]\gt ; Prof. W. H. Young .
It has also been found that the distribution of intensity found in certain celestial spectra can be approximately reproduced in the laboratory .
In any attempt to interpret the phenomena observed in connection with the series of hydrogerl , it is necessary to know the particular type to which this series belongs .
In order to decide this point a study has been made of the separations of the components of lines of the Balmer series of hydrogen , and the luean values of the } ) arations of the doublets ituting the lines and have been found to be respectively .
and .
These values are consistent with the separations appropriate to a Principal series , and the first is in precise agreement with the value deduced by Buisson and Fabry .
These results have been obtained by a Lummer Gehrcke plate with the neutral wedge , and submitting the contours obtained to mathematical analysis , by means of which the distribution of intensity in the individtlal components , and the separation of the components , can be determined .
On ultiple Integrats .
By Prof. W. H. YOUNG , Sc. D. , F.R.S. ( Received July 16 , 1916 .
) S1 .
In the modern theory of absolutely convergent integrals , as distinct from the older Riemann theory , the difference between multiple and repeated integration falls to the ground .
Every multiple integral is equal to the corresponding repeated integraIs , and the formulae , such multiple integrals , even when expressed without the repeated integral notation , can be obtained by means of the repeated ration process .
To really grasp the distinction*between multiple and epeated integrals we have to take as independent variables functions of bounded variation ; more precisely , we have to replace the ensemble of variables by a single function of bounded variation , which , indeed , takes the place of the product of the variables , the more general integration reducing to the ordinary when this product replaces the function in question .
I have already had occasion elsewhere to explain what is meant by a function of bounded variation of any number of variables , and to define * In the French language there is said to be , as yet , no word expressing this distinction .
On fultiple Integrals .
integration of any bounded function , and absolutely convergent integration of unbounded functions , with regard to such functions of bounded variation It will be sufficient here to call special attention to the fact that a function of bounded variation , for simplicity , let us say , of two variables , and , is expressible , to a value of the function pres , as the difference of two functions , which are monotone increasing in \ldquo ; in , and in , representing the positive and negative variations in the plane , increased by the differences of the variations along the axis of and along the axis of respectively .
There is , in consequence of the definition , a unique limit at each point for each of the four standard modes of approach , these four limits coinciding , except at points on a countable set of lines parallel to the axes at most .
The modes of approach here considered correspond to the four quadrants when the point is the origin , the lines corresponding and parallel to the axes being excluded .
The limits along these excluded lines may be different from one another and from the other four .
* Such functions , then , cannot in general be double integrals of ordinary functions of two variables , even if they are continuous .
The simplest case , one which constantly occurs in practice , is that in which the functions are not integrals , .
solely because they are not continuous .
This is the case , for instance , with the sum-function of the double Fourier series whose general term is , which is discontinuous at the origin and along the axes , and is the simplest type of double integral elsewhere .
The step taken in extending the theory of integration to that in which the ordinary variables are replaced by one or more functions of bounded variation , is for this very reason a matter to which no one who has occasion to use mathematical analysis can be indifferent , though he may be quite unaware of the existence of continuous functions of bounded variation which are not integrals .
The original idea of introducing integration with respect to a monotone function of a single independent variable is due to Stieltjes , and was employed by him exclusively in the case of continuous integrands .
Indeed , his definition fails in general with a less restricted hypothesis .
On the other hand , has given a definition depending on change of the independent variable , for integration of any bounded function with respect to a continuous function of bounded variation .
In my own theory , in which the treatment runs precisely parallel to that I have employed when the variable itself is the integrator ( that is , the function with respect to which we integrate ) , the integrand and integrator are both perfectly general , the latter being any function of * This could only be at the exceptional points already mentioned .
's method fails when there are more independent variables than one .
Prof. W. H. Young .
bounded variation , not necessarily continuous , and the former any bou'nded or unbounded function to which the process is applicable .
In the case in which the integrator is an absolutely convergent integral , we iain ordinary integrals , and I have utilised this fact on more than one occasion to obtain new theorems in ordinary integration .
In a recent paper , I have thus obtained eneralisation for any number of bless of the second theorem of the mean in Bonnet 's and other forms , as well as the formula for integration by parts in its most general form .
In such applications , certain fundamental theorems connecting ordinary integration with the more general type I have here in mind , are necessary .
Moreover , in the development of the new theory itself , a new type of theorem is required .
Whereas absolutely conyergent multiple integration with respect to the product is always expressible in the form of a repeated integral , this in general to be the case when the integrator is no expressible as the product of factors , each involving a single variable only .
It becomes important , therefore , to formulate tolerably wide conditions under which such a reduction of difficulty is possible .
In the present communication to the Society , I propose to give a brief account of some of the formula , which , from these two different , but closely allied points of view , are fundamental .
I by explicitly stating certain facts on which the theory of integration with respect to a function of bounded variation in two or more variables is based .
I have not introduced any discussion of integration with respect to a continuous function which is not of bounded variation .
Many of the formulae might be deduced from those of the present paper in the case when the rand is a function of bounded variation , but it seems advisable to reserve the discussion of this subject for a separate communication .
I have also it unnecessary to give the formulae for functions of more variables than two .
S2 .
The function with respect to which we integrate is supposed to be any function of bounded variation with respect to .
Every such function is , however , expressible in the form , where is the sum of the positive variations of with respect to respectively , each taken rectangle , the last two variations being taken along the axes of and , and being similarly defined for the ative variations .
On Multiple Integrats .
the definition it follows at once that only and , the positive variation and negative variation , taken , with respect to , contribute anything to the double integral with respect to , so thaC g is virtually expressible as ) difference of and , that is the difference of two functions , each of which is monotone with respect to , to and to It will be conyenient , therefore , to avoid repetition , to suppose that in the enunciation of our theorems , the function with respect to which we integrate ( integrator ) is itself such a function ( monotonely monotone increasing function ) .
It is clear from what has been said that there be no loss of generality in adopting such a course , except in cases where some additional special hypothesis is made with pect to the rator , which does not ipso facto hold for the functions of hich g is in the difference .
The condition that y ) is monotonely increasing with respect to is expressed as follows:\mdash ; , ( 1 ) and may be expressed in words by saying that the plane increment round an .
rectangle whose sides are parallel to the axes is positive , or more precisely by saying that monotone creasj with respcct to eith(of tlnj variables has ) increasing as the other varinble From these properties it follows that has a unique limit as we approach any particular point from one side or the other along the axes , or in any manner in each of the completely open quadrants .
Denoting the former four limits by , beginning with the horizontal hand axis , and proceeding round clockwise , and the latter four by beginning with the first , or -quadrant and proceeding clockwise , we have the inequalities ; ?
Hence also , that lies 'oetween and or ?
and , we can show that is continuous uith respect to , except possibly at point .
lying on a countable set of parallels to the axes of S3 .
The last result enables us to prove , as in the theory of functions of a single real variable , mutandis , that any simple or -function of can be expressed as the limit of a monotone sequence of simple or W. H. Young , " " Integration with Respect to a Function of Bounded Variation ' Loud .
Math. Soc. Proc Ser. 2 , vol. 13 , p. 139 ( 1913 ) .
Prof. W. H. Young .
-functions*none of whose discontinuities coincide with discontinuities of .
We have accordingly , as in the original theory , three methods of starting : ( 1 ) By the integral of general simple or -fnnction .
This will not be used in the present note .
( 2 ) By defining of sp jcial simple and -functions , whose continuiti arnot of .
The formula is that is where the fundamental rectangle is supposed divided up by lines parallel to the axes , the feet of which on the , -axis have abscissae , and on the -axis the ordinates in any convenient way such that all the discontinuities of lie on the dividing lines , none of which coincide with any of the set of lines parallel to the axes on which the discontinuities of y ) lie ; then denotes the constant value of in the interior of the rectangle whose left-hand bottom corner is , and is the value of y ) at ( 3 ) By defining of continuous functions .
The formula is Lt , where is any convenient point in the rectangle whose left-hand bottom corner is , and the rest of the notation has been already explained .
It may be shown that the rals we obtain are the same , whichever of the three methods we adopt .
The definition of the integral of a function which is the limit of a monotone sequence of functions whose yrals have been previously defined is ) it is the limit of the of functions the sequence .
validity of such a definitio1l follows as in one dimension , and we have , as in one-dimensional theory , the theorem that the is at the same time the lower bound of the ralsb of -functions greater than and the upper boun of the of -functions less than .
S4 .
We can now prove certain formulae for our double integrals with respect to .
The method of proof will in eneral consist in proving the * Such functions are constant in the interior of a finite number of rectangles into which the region of integration is divided by lines parallel to the axis .
On Multiple formulae from first principles for one or other of our two statldard forms ( 2 ) and ( 3 ) of the article , and then deducing the result by the method of monotone sequences , either by using generalised induction , or by deducing the truth of the formula for l- and -functions and applying the theorem mentioned at the conclusion of the preceding article .
We begin by establishing the following extremely important result:\mdash ; THEOREM.\mdash ; If is am absolutely convergent integral respect to ) , then , ( 4 ) anyJunctio of which is the , far function equal to the repeated differmtial coefficient of .
, wherever this differential coefficicnt , and equal to zero elsewhere .
Since the rectangle is the sum of the rectangles and minus the sum of the and , it follows that If then is a constant , say , the left-hand side of ( 4 ) is by definition equal to while the right-hand side , , has as has just been pointed out , the same value .
Thus the formula is true when is a constant .
Again , if is a simple or ?
-function of the special type , its integral with respect to , is the sum of the finite number of terms , by the formula ( 2 ) , each of which is the of a constant , and therefore , by what has been proved , is equal to the ordinary integral of that constant multiplied by over the corresponding rectangle .
all these terms , the equation ( 4 ) follows in this case .
But if the equation ( 4 ) holds for each of the members . .
of a monotone sequence , it holds for their function , since by definition limit , and is the limit of by a known theorem .
It follows , therefore , that the theorem is simple and -functions , as the limits of monotone sequences of these special functions , * See a forthcoming paper by the author .
Prof. W. H. Young .
is therefore true for general and -functions , as the limits of monOtone sequences of simple and -functions .
Hence eneralised induction the theorem is true for all functions which can be obtained as the limits of monotone sequences of functions , with simple and -functions .
This proves the truth of the formula ( 4 ) .
Similarly we may prove the more general theorem : \mdash ; If is integral of with respect to , is of of bounded variation , have .
( 5 ) We have in fact only to start from the identity and the argument is otherwise the same as before .
This eneral theorem , which I have not hitherto stated , even for functions of a single variable , is of very great use in practice , as it enables us to make in a very simple manner certain transformations in our integrals , which would otherwise scarcely occur to the worker .
S5 .
\mdash ; If th integrand is of one of the variables , the integration reduces to integration )respect to a .function of the other , in with the formula .
( 6 ) To prove this , let be a continuous function .
We then have , by Thus the formula is verified in the case considered .
Since , therefore , -hand and -hand sides continually eproduce themselves , when we let describe monotone ences , first of continuous functions , then of and -functions , which are the limits of such sequences , and so on , it follows that the formula holds all functions mathematically definable .
This proves the formula , using generalised induction .
If , however , we prefer to use the theorem alluded to at the end of S3 , we only use the above method to prove the ] for and ?
-functions , and then remark that is the upper bound of the rals of On JIuttiple Integrals .
-functions less than , and ( dg , being the integral of with respect to , is the upper bound of the integrals of the functions with respect to the same function ; since the formula ( 4 ) holds for in place of ( it ) , it therefore holds as it stands .
This proves the formula also .
Thus by either method the formula is verified by monotone sequences .
S6 .
.\mdash ; If the rrator g , the result integration with ?
to this given by following ( 5 ) provided only , , t ) poss)integral nith to Suppose first that ) a simple or ?
-function of the special type , whose discontinuities are different from any of those of .
Then , by , we have by ( 2 ) , , since is easily seen to be a simple function of the special type as .
This proves the formula in this case .
But if , f2 , , are functions forming a monotone sequence with as limit , we may integrate this sequence term-by-term with respect to the function J , so that And also similarly .
Denoting the function of represented by either side of the last equation by , and the function whose limit is taken on the right by , It is convenient here elsewhere to suppose positive in the proof .
; is no restriction , as we may always break up into the difference of two such functions .
Prof. W. H. Young .
increases or decreases with , according as does so , and therefore also describes a monotone sequence with as limit .
Therefore , as before , we may integrate term-by-term and write Now , supposing the equation ( 5 ) to hold for each of the functions , we have , with our present notat , ion , .
Proceeding to the limit with , and the results just obtained , we get therefore , which is identical with ( 5 ) .
If therefore the equation ( 5 ) holds for all the members of any monotone sequence of functions , it holds for the limiting function .
Thus , as before , the theorem is true , since it has been shown to hold for simple and -functions of special type .
\mdash ; If thw integramd contains ameter t to inside th , that is The proof of this theorem is the same as that in the ordinary theory of absolutely convergent integrals .
S7 .
We have already given in S5 the simplest case in which integration with respect to a function of bounded variation reduces to ordinary integration .
We prove the following further theorems:\mdash ; \mdash ; If is ( int respect to , then ( 6 ) where is any one of the derivates of , Jp ) with respect to .
In fact , denoting by the function which is equal to the differential coefficient with respect to of wherever this exists and is finite , and is zero , or has any other convenient values , at the remaining set of values of of content zero , we have .
*Absolutely convergent , or Lebesgue , integral .
On Multiple Integrals .
Now by the formula ( 4 ) , ' , and But change of order of integration is allowable on each side of the last equation* ; thus which may be written .
Using the expression already given for the integral of ?
this gives , which is identical with the formula to be proved .
In the theorem just proved the integration is reduced to lepeated integration first with respect to a function of bounded variation of one variable , and then with respect to the other .
From which cret THEOREM.\mdash ; If an o.ntegral lrith respect to and is integral with respect to a of bound omrifh .
resp to , then Here we have been able to go still farther all the tions employed are ordinary .
Again , without any assumption as to integrator , we can reduce to ordinary integration when the iegrand is a double integral .
The theorem , which is an immediate consequence of ( 4 ) when the formula for integration by parts for multiple integrals is employed , is as follows:\mdash ; ' ' Integration with respect to a Function of Bcunded Variation S32 , p. 148 .
Prof. W. H. Young .
TtlEOREM.\mdash ; If a double integral with respect to , then ' .
( 8 ) This theorem also from ( 6 ) by the formula for integration by parts for integration of fnnctions of a single variable .
S8 .
We shall now find it convenient to prove the following property of our rator g .
This is that the derivat of with respect to either are monotone asc ending ctions of the other variable .
In fact the relation ( 1 ) jves Let describe such a sequence of positive values with zero as limit that the right-hand side approaches its upper limit , the upper hand derivate of with respect to , say .
Then the left-hand approaches some limit or limits which are not greater than .
Thus .
imilarly , the sequence of values of so the left-hand side approaches its lower limit , the lower right-hand deriyate of with respect to , we .
This shows that both the hand derivates with respect to are monotone ascending functions of Similarly , taking positive and negative , in which case the symbol must , of course , be changed to in order that the plane increment may with the right-hand top corner , and proceed clockwise , we find that both the left-hand derivates of with respect to are monotone functions of By symmetry the sftme result holds when and are .
This .
therefore , proves the required result .
S9 .
If , in addition to having the property of being what we have called a monoto of , ( S2 ) , is an integral with respect , it is the integral of any one of its derivates with respect to .
These derivates agree and are finite except at a set of content zero of values of.r .
for of and , therefore , and are finite except at a set of values of of plane content zero .
Hence they ) and are finite except at a set of content zero of values of for jach fixed value of not to a certain set of content of values of .
Putting aside this exceptional On Muttiple lntegrals .
set of values of , therefore , it follows from S8 that , being con stant , exists and is finite except at a set of content zero of values of , and the values at the exceptional points be so that becomcs ascending function , defined for all values of With understanding therefore we may , when is an integral with respect to , not only write , but also we may rate a function with respect to being constant , and then rate the result with respect to , the exceptional set of content zero of values of ' at which the inside integral is undefined not the second ration .
With this understanding , we have the following theorem on the reduction to a repeated integral:\mdash ; \mdash ; If a monotonely monotone ascending function of and an ral with respect to , and therefore an integral with respect to of a monotone ncreasing function of ?
, we ' .
( 7 ) If is a constant , say , the theorem is at once seen to be true , for , by definition , the left-hand side of the equation is , and the -hand side is same , since , performing the inside integration , we get , writing for Hence , by the formula ( 2 ) , the theorem holds when is a simple l- or -function of the special type , for the integral is then a double summation of a finite number of such integrals with constant ) .
But any simple l-or -functions can be expressed as the limit of a monotone sequence of simple function of the special types just con- templated , say , , and we may integrate this sequence -by-term either with respect to , or first with respect to , being constant , and then with respect to Thus and also .
* This mode of regarding the matter is sufficient for the purpose of justifying the notation in equation 7 ) .
For fuller information with regard to the delivntes and repeated derivates of functions of bounded variation , see the forthcoming ) referred to in S4 .
Prof. W. H. Young .
But , by what we have already proved , .
From these three relations the required relation immediately follows .
Similarly , it now follows generally for any and ?
-functions , these being the limits of monotone nces of simple and -functions .
this the heorem generally by geDeralised induction , or , if prefer , as follows : the left-hand side of ( 7 ) is the upper bound of , where \ldquo ; y ) is any -function less than ; and the right-hand side of ( 7 ) is the integral with respect to of the upper bound of , which is the same as the upper bound of , so that integrating , with respect , and taking the bound , the upper of .
But we can find a for which J differs by as little as we please from , so that integrating with respect to and taking the upper bound , we see that the upper bound of , where is as small as we please .
From this the truth of the statement made at once follows .
But by what has been proved , so that the upper bounds of the two sides of this equation are also equal .
This gives us the equation ( 7 ) , and proves the theorem .
S 10 .
From the theorem of the article , we at once deduce the following : THEOREM.\mdash ; If is notoncly rtotone jnding function , an integral with respect to therefore an integral with respect to of a onotone increasing of , we hare , ( S ) provided the right-h side continuous On Multiple Integrals .
S 11 .
When the integrator , instead of the integrand , describes a sequence , the mere monotony of the sequence is not enough to ensure that the limiting process is allowable , that is , that the integral with respect to the limit is the same as the limit of the integral .
In this connection the theorem is of importance:\mdash ; THEOREM.\mdash ; If a function of , with respect to , to ?
1 , and to , and its triple ) of ' it is uith respect to , then .
( 9 ) From the hypothesis in the enunciation it follows that , is monotonely monotone increasing with resl ) to , supposing the triple increment of to be ) ositive .
For the function here considered is the limit , as positive values , of % ; , which is monotonely monotone increasing with respect to ) .
Hence if is a bounded function of , and is numerically and hence has the unique limit zero as .
This proves the theorem when is bounded .
When is not bounded , we denote by the fmlction which is equal to , where this is less than , and is elsewhere equal to as usual taken to be positive ) .
Making then .
, we see , by monotone sequences , that the theorem is true .
As a particular case of the theorem just proved , we have the THEOREM.is a'monotonely increasing of , and has , ( as limit , and generates such lonotone inrre decreasing sequence that reases , decreases , with ( 10 ) * This theorem may be utilised , among other wayS , in extending the theorem of Integration by Parts .

