Introduction
This is yet another post based on a paper of Ramanujan titled "On certain arithmetical functions" which appeared in Transactions of the Cambridge Philosophical Society in 1916. In this paper Ramanujan provided a lot of identities concerning Lambert series and thereby deduced many relations between various divisor functions. Apart from the amazing results proved in this paper, what I liked most is the very elementary approach followed by Ramanujan compared to the methods of modern authors who are seduced by the modular form.Ramanujan's Functions P,Q,R
Ramanujan introduced the following Lambert series and used them extensively in deriving many identities in elliptic function theory: P(q)=1−24(q21−q2+2q41−q4+3q61−q6+⋯)=1−24∞∑n=1nq2n1−q2nQ(q)=1+240(q21−q2+23q41−q4+33q61−q6+⋯)=1+240∞∑n=1n3q2n1−q2nR(q)=1−504(q21−q2+25q41−q4+35q61−q6+⋯)=1−504∞∑n=1n5q2n1−q2n We have already met P(q) in a previous post regarding series for 1/π. Also the R(q) is not to be confused with the Rogers-Ramanujan continued fractions introduced in the last post. Ramanujan also used the alternative notation L,M,N instead of P,Q,R. Again to simplify matters regarding manipulation of above series Ramanujan used the variable x=q2 and hence we get the following notation which will be used subsequently in this post: P(x)=1−24(x1−x+2x21−x2+3x31−x3+⋯)=1−24∞∑n=1nxn1−xnQ(x)=1+240(x1−x+23x21−x2+33x31−x3+⋯)=1+240∞∑n=1n3xn1−xnR(x)=1−504(x1−x+25x21−x2+35x31−x3+⋯)=1−504∞∑n=1n5xn1−xn The Lambert series above can be easily expressed as generating functions for divisor function. In general for any positive integer r, we can see that ∞∑n=1nrxn1−xn=∞∑n=1σr(n)xn where σr(n) denotes sum of rth powers of divisors of n.Ramanujan considered these sums and its generalization below: Sr(x)=Er+1rx1−x+2rx21−x2+3rx31−x3+⋯=Er+∞∑n=1nrxn1−xn=Er+∞∑n=1σr(n)xnΦr,s(x)=∞∑m=1∞∑n=1mrnsxmn where Er is a suitably chosen constant related with Bernoulli's numbers as we shall see later. It is easy to see that Φr,s(x)=Φs,r(x)=∞∑n=1nsσr−s(n)xnSr(x)=Er+Φ0,r(x) Ramanujan was able to express Φr,s(x), with r+s an odd positive integer, in terms of P,Q,R in a very elementary manner using trigonometrical series. This is one of the truly amazing proofs which Ramanujan provided. In a way the proof shows that a lot more can be achieved with elementary stuff than people think. To understand the proof we need to develop the series for cot(x) and there we will see the use of Bernoulli's numbers.
Expansion of cotx
We know that the Bernoulli's numbers Bn are defined by xex−1=∞∑n=0Bnxnn! Now we can see that xex−1=11+x2!+x23!+⋯=1−x2+⋯ and it is easy to verify that {x/(ex−1)}+(x/2) is an even function therefore it follows that: B0=1,B1=−12,B2=16,B4=−130,B3=B5=⋯=B2n+1=0 Next we proceed to find expansion of cotx in powers of x. We have cotx=cosxsinx=i⋅eix+e−ixeix−e−ix=i⋅e2ix+1e2ix−1=i⋅(1+2e2ix−1)=i+1x⋅2ixe2ix−1=i+1x(1−2ix2+B2⋅(2ix)22!+B4⋅(2ix)44!+⋯)=1x−B222x2!+B424x34!−B626x56!+⋯ Differentiating with respect to x we get −cosec2x=−1x2−B2222!+3B424x24!−5B626x46!+⋯⇒1+cot2x=1x2+13−3B424x24!+⋯⇒cot2x=1x2−23−3B424x24!+5B626x46!−7B828x68!+⋯A Trigonometrical Identity
Ramanujan next uses the formula for sum of cosines of angles in arithmetic progression in the following manner 2{cosθ+cos2θ+⋯+cos(n−1)θ}=2cosnθ2sin(n−1)θ2sinθ2=sin(n−12)θ−sinθ2sinθ2=sinnθcosθ2−cosnθsinθ2sinθ2−1=cotθ2sinnθ−cosnθ−1 to get cotθ2sinnθ=1+2cosθ+2cos2θ+⋯+2cos(n−1)θ+cosnθ Out of the blue Ramanujan now sets out to consider the expression S=(14cotθ2+xsinθ1−x+x2sin2θ1−x2+x3sin3θ1−x3+⋯)2 Let um=xm/(1−xm) and then the above expression can be written as S=(14cotθ2+u1sinθ+u2sin2θ+u3sin3θ+⋯)2=(14cotθ2+∞∑m=1umsinmθ)2=(14cotθ2)2+12cotθ2∞∑m=1umsinmθ+(∞∑m=1umsinmθ)2=(14cotθ2)2+T1+T2 where using (14) we can express T1 in terms of cosines as: T1=∞∑m=1um{12+cosθ+cos2θ+⋯+cos(m−1)θ+12cosmθ} and T2 can also be expressed in terms of cosines as follows: T2=∞∑m=1umsinmθ∞∑n=1unsinnθ=12∞∑m=1∞∑n=1{cos(m−n)θ−cos(m+n)θ}umun and thus T1+T2 can be arranged in a series of the form T1+T2=∞∑k=0Ckcoskθ Now it is clear that the contribution to C0 from T1 is (1/2)∑um and from T2 the contribution is (1/2)∑u2m and therefore C0=12∞∑m=1um+12∞∑m=1u2m=12∞∑m=1um(1+um)=12∞∑m=1xm(1−xm)2=12∞∑m=1∞∑n=1nxmn=12∞∑n=1nxn1−xn=12∞∑n=1nun For k>0, the part of Ck coming from T1 is 12uk+∞∑m=k+1um=12uk+∞∑l=1uk+l and the part of Ck coming from T2 is 12∑m−n=kumun+12∑n−m=kumun−12∑m+n=kumun=∞∑l=1uluk+l−12k−1∑l=1uluk−l Thus it follows that for k>0 Ck=12uk+∞∑l=1uk+l+∞∑l=1uluk+l−12k−1∑l=1uluk−l=12uk+∞∑l=1uk+l(1+ul)−12k−1∑l=1uluk−l=12uk+∞∑l=1uk(ul−uk+l)−12k−1∑l=1uk(1+ul+uk−1)=uk{12+∞∑l=1(ul−uk+l)−12k−1∑l=1(1+ul+uk−l)}=uk{12+u1+u2+⋯+uk−k−12−(u1+u2+⋯+uk−1)}=uk{1+uk−k2} The crucial part in the above proof are the easily verifiable identities: uk+l(1+ul)=uk(ul−uk+l),uluk−l=uk(1+ul+uk−l) Finally putting all the pieces together we can see that S=(14cotθ2+xsinθ1−x+x2sin2θ1−x2+x3sin3θ1−x3+⋯)2=(14cotθ2)2+12∞∑n=1nun+∞∑k=1uk(1+uk−k2)coskθ=(14cotθ2)2+∞∑m=1um(1+um)cosmθ+12∞∑m=1mum(1−cosmθ)=(14cotθ2)2+∞∑m=1xmcosmθ(1−xm)2+12∞∑m=1mxm1−xm(1−cosmθ) Ramanujan proved the above without using the symbols ∑ and um and expressed his formula directly as: (14cotθ2+xsinθ1−x+x2sin2θ1−x2+x3sin3θ1−x3+⋯)2=(14cotθ2)2+xcosθ(1−x)2+x2cos2θ(1−x2)2+x3cos3θ(1−x3)2+⋯+12{x(1−cosθ)1−x+2x2(1−cos2θ)1−x2+3x3(1−cos3θ)1−x3+⋯} The presentation we have offered above is from G. H. Hardy's Ramanujan: Twelve Lectures on Subjects Suggested by his Life and Work. The proof above may look complicated because of symbolism, but in reality it involves basic algebraic manipulations.In a similar manner Ramanujan uses the trigonometric identity: cot2θ2(1−cosnθ)=(2n−1)+4(n−1)cosθ+4(n−2)cos2θ+⋯+4cos(n−1)θ+cosnθ and establishes the following result: {18cot2θ2+112+x(1−cosθ)1−x+2x2(1−cos2θ)1−x2+3x3(1−cos3θ)1−x3+⋯}2=(18cot2θ2+112)2+112{13x(5+cosθ)1−x+23x2(5+cos2θ)1−x2+33x3(5+cos3θ)1−x3+⋯} The algebraic manipulations in this case are of similar nature but bit more complicated and hence will not be presented here. From (14) it is clear that 18cot2θ2+112=12(1θ2−3B44!θ2+5B66!θ4−7B88!θ6+⋯) and mxm1−xm(1−cosmθ)=mxm1−xm(m2θ22!−m4θ44!+⋯) Hence the LHS of (18) can be written as {12θ2+θ22!(−B42⋅4+∞∑m=1m3xm1−xm)−θ44!(−B62⋅6+∞∑m=1m5xm1−xm)+⋯}2 It is now time to relate Er of (7) with Bernoulli's numbers as follows: Er=−Br+12(r+1) and then using (7) we can write the LHS of (18) as: {12θ2+θ22!S3(x)−θ44!S5(x)+⋯}2 To compute the RHS of (18) we need to first square the equation (19). Instead of squaring the series expansion on right of (19) I prefer to use derivatives of the expansion given in (14). Thus on differentiating (14) we get 2cotx(−1−cot2x)=−2x3−2⋅3B424x4!+4⋅5B626x36!−⋯⇒cotx+cot3x=1x3+2⋅3B423x4!−4⋅5B625x36!+⋯ and using (13) we get cot3x=1x3−1x+224!(2⋅2⋅3B4+3⋅4B2)x−246!(2⋅4⋅5B6+5⋅6B4)x3+⋯ Differentiating the above equation we get 3cot2x(−1−cot2x)=−3x4+1x2+224!(2⋅2⋅3B4+3⋅4B2)−246!⋅3(2⋅4⋅5B6+5⋅6B4)x2+⋯⇒3cot2x+3cot4x=3x4−1x2−224!(2⋅2⋅3B4+3⋅4B2)+246!⋅3(2⋅4⋅5B6+5⋅6B4)x2−⋯ Adding this equation and (14) (and values of B2,B4) we get 3cot4x+4cot2x=3x4−1415+25B66⋅x22!−27B88⋅x44!+29B1010⋅x66!−⋯ Again going back to the equation (19) we see that square of its LHS is given by (18cot2θ2+112)2=13⋅26(3cot4θ2+4cot2θ2)+1144=14θ4+1480+124(B66⋅θ22!−B88⋅θ44!+⋯)=14θ4−B416+124(B66⋅θ22!−B88⋅θ44!+⋯)=14θ4+E32−112(E5⋅θ22!−E7⋅θ44!+⋯) We can now clearly see that the RHS of (18) is given by 14θ4+S3(x)2−112(θ22!S5(x)−θ44!S7(x)+⋯) and finally the equation (18) is transformed into {12θ2+θ22!S3(x)−θ44!S5(x)+⋯}2=14θ4+S3(x)2−112(θ22!S5(x)−θ44!S7(x)+⋯) For even integer n>2 we equate the coefficients of θn on both sides to obtain the following euqation \begin{align}\frac{(n - 2)(n + 5)}{12(n + 1)(n + 2)}S_{n + 3}(x) &= \binom{n}{2} S_{3}(x)S_{n - 1}(x)\notag\\ &\,\,\,\,\,\,\,\,+ \binom{n}{4} S_{5}(x)S_{n - 3}(x) + \cdots\notag\\ &\,\,\,\,\,\,\,\,+ \binom{n}{n - 2}S_{n - 1}(x)S_{3}(x)\tag{22}\end{align} where \binom{n}{r} = \frac{n!}{r!(n - r)!} is the usual binomial coefficient.
Now it is easy to see that we have P(x) = -24S_{1}(x), Q(x) = 240S_{3}(x), R(x) = -504S_{5}(x) and hence using the relation (22) we can evaluate S_{2n - 1}(x) in terms of P, Q, R for all integers n > 0. For small values of n it is easy to apply the formula and derive the following: \begin{align} 1 + 480\Phi_{0, 7}(x) &= Q^{2}\notag\\ 1 - 264\Phi_{0, 9}(x) &= QR\notag\\ 691 + 65520\Phi_{0, 11}(x) &= 441Q^{3} + 250R^{2}\notag\\ 1 - 24\Phi_{0, 13}(x) &= Q^{2}R\notag\end{align} Ramanujan does not stop here and actually uses the relation (22) to evaluate S_{7}(x), S_{9}(x), \ldots, S_{31}(x) and gives his results in terms of the function \Phi_{r, s}(x). Ramanujan did all this for his love of numbers and we show one example here which sheds light on the nature of numbers he dealt with: \begin{align} 7709321041217+32640\Phi_{0, 31}(x) &= 764412173217Q^{8}(x)\notag\\ &\,\,\,\,\,\,\,\,+ 5323905468000Q^{5}(x)R^{2}(x)\notag\\ &\,\,\,\,\,\,\,\,+ 1621003400000Q^{2}(x)R^{4}(x)\notag \end{align} In the next post we will analyze the equation (16) and the results derived from it.
Postscript: L. C. Shen provided another proof of (16) in 1993 using derivatives of theta functions which we reproduce below:
We have from these posts \begin{align} \theta_{1}(z, q) &= 2q^{1/4}\sin z \prod_{n = 1}^{\infty}(1 - q^{2n})(1 - 2q^{2n}\cos 2z + q^{4n})\notag\\ &= 2q^{1/4}\sum_{n = 0}^{\infty}(-1)^{n}q^{n(n + 1)}\sin(2n + 1)z\tag{23}\end{align} Differentiating twice the infinite series representation of \theta_{1}(z, q) with respect to z we get \frac{\partial^{2}\theta_{1}}{\partial z^{2}} = -2q^{1/4}\sum_{n = 0}^{\infty}(-1)^{n}q^{n(n + 1)}(2n + 1)^{2}\sin(2n + 1)z Again differentiating \theta_{1}(z, q) with respect to q we get \begin{align}\frac{\partial \theta_{1}}{\partial q} &= 2\sum_{n = 0}^{\infty}(-1)^{n}(n + 1/2)^{2}q^{(n + 1/2)^{2} - 1}\sin(2n + 1)z\notag\\ &= \frac{1}{2}q^{-3/4}\sum_{n = 0}^{\infty}(-1)^{n}q^{n(n + 1)}(2n + 1)^{2}\sin(2n + 1)z\notag\end{align} and thus we arrive at the partial differential equation satisfied by \theta_{1} \frac{\partial^{2}\theta_{1}}{\partial z^{2}} = -4q\frac{\partial \theta_{1}}{\partial q}\tag{24} Next by logarithmic differentiation of the product expansion of \theta_{1}(z, q) with respect to z we get \begin{align}\frac{1}{\theta_{1}}\frac{\partial \theta_{1}}{\partial z} &= \cot z + 4\sum_{n = 1}^{\infty}\frac{q^{2n}\sin 2z}{1 - 2q^{2n}\cos 2z + q^{4n}}\notag\\ &= \cot z + \frac{4}{2i}\sum_{n = 1}^{\infty}\frac{q^{2n}(e^{2iz} - e^{-2iz})}{(1 - q^{2n}e^{2iz})(1 - q^{2n}e^{-2iz})}\notag\\ &= \cot z + \frac{4}{2i}\sum_{n = 1}^{\infty}q^{2n}\left(\frac{e^{2iz}}{1 - q^{2n}e^{2iz}} - \frac{e^{-2iz}}{1 - q^{2n}e^{-2iz}}\right)\notag\\ &= \cot z + \frac{4}{2i}\left(\sum_{n = 1}^{\infty}\frac{q^{2n}e^{2iz}}{1 - q^{2n}e^{2iz}} - \sum_{n = 1}^{\infty}\frac{q^{2n}e^{-2iz}}{1 - q^{2n}e^{-2iz}}\right)\notag\\ &= \cot z + \frac{4}{2i}\left(\sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}\{q^{2n}e^{2iz}\}^{m} - \sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}\{q^{2n}e^{-2iz}\}^{m}\right)\notag\\ &= \cot z + \frac{4}{2i}\left(\sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}q^{2m}e^{2imz}q^{2m(n - 1)}\right. \notag\\ &\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \left. - \sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}q^{2m}e^{-2imz}q^{2m(n - 1)}\right)\notag\\ &= \cot z + \frac{4}{2i}\left(\sum_{m = 1}^{\infty}q^{2m}e^{2imz}\sum_{n = 1}^{\infty}q^{2m(n - 1)}\right.\notag\\ &\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, \left. - \sum_{m = 1}^{\infty}q^{2m}e^{-2imz}\sum_{n = 1}^{\infty}q^{2m(n - 1)}\right)\notag\\ &= \cot z + \frac{4}{2i}\left(\sum_{m = 1}^{\infty}\frac{q^{2m}e^{2imz}}{1 - q^{2m}} - \sum_{m = 1}^{\infty}\frac{q^{2m}e^{-2imz}}{1 - q^{2m}}\right)\notag\\ &= \cot z + 4\sum_{n = 1}^{\infty}\frac{q^{2n}}{1 - q^{2n}}\sin 2nz\tag{25}\end{align} Noting that 1 - 2q^{2n}\cos 2z + q^{4n} = (1 - q^{2n}e^{2iz})(1 - q^{2n}e^{-2iz}) and performing logarithmic differentiation with respect to q we get \begin{align}\frac{1}{\theta_{1}}\frac{\partial \theta_{1}}{\partial q} &= \frac{1}{4q} - \frac{2}{q}\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}} - \frac{2}{q}\left(\sum_{n = 1}^{\infty}\frac{nq^{2n}e^{2iz}}{1 - q^{2n}e^{2iz}} + \sum_{n = 1}^{\infty}\frac{nq^{2n}e^{-2iz}}{1 - q^{2n}e^{-2iz}}\right)\notag\\ &= \frac{1}{4q} - \frac{2}{q}\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}} - \frac{2}{q}\left(\sum_{n = 1}^{\infty}nq^{2n}e^{2iz}\sum_{m = 1}^{\infty}\{q^{2n}e^{2iz}\}^{m - 1}\right.\notag\\ &\left.\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\,\, + \sum_{n = 1}^{\infty}nq^{2n}e^{-2iz}\sum_{m = 1}^{\infty}\{q^{2n}e^{-2iz}\}^{m - 1}\right)\notag\\ &= \frac{1}{4q} - \frac{2}{q}\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}} - \frac{2}{q}\left(\sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}nq^{2mn}e^{2imz} + \sum_{n = 1}^{\infty}\sum_{m = 1}^{\infty}nq^{2mn}e^{-2imz}\right)\notag\\ &= \frac{1}{4q} - \frac{2}{q}\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}} - \frac{2}{q}\left(\sum_{m = 1}^{\infty}e^{2imz}\sum_{n = 1}^{\infty}nq^{2mn} + \sum_{m = 1}^{\infty}e^{-2imz}\sum_{n = 1}^{\infty}nq^{2mn}\right)\notag\\ &= \frac{1}{4q} - \frac{2}{q}\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}} - \frac{2}{q}\left(\sum_{m = 1}^{\infty}e^{2imz}\frac{q^{2m}}{(1 - q^{2m})^{2}} + \sum_{m = 1}^{\infty}e^{-2imz}\frac{q^{2m}}{(1 - q^{2m})^{2}}\right)\notag\\ &= \frac{1}{4q} - \frac{2}{q}\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}} - \frac{4}{q}\sum_{n = 1}^{\infty}\frac{q^{2n}}{(1 - q^{2n})^{2}}\cos 2nz\notag\end{align} Using differential equation (24) we get \frac{1}{\theta_{1}}\frac{\partial^{2} \theta_{1}}{\partial z^{2}} = -1 + 16\sum_{n = 1}^{\infty}\frac{q^{2n}}{(1 - q^{2n})^{2}}\cos 2nz + 8\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}}\tag{26} Now differentiating equation (25) with respect to z we get \begin{align}\frac{\partial }{\partial z}\left(\frac{1}{\theta_{1}}\frac{\partial \theta_{1}}{\partial z}\right) &= -1 - \cot^{2}z + 8\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}}\cos 2nz\notag\\ \Rightarrow \frac{1}{\theta_{1}}\frac{\partial^{2} \theta_{1}}{\partial z^{2}} - \left(\frac{1}{\theta_{1}}\frac{\partial \theta_{1}}{\partial z}\right)^{2} &= -1 - \cot^{2}z + 8\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}}\cos 2nz\notag\\ \Rightarrow \left(\frac{1}{\theta_{1}}\frac{\partial \theta_{1}}{\partial z}\right)^{2} &= 1 + \cot^{2}z - 8\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}}\cos 2nz + \frac{1}{\theta_{1}}\frac{\partial^{2} \theta_{1}}{\partial z^{2}}\notag\end{align} Now using (25) and (26) we get \begin{align}&\left(\cot z + 4\sum_{n = 1}^{\infty}\frac{q^{2n}\sin 2nz}{1 - q^{2n}}\right)^{2}\notag\\ &\,\,\,\,\,\,\,\,= \cot^{2}z + 16\sum_{n = 1}^{\infty}\frac{q^{2n}\cos 2nz}{(1 - q^{2n})^{2}} + 8\sum_{n = 1}^{\infty}\frac{nq^{2n}}{1 - q^{2n}}(1 - \cos 2nz)\notag\end{align} If we replace z by \theta/2 and q^{2} by x and divide resulting equation by 16 we get the identity (16) obtained by Ramanujan using algebraic manipulation.
Print/PDF Version
0 comments :: Certain Lambert Series Identities and their Proof via Trigonometry: Part 1
Post a Comment