symmetry of a Gaussian frequency .
The form of the frequency surface will now be and we have at once Assume , so that , and we have Thus 64 Prof K. Pearson .
On the Improbability of [ Apr. 5 , Next : .
is measured from the mean value , and therefore it is needful that* .
Hence Next Hence and that is .
Further , we may write Thus and differ from the values on the Gaussian hypothesis , i.e. and , by the introduction of the factor , which becomes unity when .
Still the twojgeneral propositions remain true , i.e. that .
Phil. Trans , vol. 186 , p. 374 .
; see .
cit. , p. 373 .
1910 .
] Random of the Space .
when we cut off our series at a Jiven nnitude , the of the actual lights emitted is reduced by a constant quantity , and the variability of these is constant , whatever be that nittlde .
Turniug next to the mean and variability of the arithm of the distance , we find Hence , which differs from the Gaussian frequency result solely by the presence of the factor in the denominator of the last term .
Clearly we can write leading , as on p. 59 , to the result .
Now is a constant whatever be ?
, for it equals / .
Hence the mean looarithmo of the parallax , when we increase the nitude at which we terminate our series .
decreases by ( difference of off nitudes ) .
Proceeding next to the value of , we have Substituting for in of out and .
the above value of , we find after some rather crthy reductions that VOL. LXXXIy .
Prof K. Pearson .
On the bility of [ Apr. 5 , We have then , as on p. 60 , Thus , as before , is a constant quite independent of , but exceeding its " " normal\ldquo ; value by a fixed amount .
We may add here the value of the mean } nitude .
As on p. 60 Thus our general proposition of p. 60 , that the mean nitude of a system of stars up to a given magnitude is always less by than that lmagnitude , is established on the broader basis of non-Gaussian bility .
Lastly , I tako the product moment of and whence , rating on the same lines as before , by aid of the -function , we find Bul Thus , precisely as before with the Gaussian law , we have obtained with this more general frequency the same result , namely , whatever be the nitude at which we cut off , the standard deviation of the magnitudes of the series is constant and equal to the " " normal\ldquo ; value of for an unselected universe .
We may conclude with the correlations and 1910 .
] Random Distribution of the in Space .
67 It follows therefore that all the results previously obtained for a Gaussian distribution hold for the much more general type of frequency now adopted .
The only differences depend upon slight involving the constant factor .
The distribution I have selected for the emitted enables us to suppose bnlliaut stars or even dark stars to be in a majority .
It is sufficiently wide , I think , to cover any probable distribution of the intrinsic brilliancies of the stellar universe very closely .
The main conclusions , however , that I suspect to be capable of simple proofs independent of any assumption as to frequency distribution .
I have not stayed to seek for such proofs , but given those by which I have myself reached these results .
They provide the mean magnitude and magnitude variability of any catalogue or random sample of stars up to a given magnitude , for example , of all the stars which could be seen with a telescope of given aperture\mdash ; of course , on the assumption that the stars are distributed at random through space .
They are thus fundamental in determining whether such a distribution is the actual one or not .
There are two points to be borne in mind here .
First our discussion is perfectly independent of direction , we have only supposed that spherical ' sbells about the sun as centre , if of equal volume , contain the same number of stars .
These sCars may or may not be clustered in individual shells towards one or more directions .
Our propositions apply to the frequency at each distance , and have no relation direction .
Secondly , we have seen that takes the value that does for a uniform spherical stellar universe .
I strongly suspect that this principle is general , namely , that the value of when we cut off at a giveu magnitude is equal to the value of the parallax constant , , for the total universe , whatever be its assumed form .
If this be true , then the values given in the table on will have additional meaning .
They show the modifications possible in when we suppose the stellar universe spheroidal rather than spherical in still maintaining a uniform distribution of stars .
As I have indicated , an ellipsoidal rather than a spherical form for the stellar universe appears to raise the value of and accords more closely , as we shall see , with experience .
I hope later to discuss how far is either approximately or absolutely true for other than spherical stellar distributions .
( 10 ) It remains now to consider how far the results reacl ) in the Prof K. Pearson .
On the Improbability of [ Apr. 5 , previous sections , ( 8 ) and ( 9 ) , are with observation .
Luckily we possess very accurate counts of the stars up to nitude F. The best of these is probably that due to Prof. Pickering , I shall make use of that here .
Even if in some of the categories the counts are not absolutely accurate , the divergencies will occur in the higher magnitudes , where they will , owing to the large numbers dealt with , make no impression on the values of the means or standard deviations .
The table below gives the frequency up to each magnitude in the second column .
the third and fourth columns are given the actual and theoretical mean magnitudes up to the nitude in the first column , the fifth and sixth columns give the square of the actual and theoretical standard deviations .
The seventh and eighth columns the differences of the observed and calculated values registered in the third to sixth columns .
The means and standard devi tions were calculated by the summation method discussed by Palin It will be obvious from this table that while our formulae and , express approximately the truth ( never by a quarter of a magnitude from it ) , yet the served values differ sensibly and almost regularly from the calculated values\mdash ; the higher the nitude the greater is the ence .
That is to say , the more stars we take into our consideration , " " Distribution of the Stars 'Annals of Harvard College Observatory , ' vol. 48 , pp. 149-186 .
Biometr ' vol. 4 , pp. 1910 .
] Random Distribvtio of the Stc ) the less accurate appears the assunlption of the ] ) ution of the stars through space .
I think we are fully justified on the basis of this in stating that a random distribution of the stRrst ] spherical is inconsistent with observed facts .
Or , if such a random distribution exists , then must be a correlation between distance and intrinsic ] ) rilliancy .
Such an apparent lation would be produced , of course , if to any extent in its transmission .
The few of lower ) the third nitude m be reasonably supposed uniformly scattered space , since they fairly closely satisfy the necessary conditions ; this nitude we have deviations ) cannot possibly be looked upon as the andom or ular variations from a chance ( 11 ) It may hal'dly be lleedful to further evidence that the stellar universe is not a landom distribution space of stars of varying size and intrinsic brilliancy .
But it may still be of interest to illustrate one or points from different material .
I chose Burnham 's ' of Double Stars ' because the selection there is of a somewhat different character to a star count , and further because there are records of stars up to a much higher magnitude . .
our attention to primaries only , have the foJlowing table : of magnitude .
Number of stars .
Up to 813 8 .
4947 \ldquo ; *Iheoretical means are placed in brackets ; of course , no stress must be laid on more than the two or three entries , and the whole evidence is not comparable with that from total It is clear that here starts a good deal higher than the star counts leave off .
But the tendency is the same , the mean does not increase fast enough , and the riability , instead of constant , tends to rapidly increase .
Very Saint donbles may easily be overlooked , but , I think , this is far from the 2 ) evidence of the above table , which is in the same direction as we have seen in the total counts of lower udes .
Double stars even more markedly than the whole stellar system do not seem to obey the condition of uniform distribution through space .
Thus from several directions we find evidence accumulating that the stars cannot , without to their size or intrinsic brilliancy , be considered as on * I owe this table to the kindness of Mr. P. F. Everitt .
Mr. D. O. Wood .
The of [ Apr. 12 , the average uniformly distributed through space .
The only way I can see to save such an hypothesis is either to assume a limited universe , not spherical in form , or to suppose light absorbed in transit .
It will be interesting to inquire further into the effect of modifying the form of a uniformly dense stellar universe , and secondly to inquire whether when we co1lfine our stars to one spectral class we get any difference in the general nature of the above deviations .
In conclusion , I must acknowledge my very great indebtedness to Miss Julia Bell , M.A. , for much aid in the laborious calculations , of which only the final results are tabled in this paper ; to Mr. E. Cunningham for a suggestion as to the best form in which to put the integral required in the case of the oblate spheroid universe ; and to Mr. C. H. Williams for mechanically integrating and tabling both oblate and prolate spheroid orrals .
The ofHelium from by the Action of By D. ORSON WOOD , B.Sc. , A.R.C.Sc .
( Communicated by Prof. the Hon. R. J. Strutt , F.RS .
Received April \mdash ; Read April 28 , 1910 .
) Introductory .
It is well known that the helium contained in radio-active minerals can be liberated by the action of heat , but , hitherto , no exact quantitative experiments have been made on the subject , and it was thought desirable to conduct such experiments with a view : ( 1 ) To ascertain whether heat may not be used as a substitute for chemical methods in the case of minerals and rocks for which those methods are not convenient , , if so , the temperatilres necessary .
( 2 ) To throw , if possible , more light on the way in which the is retained within the mineral .
A number of experiments have been made by Travers on cleveite and * Chemical methods break down ( a ) when the helium-content is very small , so that large quantities have to be treated to liberate a measurable volume of gas ; ( b ) when no simple reaction is available , or when such a reaction gives rise to large volumes of other gases .
Iron sulphide , which occurs in many geological horizons , is an example of this .
Travers , ' Roy .
Soc. Proc , vol. 64 , p. 140 .
The minerals were heated to redness in a hard glass and , therefore , to a temperature not exceeding C.

