]\gt ; Prof. W. H. Young .
On the Ordinary sugar solutions without taking up the sugar at the same time supports the author 's theory that the property of selective permeability is ( in the case of colloidal membranes ) a result of preferential adsorption .
It is also shown in the above paper that the side of a membrane in contact with pure water has a greater moisture content than the side in contact with sugar solution .
This fact supports the hypothesis\mdash ; first advanced by Graham on expe1imental grounds\mdash ; that osmosis across the membrane takes place because pure water induces a greater moisture : pressure and concentration inside the membrane than the solution does .
On the Convergence of Restricted Fourier Series .
By Prof. W. H. YOUNG , Sc. D. , F.R.S. ( Received October 11 , 1916 .
) S1 .
The necessary and sufficient condition that a trigonometrical series* should be a Fourier series is that the integrated series should converge to an integral throughout the closed interval of periodicity , and should be the Fourier series , accordingly , of an integral .
Conversely , starting with Fourier series of an integral and differentiating it term by term , we obtain .
the Fourier series of the most general type , namely , one associated with any function possessing an absolutely convergent integral .
If the Fourier series which is differentiated is not the Fourier series of an integral , but of a function which fails to be an , at even a point , the derived series will not be a Fourier series .
The series so obtained , namely , by differentiating term-by-term the Fourier series of a function which , without an integral in the whole closed interval of periodicity , is an integral in one or more sub-intervals , I have found it convenient in a recent to call ' restricted Fourier series.\ldquo ; I retain the terln ' Fourier because such series possess , in the interval or * It is convenient to suppose the trigonometrical series to have no constant term , or to regard the integrated serie .
as a series obtained from the original series by interchanging the coefficients of and nx , changing the sign of one of them , and dividing both by ? .
The integral to which the integrated series converges is accordingly a periodic function .
By integral we mean , as usual , absolutely conve.rgent or Lebesgue integral .
' ' On the Convergence of the Derived Series of Fourier Series read at the JuneMeeting , 1916 , of the L.M. S. and recently presented to that Society .
Convergence of Restricted Fourier Series .
intervals to which they are " " restricted several of the properties of Fourier series , so that they can be employed in analysis in some of the ways in which ordinary Fourier series have used .
There is , however , one important difference between the new and the old types of Fourier series .
In dealing with theformer , we can in general only employ the method of summation due to Cesaro , index unity at least .
The theorems which continue to hold for restricted Fourier series in the interval to which they are restricted are obtained by suitably theorems relating to Fourier series in which Cesaro convergence occurs .
For more information on this point I must refer my readers to the paper just cited .
It will suffice for our present purposes to refer to only one of these properties .
It is known that the behaviour of a Fourier series as regards convergence , or mode of oscillation , at a particular point of the interval of periodicity , depends only on the nature of the associated function in the neighbourhood of the point , this neighbourhood taken as small as we please , and the method of the series , or of ibs upper and lower functions , peak and chasm functions , etc. , being the ordinary classical one .
It is remarkable both that this theorem fails in the case of restricted Fourier series , and that it becomes true when bhe summation is performed in the Cesaro manner , index Thus tests for the ordinary convergence of a restricted Fourier series of a type analogous to the known ones for Fourier series cannot exist , but only tests for the Cesaro convergence analogous to those the Cesaro convergence of ordinary Fourier series .
And again only those theorems relating to the ration term-by-term of a Fourier series , after multiplication by another function , in which the con- vergence employed is of the Cesaro type , have their analogues in the general theory of restricted Fourier series .
To secure the ordinary convergence of a restricted Fourier series at a point or in an interval , and the holding good of integl.ation theorems which i1lvolve ordinary convergence in an interval , conditions must be fulfilled which concern not only the neighbourhood of the point , or the interval considered , but also the whole of the rest of the interval of periodicity .
In other words , additional hypotheses of a novel type will have to be made .
In the present communication to the Society I propose to initiate the new theory which thus arises , by considering a case which , though special in character , is probably the most important of all the cases that can prese1lt themselves .
I assume that the failure of the derived series to be a Fourier series is due to the fact that , at a finite number of points , the function associated with the series from which it is derived fails to be an integral .
In other words this function is an integral Prof. W. H. Young .
On the over every sub-interval which does not contain any of the points in question , and over intervals only .
The main neral auxiliary theorem obtained is that if is the function associated with th original series , and has accordingly , the function associated with series , for its differential coefficient almost where , then the restricted Fourier series b , haves in every interval not containone of th points , , at which ceases to be an integral , exactly like ourier s , provided : ( 1 ) is an in some interval containing , for , 2 , , to zero , as ( 2 ) if denotc function which , except in a certain sub-interval of the interval , ounding xperiodic infegrl , and in that oxceptional terval requal to , then the coefficients of the derived series of the Fourier of to , for It will be noticed that , in the case of a single exceptional point , the second of the pair of conditions , thus shown to be sufficient , is certainly a necessary condition for a ] series to converge ; its terms must , like those of any other series , themselves converge to zero .
Moreover , as is well known , if this convergence holds throughout an interval , however small , the coefficients must themselves converge to zero .
With regard to the first condition , I show in the present paper that all known conditions imposed upon , to secure the fulfilment of ( 2 ) , involve at the same time the truth of ( 1 ) .
The conditions in question are those which have recently communicated to the Society* ; they lead at once , among others , to the following theorems:\mdash ; If as , we have , 2 , , ( ii ) if fur there ( iii ) is an integr.also interval containing being the function associated with the series , from which the Fourier was derived , then latter series any point other than one of " " On the Order of Magnitude of the Coefficients of a Fourier Series 'Boy .
Soc. Proc , vol. 93 , p. 42 .
For simplicity , the conditions are stated separately for an odd and an even function , and the singular point is supposed to be at the origin .
The transition to the case contemplated in the present paper is of course immediate .
Convergence of Restricted Fourier Series .
the exceptimal points , , , depends only on the nature of in the neighbourhood of the point considered , and , therefore , on associated with the restricted Fourier series .
Again , if , as , we have for , 2 , , ( i ) ( ii ) Lt both these limits accordingly existing , ( iii ) is a bounded ftmction of , or , more ycncrally , ( iiia ) is a bounded fi of the notation being as in the theorem , same conclusion follows .
S2 .
These theorems take a very simple form if we specialise the character of the function further .
Thus , we may , for example , replace the conditions ( i ) , ( ii ) , and ( iii ) of the former of these two theorems by } ) cvle condition that is an even function of mded variation which has the unique limit zero , as , and thus obtain immediately , among others , the theorem:\mdash ; If an function of } the interral of has limit zero , as the restricted Fonrier series associated nrith ordinarily at all points of the interval .
We may also replace the corresponding conditions in the second theorem by the condition that is tien fnnctionwhich , while is itself a function of in ) interval not containing the origin , and so obtain the theorem that the Fourier series associated with such a function convcrJ Jat th origin .
Moreover , at the different points , , , of the fundamental theorem , we may , of course , secure the fulfilment of the corresponding condition by hypotheses , varying with the value of the subscript .
Closely connected with the subject of the paper are certain theorems depending on the uniform convergence to zero of that portion of the representing the nth partial summation , the treatment of which forms the main difficulty in these ations .
This uniformity is secured , like the convergence itself , by the hypotheses made .
Ihese additional theorems are those alluded to at the beginning of this introduction ; their statement and proof will be found below .
280 Prof. W. H. Young .
On the Ordinary S3 .
We begin by explainixlg some of the circumstances under which we may differentiate an integral of the form under the sign of integration .
In the succeeding theorems we encounter two such cases:\mdash ; ( i ) When has a coefficient at point of the interval , this coeficient is a bounded function of , uhile is summable .
We may then multiply the bounded sequence by and integrate term by term , which gives us the required result , ; ( ii ) when and are both integrals for the values of their arguments invotved in the integral .
In this case writing have , by parts , which may be differentiated by case ( i ) , giving , which is the required result .
The same reasoning gives the required result when we have instead of .
We have here supposed the limits of integration independent of ; in the contrary case we must , of course , add on corresponding terms to the result of integration under the integral sign .
S4 .
It will also be convenient to prefix the following two Lemmas:\mdash ; LEMMA in a closed interval , an integral which vanishes when , then the limits , as , of , can be as small as we please by making conveniently small .
Convergence of Restricted Fourier Series .
In fact , integrating by parts , the integral which appears vanishes by the Theorem of Riemann-Lebesgue , and the expression in square brackets is as .small as we please .
LEMMA \mdash ; If tf ( t ) an integral in the dosed interval of periodicity the coefficients of the derived series of the Fourier of converge to zero , .
( I ) For , denoting the integral whose limit is considered by I , nt The first integral approaches zero , being times one of the coefficients of the derived series .
The second and third integrals are of the form nt where is an integral , since , by hypothesis is an integral .
rating by parts , this becomes in which the integral approaches zero by the Theorem of Riemann-Lebesgue .
The bracket expression is zero when it involves the sine , and when it involves the cosine is whence , since , and this term appears in ( I ) with minus sign , the required result follows .
Similarly .
( II ) S4 .
The following auxiliary theorem is an immediate consequence of the known theory of Fourier series:\mdash ; THEOREM I.\mdash ; If ( i ) the associated function of Fourieseries is integral in the closed interval of its ?
of , so that it has accordingly absolutely tegrable differential coefficient , , say , cisting almost everywhere in the closed ; and if ( ii ) , Prof. W. H. Young .
On the Ordinary 1 : then the and lower functions of the derived series of the Fourier series of the point depend lie on the properties of in a neighbourhoo of the point small we please , and with be the same as those of the Fourier absolutely integrable function , agrees in the closed ' . .
Denote by the nth partial summation of the derived series in question , .
then , by the usual theory , and therefore , by condition ( ii ) , the upper and lower limits of are the same as those of They are , therefore , by S2 , the same as the limits of This proves the theorem .
In fact , if be any absolutely rable function , agreeing with in , then , by the known theory , ; and therefore the limits of the partial summations of the Fourier series of at the point are the same as those of the expression ( 1 ) .
COR .
\mdash ; If in condition ( ii ) the convergence to zero at a point or in an , the restricted Fourier series of and the Fourier series of ?
oill have the pecutiarities with regard to , or non , of vergence , oroscillation , at the point , or in the interval .
COB .
2.\mdash ; Under the nstances as Cor. 1 , we may , in integration theorems involving integration term-by-term of Fourier series when multiplied by a function which is bsolutelly integrable , substitute for the Fourier the restricted Fourier series of function which with the of th series in asub-interval , provided the range of inntegration be restricted to that sub-interval .
In fact the difference between the partial summation of the restricted Fourier series and that of the Fourier series converges uniformly to zero , and therefore continues to do so when multiplied by any absolutely integrable function , and integrated term-by-term .
S5 , We have now to transform condition ( ii ) of the theorem just given in a manner suitable for the purposes in hand .
Convergence of Rest Series .
THEOREM If , outside the interval is fwhere an inte.qral , except at the finite number of points ; and if ( ii ) each , 2 , , the expression sin , where denotes , sohere has , for brevity , been written for , has , as , limits which are as small as )please , being small , thm the expression . .
For simplicity we may suppose that the extremities of the interval of periodicity are not included among the points .
No loss of generality is hereby occasioned .
For definiteness , suppose positive .
The points fall into two classes , those which lie in , and those which lie in or in ( see figure ) .
As increases from to will pass over points of the former class only , and over points of the latter class only .
First let denote one of the first class , and consider the expression Then , supposing there are such points , we divide up the interval into intervals .
These intervals consist of ( i ) the two end-intervals and , and other corresponding intervals which do not contain the points , ( ii ) intervals of the type .
Consider the portion of due to an interval of the first class .
This is of the type S the integrand having no singularities with respect to or , so that , by S2 , this portion of may be differentiated in the usual way , iving us three terms , one of which is the integral obtained by omitting d/ \amp ; in VOL. Prof. W. H. Young .
On the Ordinary the preceding expression and writing for , and accordingly has the unique limit zero , and the other two are : S .
In the case of the end-intervals which belong to this class , one only of these two latter terms , of course , appears .
Next consider the portion of due to an interval of the second class .
This is of the type which may be written where , by S2 , the differentiation may be performed under the integral sign .
We thus get two terms , one of which has the unique limit zero , by the Theorem of Riemann-Lebesgue , while the other is Collecting our results so far , and bearing in mind that , in the statement of condition ( ii ) , stands for , and for , we see that , as far as is concerned , the singular points of the first of the two classes , into which we had divided them , give us terms precisely of the type stated in the enunciation of the theorem .
Next consider We shall have work of precisely the same kind to perform as before in dealing with the first class of intervals .
In the second class we shall write , and paying attention to the sign of , we shall find that , for ( \amp ; also , the condition ( ii ) contains expressions to which we are actually led by our process .
Thus our theorem is proved .
S6 .
It should be noticed that in the preceding theorem we have only used the condition that the singularities of are point singularities , in demanding merely that the expressions can be made as small as we please , by taking conveniently small .
If we require to converge to zero , the singularities may be any we please inside the corresponding intervals .
Convergence of Restricted Fourier Series .
In the next auxiliary theorem , about to be given , it is , however , essential that the points at which is not an integral should be finite in number .
THEOREM III .
If outside ihe interval , is everywhere an integral , except at the number of points , , ; ( ii ) is for each value of , 2 , , , and an appropriate ntly small range of values of , including , an integral which converges to zero , as ; ( iii ) The expressions have , as , limits all of be as small as we please , mahing xpressions which occur in Theorem II ( ii ) , have limits which are as as we being small .
We have to consider the value , or values , of expressions of the type , say .
Write , then where the quantity inside the second curly bracket is as small as we please , when is chosen conveniently small , since , by hypothesis ( ii ) , approaches zero , when , and is bounded .
Thus we may replace by Let us write where cos , c. pt Prof. W. H. Young .
On the Ordinary The integral last Wl.itten down being as small as we please , by Lmma 1 , since and are both integrals , so that is an integral , and has , moreover , like , the unique limit zero , when Combining this with ( 1 ) , we see that has limits which differ by as little as we please from those of an expression of the form where and , transforming in the same way as we transformed above , But , by the hypothesis ( ii ) , the expressions and have hmits as small as we please , for all values of , 2 , , .
Thus each of the expressions with which we started has limits as small as we please , which proves the theorem .
COR.\mdash ; In condition ( iii ) may be replaced by , without disturbing the validity of the theorem .
As regards the second and third members of the expression , this is almost obvious ; in fact , taking , for instance , the sine , differs from by which , by the last statement of condition ( iii ) , converges to zero , as , and therefore has , when , a value as small as we please .
As ards the integral constituting the first term of the expression , we may clearly replace the outside the integral by , since , by the Theorem of Riemann-Lebesgue , we shall only alter the term by a quantity which converges to zero , as or becomes infinite .
Suppose this change already made .
If we now change the inside the integral into , we have a difference in value given by the sum of two expressions of the form where is an integral .
These , by Lemma 1 of S4 , have , for a sufficiently small , limits as small as we please .
For is an integral which approaches zero , as , and , therefore , as shown in that lemma , integrating by parts , and , using the Theorem of Riemann-Lebesgue , our result follows .
Convergence of Restricted Fourier Series .
S7 .
We now come to our fundamental auxiliary theorem .
THEOREM \mdash ; The ( ordinary ) upper and lowerfunctions of the first derived series of the Fourier series of at particular point ( other than the points , , , to be immediately specified ) depend only on thnature of in a neighbourhood endosing the point , as small as we please , provided the following conditions be satisfied : ( i ) Except in an interval which contains one at least of acertain Jinite number of points , , , an absolutely convergent integral .
( ii ) is an absolutely ergent integral for , 2 , , , in a certain interval containing , and converges to zero , when ( iii ) If dmote function , except a sub-interval of the interval , surrounding a , and in that exceptional sub-interval is equal to , then the coefficients of the derived series of the Fourier series of to zero , for , 2 , . .
, , For , if the conditions ( iii ) hold , Hence Therefore , since is periodic , Now where the integral on the right vanishes when , by the Theorem of Riemann-Lebesgue .
Hence , as In like manner Also Adding ( 2 ) , ( 3 ) and ( 4 ) , and comparing with ( 1 ) , we get , finally , , ( 5 ) Prof W. H. Young .
On the since , and , and , .
Now , as in the preceding corollary , we may in ( 5 ) replace by , where .
The conditions ( i ) , ( ii ) , and ( iii ) of Theorem III of S 6 are therefore satisfied .
Hence , by Theorems II and III , This , by Theorem I , proves the theorem .
S 8 .
From Theorem of the preceding article , we conclude easily that whole classes of restricted Fourier series of the kind contemplated actually exist .
In a recent communication to the Society , I have virtually obtained sufficient conditions for the fulfilment of condition ( iii ) of Theorem above .
We proceed first to show that these sufficient conditions ensure at the same time the fulfilment of condition ( ii ) .
LEMbIA .\mdash ; If is an integral every intervat not cont ( xining the origin , if one of the following conditions holds some intervat containing tl , origin , ( i ) bounded ; an integral ; ( iii ) bounded ; is an the whole of per.iodicity , and approaches zero when We have , in fact , if is the integral of , , and , therefore , in case ( i ) , the differential coefficient of is mded , so that is an integral , and , therefore , is an , which , by the theory of indeterminate forms , has the same limit as , that is zero .
In case ( ii ) , if we multiply by , we still get an integral , which accordingly approaches zero , when the required result follows .
In case ( iii ) , we remark that , by the theory of indeterminate forms , has its limits among those of so that ( iii ) becomes a special case of ( i ) .
S9 .
We the conditions alluded to in the preyious article , which Convergence oj Restricted Fourier Series .
289 ensure simultaneously the holding of the conditions ( ii ) and ( iii ) of Theorem .
Taken in conjunction with that theorem , they give us the main results of the paper , stated in the Introduction .
THEOREM A.\mdash ; The conditions ( ii ) and ( iii ) of Theorem satisfied , for a particular value of if , for that value of have , as ( a ) ( b ) ; and if further for , a tion of variation ; or , more irailly , addition , the fotlowiny:\mdash ; ; an inteyral ftaining u Condition ( a ) secures that the odd function has zero as limit , when , and ( c ) secures that times the differential coefficient of the same odd function is a function of mded variation .
Hence we may apply Theorem 4 of my paper cited above , and deduce the convergence to zero of the coefficients of the derived series of the extended odd function , which is elsewhere an integral , and in the given small interval agrees with the above odd function .
Similarly conditions ( b ) and ( c ) secure that ?
times the differential coefficient of the even function is a function of bounded variation , which converges to zero , as .
Thus by Theorem 3 of the same paper , the coefficients of the derived series of the extended even function converge to zero .
Hence the coefficients both of the cosine and the sine terms of the derived series of the series of a function which is equal to in the given small interval containing , and is elsewhere an integral , to zero .
Hence condition ( iii ) of Theorem is satisfied .
With regard to condition ( ii ) , this is , by Lemma 3 , satisfied , as by condition ( c ) of the present theorem is bounded .
This proves the theorem when , ( b ) , and ( c ) are satisfied .
To see that and may take the place of ( b ) and , we remark first that , by Lemma 3 , approaches zero , as , and that , therefore , from the equation , Prof. W. H. Young .
On the Ordinary we may deduce that and , therefore , that the left-hand side of the last equation , being the difference of two integrals , is an ordinary Lebesgue integral .
Moreover we see that is itself an ordinary Lebesgue integral .
The rest of the argument is precisely analogous that employed in ealing with the set of conditions , ( b ) , and ( c ) except that , instead of using Theorems 3 and 4 of the paper cited as basis , we now use the more general Theorems 5 and 6 of that same paper .
5 .
10 .
The following theorem is based on Theorems 7 and 8 of the paper previously cited , and in its most general form , on a slight , eneralisation of the result of my ' Comptes B.endus ' paper , there utilised .
THEOREM B.\mdash ; Retaining conditions of Theorem , we may replace ( b ) and ( c ) by th ollowing : is bounded in some interv surrounding , and has limit as denoting a derivate of In fact , by Lemma 3 , condition ( ii ) of Theorem is satisfied .
More generally we repface ( b ) and ( c ) by thoe : in some 'int , and has a uniqulimit , as In fact , by Lemma 3 , condition ( ii ) of Theorem is satisfied .
Also considering , if we write we see that is bounded , say numerically less than B. Further where and accordingly bounded .
We can accordingly employ Theorems 7 and 8 of the paper already cited , and the required result follows .
Convergence of Restricted Fourier Series .
To prove the remaining part of the theorem , we have merely to use the condition that should be bounded , instead the slightly less general condition that in the test for the convergence of the Fourier sel.ies of a function and of its allied series , quoted in the paper cited from the ' Comptes Rendus.'* S11 .
In the Theorems we have had in view convergence at a point only .
That the convergence is uniform , or , more generally , that the quantities have limits which are bounded functions of , is immediate .
Thus we have the following theorem:\mdash ; THEOREM pu Su of Theorem I converges uniformly to zero dosed interval to a completely open interval , in which the of hold .
We have , in fact , only to examine the proofs of Theorems II , III , and to see firstly that , for such a closed interval , there is a finite upper and a finite lower bound to each of the factors which involve , e.cJ , or in Theorem III ; and , secondly , that wherever the Theorem of Biemann-Lbesgue is used , the conyergence is , by a known property .
of the integral in question , uniform convergence to zero for values of in our closed interval .
Finally , where the argument introduces a small quantity which , when chosen conveniently small , ensures certain auxiliary quantities being as small as we please , we see that these auxiliary quantities may be made less than an assigned small quantity independent of , since , point of a closed interval inside a completely open interval in which the conditions hold , may be chosen independent of , and , by the point first referred to , the auxiliary quantity in question may be made to depend only on .
Thus the theorem is proved .
S12 .
Referring then to Cor. 2 of Theorem I , we at once obtain the following results in the theory of term-by-term integration of restricted Fourier series when multiplied by another function:\mdash ; THEOREM C.\mdash ; If function which in certain sub-interval of the The proof of the extended form of test is almost exactly the same as that given in the ' Comptes Rendus .
' It is given in extenso in a paper on " " The Convergence of the Derived Series of a Fourier Series cited above .
292 On the Convergence of Restricted Fourier Series .
of periodicity has its squar.e summable , and is another whose square ?
, then , if we ultiply the restricted Fourier series of , integrate term term over the sub-intervat , and result be the of th ( product the same interv , provided that\mdash ; ( i ) estricted / of series is to a finite number of points at colbich the primitive of which is a derivate is not an integral ; ( ii ) at such points , is an ( iii ) if any function which , eircept in certain sub-interval of the interval su a periodic intgal , and in that exceptional equal to , th the coefficients of the derived series of the series of to zero for , 2 , THEOREM D.\mdash ; If a frmction which , in a certain sub-interval of the of has bounded , and summable , then , if multiply th restricted Fourier series of by , and integrate term term over the sub-interval , the result be the integral of the product over the same provid , only the same conditions hold as in Theorem E.\mdash ; If is any function of variation , then , ltiply the restricte of function by , we may integrate term overthe -int to which it is , and the result will be tegral of the over the same -interval , conditions hold as in Theorem In these theorems , as elsewhere , the integration is supposed to be over a closed !
nterval .
By the term therefore integration over the sub-interval , \ldquo ; which is itself necessarily open , we mean that we may rate between any two points of that open interyal .
We may , of course , substitute for the conditions ( i ) , ( ii ) , ( iii ) any of those sets of conditions stated in Theorems A and as sufficient to ensure heir fulfilment .
It should be noticed that in Theorem we use the fact that the ence of the restricted Fourier series in an interval to which it is restricted is necessarily bounded , being of the same character as that for a proper Fourier series , in virtue of Cor. 2 to Theorem I. * In the present communication , as in the paper which preceded it , I have confined my attention to first derived series of Fourier series .
It is open to us also to consider the circumstances under which the coefficients of the second or higher derived series to zero , while the series themselves converge at isolated points , or throughout intervaIs .
No new principles are introduced into the consideration of such higher derived series ; I have therefore , thought it undesirable to extend the length of my communications by dealing with them .

