]\gt ; [ 57 ] III .
A Study of Spectral Series.\mdash ; Part I. Alkaties and He .
W. M. HICKS , Sc. D. , F.R. Received October 25 , \mdash ; Read December 9 , 1909 .
AND number number of waves per cm .
Adopted formula , A is the " " limit\ldquo ; when m RYDBERG'S notation for the series is adopted .
Thus\mdash ; denotes a principal series .
the two principal series of Na .
is the first principal series of Na .
is the second line in , \amp ; c. denotes the satellite to the fourth line in the first diffuse series of Cs .
referred to as except when there may be an ambiguity with diffuse , when den .
is used .
is written ( i.e. , variable part of series ) .
Thus\mdash ; stands for the value of in .
is the symbol for the new series ( fundamental ) as P.S.D. stand for those already known .
by itself stands for a fraction .
In ring to observers\mdash ; K.R. stands for KAYSER \amp ; RUNG .
L. \amp ; D. , , LIVING \amp ; DEWAR .
B. , , BERGMANN .
K.H. , , KONEN \amp ; HAGENBACH .
L. , , LEHNIANN .
LE .
, , LENARD .
P. , , PASCHEN .
R. , , RAMAGE .
S. , , SAUNDERS .
W. , , WOOD .
THE following investigation was commenced some years ago , at a time when the discussion as to the atomic weight of Radium raised the question as to the dependence on their atomic weights of the wave-lengths of corresponding spectral lines of different efVOL. .
the same group .
TheI arguments were necessarily vague a PROF .
W. M. HICKS : A CRITICAL STUDy OF CTRAL SERIES .
unconvincin in the absence of any exact knowledge as to the connection of wavelength with atomic weight , even supposing such connection existed .
Our knowledge of series spectra is chiefly\mdash ; one might say almost wholly\mdash ; due to the sets of very exact measurements of KAYSER and RUNG , and of RUNG and PASCHEN , supplemented by extensions to longer and shorter wave-lengths by BERGMANN , KONEN and HAGH : NBACH , LEHMANN , RAMAGE , and SAUNDERS .
These have been only quite recently added to by and by the remarkable extension of the Sodium Principal series up to 48 terms by WOOD .
A most valuable feature of KAySER'S work was the publication of possible errors of observation .
This has rendered it possible to test with certainty whether any relation suggesting itself is true within limits of observational error or not .
In fact , without this , the investigation , of which the present communication forms a first part , could not have been carried out .
So far as the author knows , SAUNDERS is the only other observer who has accompanied his observations with estimates of this kind .
Others have given probable errors\mdash ; practically estimates of the exactness with which they can repeat of that feature of a line which they take to centre\mdash ; an estimate of little value for the present purpose .
In deducing data from a set of lines it is thus possible to express their errors in terms of the original errors in the observations , and limits to the latter give limiting variations to the former .
We therefore know with certainty what latitude in inferences is permissible , and are often enabled to say that such inference is not justifiable .
The formulae of RYDBERG or of KAYSER and RUNG are sufficient in general to identify lines as to a given series , except in the cases of lines in the infra red , where the order is 1 or even 2 , but they are not sufficiently accurate for our purpose .
It is necessa1y to obtain formulae for the yarious series which can reproduce the known series within limits of error or , if this is impossible , with as few outside as possible .
The fact that RYDBERG'S formula contains two arbitrary constants , whereas K.R. have three , and the well-known relationships brought to light by RyDBERG'S formula naturally suggest it as the basis for a second approximation .
In the absence of any definite theory as to the nature of the vibrations giving rise to spectral lines , it is necessary to make some assumption and to test it by its results .
The most natural one clearly is to suppose that in RyDBERG'S form where is some function of which can be expanded in a series of which is the first term .
Comparison with observation shows that his formula becomes exact with increasing , and that , therefore , the function should be expansible in negative powers of .
In other words , where are small .
' Ann. .
Phys 27 , p. 537 .
'Astro .
Jour xxix .
, p. 97 .
PROF .
W. .
HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
If RYDBERG'S opinion , that the true formula should be of the form , is correct , then would be of the form This , however , would be included in the previous formula with sufficient exactness except possibly for or 2 .
Any term added to will , of course , by giving an additional arbitrary constant , produce a formula capable of reproducing spectral series with greatel accuracy .
The criteria justifyin the form of a new term should be a very considerable increase of accuracy over a wide range ( and particularly for , and that it should brin to light further relationships .
RITZ , from certain theoretical ideas , has taken or , and with a large number of series has obtained very close agreement .
It was not until I had made some progress with this work that became acquainted with his paper .
I believe this to have been fortunate , for had I his work , and the great increase of accuracy obtained by his forms , I should probably not have attempted to essay any further improvement .
As it was I had proceeded so far as to feel certain that I was on correct lines , and had already obtained even more accurate results than his , besides some of the relationships presented below .
The first important result was that it was rarely necessary to go beyond the term\mdash ; in fact , not at all for the alkali metals\mdash ; that if were both included , and , of course , an additional line used for calculation , always came out a small fraction of , and that with alone the agreement was much better than with alone ( RITZ ) .
Evidence of this is given below .
As will be seen later , there are indications that a form may be the true form , at least for certain series .
These indications will be mentioned in due course , but as the metals of Group II .
are not discussed in the present communication , the case may be further illustrated from a special series in Mg , which will at the same time serve to show that too much stress must not be laid in favour of a particular formula on the mere fact that it reproduces the obseryed lines , unless the observations are very exact , or comprise the first members and a considerable number of the others .
The series in question was first discoyered by RYDBERG in K.R. observations , and consists of six observed lines contained between 5528 and 3987 , which he called a " " new kind of series\ldquo ; of the Mg spectrum .
Neither his own nor K.R. formula reproduce the lines , and he suggested a combination of the two , As will be seen , this involves four constants and requires four to find the constants .
The formula reproduces the two last lines within error of observation .
'Ann . .
Phys 12 , p. 264 .
'Ann . .
Phys 50 , p. 625 .
To anyone using the form , RYDBERG'S new formula suggests at once the form , which only requires three constants .
Using the fi.rst three lines for these there results which reproduces the last three lines well within the limits , as the table below shows .
I felt no doubt but that this was the true form , until it occurred to me to test the simpler form .
The formula reproduces the last three quite as well as the other , and , in addition , possesses all the criteria of a diffuse series .
The series is not , therefore , " " new kind of series as it was called by its discoverer .
The following table gives the series with the excess of calculated over observed wave-length for each formula , and the limits of observational error:\mdash ; Used to calculate constants .
An interesting remark may be made here .
If a series is represented by a slight alteration will represent it equally well by putting a continued fraction , viz. , in other words , or which looks quite different , and points to the depending on the roots of a quadratic .
In RYDBERG'S formula the value of ( or ) has a certain PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
may be a positive or negative fraction , or may be .
If we know that the first observed.line of a series is actually the first member for that line and is definite .
If we assume , for instance , that the first doublet of a principal series has the same separation as those of the associated series , the question is settled ; or if such a line has not been observed , then the degree of convergence of the separation of any doublet will give its order at once , , the fact that the lines in Na have the same separation as the sharp series , would show that the lines are actually the first members of the series .
If these had happened to have been outside the region of observation , however , and the doublet had been the first observed lines of the series , their separation would have at once shown that for them , and again would be definitely determined .
In spectra with no doublets or triplets , as in the singlet series of He or , this criterion would not be available .
Recourse might then be had to the fact that in all known cases the first lin of a series is not very different from the first line calculated from any approximate formula of the sharp series with sign reversed .
It was this consideration which led RYDBERG to postulate a value for in the principal series of the alkali metals .
In all cases there is nothing to distinguish the order for a diffuse series .
If , , we wish to discuss how the value of changes from element to element , RyDBERG'S formula can only give an approximation to its fractional part alone .
As soon , however , as a more approximate formula , such as that used in this paper or that used by RITZ , is applied , becomes definite at once\mdash ; e.g. , in the principal series of the alkalies the denominator is ( denoting a fraction ) and not , for the latter will not reproduce the series within the limits of observational error , and consequently ; in the sharp series In the sharp series , however , of Group II .
or the group , from which it may be remarked incidentally it follows , using RYDBERG'S law , that the wanting principal series must be looked for in the ultra-red .
The unique determination of is a matter of the first importance for comparative study .
It only fails when the lines are so few or the measurements so bad that or will either of them reproduce them within the observational limits .
In order to draw safe conclusions as to any relationships between the constants for various elements , or between the different series of the same element , we need to know their limits of possible variation .
This is possible , as mentioned above , when the data are based on the measurements of KAYSER and RUNG or of SAUNDERS .
It is not proposed , in what follows , to give complete expressions for these variations in terms of obseryational errors , but the method can , perhaps , be understood best by taking an actual example\mdash ; say that of K.S. The wave-numbers of .
( 3.4.5 ) reduced to vacuum from the wave-lengths given by K.R. are , with possible errors , as deduced from wave-length errors of , 15 , 53 respectively .
The true values are taken to be , where may have any values from to .
The lines as PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
observed give .
The variations due to the errors are found to be SA S S If , then , the constants calculated from three observed lines give a formula which brings a particular line outside the limits , it is possible to determine whether a permissible change in the original can bring it within .
It is used also to test the validity of RYDBERG'S law as to the limits of S. and P. A further valuable result is that it enables us to give the limits of possible variation of , e.g. , in the example above cited by putting .
The utmost possible , though unlikely , variations are then for for , and for When in the followin pages limits of this kind are given , it will be understood that they have been arrived at in the above manner .
We shall have to apply certain laws and relationships already known , and it is necessary to have some clear idea of the degree of exactness with which they represent facts .
Some of these depend on direct observation , others on relationships arising from the formulae .
Of the former the most important relate to the doublet and triplet separations of the various series .
They ( A ) The value of is the same for all the members of either the sharp or diffuse series\mdash ; the separation in the diffuse series being taken between the satellite of the first and the second , and in the case of triplets for between the satellite of the second and the third .
( B ) The values of for sharp and diffuse are the same .
( C ) The value of for the first lin of the principal series is the same as for the sharp or diffuse .
It may be said at once that these statements are true within the limits of observational errors , the only exceptions that I have met with are in the case of the sharp series of oxygen and the diffuse of sodium .
O. gives as lying between and , and O. as between and , or say by the former and by the latter .
One is naturally tempted to put this down to a larger observational error than the estimated one .
There are cases , however , in which a doubt may arise as to whether these statements are absolutely exact .
For instance , gives the following values for ( the observational errors are given in brackets):\mdash ; D ( 3 ) .
Tt is to be noticed the least possible from is , and the greatest from is .
If it is real , ( A ) , ( B ) are not true , but and are bad measurements , and the limit of accuracy ( 5 ) given for each is probably a rough way of saying very large .
On the other hand the gradual rise with increasing order PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
raises a suspicion that the rise exists and that is not constant .
It is just possible , as we shall see later , that the sodium series , which has generally been taken as the typical diffuse , may not be so .
As regards ( B ) , the following table gives the values of for the alkalies , as determined by least squares ( duly weighted ) for Na and and arithmetrical means for Rb and Cs from the observations of K.R. , SAUNDERS , and RAMAGE:\mdash ; Any doubt , however , as to the exact truth of ( A ) and ( B ) must vanish when we consider those elements in which is very large .
As examples , take Tl for doublets and Hg for iplets .
In the case of Tl we find the following values of :\mdash ; Sharp .
Diffuse .
( 31 ) ( 63 ) ( 73 ) ( 87 ) Most probable .
Most probable Most probable supposing both same The corresponding values for Hg Sharp .
Diffuse .
If there were real variations , we cannot but believe that they would bear some relation to the absolute magnitude of .
Yet here , even with such exceedingly large values of , the variations do not exceed the corresponding values for small .
We need , therefore , feel no hesitation in imposing the laws ( A ) and ( B ) as conditions which our formulae must fulfil .
As to ( C ) there cannot be quite the same degree of certainty .
The principal series have only been observed in the alkali metals and He among doublets and and amongst the triplets .
In the alkali metals for is known with extreme accuracy , ( from interferential measurements ) .
For the others the measurements are very bad .
For , KAYSER and RUNG give ( very large possible errors ) , SAUNDERS , and LEHMANN .
For Rb , SAUNDERS gives , LEHMANN .
For Cs , LEHMANN gives .
In every case ( C ) is within limits of errors of the measurements , but those limits are very wide , and it may well be that exact equality does not hold .
There can be little doubt but that the value of from is very close to , and is certainly exact for .
The He lines have been measured with great exactness , but unfortunately is small and is far up in the ultra-red .
In is of the order , and OP ( 1 ) is a bad observation , whilst in , SP ( 1 ) is outside the region of observed lines .
The eyidence of the exactness of ( C ) is , therefore , not conclusive , but in spite of the indications of Na to the contrary , it is probably correct .
The doubt , however , should be kept in remembrance .
* Observation also tends to show that : ( D ) The corresponding and series tend to the same limits .
( E ) The doublet or triplet series of the type tend to the same limit .
The exact truth of ( D ) and ( E ) has not been proved from direct observation , for the actual limits have been seen nor have a sufficiently high order of terms been observed .
The HD has been measured up to , but HS star by only a few to compare with it .
has been measured by WOOD up to as an absorption spectrum , but only two terms of are known .
With the exception of these two , no series has yet been observed nearly up to its limits , and the strongest eyidence in favour of ( D ) and ( E ) is based on the limits as calculated from formulae\mdash ; in which again the actual limits will vary slightly with the form adopted for the formula .
The lists of values of limits given by show that the approximate truth of ( D ) is certain .
Amongst the relationships depending on the formulae , the chief are those which will always be associated with the name of RYDBERG , , the remarkable connection between the sharp and principal series , and the constancy of N. The former may be expressed in the form : ( F ) of ( G ) of .
The discussion of the evidence for or against them will come later .
Their approximate truth is so clear that it would require very strong evidence to justify the use of any formula which did not conform to it .
RITZ has imposed the law on his formula , using it in calculating the constants , and supposes that the results prove * Since the above written , PASCHEN has published further observations in the ultra-red , giving what he takes to be the principal series for several elements .
When discussed , these will probably give considerably more light on the question .
'Ann . .
Phys vol. 29 , p. 625 .
'Astro .
Jour vol. xxix .
, p. 97 .
'Rapports Congres Internat .
de Phys. ' II .
, p. 210 .
expression " " of\ldquo ; will have to be employed so frequently that it will bc convenient to use a symbol for it .
will stand for the term of the formula in which varies .
PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTEAL SERIES .
its truth .
When , however , his values are contrasted with actual measurements ( see below ) , it would appear rather to show that if the law is correct his formula is wrong , or if his formula is correct the law is not exact .
( H ) The value of is the same for all series and for all elements .
There can be little doubt but that is very nearly the same for all known series , and it is a tempting hypothesis that it is actually so .
We have no evidence , however , to prove that it is either the same for all series , or has different values for the different types of series , which may well be the case .
If it be found that all observed lines come into the formula with the same value of , and further if thereby the validity of ( F ) and ( G ) can be established , or if other relationships which present themselves are more clearly brought out , then the evidence that is a constant of nature ( as RYDBERG calls it ) will be very strong .
We shall start , therefore , by taking to be the same for all series , and the same which RYDBERG assigns to it , , 109675 , when referred to yacuum .
This value , however , may require redetermination .
It was determined by RYDBERG from hydrogen by least squares applied to BALMER'S formula for HD and using the observations of AMES .
It was BALMER'S form which led to RYDBERG'S modification , putting in place of , where depends in some way on the properties of the substance .
To suppose , therefore , that BALMER'S formula is exact , i.e. , , seems like saying that hydrogen has no individual properties .
In any case , it would seem extremely unlikely that , as every other substance has a finite value of should have .
It is to be expected that its would be small .
All the low atomic weight elements have small ( or very nearly unity ) for a series , and we should expect the lightest element of all to have a very small one ( or nearly ) .
To find , then , the series of RYDBERG should be used as a basis , and and both determined .
Unfortunately the HD spectrum is not known with great exactness of measurement .
The most accurately measured spectra as wholes are without doubt those of He by RUNG and PASCHEN , and it would preferable to use these to determine N. At first sight , then , it might appear more logical to consider this point of the value of as determined from and He spectra before going further .
It is preferable , however , to postpone this until the alkali spectra have been discussed , and some information obtained as to the connection of the different types of series .
Moreover , by keeping at present to RyDBERG'S value , it will be possible to directly compare the results from the formula adopted in this paper with those from that of RITZ , which is the only other comparable in accuracy .
of the Metals .
As SAIJNDERS has given very complete tables of the observed spectra of the alkalies , it will be sufficient to refer to them for the material at disposal at that date .
* ' Astro .
Jour , p. 188 .
VOL. CCX.\mdash ; A PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
Since then we have had the very exact measurements by PASCHEN* in the ultra-red of Li , Na , and , and the remarkable discovery by WOOD of the absorption by its own yapour of the series , lines being obseryed down to A superficial examination of the series of the different alkali elements shows that Li apparently occupies an anomalous position , and leads to the surmise that the series are not homologous with those of the other elements .
Thus , if the first lines of the principal series in each be compared , Li .
Na .
K. Itb .
Cs .
it is seen that the wave-lengths continually decrease with decreasing atomic weight until we arrive at Li , when the wave-length jumps back above Na .
Further from with the others should be a doublet with about , and the and should be all doublets of the same , but , as a fact , they are all single , with the possible exception of 4602 , which KONEN and HAGENBACH believe to be a double line .
For this reason in my first examination I left Li for separate consideration , and as this was fully justified by the result , it may be well to proceed on the same lines now .
We shall discuss then first the well-known principal , sharp , and diffuse ries of the other alkalies , then the additional series which appear in the spectra , and finally return to Li in the light of knowledge gained from the others .
The first step is to justify the use of the form for the denominator , by showing how closely it reproduces the obseryed lines .
This is done in Table I. For comparison the results as calculated from RITZ ' formula are also given .
In each series the first column ( O ) gives the possible obseryational error , the second ( H ) the excess of observed over the calculated ength , and the third ( R ) the corresponding values ffom RITZ ' formula .
Whenever this is outside the observational error it is in thick figures .
For the present purpose the values for Li are added .
The values of , \amp ; c. , are given in Table II .
The results for RITZ are givell from his paper , or , for those lines not observed at the time , are calculated from his constants .
It should be remembered that the estimated possible errors are considerably greater than the probable errors , and consequently we ought to look for a closer agreement between calculated and observed values the possible errors show .
It should further be noted that for high orders any formula giving rough approximation only will give some information as to the excellence or otherwise of the observations themselves .
If the deviations show gradual change as increases , the formula is probably in fault ; if , however , a deviation makes a sudden change and comes back to its former course as increases , the observations are probably the cause .
Where two values are given under they refer to the doublets .
All the have 'Ann . .
Phys 27 , Astro .
xxix .
, p. 97 .
PROF .
W. M. HICKS : A CRITICAL SPECTRAL PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
been found by correcting to constant , also in has been taken to be the same as .
Remarks on Table .
It will be seen that agreement is perfect in This is the only series in which the first three lines ( from which the constants are determined ) are known with great accuracy .
It has not been thought necessary to give the comparison with all WOOD'S measurements , but after they are given for every interval of 5 .
The deviations from WOOD show that his observational work ( apart from instrumental ) have mean errors of about , and if his instrument constants are comparable with K.R. , the constant in the formula should be about higher .
RITZ has two outside , one times and the other more than twice the amount allowable .
is not a first-class set of measurements .
Nevertheless , there is good agreement except for .
For PASCHEN'S estimate of possible error is 1 .
For this a very slight alteration in the value of , or an error of 01 in ( possible error ) , would bring both mine and RITZ ' values within limits of error .
The deviation in , however , is serious .
It has the appearance of being chiefly due to a measurement error , for successive deviations , as increases from 4 to 8 , are , and RITZ ' formula suggests the same .
If the observations are looked at more closely gives instead of .
LIVING and DEWAB , in their early measurements of this doublet , say : " " The lines in the Normal Solar spectrum probably represent this pair These give .
L. and D. measurements are , of course , not to be compared with those of the later observers , but apparently their method of producing the light had advantages over those of Their observations generally give wave-lengths less than those of K.R. , but in this case they are much less , pointing to a suspicion that instead of one nearer would be more correct .
is that demanded by the Although , therefore , this case is a serious one against both formulae , there are yet indications that it may be closer to the real value than the observations show .
are from L. and D. observations .
The reading for is given by them for the middle of a narrow band which they have not seen divided .
For this .
I have taken the readings for the two lines to be , and RITZ ' result is also compared with this . .
The observation errors for all the series are .
The deviation from constant has been referred to already , especially the excessive of , which renders the readings for this very doubtful .
The constants have been determined om L. readings for and K.R. for .
There is some uncertainty as to whether for the series should be a fraction or fraction .
The results here are calculated for the form , but in all cases for lines in this table the order of a line is taken to be the integral part of the denominator whether we use or .
The doublet at is outside the limits , and it ought to be if is to be constant , also the reading for , but both and can be brought within PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIE@ .
by a permissible change in , thus leaving outside , which must necessarily be the case with so large as 28 .
We may regard , therefore , the formula as satisfying the conditions for .
RITZ gives no constants for his formula for this series . .
The observations for are very bad .
K.R. give and ( possible errors 5 ) .
L. gives and , R. 7664 , and S. and 769908 . .
The votes carry it for the lower value of and one gets an almost superstitious belief in K.R. measurements after long use of them in this kind of work .
On the contrary , L. 's , result appears to agree better with those computed from K.R. values of , the case here adopted .
RITZ ' constants are calculated from L. result for and K.R. for .
In the table , then , L. result is used .
It will be seen that the formula adopted here gives excellent agreement except in , the deviation is twice that allowable .
The order from to 7 is , 21 , , 01 , a striking agreement with a sudden jump at , pointing to observational error , but there is no evidence as in the case of .
RITZ ' are all outside , except for , 9 .
The are corrected to constant .
The agreement is very good except for , which is far out .
RITZ has all outside .
In his paper he must have given for .
I have calculated his on this supposition .
The numbers for KD are calculated from a form It appears is not possible .
As is seen , there is good agreement .
In this case , in order to bring within limits , errors have been allowed to and .
This is the only diffuse series for which RITZ has given constants , and I have calculated from them the values for 2.3.8.9 .
The formula in this case is quite out . .
The only accurate measurements for this series are those of K.R. for .
is as uncertain as in the case of KP ( 1 ) .
K.R. give 7950 , 7811 ( observational error ) with ; at least one , therefore , is certainly wrong .
L. gives and ; S. gives and ; whilst R. has 7799 for .
Using for constants gives close to L. value and close to obseryed value , pointing to L. yalue for as being very close to the truth and that K.R. error is almost wholly in . .
and L. agree practically for .
S. gives a line at which is closer to than ; it has been taken for both lines in the table .
For the limit has been made the same as for , and the lines used for calculation .
is just outside , but a slight permissible alteration will bring it inside . .
The text for this series is very corrupt .
K.R. only observed and with possible errors 2 for and 15 for .
For L. 's measurements give and S. .
For the constants I have used corrected to .
The formula gives S. result for as against L. 's .
are given both by R. and by S. : for lack of any other guidance I have taken 'Ann .
16 , p. 684 .
PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
that which agrees most closely with that calculated from the formula , and I have treated RITZ in the same way , calculatin his ( 2.6.7.8 ) .
The observations for are by B. They are in no way comparable with the corresponding values by P. in the case of Na and K. In fact , B. readings give , or an error of about 16 A.U. in the wave-length differences of and .
Probably RITZ is closer to than the formula in ; on the other hand , he has errors from nine to ten times the allowable in two cases . .
In the series it has been thought that Rb has a satellite to 7753 ( L. ) , viz. , 7759 ( S. ) .
If we take and use K.R. values for , it can be shown that it is impossible to get within wide limits of either S. or L. , or of BERGMANN'S ; but it is possible to satisfy conditions for if what has been supposed to be the satellite is actually the line , i.e. , 7759:5 S. instead of L. , but still comes far wide of B. value on the side of longer W.L. With the calculated place of is close to the supposed satellite , and it can only be modified to bring it to the supposed , i. e. , 7753 , by giving large possible errors , alternately positive and negative , to K.R. observations of .
Also the lin , as calculated , falls half-way between B. and .
It can be made to fall on his , but not on his .
I am inclined to think , first , that B. has a large error here , and , second , that Rb has no satellite series .
There are two additional considerations pointing to the same conclusion .
One is that , if a satellite series exist , the doublet separation for will be considerably less than the normal .
Now B. two lines , 14830 ( those now taken to be and giye , about 40 A.U. aboye the normal ; whereas , if there were satellites , it should be expected to be somewhere about 40 A.U. less .
Although B. actual wave-lengths in this region might possibly err to this amount , it is not probable that the difference of two lines so close should be wrong to the extent of 80 A.U. as they would be if satellites existed .
The other reason against Rb having satellites is considered later in p. 86 .
The agreement of the formula with observation for the whole spectrum when no satellite is supposed is remarkable .
The values are given for this case in the Column Rb II .
, and for the case of satellites in Column Rb I. .
have only three lines , of which one alone is good .
R. has given to ; L. has given with .
R. practically agrees with K.R. for , but deviates in from them by four times thei1 estimated possible error .
The observations of R. were made with the flame , and the spectra were probably better developed than in K.R. 's case .
To compare the two , I have calculated out the doublets , taking for the first .
For this purpose any approximate formula will serve .
The result is , 184 , .
K.R. 's give , and R. , which favours R. values for and .
I have therefore calculated constants from E. and L. , and those for from , L. 's and R. 's .
In the table I have entered ths deviations from K.R. 's values for ( 2.3.4 ) .
Really , therefore , for our formula the PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
apparent deviations are the deviations between K.R. and R. I have entered them thus because o'bservation errors are entered , and it would be misleading to enter against them .
It reproduces the whole of RAMAGE'S results very closely .
RITZ calculated from L. and K.R. His difference for , though actually greater than the limit , is probably not due to the form of his formula , but to K.R. errors , and should not be counted . .
Here is by L. , by S. , and by R. The values of of SAUNDERS corrected to constant were used to determine the constants .
There are no estimated possible errors to compare with .
by B. gives If his readings share the error equally , then the difference between observed and calculated wave-length is 74 .
It is probable that this is chiefly due to imperfection in the formula .
RITZ has given no formula for this series . .
The series differs in Cs from the other elements in the clear possession of satellites , and consequently there are two distinct series to find , and no longer differing by the same amount .
The form gives the best result .
The tabulated numbers are those determined for the and series , as the are always stronger lines than the .
The formula calculated from the lines ( the so-called satellites ) differs only very slightly from that calculated from .
They should , of course , be the same ( limits excepted ) .
If the values of for ( the top line ) be calculated from the two formulae they are found to be respectively and .
They should be the same ; their mean is .
The difference between and this is .
The significance will be seen when the additional series are discussed .
The table also contains the corresponding deviations for the Li spectrum , but the consideration of details is better postponed until that element is dealt with specially .
The following list of excessive deviations of the formula is excerpted from Table I. \mdash ; only in place of the actual deviations the ratio of the deviation to the maximum allowable is given : H. R. H. 2 ( 5 ) 1 ( 5 ) 4 ( 5)2 ( 6 ) 2 .
( 7 ) 1.2 KS ( 2)16 KS ( 4 ) 4 KS ( 5 ) 1.5 KS ( 6 ) ( 4 ) KS ( 7 ) 1.1 ( 5 ) 2 ( 3 ) 9 ( 3 ) ( 5)10 ( 4 ) 3.6 PROF .
W. M. HICKS : A CBITICAL STUDY OF SPECTRAL SERIES .
The case is not included , as there is strong evidence that the readings are not correct .
There is some indication also of the same for KP ( 5 ) .
The formula is bad for the top line of the KS series , and this must be specially considered in the sequel .
With this exception the agreement is remarkable .
Preliminary Discussion of Formuloe Table II .
gives the various constants determined under the conditions described above .
The possible maximum variations are given in all cases where K.R. give possible errors of observations .
Where these are given in brackets the numbers show how the last digits in the constant are affected .
Thus in .
This means that may differ from by any amount less than .
It is to be understood that in any one formula the top and bottom signs go together .
Thus if be taken as must be , and so on .
It should also be noted that the variations in are not necessarily proportional , i. e. , if the error in is one-half of , the corresponding errors in TABLE II.\mdash ; Formulae Constants .
VOL. CCX.\mdash ; A. TABLE II.\mdash ; Formulae Constants ( continued ) .
PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
are not necessarily one-half of -(477 ) and .
The variations given are the greatest allowable subject to obseryational errors alone , and the assumed value for but that they should be so great as these must be regarded as highly improbable .
The variations for KP are so large because the constants are determined from instead of , a less accurate value of being more effective for correct determination of or than a more accurate one of .
Two estimates are inserted for ( which were deduced from ) , the larger are based on K.R. estimates of possible error , the smaller on L. estimate of probable error ; the truth probably lies between the two .
Different positive and negative errors appear under and .
This is due to the fact that small permissible 1T0rS were introduced into the observations giving the constants in order to bring all the calculated series within limits .
The values of inserted in the table give the values used in correcting the two sharp series to a constant .
They are not the same as those given elsewhere as the most probable values , but as they were used in the original calculations , and as nothing was to be gained by recalculating , they are left as they were .
The first column under each series gives the value of the limit ; the top number being its value and the lower the value of the when it is written in the form Thus in The third line under gives the values for on the supposition that .
The top number in the second column gives the value of and the lower the value of the denominator when , i.e. , the value of .
Thus in and the denominator of The variations are , of course , subject to the correct value of being 109675 and to being integral .
Alterations in or taking , say , to be would produce further consequent alterations in A. Relation \mdash ; An inspection of the values given for and , with their possible variations , show that in all cases their values are the same within limits of observation .
Relation ( E).\mdash ; The equality of and is so far justified by the fact that its assumption in calculating gives constants which reproduce the observations within possible errors .
There however , still room for doubt as to exact equality , owing to imperfection of data .
.\mdash ; Viz .
: If be written in the form and be the denominator of , this relation involves the equality of and .
The values are given in the table .
From them we get the following values of Rb Cs ?
All are easily zero within limits of observational error .
It will be further necessary to discuss whether the alterations called for wil ] affect the permissible exactness of the formulae for all the lines .
Its discussion is bound up with that of the next relation .
Relation ( G).\mdash ; Viz .
: .
As before , write .
The table gives the following values of :\mdash ; Rb Cs . . .
This at first sight appears distinctly against the truth of ( G ) .
Taking the actual errors of observation in the three lines used to calculate the constants of to be respectively times the possible amounts , and to be the corresponding ratios for the series , the constants of he formulae were determined in terms of them as explained before .
Then the conditions ( F ) and ( G ) were imposed , and the question discussed whether it was possible to satisfy the equation by values of the numerically less than unity .
Starting with Na , the result was that it impossible .
As , however , this might be due to the fact that another term was required in the denominator , the work was repeated , using the formula .
As we have seen above , is in all probability a bad reading .
If this be put aside , it was found that it would be possible to satisfy the conditions and still bring in all the other lines within observational errors .
For instance , putting and have inappreciable effect ) gives For These satisfy both ( F ) and ( G ) exactly , and reproduce all the lines , of course , excepted ) well within the limits of error .
In it is to be noted thal the are practically equal , and that for is negligible , i.e. , so small that a very small error in one of the determining lines would wipe it out .
The case of was then discussed .
Again , the relations could not be satisfied with the form , and the work was repeated with .
As in the previous case , it was found possible indeed , but it was now necessary to take close to K.R. value instead of that of L. It could be done , for instance , by putting , and .
This gives For PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
These bring all the lines ( , excepted as before ) well within the limits of error .
Again , it is to be noted that in the formula , and the of is negligible .
The fact that this relation occurs twice suggests its being a real relation , and not due to the chance choice of the suitable values of , \amp ; c. The limiting errors in the cases of Rb and Cs are so wide that it did not seem worth while to discuss them in the same detail , especially in view of considerations which will appear later .
The evidence so far is strengthened as to the correctness of ( F ) and ( G ) in fact ; but , on the other hand , the impression is intensified that there is something not quite correct with the form for the series .
RITZ uses instead of , although there does not seem any experimental evidence in its favour .
As a fact it can be shown that the use of in place of in our formula makes it still more difficult to fulfil conditions ( F ) and ( G ) .
( H ) .
The difference between the 's of P.\mdash ; If we write down the value of of of , we get Na ( 5800 ) or between .
and487085 ( 6647 ) , , , , 477341 , , 464047 Rb , , , , 496331 , , 478911 ( 6842 ) Cs 487 would include all except K. The values point to the possibility of here bein a constant difference , if the correct formula be found .
Also it shows conclusively that such difference cannot be 5 , a supposition which has suggested the idea that the and are similar series , with eyen numbers and with odd .
The supposition , moreover , is not borne out by the evidence of other spectra .
It is very nearly 5 in 153 in He ' , 445 in O. The is always very close to UP ( 1 ) wherever it can be tested .
If this hold for Mg and Ca , the difference in their cases would be about34 for Mg and for Ca ; , Cd , about the same .
There are clearly no valid grounds for the supposition in question .
( I ) .
The 's of P.\mdash ; A first inspection of the values of given in Table II .
makes evident a remarkable approximate relation between the values of of the various elements .
They may be written as follows:\mdash ; 5 The possible variations show that , with the exception of Rb ( which , however , very nearly falls in with the others ) , the values of are multiples of a number not far from .
The case is not really so strong as it looks , for the limits of variation are almost certainly much less than the maximum possible ( which has been mentioned before ) .
On the other hand , the relation is so close that a more correct value of or the true form of the formula might well make the ratios exact .
It is clear that the have no direct relation to the atomic weight , but depend on some other physical property which obeys a similar law .
In searching for such property it occurred to me to try the atomic volume , and I found an almost precisely similar relation for these .
The densities of the alkalies are not so exactly known as those of other metals , and , moreover , they ought to be compared at corresponding temperatures .
If , however , they are compared at about to C. , and the values used which are given in LANDOLT and BoRNSTEIN'S tables , viz. :\mdash ; Li Na Rb Cs the atomic volumes come out as follows:\mdash ; Li Na Rb Cs As the density of Cs is taken near its melting point it is too small , and we should expect a rather lower value for its atomic volume to compare with the others .
To settle whether the atomic volumes are multiples of the same number requires a more exact determination of densities .
Mr. SWANN is at present investigating this question .
The values of are Na Rb Cs They indicate the same value for the ratio 2120 ) , but only by stretching Na and Rb to the extreme limits .
It is noticeable that the ratio as determined from the most probable value continually decreases with increasing atomic weight .
When these relations were first observed the for used had been determined on slightly different data , and the value of was 4 , well within the maximum limits above .
In this case the series of values of showed a superficial agreement with the atomic volumes , but it was not possible to correlate them with complete exactness .
It was attempted to do so by adding a constant to .
It was then found to give good agreement if this constant were about 014 .
In other words , add 014 to the fractional part of and deduct it from the integral , i. e. , instead of writing , write ; then this fraction was proportional to the atomic volume .
It will be remembered that the difference found between the 's of the and series was about 487 , so that the 's of and are of the form and .
This agreement for about 987 from two lines of agreement strengthens the case for each .
Again , if the values of be calculated , there results Na Rb Cs PROt .
W. M. HiCKS : A CRITICAL STUDY OF SPECTRAL SLRIES .
a much nearer approximation to equality than before ; in fact , 1950 well within all limits .
We are therefore led to surmise that the series of all the alkalies can be represented by a formula , VS ( 1 ) atomic yolume the numbers inserted being only very roughly approximate .
In other words , the knowledge of four constants gives all the lines of all the series of the alkalies ( Li not considered ) .
On the other hypothesis , , the excesses of over 1 , exact multiples of a constant , and the ratio of also a constant , ought to exhibit a similar relation .
Now is determined with a far less error than either or separately , and hence the values of ought to show the relationships more clearly .
The respective values are 6 which is very strong evidence in favour of the multiple law for the first three at least .
The limits of possible variation are so close , the exactness of the equality within those limits so absolute , that there can be no doubt that the relation is an absolute law .
We have found already that the same law is indicated for and , but with greater deviations ; in the case of Rb outside the permissible limits .
It is clear , therefore , that the inequalities of and must counterbalance one another when , i.e. , there is probably a term in addition to the term .
( Here and for the future stands for the integer proper to the particular element , 2 , 4 , 5 , 6 , for the factor , and for the constant ratio of to ) Later , evidence will be given of a new type of series the based on an atomic weight term which ( and multiples of which ) , as will be seen immediately , determines the second ( or second and third ) of the doublet ( or triplet ) series .
The values of these terms are determined in the next paragraph .
If in the above be taken , one-half of these values , and deducted , it will be found to give the perfect agreement required for Na , , and Rb .
If denote this atomic weight term , for the four elements are 000370 , 001466 , 006444 , 016267 , and the values of for these elements are then following terms:\mdash ; .
2 .
5 ( 89 ) ( 85 ) .
6 ( ?
) ( ?
) PROF .
W. M. HICKS : A CRiTICAL STUDY OF SPECTRAL SERIES .
For the first three , the factors are easily equal within limits , If we remember that the limits for Rb were based for on LEHMANN'S estimates of error , which are rather too small , we may represent the degree of closeness of the above numbers by attaching a possible error of about ( 80 ) .
With the above numbers the ratio is 21521 .
The formula for Na , , Rb , is therefore with uncertainties in the two last digits in each number .
The new arrangement has , however , thrown Cs quite out of order .
Not only is the factor too large , but the ratio of to it is also too large , viz. , 2276 .
If the term in be omitted .
To bring this into agreement with the above would require a change in of about 003600 , which might just be possible .
As we shall see later , .the series appear to depend on fundamental types in which where are integral .
It is just possible that in Cs the type changes from one to the other .
The preceding arguments may appear unsatisfactory , as they apparently lead to two different results with equal evidence , but there is reason to think that if the values for the atomic volumes are known with greater accuracy , they will be found to agree .
If not , then the second argument stands , and the former falls , as based on incorrect values of the atomic volumes .
( J ) .
The 's of .\mdash ; A glance at the table shows that the are practically the same for both and .
The doublet separations , therefore , are due to a change in alone .
In.determinin the values of use may be made of the values in the table ; but as the doublet separations are much more accurately known than the wave-numbers of the lines from which the constants have been determined , it will be better to calculate the differences between the 's of and on this basis , assuming the of correct .
Any small error in the latter will have an infinitesimal effect on the difference in question .
Taking the following values of as the most probable , Na , Rb , Cs and taking , there result for ths differences of the 's of and Na .
K. Rb .
Cs .
or 01406 ' PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
We might have expected that the differences would have colne out to the squares of the atomic weights .
It is not so , however ; nor is there a regular progression , as the order of increase is Na , Rb , Cs , K. In the original calculation the limits of were found from the observations , and not by making , except , of course , for Na , where only two lines of had been measured .
In this case the difference of the 's of and involved as well , and an inspection showed that the ratio of S to S was about for for Rb , and for Cs .
If this apparent rule be made exact and put in the form where , as before , denotes the element multiple , we find the following values of :\mdash ; K. ; Rb ; Cs .
, whilst the result for Na is without a term in .
This points to the probability that the change in is accompanied with a change in , and that the change in is a constant multiple of , whilst the change in depends in a more complex way on the properties of the element .
The term looks like a first approximation from a complete formula .
In many cases , for instance , the form of denominator reproduces the lines with great accuracy .
If the were decreased by the change in would not be in a constant ratio to , but would involve and , i.e. , .
The doublet separation is given by the difference , when .
In the approximation arrived at above it is for Na , and for and Rb , and Cs , but second order terms are required for the large of Cs .
( K ) .
's of .\mdash ; Since the are very close to UP ( 1 ) , we should expect to find a similar relation between the 's of as for the series .
The actual numbers are , writing down only the fractional parts , Na Rb Cs ?
but there is here an additional uncertainty owin to the fact that , in calculating the series , the means have been taken and the , determined from their means by an estimated value of .
The limits of the series are much closer to the .
Naturally , therefore , they bring out the relationship rather more clearly .
( L ) .
The 's of S.\mdash ; The 's of , as we have seen , suggest that we have to do with .
If we write down the fractions , we get Na Rb Cs ? ?
VOL. CCX .
\mdash ; A. PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
The numbers do not suggest the relationship so strongly as those of the preceding cases .
The limits , however , are so wide that it is the only one in which the absolute identity of the common factor is possible without overstraining them .
The values of the are for all the elements less than for the series .
They are , for the elements in order , easily 0120 within limits of error , but the limits are wide .
The ratios of are 0170 , 0140 , 0164 , 0140 .
They are not constant , but their irregularity points to a want of law in deviations from equality , or , in other words , that equality exists in the true formulae .
There seems something defective in the form for the series , as has already appeared in its failure to give the first lines , as well as to reproduce the values of exactly .
The Diffuse Series .
The observations for the series are not sufficiently exact to enable such certain conclusions to be drawn as in the case of the other two types .
Besides , there seems at first sight to be a change of type as we go through the group of elements .
Satellites clearly show themselves only in Cs .
There is some indication of satellites in Rb , but , as we have already seen , there is considerable doubt as to their actuality .
shows no sign whatever of them .
On the contrary , Na does , not by offering satellite lines themselves , but by a regular increase in as the order increases\mdash ; a poin already discussed under the remarks on Table I. The satellite series in general plainly suggest an analogy with the series , in which the lines are satellites to the , and it is natural to search for series whose limits are the and giving corresponding doublets .
Working on this basis for the clearly defined satellite series of Cs ( satellites for ( 3.4.5.6 ) obseryed ) , we get two values for the of the top lines of the formulae determining them which give a separation of , and , as we shall see , determine the limits of an additional doublet series .
The satellite series are more fully and more numerously developed in other elements than the alkalies , and their general discussion is better deferred until the other elements are discussed in a succeeding communication .
In determining the value of the question arises whether should be taken to be a fraction or .
If gives a line too far in the ultra red to have been observed .
The observations of kno lines by themselves are not sufficiently exact to decide the question\mdash ; as , e.g. , in the series .
If we regard the doublet separation of an associated series as determined by the top lines of its principal series , then , as we shall see shortly , the evidence in the case of Cs points to the fact that the denominator of the top line , i. e. , ; but there may be some uncertainty here also in view of RYDBERG'S suggestion that each V of a series may form the limit PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL of a new series\mdash ; e.g. , not only is there the sharp series UP ( 1 ) , but there may also be UP ( 2 ) .
RITZ has given several examples of this .
As the matter is thus open formula as determined on each supposition is entered in Table II .
In the and series is negative .
It is positive in the series , only in the alkalies , but in other typical series ; in fact , so general is the rule , that if new series gives positive it may be taken at least provisionally as a series .
( M).\mdash ; If we write down the values of of of and of for , we find the following:\mdash ; Na . . . . . .
Rb . . .
Cs . . . ? ?
numbers for Na look abnormal .
We shall see later that the series belongs to a sequence of a new type , for which the notation is proposed .
It need not , therefore , be here further considered .
The values of for the others ) oint to a constant value , which suggests that is of the and of the form .
Now we have seen that is of the form .
If we calculate the values of we get the following values for , Rb , Cs , using the mean value of , 074560 , .
The first two agree with the values found for , for , and ) well within the limits of error .
The third is obtained on the supposition that the atomic weight is as for the others , and it is clearly too large .
But in the Psel.ies we saw that apparently the term had fallen out .
If this missing term be put in here , i.e. , deducted from unity in place of , the value of comes out to 520205 , well permissible limits of the of or .
The of is 008168 greater than that of as calculated from the obsel.ved lines ( see Table IT ) .
Now is Hence , with the satellite series of Cs , the 's are given by for or the values for and Rb being The cases of and Rb , both for and , falling into analogous forms , while in that of o both and show a difference , although depending on the same kinds of PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
constants , forms another argument in favour of the view that Rb and actually do not possess satellite series like Cs .
They have not been observed\mdash ; not through the observer 's fault , but because they do not exist .
The limits of errors are so much wider in the case of these series than in that of the that the evidence in favour of the above is not , perhaps , so as for the latter .
I think , however , that there can be no reasonable doubt that it is correct in the main .
It is quite possible that , in place of unity from which the terms are deducted , the true quantity may be a number near unity for the alkalies .
The values of are so uncertain that it is hopeless to expect to draw any safe conclusion from them , beyond this that apparently the simple law observed in the case of the series does not now hold .
For the value of is very close to .
For Rb it may be written For Cs the is very small for both satellites and practically equal , the 's differ only in .
The constitution of for must therefore , unfortunately , with the extremely inexact measurements , be left unsettled at present . .
Associated Series .
In discussin the remaining observed lines of the alkalies.not included in the three typical groups , it will be simpler to take the elements in the reverse order , commencing with Cs .
Cs .
The lines to be accounted for are the following , in which the letter refers to the observer:\mdash ; 96 .
98 .
97.38 .
5209 .
97 .
98.38 In the above denotes the difference of wave-numbers of the lines and below it , the top line being taken when two observers are in question .
The pairs are PROF .
W. M. HICKS : CRITICAL ) lecognised as formin a doublet series .
We have seen that this value is the difference between the and calculated from the formula , not having yet been observed .
It is natural , therefore , to look to the series as those in question in the same way as the depends on the , i.e. , the limits of the new series should be and or , , about 17000 .
If we employ this , however , it is at once seen to be too .
If the lines 8082 , 7280 , C872 are used to calculate the constants , there results which is of the type .
The formula does not look right , the limit is rather far from , and , moreover , it is the first case where there has appeared a value of which passes from to as increases .
The formula which bounds , so to say , the change from the type to the type , i.c. , , and satisfies the two lines 8082 and 7280 exactly , is and it gives a better agreement with the other lines , as is shown by the following table , in which only the values the first of each doublet is given for the former formula .
The differences are for obs.-calc .
and is taken to be For the last , the measurements of give or 13 too small , i. e. , a wave- length difference .
too small .
If the second line is correct , showing good agreement with the formula , the first would therefore be wrong instead of As is seen , the second formula gives extremely good agreement .
If the limit is higher than 16810 the series becomes a type .
In these inexact measurelnents , however , it is probably best to use RYDBERG'S form without the now the limit is not that calculated for ; there may be a small error in the latter , but hardly to the extent of 200 .
We may feel as certain , however , that the limit and VD ( 2 ) are the same as for the corresponding case of and , in which the " " error\ldquo ; is also considerable .
The strongest evidence is in the striking fact that the and series generate the doublet separation .
For , which would be far outside BERGMANN'S region of observation .
The only lines now left unaccounted for in Cs are 13711 B. and 5209 R. The wavePROF .
W. M. HICKS : A CRITICAL STUD OF SPECTRAL SERIES .
lumber of 13711 is roughly , , so that .
There is just a possibility , therefore , that 13711 may be the latter ; but , if so , it is curious that the should come in and not , which we should expect to give a stronger line .
Rb .
The lines to be accounted for are 13442 B. , 10069 B. , 8872 B. , 8271 B. , L. , R. , R. , 4967 R. , to which should possibly be added that hitherto supposed to be , L. The first four form a series .
Using the first three for the constants , there results This reproduces the fourth with an error 10 A.U. , which is not excessive when the inexactness of the whole measurements is considered , but they are sufficient to settle that the limit is close to 14295 .
The value of as calculated is 14270 .
BERGMANN'S value for has a wave-number 6487 , which with the reliable value of gives VD ( 2 ) , but we have seen that BERGMANN'S value is probably considerably .
The evidence , howeyer , is sufficient indicate that Rb also has a series in which the limit is VD ( 2 ) .
The corresponding series in Cs was a doublet owing to the series satellites .
The absence of duplicity here , , favours the view put forward on other gro.unds , that is not a satellite series .
On the contrary , it is possible that , if there were a doublet series , the doublet separation would be too small for BERGMANN to have separated : for instance , he did not separate , which he observed as one .
The formula gives the next two lines to be and .
The first , allowance being made for its uncertainty , probably comes close to 7950 , which is , and the latter may well be 7753 , the line which has genera ] been taken to be , but which we have had reason to suppose not to belong to the series .
The lines L. , R. , 4967 R. satisfy the equation for and 7 .
is ractically the same as for the series .
The missing members should be at 5602 , 5270 , 5084 .
The first has not been observed , the others are in the neighbourhood of ( 5260 ) and ( 5089 R. or 5088 S The limit , so that it is not apparent how it arises .
The arrangement does not give much confidence , although the connection with is curious .
Of the other line , 5165 R. , SAUNDERS has pointed out that it is near the ed ge of a carbon bond , and is therefore a doubtful Rb line .
17040 is extrapolated from a not very accurate formula .
is probably somewhat greater than the limit in the ssries just discussed , , 16810 , making about 16910 , which would give 7275 ?
agreeing .
PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
K. The lines to be accounted for are the P. 4829 R. .
P. 4767 R. . .
9590 B. 5 S. 8908 B. 8500 B. R. .
The lines marked with form one series .
The constants calculated from the first three ( third inexact ) give .
This reproduces the lines with the following values of obs.-calc .
:\mdash ; 3 4 5 6 7 8 9 10 27 The lines corresponding to would be , which lie below BERGMANN'S lowest region of observation and ' upper limit .
two observations are good , BERGMANN'S may quite possibly err to the amount indicated .
* There can be no doubt about the series\mdash ; the formula is only doubtful in so far as the third line from which it is determined may be several units out .
The value of the wave-number of is by PASCHEN'S observation .
The limit of the series is ( see Table I Hence the value of VD ( 2 ) is .
The most probable value is .
The limit of the series as found is so close to this that there is some justification in taking it to be exactly so .
If this be assumed , there are three reliable data to go upon , , PASCHEN'S two lines and the limit .
These give the formula giving the following values of obs.-calc.:\mdash ; The agreement is much better for BERGMANN'S lines , but the last is too far out to allow to be included in the series , nor can this be set down to uncertainty in the constants if the supposition of limit is exact for the kind of formula .
If it be remembered that , so far as the matter has been yet considered , the value of is never exactly VS ( 1 ) , and that a similar cause for the apparent difference may occur here , we may still hold that in fact the limit is the true , but that the first formula is the most correct to use with the actual form of the formula *E.g .
, in the lines 27215 and his differences from PASCHIBN were and 16 .
PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
( which involves a particular value of N ) .
This gives a better agreement , and clearly brings into the series .
If we employ ] formula , with errors clearly the limit is too low .
PASCHEN has pointed out that and 27215 are respectively and VS ( 2 ) .
How closely this is the case may be illustrated by calculating these values with the limits of errors .
and are and , with possible errors .
Using the calculated limits of the series ( Table I. ) and PASCHEN'S value of , the value of VS ( 2 ) is , giving respectively from the aboye relations and .
The wave -numbers of the actual lines are and .
Still it is a single agreement , and might possibly be fortuitous .
The measurements of wave differences are much more reliable than those of actual lengths .
Now the difference for the two lines in question gives , with PASCHEN'S estimates of his errors .
The difference for and is which is decidedly against the conclusion .
In favour of it , on the other hand , is the fact that a similar relation shows itself under Na .
There is another unallotted pair , R. and R. , with a separation , where denotes the wave-length difference error of observation .
The first has a wave-number about , as against .
It is difficult to see where a doublet with about 17 to 20 could come in unless the series was a satellite one , in which case the first series here considered should be a doublet one .
If so , however , it would have been seen by PASCHEN , who separated much closer lines in the ultra-red Na spectrum .
The other lines left over are 4829 R. , 4808 L.D. , and 4767 R. Na .
The lines observed in Na not belonging to the regular series are much more numerous than in the other elements .
The following is the list : . . .
1 . . .
P. 5100 LE .
S. P. 6 .
4975.0 . .
7418.3 S. 7410 S. ( 7)4372 S. S. S. PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL RIES .
From analogy with the preceding elements it is natural first to seek for the presence of a series whose lilnit is VD ( 2 ) .
This is The lines 18459 , 12677 fall in with this , but no others of the series have been found .
If the limit is actually 12267 the formula can be found at once .
In all the other cases , however , it has been slightly different .
It will , therefore , be best to use the old RYDBERG formula , whence .
The next three lines given by this are , 9963 , 9469 , but these come in a region in which there are no records .
It is below lowest and ) ' highest .
The limit falls about 8147 , close to 8210 , where observed a faint group .
rfhe latter might , therefore , represent the last faint lines visible .
The four lines aboye these in the list are doublets and PASCHpN points out that they satisfy the relations , , .
To test this , as in the case of is .
Hence , whence the waye-numbers compared with observed would be Calc .
, ; Obs. , Here also , contrary to the case in , the test as to doublet separation holds .
The observed for the lines is , agreeing with the separations in this case .
The lines marked with A in the list form a doublet about 147\mdash ; which has been called LENARD'S series , who firstdiscovereds ; ome of the lines , later measured by K.H. and by S. The lines were arranged in the by SAUNDERS , who , however , could not fit in any of the usual formulae for it .
The readings are too uncertain for any definite assertion .
If the first three corrected to constant be used to determine the constants , each of the second doublet being means of K.H. and S. , the resulting formula is , 85.4 giving the following values for obs.-calc .
:\mdash ; 1 2.5 VOL. CCX .
\mdash ; A. PROF .
W. JM .
HICKS : A CRITICAL STUDY OF taking values , except for the last but one , where LENARD'S is taken , SAUNDERS ' reading would give a difference about greater , or When we the remainder , considerable difficulty arises owing to the number of ill-defined lines so close together .
In some cases K.H. and S. give very discordant readings for doublets , which yet are so close that they look as if the measurements might refer to the same lines , whilst in other cases the same observers agree much more closely .
It is to be remembered that apparently the method of illumination has a great deal to do with the development of some of these lines in the spectra\mdash ; in fact , they appear to come from different sheaths of the flame or arc , as if it were a question of temperature .
The doublet at 5675 was observed by K.R. , and was the only observation of theirs which did not fit in with the ordinary series .
It is peculiar in being diffuse towards the violet .
Diffuseness to the red is a general peculiarity of the diffuse series .
If we look for the seat of this diffuseness in the formula , it would appear to lie in the VD part of the formula through diminishing denominator .
If a series were such that the VD came to the limit\mdash ; as , for instance , in many of the new series\mdash ; the VD would be additive instead of subtractive , and the change in the denominator would now produce a diminishing wave-length or diffuseness towards the violet .
RYDBERG'S tables show that this doublet 5675 , the doublet at 4976 K.H. , and 4660 range near his formula .
They give , if K.H. value for the second is taken , , being respectively 3 , 4 , 5 .
On the contrary , if we take 7418 S. , 5675 K.R. , and 5100 , we get a series including in addition 4820 LE .
, 4660 S. As LENARD'S readings were very rough , we may take the ] and the last .
There results giving for obs.-calc .
:\mdash ; which is sufficiently good agreement .
The lines 5100 and 4660 , however , have been observed by LENARD alone , and neither by S. or K.H. It does not look as if such lines , if really Na , would come next to so marked a line as 5675 K.R. The allocation of these lines to one series is therefore very doubtful .
It is against the former that no other lines come in ( a preceding line should be about 8025 , and not likely to have been observed if existent ) ; but , on the contrary , the nature of the lines is more correspondent , and it should be noted that the limit is within the limits of error of the LENARD series .
If the apparent agreement is real , the limit would be very close PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL to 24558 in both cases .
If the LENARD series be recalculated with the limit giving obs .
-calc .
:\mdash ; S. making the fourth lie between the readings of K. and S. The agl.eement is improved .
There remain 7418.3 S. , 5100 LE .
, 4820 LE .
, 4730 LE .
, and possibly the doublets in which K.H. and 4633.1 4910 . .
4629.4 K. H. Of these , , 4633 , 4470 are shown by RYDBERG'S table to have a limit about 24466 , i.e. , the S. and D. series and 's respectively , which clearly refer to the 's of .
To test this , find the differences between the wave-numbers of these lines and the corresponding values of , and suppose K.H. and S. values may have observational error 1 A.U. and LENARD 5 .
Obs. for K.H. LE .
S. S. S. 52:69 K.H. LE . .
The numbers would seem to show that where K.H. and S. give doublets not quite the same , that of highel wave-number must be taken .
The differences are of about equal value , and their mean is ; using this for other lines gives :\mdash ; . . .
13324 has not been observed , is near the LENARD ) is also near the line S. .
limit of the series is * It was originally done with this , and nothing is gained by culation , \mdash ; 169476 4144676169\mdash ; 169476 , say .
This , of course , suggests the same limit as the S. and D. series , which is .
But the limits of error show that this is not the case , but that the difference is real .
The limit of is higher , or .
Another higher would give .
In view of the actual limit , we might provisionally fix it at 24506 .
The second line of the doublet would then be .
An additional lands us , which is close to the ?
of the two series considered above .
These results can hardly be fortuitous .
The limit of is generated by the series , bein ; the limit of is found from this by deducting the atomic weight term , , from the of .
The limits will therefore be generated by writing in has integral values from to 5 , and possibly to 7 ( LENARD'S series ) , both inclusive .
RITZ has given the doublet series just considered as , .
We have seen , however , that this cannot be the case .
* Moreover , for would give an infinite time of vibration , and would suggest instability .
In fact , the configuration giving this series , containing one very long period , is apparently very slightly stable .
It is possible that the ought also to be slightly different , but the observations are far too inexact to settle this .
Again , RYDBERG'S tables show that 7418 S. , 5 ] 00 LE .
, 4472 S. have denominators about , and , with a limit 27026 .
Working from this they satisfy the formula .
27034 is not from the value of VD , which is .
But the interest of this allocation lies chiefly in this , that its is of the form we should expect for the real series , if were formed on the same plan as for , Rb , Cs , in which we have seen that the is with undetermined .
The closeness of the agreement is chance ; the errors in observation are too great to give with certainty , but it is suggestive .
It will be convenient , for reference , to attach letters to the various series just considered .
In the list the LENARD series is denoted by , the second by , the third by , although a very doubtful one , that parallel to by and the last by G. Collecting , we find { VD ( 2 ) } * Unless S. is 4 A.U. in error , which is not likely .
PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
All the terms in have little significance , as the lines are too uncertain to enable to be determined with any approximation .
The only lines not allocated are LENARD'S 4820 , 4730 , which have not been seen by any other observer .
Comparison of the Different Elements.\mdash ; The first certain result is the appearance of a strong well-deyeloped series in the ultra-red , whose limit depends on the value of VD ( 2 ) , giving doublet series when VD has two values as in Cs .
The limits , as calculated from three observed lines , differ from the exact values of VD by amounts increasing from Na to Cs in somewhat the same way as in the corresponding cases of UP ( 1 ) and , and will probably be accounted for in a similar way .
If the limits are to be VD ( 2 ) exactly , the resulting formulae are of the type , as found from two lines .
Na and give the type , Rb and Cs the type .
The observations are not sufficiently accurate to definitely settle this question .
Under the circumstances it may be best to use RYDBERG'S formulae , i.e. , .
In this case the is always not far from unity , and the limits are closer to VD ( 2 ) .
The values of for Na , , Rb , Cs come out respectively 9986 , 9965 , 9798 , 968 , the first two only being reliable as deduced from PASCHEN'S observations .
Takin his estimates of observational errors , the values for Na and are respectively and where are the fractions of his possible error .
It is seen that they may possibly be equal , and the possible errors for Rb and Cs do not contradict , that they also may have the same , about 997 .
Its possibility , also , may be further illustrated by finding the difference of wave-numbers between the lines of , Rb , Cs for and ?
, which have been measured for all three . . . . .
Rb Cs . . . .
denotes B. possible error .
It has been taken about 16\mdash ; his actual error as against PASCHEN in KF ( 4 ) .
It was my first impressio1l that this would be the same for all the alkalies and the 987 of the earlier discussion .
It would be the basis of a fundamental series for each element , and the letter which has been attached to it had its origin in this idea .
But RITZ , who has discussed these series very fully and compared with those in other elements , oposes to represent the denominators by the difference of the nators of the and series .
At first sight this may appear a fanciful relation , but if it be remembered that this difference is merely the atomic weight term , whose subtraction from the atomic volume term in gives the of , the supposition is quite reasonable .
His proposal , then , comes to this : to put where is the atomic weight term .
Now , the values of are about , 002933 , 012887 , 032435 , their differences from unity giving ( to four figures ) 9993 , 9971 , 9871 , 9676 , which march closely with the values of obtained from the lines as observed .
There is something to be said , therefore , in your of the view that the value of for this series is 1 -atomic weight term , though it cannot yet be regarded as certain , against the supposition of constancy .
The following table is interesting as throwing light on this point .
It gives the differences of wave-numbers between successive lines for the elements indicated:\mdash ; TABLE III .
What tion of change there is , is that there is an crease with increasing atomic weight .
It is just possible , however , that they might be the same as the possible limits of error inserted below each number shown .
Where no number is inserted ( and He excepted ) the limits are very large , but no definite values can be assigned .
At the same time the evidence is in favour of RITZ ' .
In accordance with what has been learnt as to the formation of the and sequences , I was inclined to think that the true denominators of the sequence were , the alkali and series depending on its second term , giving ] .
the basis of the and D. Such a form rees well with the observed fo and where the measul.es are llot exact .
When , however , it is applied to PASCHEN'S observation of the two first lines in Na and it PROF .
W. M. HICKS : A CRITICAL STUI ) OF SPECTRAL is seen not to hold , within at least PASCHEN'S estimates of his ( possibly estimated as compared with K.R. scale ) , but that RYDBERG'S form gives much closer approximation .
Thus , Na with , the differences of wavenumbers for , , would be , and without the term with they would be , 77 , both within the limits of error for the second line .
The value of the limit is nearer VD ( 2 ) in the first case .
There is nothing to decide between the two theories here .
But a similar treatment for , where , gives in the first case with differences 88 and , both close to PASCHEN'S estimates of error and , whilst without the term and the differences are .
This definitely points to the RYDBERG form as the preferable one .
On the other hand , with Cs , , where the term is large enough to be really important , agrees better with the whole set of observations .
Comparing the wavelength differences with those for the RYDBERG form we get the following table:\mdash ; For Rb , , giying Thus , as between the two suppositions , Rb gives no indication ; Cs is in favour of and its large value of makes it important , but it is not decisive ; Na is satisfied by both , but is better for RYDBERG ; is the only one strongly in favour of RYDBERG .
The question must therefore be left open at present , with leanin in favour of RYDBERG .
The validity of the supposition , that lines exist whose wave-number is the difference of the V part of the wave-numbers of two other lines , may be regarded as fully established by RITZ , calls it the Principle of Combination .
Examples have occurred amongst the lines just discussed .
It is possible , however , that what really happens is not an addition of unity to , but an addition of our alkali constant .
987 ?
to the , e.g. , while we write a new limit as , it may actually be .
To settle this requires greater accuracy in the obscure lines than is to hand .
* ' Ueber ein neues Gesetz der Serienspektren ' Phys. vol. 9 , p. 621 .
PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
It appears that , whatever the kinetic configuration may be which is the source of the vibrations , the light periods depend on the rence of frequency of ) systems , each with distinguishing train of ncies .
This train , or sequence , we have denoted in the foregoing by putting the letter V before the letter representing the series .
In what succeeds we will refer to them as the -sequence , -sequence , and so on .
We recognise four typical sequences , the , and the new type .
The evidence adduced in the preceding goes to show that the type is the fundamental or primitive type .
The is formed from it by adding , where are constants , or atomic volume term , the by deducting the same term or the atomic volume term , with some as yet unknown law for the term in The term is found roughly by deducting a constant from the of P. The theoly will fall in , as we shall see , with the Li , , and He spectra also .
Regarded from this point of view , we may look upon as standing for positive , for difference , and for semi .
Lithium .
It is now time to take up the discussion of the spectra of lithium in the light of the knowledge gained from the consideration of the other alkali metals .
The spectra of Li have been arranged like the others in three typical series , called the , and series .
Taking these letters at present to stand for the arrangement hitherto accepted , and using the first three lines of each to calculate the constants , there result the following formulae :\mdash ; P. D. The first glance shows at once that the lines ascribed to the series follow a type , and those of the look like a type .
Also the values of the constants are very different from what we should be led to expect from the relationships discovered amongst the other elements .
We should all the more expect to find these relations to hold , since Li falls in so closely with the analogous relationships for the atomic volume .
In the above formulae of ( 21 ) , , ( 13 ) of ( 23 ) It follows at once extremely closely that the laws ( D ) , ( F ) hold , viz. , UP ( 1 ) , PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
07 also ( G ) is nearly so .
We note that Li agrees with the her elements in making , and it differs also in the same direction .
The form of recalls at once the formula for the sequences found in Cs and Rb .
In fact , a RYDBERG form would well serve it .
To test this we can find the differences in successive wave-numbers .
They are entered in the table above with the other elements , and an inspection leaves no doubt but that here we have the type .
is clearly a type , although we should have the to be somewhat smaller .
The third series is thus VD ( 1 ) , and the first VS ( 1 ) .
The usual series , however , gives , and we ought to expect to find it .
Now , certain additional lines have been observed by K.H. and S. , S. , S. , 3924 K.H. If they are the series in question their wavenumbers ought to differ by the same amount from those of the corresponding lines in the old P. The differences are In finding these the means of K.H. and S. used , except for the last , which S. has not observed , but EXNER and HASHEK have .
The permissible errors have beeIl taken wice the difference of the readings of each two observers .
The difference is close to 14904 .
Since the limit of the old is , the limit of the present is the difference or , the same as for the old S. and D. It should clearly .
the same .
Hence the formula is , the ordinary type .
In all the other elements we have met with series of the type with limit VD ( 2 ) .
Now VD ( 2 ) ( extrapolated from the formula ) is 12557 , and lines ol the new series , if existent , ought to have wave-numbers below those of the old D. The wave-numbers calculated in this way would be 5699 , 8169 , 9511 , 1032 ] , 10859 , sub.iect to uncertainties of a few units owing to the want of exact knowledge of .
The first two lines clearly appear in PASCHEN'S list of new ultra-red lines with waye-numbers and , or and There are two lines of PASCHEN'S , whose difference of wavenumbers is .
Table III .
shows that they must have reference to the sequence , and must be of the form A-VF ( 3 ) and A-VF ( 4 ) .
The actual values of their wave-numbers give Now the of Hence these two lines are VF ( 2 ) and VF ( 2 ) .
There are no apparent signs of series of the positive type .
If we apply the multiple laws deduced in the foregoing we should get for the of VOL. CCX.\mdash ; A. taking the atomic weight term as about that of Na , i.e. , giving UP ( 1 ) .
The of the vould be , about more than of , and the about 8 that of , judging from analogy with the other series , i.e. , whence The values for are more difficult to estimate , owing to uncertainty as to the The , and the denominator .
Whence VD ( 1 ) .
These give The or the first of the typical principal series would give would be about .
Neither have been observed .
The and should be doublets with about .
The , is clearly the same as PASCHEN'S 3719 .
If we make it so , must be given by error in , or error in is probably considerable , which would make a small positive quantity .
Not only is the deduced line close to the observed one , but the observed is a doublet of the proper amount .
The doublet would be at , and has not been seen .
The foregoing is professedly speculative .
If the relations found for the other elements really hold for Li , it shows that while in the others all the energy goes into vibrations of the positive type , in Li it goes into those of the negative type .
It might be objected in the above , that there has been no evidence that the and forms , when used in a positive group of spectra , are different from the corresponding forms in a negative group .
A similar difference will , however , be shown to exist when the spectrum of helium is considered .
As we look down through the different elements we find traces of the negative type , ( VD-VP ) .
In Na the series is not the ordinary but the type , but we have seen indication of three lines PROF .
W. M. HTCKS : A CRITICAL STUDy OF SPECTRAL of the typical form .
In Li all the negative rroup come only one indication of lines of the positive .
It is apparent , not that the fol.lnulae have ceased to follow regular laws , but that the has passed from one set to another .
It is necessary to consider whether the identification of PASCHEN'S doublet with the first line of the series will hold if we the and forms to be the same for both positive and negative groups .
The difficulty is , there is no appearance of the form .
It is possible , however , to determine UP ( 1 ) so as to make the identification correct , and see whether it leads to a likely value .
Now VD ( 1 ) Hence UP ( 1 ) or UP ( 1 ) or 24862 , giving values of the denominator of respectively or .
The former is quite out of character , but the'latter is more in character with analogy .
It makes the of only less than that of Na , and the fractional part greater than the of the set , which for is 089 .
This fractional part is greater in all the previous cases .
The identification , although possible , does not commend itself .
Hydrogen .
For our present purposes the spectrum of hydrogen is chiefly important , because on it.is based the determination of the constant N. The value was determined by RYDBERG on the supposition that BALMER'S formula was correct , by applying the method of least squares to ' measurements of the first 13lines .
I have already given reasons why , in deducing IiYDBERG'S formula should be used preference to BALMER'S\mdash ; the probability being that would be so small that the term in would be quite negligible .
Also , since AMES ' esults , EVERSHED has given measurements from solar prominences down to .
For these reasons it seemed desirable to recalculate the value of with RYDBERG'S formula .
ROWLAND'S measurements were used for HL ) ( 3.4 ) .
For HD ( 5 ) the of various observers differ greatly .
The work was done twice\mdash ; first for AMES ' value of HD ( 5 ) , weighting it one-half the others ; and second , leaving out the line altogether .
HD ( 6.7 ) have been very carefully and specially measured by The remaining 24lines used were from EVERSHED'S measurements , all of course corrected to vacuo .
In absence of any indication to the contrary , all observatioris were treated as of equal value , although as they are prismatic measurements it might have been better to 'Phil .
Mag , p. 33 ( 1890 ) .
'Phil .
Trans vol. 197 , p. 413 .
Astro .
Journ xxviii .
, p. 72 .
take wave-numbers as of equal value .
It would , at any rate , have made the calculation simpler .
The results are as follows:\mdash ; With AMES ' HD ( 5 ) without The limits are practically the same .
They are .
From the latter formula the values of obs .
-calc .
up to , and then at intervals of 5 to 30 , and the last 31 are as follows:\mdash ; 3 4 5 6 7 8 9 10 15 20 is the doubtful measurement .
In spite of the good agreement , however , the observations on which the determination of is based are not at all as accurate as is desirable .
PICKERING'S measurements of HS in Puppis are far too inexact to give any information as to the value of N. The most accurately measured spectra are those of helium by RUNG and PASCHEN , to which we next proceed .
The HD and series should be almost identical , so that the observed lines should be close multiple lines , for which , as is well known , there is good evidence .
The combination principle in has recently been illustrated by the discovery of two ultra-red lines by PASCHEN belonging to UP ( 3 ) , , 5 .
Helium .
The observations for the spectrum of helium are due to RUNG and PASCHEN* for the visual and photographic region , PASCHEN for the ultra red , and measurements of certain visual lines by interferential methods by Lord RAYLEIGH .
R. and P. estimate their possible errors to amounts of the order of .
There is no doubt but that their measurements give by far the most accurate values for any .omplete spectrum yet obtained , but the possible errors are certainly several times larger than their estimates .
They estimated their possible errors for and to be respectively 80 and 48 .
PASCHl ] later and very accurate determinations show that actually those errors were respectively 182 and 340 .
It is true that these were bolometric measurements , and naturally cannot be compared with the others , still * ' Astro .
Journ iii .
, p. 4 .
'Ann . .
Phys 27 , p. ( 1908 ) .
'Phil .
Mag .
( 6 ) , p. 548 ( 1908 ) .
PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
RAYLEIGH'S interferential measurements allow us to compare some of these lines with what may be considered their actual values , and comparison shows that the degree of accuracy is not nearly so high as that estimated by R. and P. The lines are given in the following list , together with the estimates of error by R. and P. :\mdash ; If R. and P. measurements are correct as a set they ought to show a constant ratio to the corresponding ones of RAYLEIGH , but a glance at the differences shows that this is by no means the case .
With so large a number of lines for comparison as seven , it would seem legitimate to deduce what this ratio ( k ) should be for the whole of R. and P. scale , and thus to correct these seven at least .
I have done this , making the mean square value of ( correct-actual ) least , treating first all observations as of equal value , and secondly as weighted according to R. and P. estimates .
In both cases the result is the same , .
If , as might possibly seem preferable , the ratio of difference to wave-length is to be a minimum , then .
On these suppositionoe we obtain the following corrections to apply to R. and P. lines to bring them into their true positions on their scale:\mdash ; dnII .
The List I.\mdash ; actual errors a minimum\mdash ; appears the better , in view of the fact that some of the corrections are within R. and P. estimates , and that are generally smaller than in II .
It would , perhaps , have been advisable to use these corrected values in determining by least squares , but the work had been done before results were published .
The work of applying the method to the six series is exceedingly laborious , and as th corrections themselves not quite free from doubt , it did not seem worth to recalculate .
any case , it is not probable that the actual values of would be greatly altel.ed .
All six series have been treated in the same way , the observations being weighted in accordance with R. and P. estimates .
In the case of the series , however : the top line has been omitted .
This was done because there is some doubt as to the correct form for the formula , and any term has much less effect on the lower lines .
Consequently the value of is less likely to be affected .
The resulting formulae Singlet Series . . . . . . . . . . . . .
Doublet . . . . . . . . .
The values of obs.-calc .
are given in the following table , with the corresponding values from RITZ ' formula , although they are not comparable on the same basis , except possibly in and where the is practically the old , but RITZ gives no formula for the series:\mdash ; R. PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL * SCHNEIDERJOST , 'Beiblatter , ' xxviii .
, p. 1175 .
The agreement between observed and calculated is very good except for and , which is in accordance with what has been found in the previous cases .
As mentioned above , these were left out in finding the constants , as the chief end was to get accurate values of , and there was some doubt as to the form of the formulae being correct .
In comparison of RAYLEIGH'S lines we found probable errors in R. and P. measurements , amounting to eight or ten times their estimates .
If we take their possible errors as only three times their estimates , very few of the calculated values overstep these limits .
It is interesting to note that the least square treatment corroborates independently the errors determined from RAyLEIGH'S interference standards .
Thus , in , the differences in for 1.2.3.4 are , pointing to considerable error in .
If the calculated correction of II .
be applied to R. and P. , the difference should have been .
This would have been in the course , and then the method of least squares would have brought these into close agreement .
In fact , the considerable error in has pulled the calculated values for out of their proper places .
A similar result is shown in S If the calculated waye-length values of be compared with R. and P. corrected values from II .
, the differences would be , 007 , 007 , instead of , 007 , and if they had been used to calculate the constants the agreement would have been much closer .
So the differences for would be diminished from to 525 , but increased from to increased from 006 to The values of may be expected to depend to a few units on the weights allowed , and the accuracy of the measurements .
Allowing for this , the evidence , with one ears out what we have already had reason to infer , , that the and series have about the same value of , and that that of is larger .
The exception is .
The mean of of is , or practically old value obtained from H. We may feel , therefore , reasonable certainty that the actual value for these series is very close to 109675 .
The for the two series are again practically the same 109725 , but there is some uncertainty owing to the top lines having been omitted .
The case of is exceptional with the large value of 109815 .
The agreement of the calculated with the observed values is not so good as in the other cases .
It must be left at present as an apparent anomaly , but it may be noted that is a singlet series , like that of Li , in which a similar case arises .
In the alkalies we saw that for increased with atomic weight , and if this were the general rule we ought not to expect a large value of for helium .
For this reason He is not a good test .
We shall find better ones when we come to deal with other spectra .
It is seen that the limits of the and series for both sets are very closely the same .
Also and , whilst and , so that , as in the other elements , relation ( F ) is very closely verified .
On the other hand again , as in the other cases , relation ( G ) is only roughly satisfied .
, which differ from and by and respectively .
The two series run nearly parallel to one another .
The differences of corresponding wave-numbers are .
The two series run down to limits which will be very close to the true values for the last six lines , which give a mean of this , the values of for the two series differ by numbers which are given in the first row of the table below .
The denominators of VD are or roughly .
There are three cases to consider as to the origin of these differences : ( 1 ) an alteration in , the differences will then be SN/ ; ( 2 ) an alteration in , when the differences will be ; ( 3 ) an alteration in , when they will be .
The second , third , and fourth give numerical differences in each of these cases , the first difference being that observed , .
The fifth row gives the result of , dividing successive differences by 2\mdash ; in other , making the difference depend on a term 2 .
3 .
4 .
5 .
6 .
7 .
8 .
9 .
10 .
11 .
12 .
II . . .
84 62 47 37 III . . .
73 42 27 12 IV . . .
44 V. . .
84 42 21 10 05 VI . . .
92 53 33 09 04 02 PROF .
W. M. HICKS : A CRITICAL STUDY OF SPECTRAL SERIES .
The for in I. is clearly wrong , the mean of and 8 gives , so also is that for , and beyond this the measurement errors are clearly greater than th1 differences themselves .
It will be seen by insp ection that a change in alone gives too slow a decl.ease and a change in alone too quick .
A change in gives about the correct average descent with increasing , but , 4 are too much out to be due to mere errors of observation .
Row V. , term in , marches closely with I. On the other hand , VI .
is extremely close .
It gives the differences as calculated from the formulae obtained by least squares , which has met the difficulty by increasin by 17 and diminishing the denominator by .
Now for He is 000022 , which is evidence that we have here to do with the regular atomic weight term .
So far nothing has been said as to the true form of the series , but the usual classi. .
fication has been followed .
With our present knowledgge , however , a glance at once shows that the third series in each set belongs to the sequence , and that the sets of series should be classified as follows:\mdash ; glet seriespositive tublet series The sets are apparently independent , except as depending on the same He fundamental sequences .
In the singlet series the negative type is wanting ; in the doublet the positive is wanting .
They are , as is well known , developed in ways , and it is not to be expected that the sequence of the first is also that of the second , and the sequence of the second the same as that of the first .
The singlet series stands out by itself also as having a much larger value of for the sequence than in any of the other spectra .
In the doublet series , however , it might be expected that similar relations to those found in the other elements might exist .
For instance , the of VF ( 1 ) is 998547 .
The of VD is 9294 or 0691 ( say 0700 ) less .
If this depends on the atomic volume constitution of the term , as indicated in the alkalies , it should be possible to determine the atomic volume of solid He by comparison with any of the alkali spectra .
Now , with an atomic volume of gives for the corresponding difference about .
On this supposi$ion , therefore , the atomic volume of He would be .
The denslty of solid He should , therefore , be 38 .
KAMIIRLINGH ONNES gives that of liquid He at its condensing point as 154 .
The density of the solid would probably be considerably greater , but scarcely so large as 38 .
The deduction , of course , is not reliable , but is interesting as deducing a value of the right order of magnitude .
It may be noticed that the of differs from that of by of of by ; and of of by , all approximately multiples of the 075 which go VOL. CCX .
\mdash ; A. PROF .
W. M. HICKS : A CRITICAL STUDy OF SPECTRAL SERIES .
to build ) the alkali series .
Whether this has any real significance , however , must at present an open question .
Discussion .
From the foregoing discussion a few points stand out for further co1lsideration .
The most important is the doubt as to the form for the series .
Tho dissatisfaction with the form in for the series arises from the fact that VS ( 1 ) does not give , and that in some cases the value of calculated by it from constants determined from is considerably the limits of observation , certainly in the case of , and probably also for Rb and Cs .
This second failure , however , clearly is a consequence of the first , and will probably be corrected if the first is .
In addition , the of the is so close to the value of the atomic volume term in as to point to a real equality if the true form be obtained , whilst the value of in is always considerably less than that in P. It has been shown that the VS ( 1 ) cannot be made without introducing a term in , and that in the two llumerical examples given for Na and , the and come about equal\mdash ; in other words , a denominator would satisfy all the conditions .
Also the is brought nearer to the atomic volume term , but becomes about one-half that in the series .
Before considering this clue further , two other ways out of the difficulty may be mentioned .
RlTZ has systematically used ?
in place of in the series .
But the adoption of in place of makes the between VS ( 1 ) and greater , although it brings the more into agreement with that of P. On the other hand , , which makes the of in the alkalies practically the same as that in the difference the other way .
It is clear , then , that since integer makes VS ( 1 ) , and makes VS ( 1 ) , that a value of can be found , such that a denominator would make the limits come correctly .
In all that precedes , not only has the value of been taken to be a universal constant , but it has had the special value 109675 attached to it .
The evidence for the iversality of its value is not , however , conclusive , and it is clearly possible , if it be not so , to assign values to so that of , , and such contingency at least requires examination .
These three possible solutions will now be taken in order .
Addition of Term in -We have seen that it is possible to satisfy the required conditions in the case of Na and , but only by allowing actual errors in the measurements alternately positive and negative , and each a considerable fraction of the possible .
The necessity of taking in each case the errors alternately positive and PROF .
W. M. HICKS : A CRTTICAL STUDY OF SPECTRAL RIES .
negative makes this way out of the difficulty very doubtiul , and any one using KAYSER and RUNGE'S observations gets to doubt whether many of their actual errors are more than small fractions of the possible .
It would probably be possible to satisfy the conditions for Rb and Cs within the present possible errors , but only because they are so considerable .
I think , therefore , on the whole , the evidence is against this explanation .
As , however , the coefficients and were nearly the same in ) numerical examples , this point was tested for Na , usin for , PASCHEN'S , and K. and .
The resulting formula was ; but the deviations for the other lines were far too great , being several times the possible . .
\mdash ; The value of , as from the observations , will depend so much on the exactness of the measures of and that we can expect to learn nothing from considering the cases of Rb and Cs .
For Na and the resulting formulae are For Na and for These give values of obs .
-calc .
as follows:\mdash ; 1 .
2 .
3 .
KP KS 4 .
5 .
G. 42 05 18 By comparison with Table I. , it will be seen that now comes outside the limits , is slightly bebter ( but can be brought inside as before ) .
is much worse , and KS ( 2 ) improved , but the relation between the limits now holds .
No relationships are apparent between the various constants , except that , allowance being made for possible variations , the of in both cases is one-half .
The inequality between the for the and series is increased .
constant.\mdash ; In this case the values of are supposed to be different for both the and series .
Each series , therefore , requires as before three lines to determine its constants , and the relations VS ( 1 ) and give two additional equations to determine the two values of N\mdash ; say , and .
Unfortunately the values of the first lines are so uncertain ( with exception of ) that the values of and very largely with the errors in the observed lines .
For instance , in the case of Rb , if L. values for and be used , IfS.'svalues forP(l)andS(3.4)are used , , N , -N In the first case is reproduced as well as the original , but not so well ; in fact , comes out too large .
In the second , is not so good , and is better than with the 1300 difference .
The discussion shows , however , that within easy limits of error the of may be taken as constant , and for a provisional value the usual value of will do very well .
For Na , using K.R. values for ) , and the calculated lines for agree better than the original except , however , , which is wrong in place of 137 .
If , however , PASCHEN'S is used and K.R. , then , the limits are considerably altered , and the calculated results bad .
If we make and use PASCHEN'S and K.R. , , and the other calculated lines are all outside the limits .
The most reliable result appears , thel.efore , to be that calculated from K.R. values for , which improves all the other lines but requires to be .
larger than PASCHEN'S observation , which can hardly be the case .
For , using K.R. , .
rfhis brings within 22 of PASCHEN'S value ( original difference ) , and all the others within limits .
For Rb , using S. values for and K.R. , gives , but the series is bad , although is better than that referred to above where .
It is probable that a value between 651 and 1300 would be preferable .
For Cs , using L. for , K.R. 's for , and S. 's for , gives .
These bring all the lines within limits , better than the ordinary .
It is to be noticed that although Np may be kept.constant by giving permissible variations to the lines from which the constants are determined , yet there is a tendency shown to increase with increasing atomic weight .
In all increases with increasing atomic weight , but the calculated values vary so much with the actual errors of observation , that it is useless to attempt to find any relation between them .
for Na is probably not far from 220 ; for the others it is larger .
The values of for , Rb , Cs might be the same , but in view of the smaller value required for Na , this is not likely .
One important fact brought out , however , is the raising the value of for , so that its ratio to becomes much closer to that for the series .
PROF .
W. M. HICKS : A CRITICAL STUI ) OF SPECTRAL SERIES .
If the condition that should be the same as for be applied , say about , the values of come to about 382 for Na and 650 for the others .
In the present state of the observations , it is useless to attempt to arrive at anything more definite than that a very small proportionate increase of for the series will satisfy relations ( and G ) , make better agreement in general for the lines bring ratio into close agreement with that for P. Possibly some constant value for would make the same as for the series , in which case the of would be formed , not by adding the term to 5 , but to a number .
For these combined reasons it would appear , so far as our knowledge at present extends , that the required improvement in the formulae should be looked for in a change of the value of N. The evidence shows that at least the greater part of does not depend , like , on the square of the atomic weight .
It might , however , depend on .
Thus apparently SN for is about 300 , for KS about 600 , and the corresponding multiple law would make it 750 for Rb and 900 for Cs , quite in agl.eement with the actual numbers found .
If this were so , the sequence would be formed by adding a multiple term to the denominator , and the sequence by adding a corresponding multiple term to the numerator .
A further argument in support of this theory is found from the discussion on the various He series .
It was there found that using least squares to these very accurately measured spectra , rich in lines , that the and sequences have of very close to the value 109675 generally adopted , but that the series in both cases require a value of about 50larger , and this when the lines corresponding to were not used .
In this case they do not reproduce the values of .
It is probable , therefore , that if they were introduced the values of would be still more considerably increased .
The case of requires a still higher value of , and in this case agrees with the observed .
It seems a clear case for an increased if the form is absolute .
The argument that it is so is that in cases of elements where and are comparatively large , it is sufficient to give practically all the lines within limits of observation errors , and that if be added , comes always a small fraction of and capable of being wiped out by errors of observation .
Now the in the He spectra are all very small , and would be very much smaller , and exert scarcely any influence on the lines beyond the first two or three .
Nevertheless is increased for the sequences .
Summary of Results .
The principal results arrived at in the foregoing discussion may be summarised as follows : 1 .
A modified RYDBERG'S formula , in which the denominator is of the form , is found to be capable of representing practically all the observed lines within errors of observation .
The chief exception is for the first line of the sharp series , which is shown to be probably due , on other evidence , to a change in N. 2 .
The existence of a new fundamental sequence whose denominator is either ( RITZ ) or , where , the atomic weight term , is a small fraction of the square of the atomic weight .
The dependence of the fraction on the element is not at present determined , but if denote the atomic weight divided by 100 , the fraction of is 0141 for Na and about for the others .
3 .
That the of the sequence is found by adding 074560 to , as found from the sequence , where has the values 2.4.5.6 for Na , , Rb , Cs .
In the case of Cs , however , in which satellites appear in the series , the term in does not occur .
4 .
The of the sequence is found by deducting , or approximately , from that of 5 .
The of the sequence is found by deducting the same quantity from .
The form of the term in has not been determined .
When , however , satellites occur , and there are two sequences , as in Cs , the missing in the sequence must be put in here , and the above quantity deducted from instead of from .
This gives the for or .
The of is found by adding to that for , i.e. , .
6 .
In the same way as the sequence forms a doublet series , , , so do the sequence produce a series\mdash ; doublet when has satellites\mdash ; viz. , , .
7 .
The values for the series show that its is obtained by deducting quantities somewhat smaller than from that for the , and that its is always considerably less than that of P. Also , although RYDBERG'S relation is very closely verified , the other , cannot be true with the formulae used .
Moreover , the formulae fail to reproduce within limits of error .
Evidence is given that this is due to a change in value of in the sequence .
A suitable change makes VS ( 1 ) , the difference of the of and the same for each element , the same for and , and the values of within the limits .
Moreover , in using the He spectra to obtain a more exact value of than can be expected from , it is found that is greater than for , by about the same amount in both the singlet and doublet series , whilst close to it for the and P The evidence in favour of this explanation is therefore very strong , and the of would then be found by subtracting , where is small of order from that of P. 8 .
The spectrum of Li is abnormal .
It is a singlet series in which what has been PROF .
W. M. HICKS : A CRITICAL STUDY OF spECTltAL SERILS .
called the series is formed on the sequence , and what has been called the on the sequence .
The set belongs to what has been here called the negative type .
The first , the so-called Principal , is the only one in which the formula does not reproduce all the lines .
It is possible that it a larger is for the lines in the other singlet series discussed , , He ' .
9 .
As we pass down through the alkali metals we find a change of type occurring , first in Na , and then in Li .
In Na , what is called the series , has much closer to the type than to the type .
There seems to be a certain instability , and a large number of extra lines appear whose development appears to depend on narrow limits of temperature , amongst these are possibly three lines which show the type normal to the other elements .
In Li all the normal type have disappeared , nd we get a singlet system depending on a new kin of sequence for the Principal series , and the sequence for the Dseries .
In addition , we get lines conforming to the type based on the new sequence .
Also there are indications of the no1mal series in a doublet spectrum , with the doublet separation to be expected in the normal type .
But no or lines of the normal type have been seen .
10 .
In He the singlet sets of series are of the type , with the associated based on the sequence .
The doublet sets are of the type with the associated also based on the type .
These spectra have been used to determine accurate values of by the method of least squares .
The important result comes out that the value of is not same for each series .
It is practically 109675 for the and sets , for and ( probably higher if the lines had been employed in the determination ) , and 109810 for The relations found strengthen the evidence in favour of Lord theory as to the kinematical origin of these spectra .
A general consideration of questions relating to origin is , however , postponed until the spectra of the other elements are discussed in a future communication .

