]\gt ; SLIP .
, W. H.\mdash ; On tho of Integration .
PhiLTrns .
, , vol. 204 , 190S , pp. 221-252 .
Theory of , with reepect to of Points .
Youue , W. H. PhiL Prans .
, , 1905 , pp. 221-252 .
Variable , Functione of , umnable , Semi-oontinuoue , etc. Youwo , W. K. Phil. Trans. , , vol. 204 , 1906 , pp. of Points , Content of\mdash ; Integration with respect to .
YOUNG , W. .
Trans. , , vol. 204 , 1905 , pp. VI .
On the General Theory of Integration .
W. H. YOUNG , Sc. D. , St. 's College , by : E. W. HOBSON , F.R. Received April 2.3 , \mdash ; Rend May 19 , 1904 .
RIEMANN was the first to consider the thebry of integration of non-continuous functions .
As is well known , his definition of the integral of a function between the limits and is as follows :Divide the segment into any finite number of intervals , each less , say , than a positive quantity , or norm ; take the product of each such interval by the value of the function at any point of that interval , and form the sum of all these products ; if this sum has a limit , when is nitely diminished which is independent of the mode of division into intervals , and of the choice of the points in those intervals at which the values of the function are considered , this limit is called the integral of the function from to The most convenient mode , however , of defining a Riemann ( that is an ordinary ) integral of a function , is due to DARBOUX ; it is based on the introduction of ?
lower integrals exces , defaut , oberes , unteres Integral ) .
The definitions of these are as follows : may be shown that , if the interval , b ) be divided as before , and the sum of the products taken as before , but with this difference , that instead of the value of the function at an arbitrary point of the part , the upper ( lower ) limit of the values of the function in the part be taken and multiplied by the length of the corresponding part , these summations have , whatever be the type of function , each of them a definite limit , independent of the mode of division and the mode in which approaches the value zero .
This limit is called the upper ( lower ) integral of the function .
In the special case in which these two limits agree , the common value is called integral of the function .
The progress of the modern theory sets of points ( Theory des ensembles ; Mengenlehre ) , due , as is well known , chiefly to G. ANTOR , though taking its origin in RIEMANN'S paper 'Ueber die Darstellbarkeit einer Funktion durch eine trigonometrische Reihe , ' naturally leads us to put the question how far these definitions * An abridged statement of the contents of this memoir will be found in the " " Abstraet\ldquo ; published the ' Proceedings of the Royal Society , ' vol. 73 , pp. 446-449 .
377 .
22.2.05 can be generalised .
This theory has in fact us on the one hand that many of the theorems hitherto stated for finite numbers are true with or without modification for a countably infinite number , and on the other hand that closed sets of points possess many of the properties of intervals .
We may , in accordance with these facts , divide the segment ( 1 ) into an infinite number of non-overlapping intervals , in which case , however , seeing that such a set of intervals always has points which are external or semi-external to them , we must in general add a set of points to the set of intervals , if the division of the segment is to be properly performed , that is , if all the points of the segment are to be accounted for in our division ; or , more generally , ( 2 ) into a finite or countably infinite number of sets of points .
What would be the effect on the Riemann and Darboux definitions , if in those definitions the word " " finite\ldquo ; were replaced by " " countably infinite and the word " " interval\ldquo ; by " " set of points\ldquo ; ?
A further question suggests itself : we at liberty to replace the segment itself by a closed set of points , and so define integration with respect to any closed set of points ?
Going one step further , recognising that the theory of the content of open sets quite recently developed by M. has enabled us to deal with all known open sets in much the same way as with closed sets as regards the very properties which here come into consideration , we may attempt to replace both the segment and the intervals of the segment by any kinds of measurable sets .
In the Riemann and Darboux definitions it is tacitly assumed that the interval is finite , and that the function is throughout the interval finite and possesses finite upper and lower limits .
The discussion of the integration of a function which is not necessarily finite , over an interval not necessarily finite in length , requires separate consideration , and the definitions of such rals , called improper integrals , are of the nature of extensions of the definitions of ordinary integrals .
Bearing in mind the somewhat sfactory and artificial character of such extensions , we may hope finally that our discussion may throw light on improper integrals also .
In M. LEBESGUE'S valuable memoir , already referred to , a striking addition has been made to the previously existing knowledge of the subjects dealt with .
He has shown that a more general definition than that of RIEMANN , available for all known functions , one moreover coinciding with that of RIEMANN in the case of all functions integrable in the Riemann sense , may be given ; a definition possessing , among others , the remarkable property of permitting passage back from the derivatives of a continuous function to the continuous function itself .
In the present paper I attempt to discuss the whole matter , and take occasion in the proper place to bring LEBESGUE'S work into connection with my own .
Some special cases of the results I obtain have been given by me in the paper presented to " " Integrale ; Longueur ; Air 'Ann .
di Mat 1902 . .
also a paper by the Author , " " Open Sets and the Theory of Content ' Proc. Lond. Math. Soc Ser. 2 , vol. 2 , Part I. , p. 16 .
In the instalment now presented to the Society I confine my attention however to proper integrals . .
W. H. YOUNG ON TH GENERAL OF 2 .
23 the L.M.S. already cited , and I apply one of the results of this paper , viz. , that the content of any closed dimensional set may be expressed both as an ordinary integral in terms of an dimensional content , or as the upper integral of the content of sets of lower dimensions , to obtain the corresponding theorem .
any measurable set ( contained in a closed set of finite content ) .
PART I.\mdash ; .OF FINITE PROPER INrEGRALS .
S1 .
The necessary and sufficient condition that a function should the requirements of RIEMANN'S definition is simply that the content of the points of continuity of the function should be equal to the length of the segment , ( or , in the case of a multiple integral , to the content of the region over which the integration is extended ) , that is to say , when this condition is satisfied , the summations referred to in the Riemanm definition have a definite limit , independent of the mode of division , \amp ; c. Is this still when for " " interval\ldquo ; the expl.ession " " set of points\ldquo ; is substituted , for " " of \ldquo ; the " " content of the set of points\ldquo ; The following exalnple shows that , when these substitutions have been made , the definition , as it stands , ceases to have meaning , even in the case of continuous functions .
.\mdash ; Take as the function .
The iemann integral has the value .
If , however , after dividing the segment into equal intervals , we abstract the points and add them singly to what remains of the intervals , we obtain able sets of points , each of the same content as before , ; in , however , there is a poin at which the function has a value greater than , except possibly in one of the sets in which there is a point at which the function has the value .
The summation is therefore always greater than S , so that it is clear that we do not get the same limit as before when is indefinitely increased .
The principle of this example shows that in the general case also , except in the single case when the function is a constant , rent modes of division of the segment into a finite number of sets of points each of content ] than a given different modes of proceeding to the limit , will certainly not lways give the limiting value of the summations .
Thus , suppose the function to be uous at its upper limit , then we can arrange that mode of division is such that to every partial set at every stage a point belongs for which the function from its *I shall always , except when the contrary is stated , suppose that the sets employed are measurable sets , so that the sum of non-overlapping sets for content the sum of their separate contents .
maximum value by a quantity as small as we please , and there is nothing to prevent our taking precisely this point as.that at which the value of the function is to be taken in forming the summation ; similarly for the minimum and it is clear that the limits obtained in these two cases could only agree when the fUnction is a constant .
It is plain , moreover , that there is no room for discussion of the range of possible values of limits corresponding to the various conceivable modes of division .
Speaking generally , the range will be from SM to , where is the length of the segment , or more generally -dimensional volume of the region , and and are the upper and lower limits of the values of the function .
Thus the Riemann definition completely breaks down when we attempt to generalise it in this direction .
92 .
There is another , however , in which an important generalisation of the Riemalm definition is possible .
If we change the words " " finite number of intervals\ldquo ; into " " set of intervals\ldquo ; and add " " such that the content of the external points is zero the definition still holds good ; we get a pel.fectly definite limit , which is , of course , the Riemann integral .
To this , we tice hrst that the content of the set of intervals is in this case , and this case alone , the same as that of the segment ( or under discussion , say Let the intel.vals be arranged in any way countable order .
Then , since the 's are all positive , their sum is an absolutely convergent series ; therefore the same is true when the content of each is multiplied by a quantity which , for all values of , lies between finite upper and lower limits , say between , as is the case in forming our summation If we consider of these intervals in order , these leave over a finite numbel of complententary intervals , say , and we can so choose that the sum of these latter intervals is less thau while the contribution to our summation over the remaining intervals .
is numerically less than If now we form the summation in RIEMANN'S way over the finite number of intervals , and compare it with the corresponding summation ove the set of intervals , we see that the difference between the two summations is less than .
Since is at our disposal , and we can insure that both the intervals and are less assigned norm , this the statement embodied again in the following theorem:\mdash ; * I take this first opportunity of emphasising the fact that , though it is convenient , indeed necessary , in forming the sum of an infinite number of terms to arrange them in some sort of order , in doing here we do not introduce the idea of order into the concept of integration .
Indoed , from the definitions it is evident that the concept of integration no more of itself involves the idea of order than do the concepts of length , area , and volume .
The distinction of the two notions has , perhaps , not always been sent to the mind of some DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
Theorem 1.\mdash ; If the interval be divided into a set of intervals set of points of zero content , length of interval being less than some assigned norm : and if the product of the length of each by the of given integrable function at any point of that be formed and the summation of all such proclucts calculated , this summation has definite limit when is indefinr.tely diminished ; this limit is , of course , the integral of the function .
Corollary.\mdash ; The value of the integral of an is unaltered if , the points of a set of zero content , the values of the function .
In other words , if we add to the function an integrable null function , we leave the integral unaltered .
It will be convenient to prove the following theorem , which is , in practice , indispensable ; from it the theorem on which DARBOUX'S definition of upper ( lower ) integration is based , can be at once deduced:\mdash ; S3 .
Theorem 2.\mdash ; Given any small positive quantity , we can determine positive quantity , such that , the segment be divided up in any into a finite number of non-overlapping intervals , then , provided only the length of each interval is less than , the upper ( lower ) of any function over these intervals differs by less than from definite limiting , the upper ( lower ) integral .
The following is the proof for the case of the upper integral ; with slight modification it holds for the lower integral .
Let I be the lower limit of all such summations ; then we can determine a division of into a finite number of intervals , such that the upper summation over these intervals lies between I and Let be chosen to satisfy the following equation where is any quantity greater than the greatest value of the function , and let us consider any division whatever , into a finite number of non-overlapping intervals each less than .
The number of such intervals which do not lie entirely in one of the intervals previously determined , is at most , so that the sum of the terms corresponding to these intervals is less than , that is In each of the remaining intervals the upper limit of is not greater than the upper limit of in that one of the intervals in which it lies , so that the upper summation over the remaining intervals is not greater than that over the intervals .
Hence the summation over our intervals , each being less than , is less than [ Q.E.D. ] S4 .
We have now to discuss the Darboux form of the definition of integration , that is , in the first instance , to consider the effect of the modifications proposed on DARBOUX'S definitions of upper and lower integration .
The following example*shows that the theorems stated in these definitions no * Example 1 , of course , shows this in the oase of an integrable function .
VOL. CCIV.\mdash ; A. 2 longer hold , that is to say , it is no longer true that definite limits , independent of the mode of division , \amp ; c. , exist , when sets of points are substituted for intervals:\mdash ; Example 2.\mdash ; Let be a function of , which is zero evelywhere , in the segment , except at every point of a perfect set , nowhere dense , whose content is I , and at every point of let If we divide the segment into equal parts , each of length less than , the upper summation , that is , the summation which corresponds to the upper integral , is always greater than I , but approaches the value I as is indefinitely decreased .
If I is zero ( in which case the function is integrable in RIEMANN'S sense ) , the limit is always zero , and in any case it is less than 1 .
We can , however , arrange so that we get 1 as limit .
Let the semi-external points of the intervals be arranged in countable order , . .
; take the same division as before , abstracting , however , the points , . .
, and then adding these points singly to the single parts ; thus we have a division into measurable sets , each of content less than , in each of which the function has the value unity , so that the summation always has the value unity however small may be ; therefore we get 1 as limit . .
If , on the other hand , we retain the intervals , but drop the restriction that they should be finite in number , there remains over a complementary set of points of content , say , I. The following example shows that the upper summations over a set of intervals have not then in general a definite limit , so that this extension of the definition of upper integral cannot be made without some restriction:\mdash ; Example 3.\mdash ; Take a perfect set , nowhere dense , of content I in the segment , and let the function be zero everywhere except at the points of this set where the value of the function is unity .
The value of the upper integral as defined by DARBOUX is I. Now let be any assigned norm , and let an odd integer be determined so that Divide the segment into equal parts , and blacken the middle part .
Then divide each of the remaining equal parts of the segment into equal parts , and blacken each of the middle parts , and so on , in the usual manner .
The set of intervals , each less than , which we thus obtain , have content which is less than .
The complementary set of points is a perfect set nowhere dense of content greater than .
The upper summation of the function over these intervals is then at most equal to , and is , therefore , less than the upper integral as soon as the norm is less than I , and has the limit zero .
The principle of this example shows that , if omit the condition that the number DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
of intervats should be finite , the upper summation can always be made as small as we please .
The same is true of the lower summation .
Theorem , however , we make the restriction that the set of points complementary to the intervals has zero content , or content which decreases without limit , as does so , the limits approached by the upper and lower are perfectly definite , and , of course , the upper lower The proof of this statement is identical in character with that given in S3 for the corresponding theorem regarding the integral of an integrable function , which theorem is , of course , a special case of the above .
We should naturally ask whether we cannot correct this discrepancy by adding to the upper or lower summation over the set of intervals the product of the content of the complementary set of points into the upper or lower limit of the function for points of that set ; or , if this does not suffice , by dividing up the complementary set itself into components and adding the sum of the corresponding products .
That neither of these corrections suffice is shown by the following example:\mdash ; Example 4.\mdash ; Take the same set as before at which the function has the value 1 , and inside the largest of its black intervals place a similar set of content I ' , at every point of which internal to that black interval the function has the value 2 .
Everywhere else the function is to be zero .
The upper integral is If now we merely subdivide the black intervals of which lie inside the largest black interval of , and subdivide the remaining black intervals of , the product of the content of the complementary set of points into the upper limit of the function for points of that set will be at least 2 while the summation over the intervals will be zero .
Thus the addition of the term in question would not correct the result .
If , on the other hand , we subdivide the complementary set into components whose contents are themselves less than the norm , we could , on the principle which has already been employed , insure that each component contained a point at which the function had the value 2 , and the result would be the same as before .
In neither case do we obtain the upper integral .
S6 .
As we have seen in S4 and S5 , the Darboux definitions of upper and lower integration require modifications , if they are to be generalised in the manner proposed .
We are naturally led to define upper and lower integration tentatively as follows:\mdash ; Let the division of the segment into ( measurable ) sets be performed in any conceivable way , and let the upper limit of the values of the function in each partial set be multiplied by the content of that set , and let the sum of these products be formed ; then the upper integral is defined to be the lower limit of all such sums .
Similarly the lower integral might be defined , the words " " upper\ldquo ; and " " lower\ldquo ; being throughout interchanged .
There is clearly no logical reason to prevent our considering these limits .
The names upper and lower integrak will , however , not be suitable , unless ( 1 ) the upper integral , so defined , is greater than , or equal to , the lower integral ; ( 2 ) these definitions agree with DARBOUX'S in all cases .
It is easy to prove that ( 1 ) holds:\mdash ; Theorem 4.\mdash ; The lover limit of the upper summations is not less than the uwer limit of the lower summations .
For otherwise we could clearly take a quantity lying between the upper and lower integrals so defined , and find two divisions of the segment so that the upper summation for the first division is less than and the lower summation for the second greater than L. If , now , we consider the division of the segment got by combining .
these two divisions , that is , if we divide each of the former sets up into the components which it has in common with each of the latter sets , the upper summation is not increased nor the lower summation diminished , thus for this division the upper summation is less than and the lower summation greater than which is impossible .
This proves ( 1 ) to hold .
In regard to ( 2 ) the following simple example shows that , in the general case , there is no agreement between our tentative definitions and those of DARBOUX .
By dividing up into sets of points , instead of into intervals , we get a lower value for the upper integral than that given by DARBOUX'S definition , and a greater value than the lower integral .
Example 5.\mdash ; Take the function which is 1 at all the rational points of the segment , and zero everywhere else .
The Darboux upper integral , from to ] , has the value 1 ; the lower limit of all possible upper summations is however , since the rational points can be enclosed in a set of intervals whose content is as small as we please .
S7 .
I now proceed to show that in the case of a function which is integrable , in the ordinary sense of the word , so that the ordinary upper and lower integrals coincide , the division into sets of points , instead of merely into a finite number of intervals , leads to the same limit , viz. , the integral of the function .
To see this , we have merely to remark that the upper and lower summations are respectively less than the ordinary upper integral and greater than the ordinary lower integral , except in the case of equality , and that , as shown in the preceding article , the lower limit of the upper summations is not less than the upper limit of the lower summations .
Thus we have the theorem:\mdash ; Theorem 5.\mdash ; If a function be integrable in , given segment , the value of the integral is equal to the limit obtained as follows : divide the segment into any finite or countably infinite number of measurable sets of points , multiply the content of each set by the upper ( lower ) limit of the values of the function for points of that set , and sum all such products ; then the lower ( upper ) limit of such for every conceivable mode of division is the integral of function .
S8 .
I next proceed to show that in the case of an upper ( lower ) semi-oontinuous DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
function , the upper ( lower ) integrals may be obtained in a similar way , so that , in the case of an upper semi continuous function the tentative , definition discussed in S 6 of the upper integral be retained , and the same is true of the tower integral lover semi-continuous function .
To prove this let us suppose the segment divided up into a finite or countably infinite number of sets of points , E2 , , so that the corresponding upper summation differs from the minimum X by less than some assigned small positive quantity Now round every point of describe a small interval , in which the maximum of differs from the value of at by a quantity less than some assigned small positive quantity .
We thus get a definite set of intervals enclosing all the points of ; their content is therefore not less than the content , but we can so construct them that it is less than , where is another small positive quantity as small as we please .
In each of these intervals the maximum of the function is less than Let us do likewise for each set , choosing the small quantities so that Applying the Heine-Borel theorem to all these intervals , which enclose every point of , , that is of , we can determine a finite number of these intervals enclosing every point of , and we can insure that the content of these intervals differs from their sum by less than To each of these intervals we can attach the index of the first of the sets , from which it was constructed ; the content of those which have the index will then be less than and the maximum in each will be less than .
Hence the upper summation over the non-overlapping intervals , consisting partly of the simple parts of these intervals and partly of the overlapping parts , is less than where is a positive quantity greater than the numerically greatest values of That is , this upper summation is less than , which , since is at our disposal , proves the theorem .
A similar proof can be given for the case of the lower integral of a lower semicontinuous function , or we may deduce the corresponding theorem in this case from the fact that a lower semi-continuous function becomes an upper semi-continuous fUnction when its sign is changed , and at the same time the lower integral becomes the upper integral .
Bearing in mind now that the upper integral of a function is equal to that of .
" " An extension of the Heine-Borel Theorem ' Messenger of Mathomatics , ' New Series , No. January , 1904 .
its associated upper limiting function , and that its lower integral is equal to that 01 its associated lower limiting function , have the.following theorems , corresponding to that given at the end of S7 .
Theorem 6.\mdash ; To find the upper integral of function we may proceed as follows : divide the segment or region into an.finite or countably infinite number of surable sets of points , multiply the content of each set by the upper ( lower ) limit of the of the function at points of that set and sum all such products ; then the lower ( upper ) limit of all such summations for every conceivable mode of division is the upper ( lower ) integral of the function for the segment or region .
S9 .
We are now able to define upper and lower over a set of points , instead of merely over an ordinary region ; we may , indeed , if we please , suppose the function defined only for the set of points .
The set will be assumed to be measurable , so that , like an ordinary region , it can be divided into component sets , each of which is measurable , In the special in which the content of the set is zero , define the upper and lower integrats to be zero also .
In the general case the definition will be as follows:\mdash ; Divide the set into any finite or countably infinite number of measurable , multiply the content of each componaent by the upper ( lower ) limit of the ( minima ) tnith respect to of the function at points of that component sum such products ; then the lower ( upper ) limit of all such for every conceivable mode of division is the upper ( lower ) integral the function for the mental set S. Further , when the upper lower integrals for are equal , the function may be said to be integrable over that fundamental set S. Thus we have defined upper integration , lower integration , and integration over any measurable set in such a manner that , in the particular case when the set is a segment or region , we get the ordinary .
Riemann and rboux integrals .
S10 .
Summing up our results so far , we saw that .
though DARBOUX'S form of the definition was preferable to RIEMANN'S , it did not at once lend itself to generalisation .
I then showed how to modify it so that the number of intervals should not necessarily be finite , provided that their content was equal to that of the segment .
We then saw how the introduction of the maximum ( minimum ) at a point , instead of the value .
Upper and Lower Integration , 'Proc .
Lond. NIath .
Soc Ser. 2 , vol. 2 , Part I. , p. 55 , also S11 , below .
It should be carefully noted that the maximum at a point of a set is the lower limit of the upper limit of the values of the function in a small interval or region contaimng the point , when that interval or region is indefinitely decreased .
Similarly for the minimum .
In the enunciation of the above theorem , therefore , the word ' maxima \ldquo ; has hing to do with the particular set to which the point belongs , the upper ( lower ) limit , however , is taken with respect to that set , that is , it is the upper limit of the maxima corresponding to the various points of that set .
Similarly in the defi1tition of S9 the set takes the place of the segment or region .
DR. W. H. YOUNG ON THE THEORY OF INTEGBATION .
at a point , permitted of division of the segment into sets of points , instead of merely into sets of intervals ; and , finally , that in this form the definition is applicable to integration over a set of points .
There is , however , another mode by which we can define integration with respect to a set of points in such a manner as to get the Darboux integral when the fundamental set reduces to a segment or ordinary region .
This new generalised definition is in some respects more closely analogous to the original Darboux definition and brings out , more clearly than the one just given , the distinction between an interval and a set of points in general , throwing light , as it does , on the question why it is that the limit of the upper ( lower ) summations over a segment is different when the segment is divided into intervals from what it is when the division is into set of points .
Like the Darboux definition , it concerns itself with the actual value of the function at a point instead of the maximum ( minimum ) there , while it divides the fundamental up into components , closely analogous to the intervals in the Darboux definition .
It will be convenient , and conduce to clearness , to give first a few preliminary explanations and theorems .
We require to define and give one or two properties of semi-continuous functions , when the region of existence is a set of points .
It will be found that the introduction of these functions materially simplifies the eatment- of the subject .
S11 .
The definition of an upper ( lower ) semi-continuous function , defined for any fundamental measurable set , does not differ from the usual definition for a segment , or ordinary region , the maximum ( minimum ) in each case is to be estimated with respect to the fundamental set alone , in the usual case the fundamental set being that segment or region , and in the general case that measurable set .
Theorem 7.\mdash ; If a function , defined uith respect to a measurable set of points , be an upper ( lower ) semi-continuous function , the points at which the value of the function form a set .
Complete the set of points , i.e. , form the smallest closed set of which is a component .
Attribute to the function at the points of which are not points of the upper ( lower ) limit of the values of the original function at points of in a small neighbourhood of the point , when that neighbourhood is indefinitely decreased .
We thus get a fUnction which is upper ( lower ) semi-continuous with espect to a closed set H. By an argument of precisely the same nature as that used for the case when is a segment , it follows that the set of points at which the new function is is a closed set , say Q. ce both and are measurable , the same is true of their common component , which is none other than the set of points at which the given function , upper ( lower ) semi-continuous with respect to , is .
[ Q.E.D. ] Definition.\mdash ; If at every point of the fundamental set we take , as the value of a new function at any point , the maximum ( minimum ) with respect to of a given function , I call the new function the associated upper ( lower ) limiting function of the given function .
Using the term oscillation at a point for the excess of the maximum over the minimum , we get a third associated function by taking as value at a point the value of the oscillation there , this I call thoe associated llation function .
Theorem 8.\mdash ; The ssociated upper ( lower ) limiting function of any function unth respect to set is an upper ( lower ) semi-continuous function with respect to S. For , complete the set , and let be the closed set so obtained .
an extended upper ( lower ) limiting function as in the proof of the preceding theorem .
The original discontinuous function may be supposed to have the same values in the points of ( H-S ) as this extended upper ( lower ) limiting function .
A proof precisely similar to that for the continuum*proves that the new upper ( lower ) limiting function is upper ( lower ) semi-continuous .
Now it is plain that , though the points of may have points of for limiting points , the upper ( lower ) limits of the values of the function in the neighbourhood of points of are the same for the old and new upper limiting functions .
Therefore the values of the extended upper ( lower ) limiting function at the points of are the maxima ( minima ) with respect to either or , so that the old upper ( lower ) limiting function is upper ( lower ) semi-discontinuous with respect to S. [ Q.E.D. ] Corollary .
oscillation function , being the sum of two upper semicontinuous functions , is itself upper semi-continuous function .
Theorem \mdash ; If be a component of the set , such that au the limiting points of which belong to are contained in , then the upper ( lower ) limit of a function in ( S-S ' ) is the same that of the associated upper ( lower ) limiting function in ( S-S ' ) .
This follows from the fact that the maximum ( minimum ) at any point of ( S-S ' ) is unaffected by the values of the function at the points of , since is not a limiting point of .
Whence the result is easily deduced .
The fact that the value at a point of an upper ( lower ) semi-continuous function is the maximum ( minimum ) at that point in the case of such a function , enables us to substitute the word instead of maximum ( minimum ) in the definition of integration .
The definition in this case takes the following simplified form : Divide the fundamental set into a finite or countably infinite nurnher of measurable components , multiply the content of compcnoent by the ( lower ) limit of the values of an upper ( lower ) semi-continuous furwtion at points of that component sum all such products ; then the lower limit of all such * BAIRE , ' Ann. di Mat .
' , vol. This property is evidently unaffected by the substitution of a fundamental set instead of a segment or region .
DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
summations for every mode of division is the upper ( lower ) integral of the semi-continuous function for the fundamental set S. Taking this as the definition of the upper ( lower ) integral in the case of an upper ( lower ) semi-continuous function , the definition in the general case is equivalent to the following.\mdash ; Theorem 10.\mdash ; The upper ( lower ) integral of any function rvith respect to an .
measurable set is the upper ( lower ) integral of its associated upper ( lower ) semicontinuous function .
S12 .
The division of the continuum adopted by DARBOUX is , as has been proved , special case of a more general division of the continuum into intervals , by means of which we obtain an upper summation differing by as little as we please from the upper integral .
This division was such that the sum of the intervals was equal to that of the segment , while each interval had to be less than a quantity which depended only on the degree of approximation desired .
When the fundamental set is not the continuum , but merely any measurable set , we can , given any small positive quantity , find a set of intervals enclosing every point of , that is , having every point of as an internal point , the content of the intervals lying between and .
If we assign any small positive quantity , there will only be a finite number of the intervals which are not less than , since their content is finite .
Each of these we can divide into a finite number of parts less than , or we can in any other way determine inside the intervals a set of intervals enclosing all points of , with the possible exception of a component of of zero content .
In each of these intervals there is a measurable component of of content less than , and the sum of all these components is S. This division of the set will be found to correspond very closely to the division of the continuum contemplated above , which is , of course , a special case of such a division ; in particular it will be shown to lend itself conveniently to form approximations to the upper integral .
I shall , for convenience , refer to such a division as a division of by of segments ; and the upper summation of any function over these components I shall call the upper summation with respect to over the S13 .
heorem ll.\mdash ; Given any }positive quantity can determine a positive quantity , such that , if the fundamental set be divided in any by means of segments then , provided only , and , the upper summation of the function uith respect to over these differs by less from a definite quantity I , the lower limit of such upper summations , when is divided by means of segments .
For , I being a lower limit , we can determine a set of intervals enclosing , such that the upper summation with respect to over these intervals lies between I and , where . . .
( 1 ) .
VOL. CCIV .
Also , by the definition of content , we can determine a set of intervals containing , whose content lies between and , where . . .
( 2 ) , being any quantity greater than the greatest numerical value of the function in question for points of S. The set of intervals consisting of the common parts of these two sets will contain , have content lying between and , and also be such that the upper summation with respect to over it has a value lying between I and .
From among the intervals of this set , we can determine a finite number , such that the sum of the remaining intervals is less than .
The content of the component of which is not external to these intervals is not less than .
The upper summation over the remaining intervals is , by 2 , numerically less than ; therefore the upper summation over these intervals is less than Let us now consider any division of by means of segments There cannot be more than of these which lie partly inside and partly outside the intervals .
The contribution of these to the upper summation will therefore be numerically less than Since , by what has been shown , the segments which are not entirely external to the intervals , enclose a component of of content not less than , the sum of those segments which are external to the intervals is not greater than , that is ; their contribution therefore to the upper summation is less than numerically .
Finally the contribution made by those segments which lie each inside one of the intervals cannot be greater than the upper summation over those intervals , that is cannot be greater than .
Hence the upper summation over these segments is less than .
Using ( 1 ) , it follows that we have only to take so that the upper summation over the segments may differ from I by less than ; this proves the theorem .
S14 .
It is clear from the above proof that , when we divide by means of segments it is immaterial whether , in estimating the upper limit with respect to over any segment , we include the end-points of that segment ( supposing them to belong to S ) , or not ; the potency of the points of not included will in either case be the same , and , as the theorem shows , in either case the upper summation be within of the limit I , provided only and are sufficiently small .
Suppose , then , we exclude the end-points of every segment .
Then the upper limit of the values of the function for points of inside any segment being the same as that of the associated upper limiting function , we have the following result : .quantity I is the same for function its upper lirniting function .
DR. W. H. YOUNG ON THE GENERAL THEORy OF INTEGRATION .
S15 .
quantity I is the upper of the function with respect to S. The result of the preceding section , together with the fact that the upper integral of a function is the same as that of its upper limiting function , show that it is only necessary to prove the present theorem for upper semi-continuous functions .
Lt be an upper semi-continuous function , and let us suppose the fundamental set divided up into a finite or countably infinite number of sets of points , so that the corresponding upper summation differs from the upper integral X by less than some assigned small positive quantity Round every point of we can describe a small interval , in which the maximum of with respect to differs from the value of at by a quantity less than , say , where ; and this interval may be decreased indefinitely .
That is , we have a set of tiles , *each ofwhich may be chipped as much as we please , and their points of attachment fill up Applying the Tile Theorem , we obtain a finite or countably infinite set of the tiles , each less than , covering up every point of , and the sum of the tiles is less than , where like manner we get a set of tiles from each set .
Applying the Tile Theorem to the set of all these tiles , since their points of attachment fill up , we obtain a finite or countably infinite set of them , covering every point of , and the sum of all is less than To each of these tiles we make correspond the lowest integer such that its point of attachment belongs to ; the sum of the tiles corresponding to any particular integer will then be less than , and the maximum of in each will be less than or equal to Now , if we include the boundary points of any interval , the set of intervals consisting of ( 1 ) the simple parts and ( 2 ) the overlapping parts of the tiles , leaves no points of over , and gives us therefore a division of by means of segments Of these , the sum of the overlapping parts is less than , since the content of the tiles is not less than , and their sum not greater than ; thus the contribution of the overlapping parts is numerically less than Me ' .
The contribution of the simple parts , on the other hand , is less than the upper summation over the tiles themselves , that is , less than Thus the upper summation over these segments is less than .
" " The Tile Theorem ' Proc. Lond. Math. Soo Ser. 2 , vol. 2 , Part I. , p. 67 .
Choose and so small that 1 ) this is less than , and 2 ) it differs ffom I by less than ( by Theorem 11 ) , being an assigned small positive quantity as smau as we please , which proves that in this case X and I must be identical , and so , as already pointed out , proves the theorem .
S16 .
It is now possible to give the promised alternative definition of integration ; the preceding theorems prove the equivalence of the two definitions:\mdash ; Definition.\mdash ; Let the fundamental set , excluding at most a set of points of zero content , be enclosed in or on the borders of a set of non-overlapping segments each less than , and of content less than .
Then let the content of that component of 8 in any segment be multiplied by the upper ( lower ) limit of the values of the function at points of that component , and let the summation be formed of all such products .
Then it may be shown that this summation has a definite limit when is indefinitely decreased , independent of the mode of construction of the segments and the mode in which approaches the value of zero .
This limit is called the upper ( lower ) limit of the function with respect to the fundamental set S. In the speci(case when the and lower integ ) als agree , the value is cnlled the integral of the function .
S17 .
To find the condition of integrability of a function with respect to a fundamental set , we require the following theorem:\mdash ; Theorem 13.\mdash ; The npper ( lower ) integral of the sum of any finite numbetof upper ( lower ) semi-continuous functions with respect to fundamental set the of the upper integrals of those functions .
Let be the sum of two upper semi-continuous functions and ; then is itself upper semi-continuous .
Then , by Theorem 11 , being assigned , we can determine so that if be divided by means of segments and the corresponding upper summations formed , then provided only and are each less than , the summations , say , ( 1 ) , . . . . .
( 2 ) , ( 3 ) , differ from the corresponding upper integrals by less than Since , and are all upper semi-continuous with respect to , we can , to each point of , determine an interval round , such that the maxima of , and in that interval differ from the corresponding values of , and , by less than some assigned small positive quantity , and this interval may indefinitely decreased .
Applying the Tile Theorem , we see that every point of is covered by one at least of a countable number of these tiles , each less than , no point of attachment being covered by any other the , and the sum of the tiles differing from by less than .
Thus we may take as our segments those respectively covered by the simple and overlapping portions of these tiles , if we include the boundary points of each such segment in it .
DR W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
The upper summation over the overlapping parts is in each of the three cases numerically less than Me ' , where is a quantity greater than the greatest numerical values of , and F. The simple part of any the contains the point of attachment , therefore the upper limit in that part differs from the value at by less than e3 .
Thus the upper summation over the segments lies between and Thus , denoting by with any distinguishing suffix or accent a quantity numerically less than 1 , with similar equations for the other two functions .
Now the summation is the sum of the corresponding summations for.fi and ; whence it follows that the upper integral differs from the sum of the upper integrals of and with respect to by a quantity which is smaller than any assigned positive quantity , which proves the theorem .
S18 .
Denoting by , and the associated upper , lower , and oscillation functions of a given function , the three functions , and are all upper semicontinuous with respect to the fundamental set , and is the sum of and The upper and lower integrals of with respect to the fundamental set are , by 2 Theorem 10 , the upper integral of and the lower integral of respectively , and the latter is minus the upper integral of .
Thus the excess of the upper over the lower integral of with respect to the fundamental set is the sum of the upper integrals and , that is , by Theorem 13 , the upper integral of with respect to the fundamental set .
Thus will be integrable with respect to the set if , and only if , the upper integral of its associated oscillation function be zero .
Let denote the component of the fundamental set at which has a value greater than , or equal to , ; by Theorem 7 , is a measurable set , let its content be Then if , omitting at most a component of zero content , we enclose the fundamental set by means of segments since the common part of two measurable sets is measurable , there will be at most a component of of zero content not included in the segments , and the remaining component of will have content .
Thus the content of the components of the fundamental set in segments containing points of will not be less than , and the upper limit of in each will not less than Hence the upper summation will not be less than Thus it is clear that the upper integral of cannot vanish unless , for all positive values of is zero .
Theorem 14.\mdash ; Thus the necessary sufficient conditions that a function should be integrable with respect to measurable set , is that the content of that component of at every point the oscillation of is greater than , or equal to , should , for every value of , be zero .
Remembering that the outer limiting set of measurable sets is measurable , and has for content the limit of the contents of the defining components , this gives us the following alternative statement:\mdash ; The necessary sufficient condition that function should be integrable with respect to a measurable set is that the content of that component of , at every point of which the function is continuous with respect to , should be to the content of S. S19 .
With the ordinary definition of upper ( lower ) integration , or of integration , it was at once evident that if the segment or region of integration were divided into two parts ( segments or regions ) , the sum of the upper ( lower ) integrals over the separate parts was the upper ( lower ) integral over the whole segment or region .
Theorem 15.\mdash ; More generally , it is evident from Theorem 1 that the sum the upper ( lower ) integrals over set of non-overlapping segmaents or regioms is equal to the upper ( lower ) over the whole segment or region , provided only the content of those segments or regions is the same as of the fundamental segment or region .
That this is not so for the general case when the fundamental set of points , whether a segment or not , is divided into component sets is shown by the following simple example :\mdash ; 6.\mdash ; Let be zero everywhere except at the rational points of the segment , and let have the value unity at the rational points , and consider the integrals over the rational and irrational points separately ; both of these are zero .
The upper integral over the whole segment is however unity , and the lower integral is zero .
The alternative definition of upper ( lower ) integration , given in S11 , shows that when the two component sets consist of parts of obtained by means of segments , the sum of the upper ( lower ) integrals over the two components is the upper ( lower ) integral over the whole fundamental set .
Theorem 16.\mdash ; More generally , the sum of the upper ( lower ) integrals over any finite or countably infinite number of non-overlapping components of obtained by means of segments integral over the whole fundamental set , provided the content of the components is equal to that of the fundamental set .
An upper ( lower ) semi-continuous function stands here again in an exceptional position .
We have , in fact , the following theorem:\mdash ; Theorem 17.\mdash ; The upper ( lower ) integral of an upper ( lower ) semi-continuous function over any mental set is equal to the sum of its integrals finite or countably infinite number of sets into wfiioh be divided .
DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
This follows from the definition at the end of S11 ; for if we divide the fundamental set into ( for instance ) two measurable components and , and then subdivide all three by means of segments , the definition shows that , in the case of an upper semi-continuous function , the sum of the upper summations over the segments with respect to and is not less than the upper integral .
But in each segment the sum of the contents of the parts of and S2 is the content of the corresponding part of , while the upper limit corresponding to is the greater of the upper limits corresponding to and S2 , so that the sum of the upper summations for and is not greater than the upper summation for S. Thus the sum of the upper summations with respect to and lies between the upper integral with respect to and the upper summation with respect to S. And since , by properly constructing the segments , we can make the upper summations differ by as little as we please from the corresponding upper integrals , this proves the theorem .
The above method of proof shows at the same time why it is that the theorem does not hold for every function , for , if we form the associated upper limiting functions with respect to and , the values at the different points of are not always the same .
S20 .
If in finding upper ( lower ) integrals we wish to divide the fundamental set up into convenient components , we must first replace the function by its associated upper ( lower ) limiting function .
Example 6 is a particularly instructive one ; the function is integrable over each of the two component sets into which the segment which is the range of variation is divided , and is not integrable over the segment itself We easily see , however , that the following theorems hold : Theorem 18.\mdash ; If function be integrable over the set , it is over every component set of S. From Theorem 17 it now follows that lbaeorem 19.\mdash ; The integral of an integrable functio over its set is to the sum of its inte.over every finite or countably infinite number of components into which may be divided .
For , as has been shown in the preceding section , the upper integral of a function over the fundamental set cannot be less than the sum of the upper integrals over the component sets ; and the lower integral of the function over cannot be greater than the sum of the lower integrals over the component sets .
Since , however , the function is integrable over , it is also integrable over each of the components .
Therefore the integral over calmot be less nor greater than the sum of the integrals over the components , so that it must be equal to this sum .
[ Q.E.D. ] In this connection , it should be noted that the integration of an integrable function ( which has finite upper and lower limits ) involves nowhere the idea of order , even when , for convenience , we determine it as the s'um of a countably infinite number of integrals .
The series of such integrals is an absolutely convergent one , and it has the same sum however it be arranged .
Theorem 20.\mdash ; The sum of the of finite number of integrable functions taken over the same fundamental set equal to the integral of the sum of the functi)taken over the set .
This follows from Theorem 13 , which asserts the same for upper ( lower ) integrals of upper ( lower ) semi-continuous functions .
The proof is of precisely the same nature as the preceding .
S21 .
We now proceed to obtain a formula for the upper ( lower ) integral of an upper ( lower ) semi-continuous function over any measurable set of points , in terms of an ordinary integral .
These formulae of course at the same time give us the upper ( lower ) integral of any function , by reason of S11 .
Let be any quantity not less than the greatest , and not greater than the least , value of an upper semi-continuous function defined for a fundamental set S. Divide into parts , and consider the sets of points , where denotes all those points of for which is , and is , so that .
By Theorem 7 , , are measurable sets ; let , be their contents .
Then by the definition at the end of S11 , the upper integral , being not greater than any upper summation , is not greater than - - i. e. , however great may be , and is therefore since I is a monotone function of and therefore an integrable one .
Here is the content of the fundamental set , and I that of the set of points for which the values of the function are greater than or equal to .
But it was shown , in S6 , that for any function whatever the greatest value of the lower summations is certainly not greater than the least value of the upper summations , therefore the upper integral is not less than all the lower summations .
Hence the upper integral is certainly not less than , DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
that is , not less than and therefore not less than Hence , finally the upper integral is actually equal to .
In precisely the same way it may be shown that the lower integral of a lower semi-continuous function can be expressed as an ordinary integral .
For variety I give the following proof Let be a lower semi-continuous functio1l , and ; then is an upper semicontinuous function ; and if be not greater than the lower limit of , and not less than the upper limit of , then is not greater than the lower limit of , and not less than the upper limit of .
Hence , by what has just been proyed , where is the content of the set of points of at which has a value Writing , we have for the lower integral of the sion K where is the content of the set of points of at which has a value Summing up , we have the following theorems:\mdash ; Theorem.\mdash ; The upper integral of an upper function with respect to measurable set is , where I is the content of that ) of at every point of which the fu nction has a value greater than or equad to Theorem 21.\mdash ; The lower of lower semi.continuous function respect to set is , cvhere is the of that component of every point of which the function has less or equed to In both these theorems and are quantities which are respectively less than or equal to the lower limit , and greater than or equal to the upper limit of the function for points of S. Hence also Theorem 22.\mdash ; The upper integral of any function with respect to set is Idk , where I is the content of that component of at every point of which the maximum of the function is greater or equal to J'he lower integral is is the confent of that component of at every point of which the minimum of the function is th or equed to This argument shows that in the case of an upper semi-continuous function the upper integral is equal not only to the lower limit of the upper sumrmations , but also to the upper limit of the lower summations the latter is not to be confounded with the lower integral .
Similarly for a lower semi-continuous function both limits give the lower integral .
VOL. CCIV.\mdash ; A. 2 I S22 .
By S18 the condition that a function should be integrable with respect to a measurable set is that the upper integral of the associated oscillation function should be zero .
latter unction is upper semi-continuous , by the corollary to Theorem 8 , so that we can apply the preceding theorem , putting , since this function is always positive or zero .
Thus the condition of integrability is that should vanish , where is the content of that component of at every point of whid ] the oscillation of the given function is .
This is only possible if is zero , except possibly at a set of values of of zero content ; as however , if were positive for any value of , it would be positive for every lesser value of , and therefore for a set of values of content greater than zero , it is clear that there can be no exception .
Thus we get again the condition of integrability obtained in S 18 ( Theorem 14 ) .
S23 .
When the function is integrable , the upper and lower integrals are equal , otherwise the upper is the greater , thus therefore S. Now no point can be such that the maximum there is less than while the minimum is greater than ; thus every point of belongs to at least one of the two sets I and J. Let denote the content of the set of points common to both I and then* Whence , that is the sign of equality being allowable if , and only if , the function is integrable .
Now is the content of the set of points at each of which the maximum , while the minimum is ; these points consist ( 1 .
) Points of continuity at which ; ( 2 .
) Points of discontinuity at which either the maximum is and the minimum , or the maximum and the minimum But it is clear , as at the end of S22 , that if for any value of the content of the set ( 2 ) were not zero , there would be a set of values of of positive content , for each of which the content of the set ( 2 ) would not be zero ; would then not be a null function and the giveu function would therefore not be integrable .
Thus we get a new form of the condition of integrability , and also theorems relating to the distribution of the points of continuity of an integrable function , and the values of a continuous function:\mdash ; * ' Theory of Content , ' p. 49 , ( 5 ) .
DR. W. H. YOUNG ON THE GENERAL THEORy OF INTEGRATION .
of Integrability.\mdash ; The necessary and sufficient condition for the integrability of function over measurabte set is that the content of the set of points for which the maximum of the function while the minimum should be a function of whose is zero , except for set of values of of content zero ; for each of these of the points of discontinuity of the function still form a set of content zero .
Theorem 23 .
\mdash ; The content of that component of the fundarnental set whose points are points of continuity which an integrable function , is a function of whose value is zero , except for a set of values of of content zero .
Theorem 24.\mdash ; The content of that component of the set at whose points a fnnction , is a function of wwhose is zero , except for set of values of of content zero .
S24 .
We now return to the tentative definitions of S6 , which we saw did not agree with the usual definitions .
On the other hand , the definitions we have since constructed seem more artificial than these .
It suggests itself , therefore , that the most logical plan is to throw overboard the Riemann and Darboux definitions altogether , and to define an integral as follows:\mdash ; Let the set be divided into components in conceivcpble way , and let the content of component be multiplied by the upper ( lower ) limit of the values of the function at points of component , and the sum of all such products be formed ; tloen the outer of the integral is defined to be the lower ( upper ) limit of all such summations .
If we assume either that all sets are lmeasurable , or that all functions are such that the points for which the values of the function are are measurable , or that the functions with which we are concerned have this property , the argument of S21 still applies , and we can assert that the outer measure is equal to the inner measure of the integral , and that each can be expressed in either of the two forms where I and are the contents of those components of , at every point of which the function has values respectively and It is not known whether any but measurable sets exist , or whether any functions can be constructed not having the above property .
M. LEBESGUE has , therefore , used the term summable to denote the functions under consideration .
Precisely as in S22 we can now prove the following theorem:\mdash ; heorem.\mdash ; The content of that component of the fundamental set at every of which summable fnnction has the value , is a function of whose value is zero exceptJor a set of values of of content zero .
In the case of a summable function , therefore , the outer and inner measures of the integral agree , and we may call either the generalised integral of the function .
As DR. W. H. YOUNG ON THE GENERAL THEORy OF INTEGRATION .
we have seen , functions which are integrable in the old sense are integrable in the new sense , and the integrals agree , but the converse is not true .
In the case of an upper ( lower ) semi-continuous function , the generalised integral is the upper ( lower ) integral .
S25 .
LEBESGUE gives two definitions of his generalised integral , which I shall , for convenience , allude to as the Lebesgue integral .
The first is a geometrical definition , and has the disadvantage that the positive and negative values of the function have to be considered separately .
It is as follows:\mdash ; Definition of the Lebes.que tegral.\mdash ; Let be a function defined for every point of a finite segment ; consider the plane set of points defined by the three inequalities* Let and be the two parts of respectively above and below the axis of ( the points on the axis of may be considered as belonging to whichever we prefer or .
If is a measurable set , then both and are measurable .
In this case the function is said to be summable , and the excess of the content of over that of is defined to be the Lebesgue integral of over the segment .
LEBESGUE'S second definition is analytical .
Analytical Definition of Function.\mdash ; A summable function is such that the set of values of , for which the values of the function lie between any two quantities and , is measul.able .
Conversely , if this condition is satisfied , and the upper and lower limits of the function are finite , the function is summable .
Dejinition of the .\mdash ; Let the region of variation of be denoted by , and let it be divided into / parts each less than , say , at the point , Let denote the content of those points at which ; and that of the points at which Then it may be shown } , the two summations and have a common limit , when is decreased indefinitely ; this limit is the Lebesgue integral of from to The identity of the two definitions is easily proved ; in the case of a function which is always positive , the former of the two given expressions evidently represents the * Here I have corrected an obvious misprint in paper .
This is stated without proof by LEBESGUE in S17 ; it is a special case of the theorem of S18 , p. 261 .
A more simple proof is afforded by considering as the common part of and a sufficiently large rectangle on the given segment as base and lying on the positive side of the axis , and as the difference berween and ; since the common part or the difference of two measurable sets is measurable .
similar proof apphes to the theorem of S , given below here as the analytical definition .
DR. W. H. yOUNG ON THE NERAL THEORY OF INTEGRATION .
content of a plane set containing the set , and the latter the content of a plane set contained in , so that the common limit must be the content of E. In the case when the function is sometimes positive and sometimes negative , the sum of the positive terms of the first summation is the content of a set containing , and the sum of the ative terms , taken with positive sign , is the content of a set contained in ; similarly the second summation is the content of a component of minus the content of a set containing ; thus the common limit must be the content of minus that of .
Thus the two definitions are identical .
* The second of LEBESGUE'S definitions enables us without difficulty to identify the Lebesgue integral with the generalised integral which I defined in S24 .
In fact , comparing LEBESGUE'S notation with that used by myself in $ 21 , it is clear that Whence the former of the two given expressions is equal to that is , which , as is decreased indefinitely , approaches the limit that is , the generalised integral of S24 .
Thus the Lebesgue integral is the same the generalised of S24 , the fundamental set being a finite segment .
S26 .
Contrasting the definition of S 24 ( being now a finite segment ) with the geometrical definition of LEBESGUE , we see that they stand to one another in the same..relation as the ordinary definition of integration of , say , a continuous function to its definition as a certain area .
Just as , however , the mathematical conception of area is more complex than and indeed depends on that of length , so does the theory of the content of a plane set of points depend naturally on that of the content of a linear set .
Just as the determination of area requires the application of the processes explained in the first definition of integration of continuous functions , so with the content of a plane set .
Thus the comparative simplicity of the geometrical definition is only apparent .
With regard to LEBESGUE'S analytical definition , I have pointed out that it is equivalent to what seems to me the much more convenient form in which I have expressed it as an ordinary integral ( S24 ) .
If we know I as a function of , which may be the case , all Lebesgue integration and all upper and lower integration reduce to ordinary integration .
It may , however , happen that I is not readily found as a function of .
The definition of S24 seems then the most fundamental , and is , in many respects , very convenient in the theory .
* LEBESGUE , loc. , p. 252 .
S27 .
It is instructive to show the identity of LEBESGUE'S geometrical definition with that of S24 ( being a finite segment ) directly .
At the same time it should be remarked that the argument is independent of the fact that is a finite segment , so that we have some general theorems with respect to the geometrical representation of the processes of integration with respect to a set of points to which I shall return in S32 .
Lemma at every point of a measurable set on the -axis of content I we an ordinate of length , tloe set constituted by this set of ordinates , or , as I shall call them , for deJiniteness , this set of blocks , is measurable and has content For we can enclose in a set of non-overlapping intervals of content less than .
Erecting on these rectangles of height , we have enclosed the whole plane set in a set of rectangles of content as near as we please to .
On the other hand , taking in a closed set of content greater than , the corresponding set of blocks forms a closed set of content as near as we please to .
Thus the outer measure of the content of the set of blocks is not greater than , and the inner measure of the content is not less than , which proves the Lemma .
Suppose now we are given any measurable set , and on as base any set of blocks .
Let us denote the length of the block at the point by X , and , for simplicity , let us first consider X as being always positive .
Then , if we divide in any manner into a finite or countably infinite set of measurable components , , and at each point of we replace the given block by one of length equal to the upper limit of the lengths of the blocks at points of , we get a new set of blocks which , regarded as a plane set of points , contains the given set of blocks .
Since the sum of a finite or countably infinite series of non-overlapping measurable sets is a measurable set whose content is the sum of the contents of its components , it follows that the new set of blocks constitutes a plane set of points which is measurable , and has for content , by the Lemma , the upper summation of X over corresponding to the mode of division adopted .
Hence the outer measure of the content of the plane set of points constituted by the given set of blocks is not greater than this upper summation , and therefore , since this is true for every mode of division , not greater than the outer measure of the integral of the blocks over S. Similarly for the lower measure of the integral .
If , on the other hand , X be sometimes positive and sometimes negative , we must , as LEBESGUE does , consider separately the positive and negative blocks ; thus we get .
the following theorems:\mdash ; heorem \mdash ; An upper or lower summation is the difference of the contents of two measurable sets of blocks , one on positive and one on the negatire side of thoe axis .\mdash ; Given set of blocks on a measurable set , the outer measure of tloe content of plane set of points constituted by the positive blocks minus the imoer DR. W. H. YOUNG ON THE THEORY OF INTEGRATION .
of the content of the blocks is not great \ldquo ; the outer of integral of the blocks over S. Simitarl .
, the inner ?
of the content of the positive blocks minus the outer of that of the neCjat , blocks is not tes the inner of the integral .
In symbols where and are respectively the set of points constituted by the positive and the negative blocks , and I denotes the integral , the indices and denote outer and inner measures respectively .
Denoting the sum of and by , and supposing the upper and lower limits the lengths of the blocks to be finite , and E2 are the common parts of , and the sets of blocks on as base whose heights are respectively the upper and the lower limits of the lengths of the blocks , both these sets of blocks are measurable .
Thus we see that , if is measurable , and E2 are so also , so that and are equal to one another and to the difference of the contents of and , that is to say , a summabte function is tegrable in the generalise sense of S24 , and integrat defined in LEBESGUE'S manner is the integral as defined S24 , being a finite segment .
Conversely , if and are equal , but since the outer content of a set is not less than the inner content , , whence Therefore and therefore , since the outer measure of the content is not greater than and the inner measure is not less than , the inner and outer measures of the content of are equal , and is a measurable set .
Thus function which is integrable with respect to finite segment in the generalised sense of S24 is a summable function , its integral is the Lebesgue integral , LEBESGUE'S geometrical definition .
Summing up the results of this section , we have the following theorem:\mdash ; Theorem.\mdash ; Lebesgue integration is identical cvith integration with respect to a ite segment .
S28 .
In general , without confining our attention to finite segments , we have what has been shown the following geometrical definition of generalised integration with respect to a measurable set Geometrical Definition.\mdash ; At each point of measurable set draw an ordinate ( block ) to the value of function defined for ever point of , the outer measure of the content of the positive blocks minus the measure of the negative is called the outer measure of the function with respect to the fundamental set .
Set S.\mdash ; If the whole set of blocks is , the nction is said to be integrable in the generalised sens , ( or summable ) with respect to the mental set S. In this case the positive blocks alone form a measurable set , and so do the negative blocks , and the excess of the content of the positive over that of the negative blocks is the generalised integral of the function nith respect to S. LEBESGUE ( p. 255 ) has occasion to use integration with respect to a measurable set , but only in the case when the set is contained in a finite segment .
The mode of definition adopted by him is that of completing the function by ascribing to it the value zero at all points of the segment other than those of the set in question , and then defining the integral of the function with respect to the fundamental set as equal to that of the extended function in the whole segment .
In this case the generalised integral so defined is evidently the same as that defined above and in S24 ; the mode of definition is , however , open to some objection ; characteristic properties of the function such as continuity or semi-continuity with respect to the fundamental set , which materially simplify the properties of the integral , are not maintained by the extended function ; on the other hand the definition suggests a difficulty , when dealing with fundamental sets of finite content not lying in a finite segment , which is entirely illusory , and gives a pre-eminence to the finite segment as region of operation which it does not in reality in any way possess .
S29 .
LEBESGUE'S theorem that the sum of two summable functions is a summable function , and its integral is the sum of their integrals , is equally true when the fundamental set is any measurable set of finite content .
It is an immediate result of any of the definitions that this is the case when one of the functions is a constant , thus , the upper and lower limits being as usual finite , if the theorem is true for positive functions it is true always .
In the case of positive functions the theorem is geometrically equivalent to the following:\mdash ; Theorem .any set of positive blocks forming measurable ptane set of points , the blocks may be shifted up parallel to themselves without altering the content of the set , prorided the amount of shifting at each point is a function of Divide the base up into any number of measurable components , and consider the set in the shifted position .
At any point of let the block be PQ , so that represents the amount of shifting from the position when the lower extremities were all on the -axes .
Prolong the block to , so that PR is equal to the upper limit of the length of the blocks corresponding to all points of .
In \mdash ; this way we get a plane set containing the given set .
Now , by Lemma 1 , S27 , the content of a set of blocks each of length PR , erected at all the points of , is .
Hence , since we may evidently consider as having been obtained by shifting up the ordinate the constant amount PR , the content of the set of blocks that of the blocks DR. W. H. YOUNG ON THE GENERAL THEORY OF INTEGRATION .
Now the blocks Px form a measurable set , by the usual argument , since they consist of the part of a measurable set of blocks on a measurable component as base ; therefore the blocks PR at each point of , shifted up to the position of the figure , being the difference of two measurable sets , form a measurable set , and the content is , that is , the same as it was before shifting .
Since this is true for each component , it is true for the whole set of elongated blocks , that they form a measurable set whose content is the same as it was before shifting .
Now , by S27 , we know that , before shifting , the content of this set was the upper summation of the lengths of the given blocks , that is , over the fundamental set , divided into , Thus this upper summation , being the content of a plane set containing the given set , is not less than the content of the given set ( of course in the shifted position ) .
Similarly the corresponding lower summation is not greater than the same content .
But the lower limit of the upper summations is equal to the upper limit of the lower summations , since either of them represents the content of the given blocks before they were shifted up from the -axes .
Thus the same content is itself neither less nor greater than the content of the set in the shifted position , that is to say , the content has been left unaltered by the shifting .
Corollary 1.\mdash ; In the shifted position the content is still the generalised integral of the length of the blocks .
.\mdash ; The sum of any finite number of summable functions is a summable function , its integral is the sum of their , thoe fundamental set being measurable set offinite content .
Corollary 3.\mdash ; Since the limit of sequence of measurable sets is set , it also follows that the sum of an absolutely convergent series of summabte functions is summable function , and its generalised integral is the sum of their integrals , provided , as usual , the functions have finite upper lower limits .
S30 .
By means of a theorem proved in my paper on " " Upper and Lower Integration , \ldquo ; can extend the results of S 29 still further .
The theorem quoted states if X ' be the content of the section of a closed plane set by the ordinate through the point of the -axis , the content of the plane set is , and that , further , X is an upper semi-continuous function of X. It follows , then , by S24 , that the content is the generalised integral of X ' .
The theorem now to be proved is as follows:\mdash ; Theorem \mdash ; If and be the outer and inner measures of the confent of the ordinate section of a measurable set such that the set got by closin.it ) content , by the ordinate through the point and are both fpmctions , and tloe generalised integral of either is the content of the measurable set .
Let I be the content of the set , and any assigned small positive quantity .
Let us * Proc. Lond. Math. Soc Ser. 2 , vol. 2 , Part I. , p. 60 .
VOL. CCIV.\mdash ; A. 2 take a closed component of the given set of content greater than noting by X ' the content of the ordinate section of this set , we have , by the theorem quoted , Since X ' is an upper semi-continuous function , it follows , by S24 , that upper limit of the lower summations of X ' , or , since X ' is not greater than upper limit of the lower summations of Since may be as small as we please , upper limit of the lower summations of upper limit of the lower summations of Next let the content of the set got by closing the given set be denoted by , and that of the complementary set by , so that S. Denoting by and the quantities for the complementary set and the whole closed set corresponding to X ' for the iven set , we have , as before , upper limit of the lower summations of Now is the content of the difference of two closed sets , that is an inner limiting set , *containing the set X , therefore , Hence upper limit of the lower sumtnations of , upper limit of ( a lower summation of minus an upper summation of ) .
But the upper limit of the lower sulnmations of is the generalised integral of that is , ; therefore the lower limit of the upper summations of that is , lower limit of the upper summations of fortiori , lower limit of the upper summations of Now every upper summation of a function is greater than , or at least equal to , any lower summation , so quantity cannot be less than the upper limit of the ' Tbeory of Content , ' p. 36 .
DR. W. YOUNG ON NERAL THEOliY OF GRATION .
lower summations without being less than the lower limit of the upper summations ; neither can it be greater than the lower limit of the upper summations without being greater than the upper limit of the lower summations .
Thus I must be actually equal to the upper limit of the lower summations as well as to the lower limit of the upper summations in the case of either or ; that is to say , I is the generalised integral of either or X , and both these functions are summable .
[ Q.E.D. ] Corollary 1.\mdash ; Each section of set moved on its ordinate in such a manner that thoe ( linear ) of the section is ltered , that the whole set , the content of the whole set is unaltered .
Corollary 2.\mdash ; At each point of set of points of content A dran ) ordinate , and on it set of points of content , the content of the whole set so is AB .
Here , as elsewhere , the fundamental set need not be a linear set , but may have a content of number of dimensions .
S31 .
The preceding section , coupled with S24 , give us as the content of any measurable set ( provided the set got by closing it has finite content , here I is the content of the set of points of the fundamental set at which the i1mer ( or the outer ) measure of the content is .
This , together with the preceding section , give the solution of the problem alluded to in S26 , viz. , the reduction of the calculation of -dimensional content to that of -dimensional content , and so ultimately to that of linear content .
Bearing in mind the definition of a generalised , we have the following rule for finding the content of an -dimensional Take any hyperplane section of the set , project the set on to this byperplaue , and take any measurable set containing this projection as the fundamental set S. Divide up in any way into a finite or countably infinite set of measurable components , and multiply the content of component by the upper ( lower ) limit of the values of the ( linear ) inner or outer content of the corresponding ordinate sections of the given set ; such products , the lower ( upper ) limit of all such summations is the content of the given set .
S32 .
I have explained the geometrical representation of generalised integration with respect to a set .
It is of interest to note the corresponding representations of what I call ordinary upper and lower integration respect to a set .
Consider first the case the fundamental set is a seglnent , and the set of blocks corresponding to the generalised integral of the function . .
the function is everywhere positive , the geometrical representation of the ordinary .
integral is obtained by closing the plane set of points constituted by the blocks .
If the function is not everywhere positive , we can make it so by adding a constant , which is geometrically equivalent to adding a rectangle to the representative set , part of which , viz. , , was to be considered as negative , and must be subtracted the .
W. H. yOUNG ON THE GENERAL THEORY OF INTEGRATION .
rectangle ; this addition of the rectangle is therefore equivalent to sinking the axis of to a convenient position , so that all the blocks become positive .
The upper lutegral of the given function , plus the constant , is now represented by the set got by closing the set of blocks .
From this we have now to subtract again the rectangle , or , which is the same thing , return the -axis to its original position , in order to get the geometrical representation of the upper integral of the given function .
Thus , in the general case , the geometrical repreFig .
2 .
sentation of the upper integral is not precisely a closed set , but a closed set minus a rectangle , as is shown roughly in fig. 2 , viz. , the shaded region , of which the part below the -axis , corresponding to , is to be considered as negative , and the other part , corresponding to , is closed .
Similarly , in the case of a function which is always negative , representation of the lower integl.al is a closed plane set constituted by blocks ; in the general case it is esented by the excess of a rectangle over such a set .
When the fundamental set is any measurable seb whatever , instead of closing the fundamental set actually , we do so relative to , that is , we take in only those limiting points which lie on ordinates through S. The rectangle to be subtracted in the case of the ordinary upper integral , or from which the relatively closed set is to be subtracted the case of a lower integral , is then a rectangle , that is , that part of a rectangle which lies on the ordinates through S. \mdash ;

