UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Hypergeometric functions and Mahler measure Rogers, Mathew D. 2008

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2008_fall_rogers_mathew.pdf [ 808.84kB ]
Metadata
JSON: 24-1.0066504.json
JSON-LD: 24-1.0066504-ld.json
RDF/XML (Pretty): 24-1.0066504-rdf.xml
RDF/JSON: 24-1.0066504-rdf.json
Turtle: 24-1.0066504-turtle.txt
N-Triples: 24-1.0066504-rdf-ntriples.txt
Original Record: 24-1.0066504-source.json
Full Text
24-1.0066504-fulltext.txt
Citation
24-1.0066504.ris

Full Text

Hypergeometric functions and MahlermeasurebyMathew D. RogersB.Sc., The University of California at Irvine, 2002M.Sc., The University of British Columbia, 2004A THESIS SUBMITTED IN PARTIAL FULFILMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDoctor of PhilosophyinThe Faculty of Graduate Studies(Mathematics)The University Of British ColumbiaAugust, 2008c© Mathew D. Rogers 2008AbstractThe (logarithmic) Mahler measure of an n-variable Laurent polynomial,P (x1, . . . , xn), is defined bym (P ) =∫ 10. . .∫ 10log∣∣P (e2πit1 , . . . , e2πitn)∣∣ dt1 . . .dtn.Using experimental methods, David Boyd conjectured a large number ofexplicit relations between Mahler measures of polynomials and special val-ues of different types of L-series. This thesis contains four papers whicheither prove or attempt to prove conjectures due to Boyd. The introductorychapter contains an overview of the contents of each manuscript.iiTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . viDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiStatement of Co-Authorship . . . . . . . . . . . . . . . . . . . . . viii1 Introductory Chapter . . . . . . . . . . . . . . . . . . . . . . . 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Mahler measure background . . . . . . . . . . . . . . . . . . 21.3 An overview of the manuscripts in this thesis . . . . . . . . . 3Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Lattice sums and Mahler measures . . . . . . . . . . . . . . . 102.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.1.1 Boyd’s conjectures and four-dimensional lattice sums 122.1.2 Summary of hypergeometric formulas for F (b, c) . . . 152.2 Reductions of F (1, 1), F (1, 2), F (1, 4), and F (2, 2) to inte-grals of hypergeometric functions . . . . . . . . . . . . . . . 182.2.1 More explicit examples . . . . . . . . . . . . . . . . . 322.2.2 Remarks on F (1, 3) and higher values of F (b, c) . . . 342.3 Connections with the elliptic dilogarithm . . . . . . . . . . . 362.4 Higher lattice sums and conclusion . . . . . . . . . . . . . . . 372.5 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 39Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40iiiTable of Contents3 Functional equations for Mahler measures . . . . . . . . . . 423.1 History and introduction . . . . . . . . . . . . . . . . . . . . 423.2 Mahler measures and q-series . . . . . . . . . . . . . . . . . . 463.2.1 Functional equations from modular equations . . . . 483.2.2 Identities arising from higher modular equations . . . 523.2.3 Computationally useful formulas, and a few relatedhypergeometric transformations . . . . . . . . . . . . 543.3 A regulator explanation . . . . . . . . . . . . . . . . . . . . . 603.3.1 The elliptic regulator . . . . . . . . . . . . . . . . . . 603.3.2 Regulators and Mahler measure . . . . . . . . . . . . 643.3.3 Functional identities for the regulator . . . . . . . . . 653.3.4 The first family . . . . . . . . . . . . . . . . . . . . . 673.3.5 A direct approach . . . . . . . . . . . . . . . . . . . . 703.3.6 Relations among m(2), m(8), m(3√2), and m(i√2)713.3.7 The Hesse family . . . . . . . . . . . . . . . . . . . . 733.3.8 The Γ00(6) example . . . . . . . . . . . . . . . . . . . 733.3.9 The Γ00(5) example . . . . . . . . . . . . . . . . . . . 743.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.5 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 75Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 764 New 5F4 transformations and Mahler measures . . . . . . . 794.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 794.2 Identities between Mahler measures and transformations forthe 5F4 function . . . . . . . . . . . . . . . . . . . . . . . . . 814.3 New formulas for 1/π . . . . . . . . . . . . . . . . . . . . . . 894.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 924.5 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 92Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 935 Trigonometric integrals and Mahler measures . . . . . . . . 955.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 955.2 Preliminaries: A description of the method, and some twodimensional Mahler measures . . . . . . . . . . . . . . . . . . 975.3 Relations between TS(v, 1) and Mahler’s measure, and a re-duction of TS(v, w) to multiple polylogarithms . . . . . . . . 1045.4 An evaluation of TS(v, 1) using infinite series . . . . . . . . . 111ivTable of Contents5.5 Relations between S(v, 1) and Mahler’s measure, and a closedform for S(v, w). . . . . . . . . . . . . . . . . . . . . . . . . . 1195.6 q-series for the dilogarithm, and some associated trigonomet-ric integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1235.7 A closed form for T(v, w), and Mahler measures for T(v, 1v)1345.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1405.9 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . 140Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1416 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1436.1 Computational proofs? . . . . . . . . . . . . . . . . . . . . . 143Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145vAcknowledgementsFirst and foremost, I would like to thank my advisor David Boyd. Overthe last six years I have benefitted tremendously from his knowledge andguidance, as well as from his patience and optimism. It has been a trueprivilege to work with him.I would like to thank Wadim Zudilin, Bruce Berndt, Marie-Jose Bertin,Larry Glasser, and Matilde Laln for their kind words, useful correspondence,assistance, and encouragement.I also thank Mike Bennett, Ken Ono, Greg Martin, Jonathan Sondow,and Fernando Rodriguez-Villegas for their encouragement, advice, and as-sistance.Finally, I would like to thank my family for their unfailing support.viDedicationI would like to dedicate this thesis to my mother for her love and support.viiStatement of Co-AuthorshipChapter 3 of this thesis was written jointly with Matilde Lal´ın. Section 3.1was prepared jointly, while the research and preparation of Section 3.2 is dueto myself, and the research and preparation of Section 3.3 is due to Lal´ın.The project was initiated by my rediscovery of equation (3.1.6), and by mylater proof of equation (3.1.7).viiiChapter 1Introductory Chapter1.1 IntroductionConstants such as π and ζ(3) occupy a special place in the history of num-ber theory. In the past, famous mathematicians including Newton, Euler,Cauchy and Ramanujan expended a great deal of energy searching for effi-cient ways to calculate these numbers. They were motivated by more thansimple curiosity, as they lacked the calculators that we now take for granted.Research in this direction helped spur the development of a variety of impor-tant mathematical tools, including calculus, complex analysis, the theory ofelliptic functions, and the theory of hypergeometric functions. This thesisfalls under the final category, and can loosely be described as an effort toobtain new hypergeometric formulas for special values of L-functions.The generalized hypergeometric function is usually defined in terms ofa power series involving Pochhammer symbols (which are also known asgeneralized factorials). The Pochhammer symbol is defined by(x)n :={1 if n = 0,x(x + 1) . . . (x + n− 1) if n ≥ 1,and the generalized hypergeometric function can be written aspFq(a1,...,apb1,...,bq ;x):=∞∑n=0(a1)n . . . (ap)n(b1)n . . . (bq)nxnn!.By specializing the ai’s and bi’s we can reduce many elementary and specialfunctions to hypergeometric functions. Bessel functions of the second kindare probably the most important non-hypergeometric special functions.It is a classical fact that hypergeometric functions satisfy interestingtransformations: Dixon’s, Whipple’s and Saalschutz’s theorems are all ex-amples [2]. One consequence of this fact is that new series expansions arecontinually being discovered for familiar special constants. While NicolasMercator discovered the first infinite series for log(2) in 1668, we recently1Chapter 1. Introductory Chapterfound a new identity via numerical searches:6F5(32, 32, 32,1,1, 321552,2,2,2, 1715;116)?= −211217+307217log(2). (1.1.1)In equation (1.1.1), and throughout the rest of this thesis, we will use “ ?=” todenote an unproven equality that holds to at least 60 decimal places. Equa-tion (1.1.1) exemplifies many hypergeometric formulas, in that the simplicityof the identity obscures the reason why it exists at all. The Borwein brotherstackled a similar problem when they proved Ramanujan’s seventeen formu-las for 1/π (see [22], [9], and [15]). While the theory of elliptic functionsultimately underpins Ramanujan’s claims, equation (1.1.1) is equivalent toan unproven conjecture about Mahler measures of elliptic curves.This thesis stems from the observation that Mahler identities are inti-mately tied to formulas for generalized hypergeometric functions. The (loga-rithmic) Mahler measure of an n-variable Laurent polynomial, P (x1, . . . , xn),is defined bym(P ) :=∫ 10. . .∫ 10log∣∣∣P (e2πiθ1 , . . . , e2πiθn)∣∣∣ dθ1 . . .dθn.Boyd conjectured a large number of interesting relations between values ofL-series of elliptic curves and Mahler measures of polynomials in [10]. Forexample, he showed that the following conjecture (which Denninger alsoconsidered [17]):m(1 + x +1x+ y +1y)?=154π2L (E, 2) , (1.1.2)with E having conductor 15, holds to at least 60 decimal places. Since themajority of Boyd’s conjectures translate into hypergeometric identities, thisgives rise to an entirely new class of closed form evaluations of the general-ized hypergeometric function. The L-series of an elliptic curve is a funda-mental arithmetic quantity, hence identities like equation (1.1.2) should beregarded with the same interest as most formulas for π or ζ(3).1.2 Mahler measure backgroundThe field of Mahler measure encompasses the study of both single-variable,and multi-variable polynomials. The first area is closely related to the study2Chapter 1. Introductory Chapterof heights of polynomials. Jensen’s formula provides the operative result:m (P (x)) =∑|α|>1P (α)=0log |α|,by relating Mahler measures of monic polynomials to their zeros. Single-variable Mahler measures often appear in studies of Salem and Pisot num-bers, and are especially relevant for Lehmer’s problem:Lehmer’s Problem: Does there exist a monic, non-cyclotomic polynomialP (x) ∈ Z[x], with P (0) = 0, such that m(P (x)) < m(P0(x)) = log (1.17 . . . ),where P0(x) = 1− x + x3 − x4 + x5 − x6 + x7 − x9 + x10 ?In general, the Boyd-Lawton theorem shows that multi-variable Mahlermeasures arise as limit points of sequences of single-variable Mahler measures[11]. For example, in the two-dimensional case we havelimn→∞m(P (x, xn)) = m (P (x, y)) . (1.2.3)Unfortunately, it is unreasonable to expect a single result like Jensen’s for-mula to describe all of the interesting identities for multi-variable Mahlermeasures. Formulas exist which involve special values of polylogarithms,elliptic dilogarithms, Riemann and Dedekind zeta functions, Dirichlet andelliptic curve L-functions, and even L-series of K3 hypersurfaces. In general,it is usually far easier to prove identities between Mahler measures, than toactually relate them to L-functions [23].1.3 An overview of the manuscripts in this thesisDespite the fact that no one has proved (1.1.2), several special cases ofBoyd’s conjectures have been rigorously established. Rodriguez-Villegasdemonstrated that Boyd-like identities hold for elliptic curves with com-plex multiplication [24], and Brunault proved several of Boyd’s conjecturesfor elliptic curves with prime conductors [13]. This thesis contains four pa-pers aimed at establishing results related to Boyd’s conjectures (see [26],[25], [19], and [27]). In the remainder of this section we will briefly outlinesome of the main theorems in those manuscripts.3Chapter 1. Introductory ChapterLattice sums and Mahler measuresWhile a variety of procedures exist for relating Mahler measures to valuesof the Riemann zeta function (see [18] or [12]), the situation is more com-plicated for L-functions of elliptic curves. Rodriguez-Villegas proved severalof Boyd’s conjectures for CM elliptic curves; however, his proofs dependedupon Deuring’s theorem, which only applies in fortuitous situations. In thispaper, we have used the modularity theorem to obtain a variety of formu-las relating L-functions to lattice sums. This enabled us to translate manyof Boyd’s conjectures into equivalent, although still unproven, relations be-tween lattice sums and hypergeometric functions. For example, equation(1.1.2) is equivalent to the following formula:∞∑n=0(2nn)2 (1/16)2n+12n + 1?=540π2∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4((6n1 − 1)2 + 3(6n2 − 1)2 + 5(6n3 − 1)2 + 15(6n4 − 1)2)2.(1.3.4)Despite the fact that we were unable to prove (1.3.4), we successfully usedour ideas to recover Rodriguez-Villegas’s results, and several corollaries. Forinstance, if φ equals the golden ratio, then3456√15∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4[(6n1 + 1)2 + (6n2 + 1)2 + (6n3 + 1)2 + 15(6n4 + 1)2]2=C13√φ3F2(13, 13, 1323, 43;1φ)+C23√φ23F2(23, 23, 2343, 53;1φ),(1.3.5)where C1 = 23√2Γ(12)Γ(13)Γ(16), and C2 = 3√3Γ3(23). The paper con-cludes with a brief discussion of elliptic dilogarithms and higher dimensionallattice sums.Functional equations for Mahler measuresMatilde Lal´ın and I proved functional equations for Mahler measures ofelliptic curves in [19]. For example, we proved that the following identity4Chapter 1. Introductory Chapterholds for |α| < 1:m(4α2+ x +1x+ y +1y)=m(2α +2α+ x +1x+ y +1y)+m(2iα +2iα+ x +1x+ y +1y).Our theorems enabled us to express m(2 + x + x−1 + y + y−1)andm(8 + x + x−1 + y + y−1)in terms of L(E, 2), where E has conductor 24(subject to a claim of Rodriguez-Villegas). Our proofs depended upon Ra-manujan’s theory of modular equations to alternative bases [6], as well asformulas for the Rogers-Ramanujan continued fraction [3], and q-series re-sults of Stienstra [28] and Verrill [29]. As a final corollary, we obtained new2F1 transformations which Maier also studied [20].New 5F4 transformations and Mahler measuresOften it is quite difficult to identify Mahler measures as hypergeomet-ric functions. This is true for both two-variable and three-variable Mahlermeasures. In [27] I used modular equations to equate several three variableMahler measures to linear combinations of 5F4 functions. For example, for|α| sufficiently large:m(3α +3α+ x +1x+ y +1y+ z +1z)= − 120∞∑n=11n(4n)!n!4(α33 (3 + α2)2)2n− 320∞∑n=11n(4n)!n!4(α−33 (3 + α−2)2)2n+15log(9α3(3 + α2) (3 + α−2)3).This particular identity allowed me to translate several of Bertin’s formulasfor L-series K3 hypersurfaces into explicit hypergeometric formulas [8]. Asa corollary to my hypergeometric transformations, I deduced several newformulas for 1/π involving Domb numbers, including:2π=∞∑n=0(−1)n (3n + 1)32nn∑k=0(2kk)(2n− 2kn− k)(nk)2.Similar identities for 1/π and 1/π2 have been studied in [14], [30], and [31].Zudilin summarized a variety of related results in [32].5Chapter 1. Introductory ChapterTrigonometric integrals and Mahler measuresWhile it is difficult to speculate on the nature of corollaries that mightemerge from a proof of (1.1.2), I will point to the fourth paper in this thesisfor an example. Several years ago Boyd conjectured an identity expressinga Mahler measure in terms of ζ(3)/π2. In particular, he calculated that145π2ζ(3) = m ((1 + x) + (1− x)(y + z)) . (1.3.6)Condon first established this formula using contour integration [16]. Afterconsidering Condon’s proof, I realized that (1.3.6) followed from an identityfor the 4F3 hypergeometric function, and I proceeded to discover a proofbased upon a new series transformation [25]:∞∑n=0(−1)n(2n + 1)3(2nn) ( 4r1− r2)2n+1=12Li3(r2)+ 4Li3(1− r) + 4Li3(r1 + r)− 4ζ(3)− log(1 + r1− r)Li2(r2)− 2π23log(1− r)− 23log3(1 + r)+ 2 log(r) log2(1− r),which holds for appropriate values of r. As usual, the polylogarithm isdefined byLik(x) =∞∑n=1xnnk.Transformations for sums involving binomial coefficients have attracted in-terest since Ape´ry proved the irrationality of ζ(2) and ζ(3) using such identi-ties (see [1], [4], or [5]). As a corollary to my hypergeometric transformationI also proved eight new Mahler measure formulas. Thus, the solutions ofnumerically conjectured problems often lead to unexpected and interestingcorollaries.6Bibliography[1] G. Almkvist and A. Granville, Borwein and Bradley’s Ape´ry-like for-mulae for ζ(4n + 3). Experiment. Math. 8 (1999), no. 2, 197–203.[2] G. E. Andrews, R. Askey, R. Roy, Special functions. Encyclopedia ofMathematics and its Applications, 71. Cambridge University Press,Cambridge, 1999.[3] G. E. Andrews and B. C. Berndt, Ramanujan’s Lost Notebook, Part I,Springer-Verlag, New York, 2005.[4] D. H. Bailey, J. M. Borwein, D. M. Bradley, Experimental determina-tion of Apry-like identities for ζ(2n+2). Experiment. Math. 15 (2006),no. 3, 281–289.[5] N. Batir, Integral representations of some series involving(2kk)−1k−nand some related series, J. Appl. Math. and Comp. 147 (2004), 645-667.[6] B. C. Berndt, Ramanujan’s Notebooks, Part V, Springer-Verlag, NewYork, 1998.[7] M. J. Bertin, Mesure de Mahler d’une famille de polynoˆmes. J. ReineAngew. Math. 569 (2004), 175–188.[8] M. J. Bertin, Mesure de Mahler d’hypersurfaces K3. Preprint[9] J. M. Borwein and P. B. Borwein, Ramanujan’s rational and algebraicseries for 1/π, J. Indian Math. Soc. (N.S.) 51 (1987), 147–160 (1988).[10] D. W. Boyd, Mahler’s measure and special values of L-functions, Ex-periment. Math. 7 (1998), 37-82. Academic Press, 1994.[11] D. W. Boyd, Speculations concerning the range of Mahler’s measure,Canad. Math. Bull. 24 (1981), no. 4, 453–469.7Bibliography[12] D. W. Boyd and F. Rodriguez-Villegas, Mahler’s measure and the dilog-arithm, I. Canad. J. Math. 54 (2002), no. 3, 468–492.[13] F. Brunault, explicite du thorme de Beilinson pour la courbe modulaireX1(N), C. R. Math. Acad. Sci. Paris 343 (2006), no. 8, 505–510.[14] H. H. Chan, S. H. Chan and Z. Liu, Domb’s numbers and RamanujanSato type series for 1/π. Adv. Math. 186 (2004), no. 2, 396–410.[15] D. V. Chudnovsky and G. V. Chudnovsky, Approximations and com-plex multiplication according to Ramanujan. Ramanujan revisited(Urbana-Champaign, Ill., 1987), 375–472, Academic Press, Boston,MA, 1988.[16] J. Condon, Mahler measure evaluations in terms of polylogarithms,Doctoral dissertation, The University of Texas at Austin. Availableelectronically from http://hdl.handle.net/2152/603 .[17] C. Deninger, Deligne periods of mixed motives, K-theory and the en-tropy of certain Zn-actions, J. Amer. Math. Soc. 10 (1997), no. 2,259–281.[18] M. N. Lal´ın, An algebraic integration for Mahler measure, Duke Math.J. 138 (2007), no. 3, 391–422.[19] M. N. Lal´ın and M. D. Rogers, Functional equations for Mahler mea-sures of genus-one curves, Algebra and Number Theory, 1 (2007), no.1, 87–117.[20] R. S. Maier, Algebraic hypergeometric transformations of modular ori-gin. Trans. Amer. Math. Soc. 359 (2007), no. 8, 3859–3885 (electronic).[21] Y. Martin and K. Ono, Eta-Quotients and Elliptic Curves, Proc. Amer.Math Soc. (125) 1997, no. 11, 3169-3176.[22] S. Ramanujan, Modular equations and approximations to π, [Quart. J.Math. 45 (1914), 350-372]. Collected papers of Srinivasa Ramanujan,23-29, AMS Chelsea Publ., Providence, RI, 2000.[23] F. Rodriguez-Villegas, Identities between Mahler measures, Numbertheory for the millennium, III (Urbana, IL, 2000), 223–229, A K Peters,Natick, MA, 2002.8Bibliography[24] F. Rodriguez-Villegas, Modular Mahler measures I, Topics in numbertheory (University Park, PA, 1997), 17–48, Math. Appl., 467, KluwerAcad. Publ., Dordrecht, 1999.[25] M. D. Rogers, A study of inverse trigonometric integrals associated withthree-variable Mahler measures, and some related identities, J. NumberTheory 121 (2006), no. 2, 265–304.[26] M. D. Rogers, Hypergeometric formulas for lattice sums and Mahlermeasures, in preparation.[27] M. D. Rogers, New 5F4 hypergeometric transformations, three-variableMahler measures, and formulas for 1/π, to appear in the Ramanujanjournal.[28] J. Stienstra, Mahler measure variations, eisenstein series and instantonexpansions, Mirror symmetry. V, 139–150, AMS/IP Stud. Adv. Math.,38, Amer. Math. Soc., Providence, RI, 2006.[29] H. A. Verrill, Picard-Fuchs equations of some families of elliptic curves.Proceedings on Moonshine and related topics (Montreal, QC, 1999),253–268, CRM Proc. Lecture Notes, 30, Amer. Math. Soc., Providence,RI, 2001.[30] W. Zudilin, Quadratic transformations and Guillera’s formulas for 1/π2.(Russian) Mat. Zametki 81 (2007), no. 3, 335–340.[31] W. Zudilin, More Ramanujan-type formulae for 1/π2, Russian Math.Surveys 62:3 (2007), 634–636.[32] W. Zudilin, Ramanujan-type formulas for 1/π: A second wind?, Pro-ceedings of the BIRS workshop ”Modular Forms and String Duality”(Banff, June 2006), N. Yui (ed.), London Math. Soc. Lecture NotesSer., 13 pages (accepted for publication).Department of Mathematics, University of British Columbia,Vancouver, B.C. V6T 1Z2, Canadamatrogers@math.ubc.ca9Chapter 2Lattice sums and MahlermeasuresMathew D. Rogers12.1 IntroductionIn this paper we will prove a number of formulas relating the special valuesof L-series of elliptic curves to hypergeometric functions. This paper waspartially inspired by the work of Boyd and Rodriguez-Villegas. Recall thatBoyd used numerical methods to conjecture a large number of formulasrelating the L-series of elliptic curves to special values of Mahler’s measure.The first example of such an identity was due to Deninger, who hypothesizedthatm(1 + y + y−1 + z + z−1) ?= 154π2L(E, 2), (2.1.1)where E is a conductor 15 elliptic curve. Although we will not offer aproof of Deninger’s formula in this paper, we will provide a new method forestablishing results due to Rodriguez-Villegas, and we are hopeful that ourmethod will eventually apply to formulas such as (2.1.1). The essential resultthat we will require is the modularity theorem, which supplants Deuring’stheorem in Rodriguez-Villegas’s work.Let us briefly recall that to any elliptic curve E with conductor N , wecan associate an L-seriesL(E, s) :=∏pN1(1− apps + pp2s) ∏p|N1(1− apps)=∞∑n=1anns,(2.1.2)1A version of this paper has been submitted for publication. Rogers, M. D. Hypergeo-metric formulas for lattice sums and Mahler measures.10Chapter 2. Lattice sums and Mahler measureswhich converges for Re (s) > 32 . If E has good reduction at p, then p + 1−ap equals the number of integral points on E modulo p. The modularitytheorem shows that L(E, s) has a meromorphic continuation to the entirecomplex plane, and that the sumg(e2πiτ ) =∞∑n=1ane2πinτ ,is a weight two modular form on Γ0(N) with respect to τ . In general, ifg(q) =∑∞n=1 anqn is an arbitrary power series, then we will commit a slightabuse of notation and writeL(g, s) :=∞∑n=1anns.Since an is not necessarily multiplicative with respect to n, it follows thatL(g, s) may or may not have an Euler product. However, if g(q) correspondsto an elliptic curve E, then we will always have L(g, s) = L(E, s) by themodularity theorem.In this paper we will illustrate that the Mahler measure identities of Boydand Deninger belong to a larger class of formulas for the Mellin transformsof weight-two modular forms. In particular, we can often derive identitiesbetween such modular forms and Mahler measures, irrespective of whetheror not the modular form is associated to an elliptic curve. For instance, wecan prove identities such asm(4i(8 + 3√7)+ y + y−1 + z + z−1)=16√7π2∞∑n=1ann2, (2.1.3)whereg(q) :=∞∑n=1anqn = q∞∏n=1(1− q8n)3(1− q28n)2(1− q56n) ,is a weight-two cusp form on Γ0(56). A cursory computation reveals thecoefficients of g(q) are not multiplicative, since a77 = 14, but a7 = a11 = 0.Therefore, while g(q) can not arise from an elliptic curve, we will also look atmany eta products which do possess arithmetic interpretations. Martin andOno established an exhaustive list of eta quotients associated with ellipticcurves in [49].11Chapter 2. Lattice sums and Mahler measuresMany of the Mahler measure identities in this paper can be related togeneralized hypergeometric functions. Recall that the generalized hyperge-ometric function is defined bypFq(a1,...,apb1,...,bq ;x):= 1 +∞∑n=1(a1)n . . . (ap)n(b1)n . . . (bq)nxnn!,where (z)n = z(z+1) . . . (z+n−1). One consequence of this fact is that wecan obtain series acceleration formulas for values of certain L-series. Noticethat equation (2.1.3) can be transformed into16√7π2∞∑n=1ann2= log(32 + 12√7)+12∞∑n=1(−1)n+1n(2nn)2 1(32 + 12√7)2n .(2.1.4)Given the appearance of hypergeometric functions, it should not be surpris-ing that Ramanujan’s work will enter into our proofs.2.1.1 Boyd’s conjectures and four-dimensional lattice sumsIn this section we will summarize a variety of explicit formulas relating four-dimensional lattice sums to Mahler measures of polynomials. Most of theseresults are only conjectures, although numerical calculations can be usedto verify them to any degree of accuracy. Our first step will be to invokethe modularity theorem to find explicit formulas for L-functions of ellipticcurves with conductors N ∈ {11, 14, 15, 20, 24, 27, 32, 36}.Definition 2.1.1. Let us define F (b, c) byF (b, c) := (1+b)2(1+c)2∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4[(6n1 + 1)2 + b(6n2 + 1)2 + c(6n3 + 1)2 + bc(6n4 + 1)2]2 .It turns out that F (b, c) can be used to numerically calculate a variety ofL-values. The following theorem is an easy consequence of the modularitytheorem:Theorem 2.1.2. Suppose that EN is an elliptic curve of conductor N inan appropriate isogeny class, thenL(EN , 2) = F (b, c) (2.1.5)12Chapter 2. Lattice sums and Mahler measuresfor the following values of N and (b, c):N (b, c)11 (1, 11)14 (2, 7)15 (3, 5)20 (1, 5)24 (2, 3)27 (1, 3)32 (1, 2)36 (1, 1)Proof. We are interested in cases where cusp forms of elliptic curves equalthe product of four eta functions. An exhaustive list of all such cusp formsis provided in [49]. By inspection of that list, the eta product associatedwith EN will have the formg(q) := q∞∏n=1(1− qAn) (1− qAbn) (1− qAcn) (1− qAbcn) ,where (1+ b)(1+ c)A = 24. If we recall Euler’s pentagonal number theorem∞∏n=1(1− qn) =∞∑n=−∞(−1)nqn(3n+1)/2,this becomesg(q) =∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4qA(6n1+1)2+Ab(6n2+1)2+Ac(6n3+1)2+Abc(6n4+1)224 ,and it follows immediately thatL(EN , 2) =242A2∞∑ni=−∞i∈1,2,3,4(−1)n1+n2+n3+n4[(6n1 + 1)2 + b(6n2 + 1)2 + c(6n3 + 1)2 + bc(6n4 + 1)2]2 .Since (1 + b)(1 + c) = 24/A, the theorem follows.Since we now have expressed several different L-values in terms of F (b, c),it seems logical to list all of the known Mahler measures which reduce tovalues of that function.13Chapter 2. Lattice sums and Mahler measuresDefinition 2.1.3. Let us fix the following notation:m(k) :=m(k + y + y−1 + z + z−1), (2.1.6)n(k) :=m(y3 + z3 + 1− kyz) , (2.1.7)g(k) :=m ((1 + y)(1 + z)(y + z)− kyz) , (2.1.8)r(k) :=m ((1 + y)(1 + z)(1 + y + z)− kyz) . (2.1.9)For convenience we have slightly altered the definitions of n(k), g(k)and r(k) that appeared in [48]. All of the following examples were eitherextracted from Boyd’s paper [41], or were deduced by combining Boyd’sconjectures with functional equations in [48]. While Boyd’s minimal Weier-strass models often do not coincide with the minimal Weierstrauss models in[49], the elliptic curves are presumably isogenous, and the following resultsare all numerically true:n(3 3√2) =272π2F (1, 1) (2.1.10)g(2) =92π2F (1, 1) (2.1.11)g(−4) =18π2F (1, 1) (2.1.12)m(4i) =16π2F (1, 2) (2.1.13)m(2√2) =8π2F (1, 2) (2.1.14)n(−6) = 814π2F (1, 3) (2.1.15)n( 3√2) ?=256π2F (1, 5) (2.1.16)n( 3√32) ?=403π2F (1, 5) (2.1.17)g(−2) ?=15π2F (1, 5) (2.1.18)g(4) ?=10π2F (1, 5) (2.1.19)r(−1) = 774π2F (1, 11) (2.1.20)m(2) ?=6π2F (2, 3) (2.1.21)m(8) ?=24π2F (2, 3) (2.1.22)14Chapter 2. Lattice sums and Mahler measuresm(3√2) ?=15π2F (2, 3) (2.1.23)m(i√2) ?=9π2F (2, 3) (2.1.24)n(−1) ?= 7π2F (2, 7) (2.1.25)n(5) ?=492π2F (2, 7) (2.1.26)g(1) ?=72π2F (2, 7) (2.1.27)g(7) ?=21π2F (2, 7) (2.1.28)g(−8) ?=35π2F (2, 7) (2.1.29)m(1) ?=154π2F (3, 5) (2.1.30)m(3i) ?=754π2F (3, 5) (2.1.31)m(5) ?=452π2F (3, 5) (2.1.32)m(16) ?=1654π2F (3, 5) (2.1.33)All of the results involving F (1, 1), F (1, 2), and F (1, 3) can be deduced fromRodriguez-Villegas’s paper [51]. In particular, those Mahler measures can bewritten in terms of two-dimensional Eisenstein-Kronecker series, and thenthe results follow from Deuring’s theorem.2.1.2 Summary of hypergeometric formulas for F (b, c)In general, we believe that F (b, c) can always be written in terms of integralsof hypergeometric functions, regardless of the values of b and c. In thissubsection we will translate almost all of the known Mahler measures forF (b, c) into hypergeometric functions. In Corollary 2.2.8 we will also provethat similar expressions exist for both F (2, 2) and F (1, 4), even though thosesums are apparently unrelated to the theory of elliptic curves.Theorem 2.1.4. We can express m(k), n(k), and g(k) in terms of gener-alized hypergeometric functions for most values of k:m(k) =Re(log(k)− 2k24F3(32, 32,1,12,2,2;16k2)), (2.1.34)15Chapter 2. Lattice sums and Mahler measuresn(k) =Re(log(k)− 2k34F3(43, 53,1,12,2,2;27k3)), (2.1.35)g(k) =Re(log((4 + k)(k − 2)4k2)− 2k2(4 + k)3 4F3(43, 53,1,12,2,2;27k2(4 + k)3)(2.1.36)− 8k(k − 2)3 4F3(43, 53,1,12,2,2;27k(k − 2)3)).Equation (2.1.34) is valid in C \ {0}, while (2.1.36) holds in C \ [−4, 2], and(2.1.35) is true for |k| is sufficiently small.In certain cases we can reduce these hypergeometric functions further.Suppose that k ∈ R \ {0}, thenRe(log(k)− 2k24F3(32, 32,1,12,2,2;16k2))= Re( |k|4 3F2(12, 12, 121, 32;k216)), (2.1.37)andRe(log(k)− 2k34F3(43, 53,1,12,2,2;27k3))=s(k)Re(Ak3F2(13, 13, 1323, 43;k327)+Bk23F2(23, 23, 2343, 53;k327)),(2.1.38)where A =3√2Γ( 16)Γ( 13)Γ( 12)8√3π2, B =Γ3( 23)16π2, and s(k) = 1+3sgn(k)4 .Equations (2.1.37) and (2.1.38) will often allow us to obtain convergentseries expansions from divergent hypergeometric formulas. For example,applying the results of the last theorem to conjecture (2.1.30), we obtainF (3, 5) ?=16π215∞∑n=0(2nn)2 (1/16)2n+12n + 1. (2.1.39)It is hardly coincidental that (2.1.39) bears a striking resemblance to a fa-mous formula that Ramanujan obtained for Catalan’s constant [33]:L (χ−4, 2) = π∞∑n=0(2nn)2 (1/4)2n+12n + 1.Ramanujan’s formula follows easily from Boyd’s evaluation of the degenerateMahler measure m(4).16Chapter 2. Lattice sums and Mahler measuresThe following list summarizes all of the known values of hypergeomet-ric functions which reduce to special cases of F (b, c). When possible, wehave used equations (2.1.37) and (2.1.38) to obtain hypergeometric func-tions with convergent arguments. Since no hypergeometric expression isknown for r(−1), we have simply retained that Mahler measure in our list.Finally, because Mahler measures such as g(2) and n(3 3√2)lead to identicalhypergeometric expressions, this list contains fewer entries than we mightotherwise expect. As in Theorem 2.1.4, defineA :=3√2Γ(16)Γ(13)Γ(12)8√3π2, B :=Γ3(23)16π2,then the following results are numerically true:92π2F (1, 1) =19log(54)− 1814F3(43, 53,1,12,2,2;12), (2.1.40)16π2F (1, 2) =2 log(2) +184F3(32, 32,1,12,2,2;−14), (2.1.41)8π2F (1, 2) =1√23F2(12, 12, 121, 32;12), (2.1.42)814π2F (1, 3) = log(6) +11084F3(43, 53,1,12,2,2;−18), (2.1.43)256π2F (1, 5) ?= 3√2A3F2(13, 13, 1323, 43;227)+ 3√4B3F2(23, 23, 2343, 53;227), (2.1.44)403π2F (1, 5) ?=53log(2)− 1164F3(43, 53,1,12,2,2;2732), (2.1.45)774π2F (1, 11) =r(−1), (2.1.46)6π2F (2, 3) ?=123F2(12, 12, 121, 32;14), (2.1.47)24π2F (2, 3) ?=3 log(2)− 1324F3(32, 32,1,12,2,2;14), (2.1.48)15π2F (2, 3) ?=12log(18)− 194F3(32, 32,1,12,2,2;89), (2.1.49)9π2F (2, 3) ?=12log(2) + 4F3(32, 32,1,12,2,2,−8), (2.1.50)7π2F (2, 7) ?=A2 3F2(13, 13, 1323, 43;− 127)− B2 3F2(23, 23, 2343, 53;− 127), (2.1.51)492π2F (2, 7) ?= log(5)− 21254F3(43, 53,1,12,2,2;27125), (2.1.52)17Chapter 2. Lattice sums and Mahler measures21π2F (2, 7) ?=g(7), (2.1.53)154π2F (3, 5) ?=143F2(12, 12, 121, 32;116), (2.1.54)452π2F (3, 5) ?= log(5)− 2254F3(32, 32,1,12,2,2;1625), (2.1.55)1654π2F (3, 5) ?=4 log(2)− 11284F3(32, 32,1,12,2,2;116), (2.1.56)754π2F (3, 5) ?= log(3) +294F3(32, 32,1,12,2,2;−169). (2.1.57)While most of these formulas remain unproven, a variety of partial resultsexist. For instance, identities (2.1.47) through (2.1.50) are equivalent toone another [48], formulas (2.1.40) through (2.1.43) follow from [51], andBrunault proved (2.1.46) in [42].2.2 Reductions of F (1, 1), F (1, 2), F (1, 4), andF (2, 2) to integrals of hypergeometricfunctionsIn the previous section we translated many of Boyd’s conjectures into explicitidentities between hypergeometric functions and lattice sums. This approachhas two essential consequences. Not only does it eliminate any obvious con-nection with elliptic curves, but it also allows for the construction of proofsbased upon series manipulation. We have used such an approach to reducefive cases of F (b, c) to integrals of hypergeometric functions. In this sectionwe will discuss the cases that occur when (b, c) ∈ {(1, 1), (1, 2), (1, 4), (2, 2)}.We will rely heavily on the q-series theorems contained in Ramanujan’snotebooks (see [38] and [40]).Definition 2.2.1. Recall the following q-series notation:ϕ(q) :=∞∑n=−∞qn2, ψ(q) :=∞∑n=0qn(n+1)2 ,f(−q) :=∞∏n=1(1− qn), (x; q)∞ :=∞∏n=0(1− xqn) .Lemma 2.2.3 reduces the aforementioned cases of F (b, c) to two-dimensionalsums. Such identities exist because various eta-quotients can be written in18Chapter 2. Lattice sums and Mahler measuresterms theta functions. Euler’s pentagonal number formula is probably thesimplest such indentity:∞∏n=1(1− qn) =∞∑n=−∞qn(3n+1)2 .Unfortunately, similar formulas are not known for f2(−q), f(−q)f (−q2), orf(−q)f (−q3) [45]. This fact represents the main obstruction to also provingBoyd’s conjectures for F (1, 5), F (2, 3), F (2, 7), and F (3, 5).Definition 2.2.2. We will use the following notation:F(1,1)(x) :=16∞∑n=−∞k=0(−1)n+k(2k + 1)[3(2k + 1)2 + x2(6n + 1)2]2, (2.2.1)F(1,2)(x) :=∞∑n=−∞k=0(−1)n+k(2k + 1)[(2k + 1)2 + x2(2n)2]2, (2.2.2)F(1,4)(x) :=25∞∑n,k=−∞(−1)n(3k + 1)[4(3k + 1)2 + x2(6n + 1)2]2, (2.2.3)F(2,2)(x) :=9∞∑n,k=0(−1)n(n+1)2 +k(2k + 1)[2(2k + 1)2 + x2(2n + 1)2]2. (2.2.4)Lemma 2.2.3. Suppose that (b, c) ∈ {(1, 1), (1, 2), (1, 4), (2, 2)}, thenF(b,c)(1) = F (b, c). (2.2.5)Proof. First notice that F (b, c) has the following integral representation forall values of b and c:242F (b, c)(1 + b)2(1 + c)2=∫ 10∫ q10q(1+b)(1+c)24 f (−q) f(−qb)f (−qc) f(−qbc) dqqdq1q1.We can apply Euler’s pentagonal number theorem four times (once to eachoccurrence of f(−q)), to see the truth of this last formula. Taking note ofthe following identities:q1/6f4(−q) =(q1/24f (−q))(q1/8f3 (−q)),q1/4f2 (−q) f2 (−q2) =(f2 (−q)f (−q2))(q1/4f3(−q2)) ,19Chapter 2. Lattice sums and Mahler measuresq5/12f2 (−q) f2 (−q4) =(q1/12f (−q2))(q1/3 f2 (−q) f2 (−q4)f (−q2)),q3/8f (−q) f2 (−q2) f (−q4) =(q1/8 f (−q) f (−q4)f (−q2))(q1/4f3(−q2)) ,and then employing well known series expansions:f2 (−q)f (−q2) = ϕ(−q) =∞∑n=−∞(−1)nqn2 , (2.2.6)q1/8f (−q) f (−q4)f (−q2) = q1/8ψ (−q) =∞∑n=0(−1)n(n+1)2 q (2n+1)28 , (2.2.7)q1/24f (−q) =∞∑n=−∞(−1)nq (6n+1)224 , (2.2.8)q1/8f3 (−q) =∞∑n=0(−1)n(2n + 1)q (2n+1)28 , (2.2.9)q1/3f2 (−q) f2 (−q4)f (−q2) =∞∑n=−∞(3n + 1)q(3n+1)23 , (2.2.10)we recover equation (2.2.5) in every case.We will use the next two propositions to reduce each of the two-dimensionalsums to a q-series. Then, in Theorem 2.2.7, we will reduce F(b,c)(x) to in-tegrals of hypergeometric functions for x ∈ (0,∞). For certain values of x,those formulas also translate into identities involving generalized hypergeo-metric functions and Mahler measures.Proposition 2.2.4. Assume that δ > 0 is sufficiently small, thenF(1,1)(x) =−π2i9x∫ iδ+∞iδ−∞sinh(t)1 + 4 sinh2(t)sec(√3xt) tan(√3xt)tdt, (2.2.11)F(1,2)(x) =π332− π2i32x∫ iδ+∞iδ−∞(csch(t)− 1t)sec (xt) tan (xt)tdt (2.2.12)F(1,4)(x) =−25π2i144√3x∫ iδ+∞iδ−∞sinh(t)1 + 4 sinh2(t)csc2(π3 − xt)− csc2 (π3 + xt)tdt,(2.2.13)20Chapter 2. Lattice sums and Mahler measuresF(2,2)(x) =−9π2i128x∫ iδ+∞iδ−∞sinh(t)cosh(2t)sec(√2xt)tan(√2xt)tdt (2.2.14)Proof. The proofs are all substantially the same. The idea is to use contourintegration to pick off the n-index of summation in the corresponding two-dimensional sum. We will illustrate the proof of (2.2.12) explicitly. Firstassume that 0 < δ < 1, and let C denote a closed contour which runs alongthe line (iδ − ∞, iδ + ∞) and then encircles the upper half plane. Sincecsch(t) has poles at t = πin, by the residue theorem∞∑n=1(−1)n[(2k + 1)2 + x2(2n)2]2=12πi∫C(csch(t)− 1t)1[(2k + 1)2 − x2(2t/π)2]2dt.If t = Reiθ, then the integrand has order O(R−5), and therefore the circularportion of the contour integral vanishes as R tends to∞. Next observe thatthe sumπ332sec(tx) tan(tx)tx=∞∑k=0(−1)k(2k + 1)[(2k + 1)2 − x2 (2t/π)2]2converges uniformly when Im (t) > 0, hence∞∑k=0n=1(−1)n+k(2k + 1)[(2k + 1)2 + x2(2n)2]2= − π2i64x∫ iδ+∞iδ−∞(csch(t)− 1t)sec(xt) tan(xt)tdt.Equation (2.2.12) follows easily from this last result.Proposition 2.2.5. Let χ−3(k) and χ−4(k) denote Legendre symbols mod-ulo three and four, and assume that x > 0.If q = e−πx/√12 and ω = eπi/6, thenF(1,1)(x) =2π29x∞∑k=1kχ−4(k) log∣∣∣∣1 + wqk1− wqk∣∣∣∣ . (2.2.15)If q = e−πx, thenF(1,2)(x) =π332− π28x∞∑k=1kχ−4(k) log(1 + qk). (2.2.16)If q = e−πx/3 and ω = eπi/6, thenF(1,4)(x) =25π272x∞∑k=1kχ−3(k) log∣∣∣∣1 + wqk1− wqk∣∣∣∣ . (2.2.17)21Chapter 2. Lattice sums and Mahler measuresIf q = e−πx/√8 and ω = eπi/4, thenF(2,2)(x) =9π232x∞∑k=1kχ−4(k) log∣∣∣∣1 + wqk1− wqk∣∣∣∣ . (2.2.18)Proof. All of the proofs are very similar, so we will only prove (2.2.15) indetail. Since sec(√3xt) × tan(√3xt) is periodic, we can rearrange (2.2.11)to obtainF(1,1)(x) = −π2i9x∫ iδ+ 2π√3xiδ⎛⎝ ∞∑n=−∞1t + 2πn√3xsinh(t + 2πn√3x)1 + 4 sinh2(t + 2πn√3x)⎞⎠ sin(√3xt)cos2(√3xt)dt,where the interchange of summation and integration can be justified by thefact that the summand has order O(e−2π|n|/√3x)as n → ±∞. Observethat if q = e−πx/√12, ω = eπi/6, and Im (t) = δ < x, then2π√3x∞∑n=−∞1t + 2πn√3xsinh(t + 2πn√3x)1 + 4 sinh2(t + 2πn√3x)= log(2 +√3)+ 2∞∑k=1log∣∣∣∣1 + wqk1− wqk∣∣∣∣ cos(√3xkt) .(2.2.19)This new restriction, δ < x, guarantees uniform convergence of the Fourierseries, and is consistent with the prior assumption that 0 < δ 	 1. Theproof of (2.2.19) is a straightforward exercise in contour integration whichwe will skip. Substituting (2.2.19) into our integral yieldsF(1,1)(x) = −π2i9x(log(2 +√3)I0 + 2∞∑k=1log∣∣∣∣1 + wqk1− wqk∣∣∣∣ Ik),whereIk =√3x2π∫ iδ+ 2π√3xiδcos(√3xkt)sin(√3xt)cos2(√3xt) dt=∫ iδ′+1iδ′cos (2πkt) sin (2πt)cos2 (2πt)dt=ik sin (πk/2) .The final step in this calculation follows from considering a closed rectan-gular contour with vertices at {0, iδ′, 1 + iδ′, 1} which avoids the boundary22Chapter 2. Lattice sums and Mahler measurespoints t = 1/4 and t = 3/4. With this formula for Ik in hand, the proofof (2.2.15) is complete. The proofs of the other formulas follow in a similarmanner from slightly different Fourier series expansions.At this point, a hypergeometric formula for F(1,2)(x) can be recovered.By combining equation (2.2.16) with formulas (2-9) and (2-16) in [48], it iseasy to recognize that if q = e−πx, thenF(1,2)(x) =π216xm(if4(−q)√qf4 (−q4) + y + y−1 + z + z−1). (2.2.20)In the next section we will use values of class invariants to deduce explicitexamples from (2.2.20). Unfortunately, we will require another theorem inorder to obtain useful results on the other three lattice sums.Theorem 2.2.6. In this theorem we will always assume that x > 0. Ifq = e−πx/√12, thenF(1,1)(x) =2π23√3xIm[∫ iq0f9(−u3)f3 (−u) du]. (2.2.21)If q = e−πx, thenF(1,2)(x) =π332− π216x∫ q0ϕ2(−u)ϕ4(u)− 1udu. (2.2.22)If q = e−πx/3, thenF(1,4)(x) =25π236xIm⎡⎣∫ e 2πi3 q0ϕ2(u)ψ4(u2)du⎤⎦ . (2.2.23)If q = e−πx/√8, thenF(2,2)(x) =9π232√2x∫ q0ϕ(−u2)ϕ(u4) (3ψ4(−u2)− ψ4(u2)) du. (2.2.24)Proof. Equations (2.2.22) and (2.2.23) have similar proofs, so we will onlyprove the latter identity. Notice that (2.2.17) can be rearranged to obtain72x25π2F(1,4)(x) =Re(∫ q0∞∑k=1k2χ−3(k)2ωuk1− ω2u2kduu)23Chapter 2. Lattice sums and Mahler measures=√3∫ q0∞∑k=1k2χ−3(k)(uk − u5k1 + u6k)duu=Im(2∫ e2πi/3q0∞∑k=1k2(uk − u3k + u5k1 + u6k)duu)=Im(2∫ e2πi/3q0∞∑k=1k2uk1 + u2kduu).Combining entries 10.1, 11.3, and 17.2 in Chapter 17 of [38], we deduce thatfor |u| < 1:∞∑k=1k2uk1 + u2k= uϕ2(u)ψ4(u2),which completes the proof of (2.2.23).The proofs of equations (2.2.21) and (2.2.24) will require the followingformula:Im (g (iu, t)) =1t∞∑k=0(−1)k (2k + 1)2u2k+11 + u2(2k+1)(t2k+1 − t−(2k+1)), (2.2.25)whereg(u, t) =(u;u)6∞(t−2u;u)∞(t2;u)∞(t−1u;u)4∞ (t;u)4∞,and(x; q)∞ = (1− x)(1− xq)(1− xq2) . . .Equation (2.2.25) is a direct consequence of identity (14.2.9) in [35], whichfollows from product expansions for the Weierstrass ℘-function [43].Rearranging equation (2.2.15) we have9x2π2F(1,1)(x) =Re(∫ q0∞∑k=0(−1)k(2k + 1)2 2ωu2k+11− ω2u2(2k+1)duu)=√3∫ q0∞∑k=0(−1)k(2k + 1)2u2k+1 − u5(2k+1)1 + u6(2k+1)duu.Applying (2.2.25) after letting u → u3 and t → u−2, transforms this lastintegral into3√3x2π2F(1,1)(x) = Im(∫ iq0f5(−u2) f4 (−u3) f (−u6)f4(−u) du).24Chapter 2. Lattice sums and Mahler measuresBy the following eta function identity:f5(−u2) f4 (−u3) f (−u6)f4(−u) =f9(−u3)f3 (−u) + uf9(−u6)f3 (−u2) , (2.2.26)this becomes3√3xπ2F(1,1)(x) =Im(∫ iq0f9(−u3)f3 (−u) du)− Im(∫ q0uf9(u6)f3 (u2)du)=Im(∫ iq0f9(−u3)f3 (−u) du)− 0,which completes the proof of (2.2.21). Although we will not elaborate onthe proof of (2.2.26) here, it suffices to say that it follows from algebraictransformations for the hypergeometric function 2F1 ( 13 , 23 ,1, u). We will alsopoint out that the eta-quotient f9 (−q) /f3 (−q3) curiously appears on page1734 of [34].The proof of (2.2.24) follows the same lines, but requires a few extrasteps. Proceeding as before, we find that32x9π2F(2,2)(x) =Re(∫ q0∞∑k=0(−1)k(2k + 1)2 2ωu2k+11− ω2u2(2k+1)duu)=∫ q0∞∑k=0(−1)k(2k + 1)2u2k+1 − u3(2k+1)1 + u4(2k+1)duu.If we apply equation (2.2.25) after letting u→ u2 and t→ u−1, this becomes32x9π2F(2,2)(x) = Re(2∫ ωq0f6(−u2) f (−u8)f (−u4)1(ωu, u2)4∞ (ω¯u, u2)4∞du),where ω = eπi/4. For brevity of notation let us define a new functiong(u) :=(ωu, u2)∞(ω¯u, u2)∞=∞∏n=0(1−√2u2n+1 + u2(2n+1)). (2.2.27)Since (2.2.18) is odd with respect to q, our integral can be transformed into32x9π2F(2,2)(x) = Re(∫ ωq0f6(−u2) f (−u8)f (−u4)g4(u) + g4(−u)g4(u)g4(−u) du). (2.2.28)25Chapter 2. Lattice sums and Mahler measuresNext we will reduce(g4(u) + g4(−u)) / (g(u)g(−u))4 to theta functions. Ob-serve by equation (2.2.27) thatg(u)g(−u) =∞∏n=0(1 + u4(2n+1))=ϕ(−u8)f (−u4) . (2.2.29)With two applications of the Jacobi triple product [38], we also haveg(u) + g(−u) = 1f (−u2)( ∞∑n=−∞(−ω)nun2 +∞∑n=−∞ωnun2)=2f (−u2)∞∑n=−∞(−1)nu16n2=2ϕ(−u16)f (−u2) . (2.2.30)So finally, combining (2.2.29) and (2.2.30), we find thatg4(u) + g4(−u)g4(u)g4(−u) = 2f4(−u4)f4 (−u2)[8ϕ4(−u16)ϕ4 (−u8) − 8ϕ2(−u16)ϕ (−u2)ϕ3 (−u8) +ϕ2(−u2)ϕ2 (−u8)].Recalling that ϕ2(−q16) = ϕ (−q8)ϕ (q8), this becomesg4(u) + g4(−u)g4(u)g4(−u) =2f4(−u4)f4 (−u2)[8ϕ2(u8)− 8ϕ (u8)ϕ (−u2)+ ϕ2 (−u2)ϕ2 (−u8)],=2f5(−u4)f6 (−u2)[8ϕ2(u8)ϕ(−u2)− 8ϕ (u8)ϕ2 (−u2)+ ϕ3 (−u2)ϕ2 (−u8)],and therefore (2.2.28) simplifies to32x9π2F(2,2)(x) =Re(2∫ ωq0f4(−u4) f (−u8)×8ϕ2(u8)ϕ(−u2)− 8ϕ (u8)ϕ2 (−u2)+ ϕ3 (−u2)ϕ2 (−u8) du).Next, make the change of variables sending u→ ωu, to obtain=2∫ q0f4(u4)f(−u8)ϕ2 (−u8)× Re (ω (8ϕ2 (u8)ϕ (−iu2)− 8ϕ (u8)ϕ2 (−iu2)+ ϕ3 (−iu2))) du.26Chapter 2. Lattice sums and Mahler measuresIf we recall that ϕ(−iu2) = ϕ (u8)− 2iu2ψ (u16), then we are left with32x9π2F(2,2)(x) =√2∫ q0f4(u4)f(−u8)ϕ2 (−u8)((ϕ(u8)− 2u2ψ (u16))3−4u2ϕ (u8)ψ (u16) (ϕ (u8)− 2u2ψ (u16))) du.In order to simplify this last formula, we will freely apply theta functionidentities on pages 34 and 40 of [38]. Therefore, we find that=√2∫ q0f4(u4)f(−u8)ϕ2 (−u8)(ϕ3(−u2)− 4u2ϕ (u8)ψ (u16)ϕ (−u2)) du=√2∫ q0f4(u4)f(−u8)ϕ2 (−u8) ϕ(−u2) (ϕ2 (−u2)− 4u2ψ2 (u8)) du=1√2∫ q0f4(u4)f(−u8)ϕ2 (−u8) ϕ(−u2) (3ϕ2 (−u2)− ϕ2 (u2)) du=1√2∫ q0f4(u4)f(−u8)ϕ2 (−u8)ψ2 (u4)ϕ(−u2) (3ψ4 (−u2)− ψ4 (u2)) du=1√2∫ q0ϕ(u4)ϕ(−u2) (3ψ4 (−u2)− ψ4 (u2)) du,which completes the proof of (2.2.24).The next theorem requires the signature-three theta functions. Recallthat if ω = e2πi/3, then the signature-three theta functions are defined by:a(q) :=∞∑n,m=−∞qm2+mn+n2 ,b(q) :=∞∑n,m=−∞ωm−nqm2+mn+n2 ,c(q) :=∞∑n,m=−∞q(m+1/3)2+(m+1/3)(n+1/3)+(n+1/3)2 .The signature-three theta functions satisfy many interesting formulas, in-cluding the following cubic relation:a3(q) = b3(q) + c3(q).Various other properties of a(q), b(q), and c(q) have been catalogued (see[40], page 93).27Chapter 2. Lattice sums and Mahler measuresTheorem 2.2.7. We can reduce the two-dimensional lattice sums to inte-grals of hypergeometric functions.Suppose that q = e−πx/√12, then81x2π2F(1,1)(x) =⎧⎪⎪⎪⎨⎪⎪⎪⎩3n˜(3a(iq)b(iq))+ 4√3n2(b3(iq)a3(iq))if x ∈(0, 1√5),3n˜(3a(iq)b(iq))+ 1√3n2(b3(iq)a3(iq))if x ∈(1√5,√5),1√3n2(b3(iq)a3(iq))if x ∈ (√5,∞),(2.2.31)wheren˜(k) = Re(log(k)− 2k34F3(1,1, 43, 532,2,2;27k3)),andn2(k) = Im⎛⎝∫ 1k2F1(13, 231; 1− u)udu⎞⎠ .Notice that n˜(k) = n(k) whenever |k| is sufficiently small.Suppose that q = e−πx and x > 0, then16xπ2F(1,2)(x) = m(if4(−q)√qf4 (−q4)), (2.2.32)where m(k) is defined in (2.1.6).Suppose that q = e−πx/3 and ω = e2πi/3, then144x25π2F(1,4)(x) =⎧⎪⎪⎪⎨⎪⎪⎪⎩m(4 ϕ2(ωq)ϕ2(−ωq))− 34m2(ϕ4(−ωq)ϕ4(ωq))if x ∈(0, 1√2),m(4 ϕ2(ωq)ϕ2(−ωq))+ 14m2(ϕ4(−ωq)ϕ4(ωq))if x ∈(1√2,√2),14m2(ϕ4(−ωq)ϕ4(ωq))if x ∈ (√2,∞),(2.2.33)where m(k) is defined in (2.1.6), andm2(k) := Im⎛⎝∫ 1k2F1(12, 121; 1− u)udu⎞⎠ .If q = e−πx/√8, thenF(2,2)(x) =9π2512x∫ 1ϕ4(−q2)ϕ4(q2)(3√u− 1)u3/4√1−√u2F1(12, 121; 1− u)du. (2.2.34)28Chapter 2. Lattice sums and Mahler measuresProof. The proof of this theorem follows from our ability to invert thetafunctions. First recall the classical inversion formula for the theta function:ϕ2(q) = 2F1(12, 121; 1− ϕ4(−q)ϕ4(q)), (2.2.35)which holds whenever q ∈ (−1, 1) [38]. If we use the notation α = 1 −ϕ4(−q)/ϕ4(q) and z = ϕ2(q), then many different theta functions can beexpressed in terms of these two parameters. The following identities aretrue whenever |q| < 1 (see pages 122, and 123 in [38]):ϕ(q) =√z,ϕ(−q) =(1− α)1/4√z,ϕ(q2)=(1 +√1− α)1/2√z2,ψ(−q) =q−1/8 {α(1− α)}1/8√z2,ψ(q) =q−1/8α1/8√z2,and it is also well known thatdαdq=α(1− α)z2q.Both (2.2.32) and (2.2.34) follow from applying these parameterizations toequations (2.2.22) and (2.2.24) respectively.Since equation (2.2.35) does not hold in the entire open unit disk, we willneed to generalize that result. First notice that z satisfies the hypergeometricdifferential equation with respect to α:α(1− α) d2zdα2+ (1− 2α) dzdα− z4= 0. (2.2.36)We can use the relation ddα =1dαdq× ddq , to show that (2.2.36) holds (excludingpossible poles) for |q| < 1. The most general solution of this differentialequation has the formz = C2F1(12, 121;α)+ D2F1(12, 121; 1− α),where C and D are undetermined constants. When q lies in a neighborhoodof zero, (2.2.35) shows that (C,D) = (1, 0). We can analytically continue29Chapter 2. Lattice sums and Mahler measuresthat solution to a larger connected q-domain, provided that α (and 1 − αif D = 0) does not intersect the line [1,∞). In particular, the function2F1(12, 121;α)has a branch cut running along [1,∞).If we consider values of q ∈ (0, ω) with ω = e2πi/3, then α crosses [1,∞)at the point q = ωe−π√2/3. Similarly, 1 − α intersects the branch cut atq = ωe−π/3√2. It follows that we will have to solve the hypergeometricdifferential equation separately on each of the three line segments. If u =ωe−πx/3, thenϕ2(u) =⎧⎪⎪⎪⎨⎪⎪⎪⎩−32F1(12, 121; 1− ϕ4(−u)ϕ4(u))+ 2i2F1(12, 121; ϕ4(−u)ϕ4(u))if x ∈(0, 1√2),2F1(12, 121; 1− ϕ4(−u)ϕ4(u))+ 2i2F1(12, 121; ϕ4(−u)ϕ4(u))if x ∈(1√2,√2),2F1(12, 121; 1− ϕ4(−u)ϕ4(u))if x ∈ (√2,∞).(2.2.37)The coefficients in (2.2.37) can be verified from the fact that ϕ2(u) is analyticwhen x ∈ (0,∞). For example, we can check the continuity of the right-handside of (2.2.37) by letting u → ωe−π√2/3. In that case α = 1 − ϕ4(−u)ϕ4(u)≈5.828 . . . , and we have:0 =ϕ2(ωe−π√2+03)− ϕ2(ωe−π√2−03)=2F1(12, 121;α + i0)− 2F1(12, 121;α− i0)− 2i2F1(12, 121; 1− α).This vanishing of this last expression follows from basic properties of thehypergeometric function (see problem 1 on page 276 of [47]), and thereforethe right-hand side of (2.2.37) is indeed continuous at x =√2. In practice,we simply discovered (2.2.37) numerically.We will use the theory of signature-three theta functions to prove equa-tion (2.2.31). Recall that c(q) can be expressed as an infinite product ([40],page 109):c3(q)27q=f9(−q3)f3(−q) ,and that the signature-three theta functions obey a differentiation formula(which can be derived from formula 4.4 on page 106 of [40]):c3(q)q=a(q)1− c3(q)a3(q)ddq(c3(q)a3(q)).30Chapter 2. Lattice sums and Mahler measuresIt follows immediately that equation (2.2.21) reduces toF(1,1)(x) =2π281√3xIm⎡⎣∫ iq0a(u)1− c3(u)a3(u)ddu(c3(u)a3(u))du⎤⎦ . (2.2.38)Next recall that for |u| sufficiently small ([40], page 99):a(u) = 2F1(13, 231;c3(u)a3(u)). (2.2.39)In order to apply (2.2.39) to our integral, we will need to establish a gen-eralized inversion formula which holds for u ∈ (0, i). The reasoning closelyfollows the proof of (2.2.37), except that c3(u)/a3(u) ∈ [1,∞) when u =ie−π√5/12, and 1 − c3(u)/a3(u) ∈ [1,∞) when u = ie−π/√60. Suppose thatu = ie−πx/√12, then we obtaina(u) =⎧⎪⎪⎪⎨⎪⎪⎪⎩42F1(13, 231; c3(u)a3(u))+√3i2F1(12, 121; 1− c3(u)a3(u))if x ∈(0, 1√5),22F1(13, 231; c3(u)a3(u))+√3i2F1(12, 121; 1− c3(u)a3(u))if x ∈(1√5,√5),2F1(13, 231; c3(u)a3(u))if x ∈ (√5,∞).(2.2.40)Finally, (2.2.31) follows from substituting (2.2.40) into (2.2.38) and simpli-fying.Finally, we will conclude this section by summarizing the formulas thatfollow from setting x = 1 in Theorem 2.2.7.Corollary 2.2.8. The following identities are true:272π2F (1, 1) =m(y3 + z3 + 1− 3 3√2yz), (2.2.41)16π2F (1, 2) =m(4i + y + y−1 + z + z−1), (2.2.42)14425π2F (1, 4) =m(4√θ1+ y + y−1 + z + z−1)(2.2.43)+14Im⎛⎝∫ 1θ12F1(12, 121; 1− u)udu⎞⎠ ,2569π2F (2, 2) =∫ 1θ23u− 1√u(1− u)2F1(12, 121; 1− u2)du, (2.2.44)where θ1 = 12 + i(2 +√3)4√12, and θ2 =√2− 1.31Chapter 2. Lattice sums and Mahler measures2.2.1 More explicit examplesIn this section we will use values of class invariants to deduce some explicitformulas for Mahler measures. Recall that if q = e−π√m, then the classinvariants are defined bygm :=2−1/4q−1/24(q; q2)∞ , Gm :=2−1/4q−1/24(−q; q2)∞ .It is a classical fact that Gm and gm are algebraic numbers whenever m ∈ Q,and that they satisfy the following algebraic relation:(gmGm)8 (G8m − g8m) = 14 . (2.2.45)Since most tables only contain values of gm when m is even, and Gm when mis odd, our calculations will require (2.2.45). The simplest examples that wewill consider follow from equation (2.2.32), while equations (2.2.31), (2.2.33),and (2.2.34) lead to slightly more complicated results.Theorem 2.2.9. Suppose that m ∈ N, thenm(8ig8mG4m + y + y−1 + z + z−1)=16√mπ2∞∑n=1bnn2, (2.2.46)where ∞∑n=1bnqn = q∞∏n=1(1− q8n)3 (1− q4mn)2(1− q8mn) .The following table gives evaluations of 8g8mG4m, closed forms for∑∞n=1 bnqn,and states whether or not bn is multiplicative:m 8g8mG4m∞∑n=1bnqn Multiplicative?1 4 qf2(−q4) f2 (−q8) Yes2 4√2 + 2√2 qf5(−q8)f(−q16) No3 4(2 +√3)qf3(−q8)f2(−q12)f(−q24) No7 4(8 + 3√7)qf3(−q8)f2(−q28)f(−q56) No9 4(7 + 4 4√12 + 2 4√122 + 4√123)qf3(−q8)f2(−q36)f(−q72) No15 4(28 + 16√3 + 12√5 + 7√15)qf3(−q8)f2(−q60)f(−q120) No32Chapter 2. Lattice sums and Mahler measuresProof. Setting q = e−π√m reduces equation (2.2.20) toF(1,2)(√m)=π216√mm(8ig8mG4m + y + y−1 + z + z−1). (2.2.47)Therefore, we can obtain Mahler measure formulas by appealing to tablesof class invariants [40]. If m ∈ N, we can also use the definition of F(1,2)(x)to show thatF(1,2)(√m) =∞∑n=1bnn2,where bn has the stated generating function.The only remaining task is to check the values of 8g8mG4m. In particular,we can solve (2.2.45) to show that8g8mG4m = 4(G12m +√G24m − 1)= 4√2g6m√g12m +√g24m + 1.For example, since G1 = 1, it follows that 8g81G41 = 4. While this type ofargument naturally leads to equations involving nested radicals, many ofthose formulas simplify with sufficient effort.If we consider examples involving F(1,1)(x), then we can obtain two dis-tinct types of formulas. The first class of identities occurs when Im(a3(iq)/b3(iq))=0. In two of those cases the n2 term in (2.2.31) vanishes, yielding formulasthat reduce to generalized hypergeometric functions.Theorem 2.2.10. Let φ = 1+√52 denote the golden ratio, thenF(1,1) (1) =2π227n(3 3√2), (2.2.48)F(1,1)(1√5)=2√5π227n˜(33√φ). (2.2.49)Recall that n(k) is defined in equation (2.1.7).In the case when Im(a3(iq)/b3(iq)) = 0, we can establish many inter-esting formulas by setting q = e−π√ab . The next theorem provides twoexamples where b = 1 and a is a product of small primes.33Chapter 2. Lattice sums and Mahler measuresTheorem 2.2.11. Let ω = eπi/3 and recall that n2(k) is defined in Theorem2.2.7. We have:L(g, 2) = r12π281√3n2 (θ) = r2F(1,1)(√m), (2.2.50)for the following values of m, g(q), θ, r1, and r2:m g(q) θ r1 r29 qf3(−q2) f (−q18) 9250 (7− 19 3√2w − 2 3√4w2) 3 925 q7f3(−q6) f (−q150) 11+z6, where (z2+3z+1)3z6+1= 2 15 1We will conclude this section by pointing out that F(1,1)(x) can be ex-pressed as a four-dimensional lattice sum for all values of x:F(1,1)(x) = 16∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4[(6n1 + 1)2 + (6n2 + 1)2 + (6n3 + 1)2 + x2(6n4 + 1)2]2 .While this formula for F(1,1)(x) resembles the definition of F (b, c), we haveshown that F(1,1)(x) is much easier to understand. By equation (2.2.31)we can obtain hypergeometric formulas for F(1,1)(x) whenever x ∈ Q. Forinstance, applying equation (2.1.38) to formula (2.2.49), yields an interestinglattice sum identity:3456√15∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4[(6n1 + 1)2 + (6n2 + 1)2 + (6n3 + 1)2 + 15(6n4 + 1)2]2=C13√φ3F2(13, 13, 1323, 43;1φ)+C23√φ23F2(23, 23, 2343, 53;1φ),(2.2.51)where φ = 1+√52 is the golden ratio, C1 = 23√2Γ(12)Γ(13)Γ(16), and C2 =3√3Γ3(23). Equation (2.2.51) also resembles formulas that Forrester andGlasser established for three-dimensional sums associated with NaCl lattices[46].2.2.2 Remarks on F (1, 3) and higher values of F (b, c)Finally, we will speculate on how one might reduce higher values of F (b, c)to hypergeometric integrals. Our proof of the F (1, 3) formula will be in-34Chapter 2. Lattice sums and Mahler measuresstructive. Recall that Rodriguez-Villegas demonstrated that4π281n(−6) = Re⎛⎜⎜⎝12 ∑m,n∈Z(m,n) =(0,0)χ−3(n)(3(1+i√32)m + n)2 (3(1−i√32)m + n)⎞⎟⎟⎠ ,(2.2.52)and then used Deuring’s theorem to equate this Eisenstein series to the Lseries of a CM elliptic curve of conductor 27. A different proof could havebeen constructed from numerically observing thatq∞∏n=1(1− q3n)2 (1− q9n)2=142∑j=1χ−3(j)∞∑n,m=−∞[(6m + j) + 3(6n + j)] q(6m+j)2+3(6n+j)24 .(2.2.53)The modularity theorem implies that qf2(−q3) f2 (−q9) is associated tothe correct elliptic curve, hence the Mellin transform of the left-hand side ofequation (2.2.53) will equal L(E, s). Since the Mellin transform (at s = 2) ofthe right-hand side trivially equals the right-hand side of equation (2.2.52),it just remains to prove (2.2.53). By applying limiting cases of the triple andquintuple product identities, we can show that equation (2.2.53) is equivalentto an identity between eta functions:4qf2(−q3)f2 (−q9)=qf5(−q6) f (−q36) f2 (−q54)f2 (−q12) f (−q18) f (−q108) + 3qf(−q12) f7 (−q18)f (−q6) f3 (−q36)− 2q4 f2(−q3) f2 (−q12) f2 (−q18) f (−q27) f (−q108)f (−q6) f (−q9) f (−q36) f (−q54)− 6q4 f2(−q6) f3 (−q9) f3 (−q36)f (−q3) f (−q12) f2 (−q18) .(2.2.54)Identities between modular forms, such as equation (2.2.54), are usuallyestablished by checking that both sides of the proposed formula have thesame McLauren series expansion for sufficiently many terms. If we use theinversion formula for the eta function, then (2.2.54) can be viewed as anexample of a mixed modular equation, i.e. an algebraic relation between themoduli of isogenous elliptic curves. From this perspective, it seems likely35Chapter 2. Lattice sums and Mahler measuresthat the rest of Boyd’s conjectures will follow from discovering appropriatemixed modular equations.While it is probably unreasonable to expect to find useful series expan-sions for qf(−qA)f(−qAb)f(−qAc)f(−qAbc), when (b, c) ∈ {(1, 5), (1, 11), (2, 3),(2, 7), (3, 5)} and A = 24/ ((1 + b)(1 + c)), this does not seem to rule out thepossibility of finding such results for linear combinations of related eta prod-ucts. It might be interesting to examine whether or not a series expansionsuch as equation (2.2.53) exists for the following function:N∑j=1cjqjf(−qAj)f(−qAbj)f(−qAcj)f(−qAbcj), (2.2.55)for appropriate values of cj ∈ Q. Boyd’s conjectures would most likely followfrom such a theorem, as the Mellin transform (at s = 2) of (2.2.55) equalsa rational multiple of L(E, 2).2.3 Connections with the elliptic dilogarithmIn this section we will point out that the method from Section 2.2 can be usedto establish relations between values of F (b, c) and the elliptic dilogarithm.Definition 2.3.1. Recall that the elliptic dilogarithm is defined byL (x, q) =∞∑n=−∞D (xqn) ,where D(z) = Im (Li2(z) + log |z| log(1− z)) is the Bloch-Wigner diloga-rithm.In the previous section we integrated cusp forms to obtain identitiesbetween L-values and hypergeometric functions. For example, we used thefollowing two-dimensional series:qf2(−q4) f2 (−q8) = ∞∑n=−∞k=0(−1)n+k(2k + 1)q(2n)2+(2k+1)2 ,to recover formula (2.1.13) for F (1, 2). In general, comparing different seriesexpansions for the same cusp form will often lead to relations between hy-pergeometric functions and the elliptic dilogarithm. For instance, applying36Chapter 2. Lattice sums and Mahler measuresour method to the following identityqf2(−q4) f2 (−q8) = ∞∑n,k=0(−1)k(2k + 1)q (2k+1)2+(2n+1)22 ,yieldsF (1, 2) =π2L (i,−eπ) , (2.3.1)which is comparable in difficulty to equation (2.1.13).Theorem 2.3.2. Suppose that x ∈ R is sufficiently large and ω = e2πi/3,then the following identities are true:4∞∑n,k=0(−1)k(2k + 1)[(2n + 1)2 + x2(2k + 1)2]2=π2x3L (i,−eπx) , (2.3.2)∞∑n,k=−∞(3k + 1)[(2n + 1)2 + x2(3k + 1)2]2=π2√3x3L (ω,−eπx) . (2.3.3)We will conclude this section by briefly pointing out a pair of identitiesinvolving the L-functions of modular forms. If we define f1(q) and f2(q) byf1(q) :=q∞∏n=1(1− q8n)5(1− q16n) , f2(q) := q3∞∏n=1(1− q16n)5(1− q8n) ,then we have shown thatL (f1, 2) =π216√2m(4i√2 + 2√2 + y + y−1 + z + z−1),andL (f2, 2) =π16√2L(i,−eπ√2).Despite the fact that f1(q) and f2(q) are closely related, their L-values re-duce to special cases of different functions. We will hypothesize that onlyL-functions of modular forms with nice arithmetic properties should be ex-pressible in terms of both elliptic dilogarithms and Mahler measures.2.4 Higher lattice sums and conclusionPerhaps the moral of this paper is that lattice sums are difficult to deal with.While many hypergeometric formulas have been experimentally discovered37Chapter 2. Lattice sums and Mahler measuresfor F (b, c), only the simplest cases have been rigorously proved. Noticethat our formula for F (1, 4), equation (2.2.44), involves Meijer’s G-functiondisguised as a hypergeometric integral:Im⎛⎝∫ 1k2F1(12, 121; 1− u)udu⎞⎠ = 1π2Im(G3,23,3(k∣∣ 12 , 12 ,10,0,0)).This identity almost certainly rules out the possibility of expressing F (1, 4)as a Mahler measure, and it also indicates that any explicit formula forF (b, c) should reduce to Meijer G functions in certain instances.It seems difficult to conjecture what types of formulas should exist forhigher dimensional lattice sums. Let us consider the k-dimensional latticesumF(j)k (x1, . . . , xk) :=∞∑ni=−∞(−1)n1+···+nk(x1(6n1 + 1)2 + · · ·+ xk(6nk + 1)2)j,which arises from integrating eta products. Many non-trivial linear depen-dencies exist between different values of F (j)k (x1, . . . , xk) when k  4. Forexample, clearing denominators and then integrating formula (2.2.26) leadsto a linear dependence between 13-dimensional lattice sums:F(j)13⎛⎝2, . . . , 2︸ ︷︷ ︸8, 3, . . . , 3︸ ︷︷ ︸4, 6⎞⎠ =F (j)13⎛⎝1, 2, . . . , 2︸ ︷︷ ︸4, 3, . . . , 3︸ ︷︷ ︸8⎞⎠+ F (j)13⎛⎝1, . . . , 1︸ ︷︷ ︸4, 6, . . . , 6︸ ︷︷ ︸9⎞⎠ .Although we have only proved hypergeometric formulas for F (j)k when k ≤ 6,and j ∈ {2, 3}, it might be interesting to search for higher dimensional exam-ples numerically. Lattice sums with Euler products are the best candidates(F (b, c) belongs to this class of functions when (1+b)(1+c) divides 24), sincethey seem to be the only lattice sums which ever reduce to hypergeometricfunctions with rational arguments. To illustrate this principle, we can applythe methods of Section 2.2 to show thatF(2)6 (1, 1, 1,m,m,m) =π264√mG12m3F2(12, 12, 121,1,1G24m), (2.4.1)38Chapter 2. Lattice sums and Mahler measureswhere Gm is the usual class invariant. For m ∈ {1, 3, 7}, we have the follow-ing table:m g(q) L(g, 2)1 q∏∞n=1(1− q4n)6 π216 3F2 ( 12 , 12 , 121,1 , 1)3 q∏∞n=1(1− q2n)3 (1− q6n)3 π28√33F2(12, 12, 121,1, 14)7 q∏∞n=1 (1− qn)3(1− q7n)3 π28√73F2(12, 12, 121,1, 164)In each of these three examples, g(q) is a multiplicative weight 3 cusp form.On the other hand, if we consider a similar looking, but non-multiplicativeeta product:g1(q) = q∞∏n=1(1− q3n)3 (1− q5n)3 ,then equation (2.4.1) gives a formula for L(g1, 2) involving G245/3. Since G5/3is a root of the following polynomial equation:0 = x12 − 8 4√8(1 +√52)3x9 + 4 4√2(1 +√52)x3 + 8(1 +√52)4,it is not hard to see that G245/3 is an irrational algebraic number.2.5 AcknowledgementsThe author would like to thank David Boyd, Wadim Zudilin, and M. LawrenceGlasser for their useful comments and kind encouragement.39Bibliography[33] V. S. Adamchik, A certain series associated with Catalan’s constant.Z. Anal. Anwendungen 21 (2002), no. 3, 817–826.[34] S. Ahlgren, B. C. Berndt, A. J. Yee, and A. Zaharescu, Integrals ofEisenstein series and derivatives of L-functions, IMRN, (2002), no. 32,1723-1738.[35] G. E. Andrews and B. C. Berndt, Ramanujan’s Lost Notebook, Part I,Springer-Verlag, New York, 2005.[36] B. C. Berndt, Ramanujan’s Notebooks, Part I, Springer-Verlag, NewYork, 1985.[37] B. C. Berndt, Ramanujan’s Notebooks, Part II, Springer-Verlag, NewYork, 1989.[38] B. C. Berndt, Ramanujan’s Notebooks, Part III, Springer-Verlag, NewYork, 1991.[39] B. C. Berndt, Ramanujan’s Notebooks, Part IV, Springer-Verlag, NewYork, 1994.[40] B. C. Berndt, Ramanujan’s Notebooks, Part V, Springer-Verlag, NewYork, 1998.[41] D. W. Boyd, Mahler’s measure and special values of L-functions, Ex-periment. Math. 7 (1998), 37-82. Academic Press, 1994.[42] F. Brunault, explicite du thorme de Beilinson pour la courbe modulaireX1(N). C. R. Math. Acad. Sci. Paris 343 (2006), no. 8, 505–510.[43] L. Carlitz, Note on some partition formulae. Quart. J. Math., OxfordSer. (2) 4, (1953). 168–172.[44] C. Deninger, Deligne periods of mixed motives, K-theory and the en-tropy of certain Zn-actions, J. Amer. Math. Soc. 10 (1997), no. 2,259–281.40Bibliography[45] S. Finch, Modular forms on SL2(Z), preprint, (2005).[46] P. J. Forrester and M. L. Glasser, Some new lattice sums including anexact result for the electrostatic potential within the NaCl lattice. J.Phys. A: Math. Gen. 15(1982), 911-914.[47] N. N. Lebedev and R. A. Silverman, Special functions & their applica-tions, Dover Publications Inc., New York, 1972.[48] M. N. Lal´ın and M. D. Rogers, Functional equations for Mahler mea-sures of genus-one curves, Algebra and Number Theory, 1 (2007), no.1, 87–117.[49] Y. Martin and K. Ono, Eta-Quotients and Elliptic Curves, Proc. Amer.Math Soc. (125) 1997, no. 11, 3169-3176.[50] F. Rodriguez-Villegas, Identities between Mahler measures, Numbertheory for the millennium, III (Urbana, IL, 2000), 223–229, A K Peters,Natick, MA, 2002.[51] F. Rodriguez-Villegas, Modular Mahler measures I, Topics in numbertheory (University Park, PA, 1997), 17–48, Math. Appl., 467, KluwerAcad. Publ., Dordrecht, 1999.[52] M. D. Rogers, New 5F4 hypergeometric transformations, three-variableMahler measures, and formulas for 1/π, to appear in the Ramanujanjournal.Department of Mathematics, University of British Columbia,Vancouver, B.C. V6T 1Z2, Canadamatrogers@math.ubc.ca41Chapter 3Functional equations forMahler measuresMatilde N. Lal´ın and Mathew D. Rogers23.1 History and introductionThe goal of this paper is to establish identities between the logarithmicMahler measures of polynomials with zero varieties corresponding to genus-one curves. Recall that the logarithmic Mahler measure (which we shallhenceforth simply refer to as the Mahler measure) of an n-variable Laurentpolynomial P (x1, x2, . . . , xn) is defined bym (P (x1, . . . , xn)) =∫ 10. . .∫ 10log∣∣P (e2πiθ1 , . . . , e2πiθn) ∣∣dθ1 . . .dθn.Many difficult questions surround the special functions defined by Mahlermeasures of elliptic curves.The first example of the Mahler measure of a genus-one curve was studiedby Boyd [63] and Deninger [65]. Boyd found thatm(1 + x +1x+ y +1y)?= L′(E, 0), (3.1.1)where E denotes the elliptic curve of conductor 15 that is the projectiveclosure of 1+x+ 1x + y+1y = 0. As usual, L(E, s) is its L-function, and thequestion mark above the equals sign indicates numerical equality verified upto 60 decimal places.Deninger [65] gave an interesting interpretation of this formula. He ob-tained the Mahler measure by evaluating the Bloch regulator of an element{x, y} from a certain K-group. In other words, the Mahler measure is given2A version of this chapter has been published. Lal´ın, M. N. and Rogers, M. D. (2007)Functional equations for Mahler measures of genus-one curves. Algebra and NumberTheory 1:87-117.42Chapter 3. Functional equations for Mahler measuresby a value of an Eisenstein-Kronecker series. Therefore Bloch’s and Beilin-son’s conjectures predict thatm(1 + x +1x+ y +1y)= cL′(E, 0),where c is some rational number. Let us add that, even if Beilinson’s conjec-tures were known to be true, this would not suffice to prove equality (3.1.1),since we still would not know the height of the rational number c.This picture applies to other situations as well. Boyd [63] performedextensive numerical computations within the family of polynomials k + x+1x +y+1y , as well as within some other genus-one families. Boyd’s numericalsearches led him to conjecture identities such asm(5 + x +1x+ y +1y)?= 6m(1 + x +1x+ y +1y),m(8 + x +1x+ y +1y)?= 4m(2 + x +1x+ y +1y).Boyd conjectured conditions predicting when formulas like Eq. (3.1.1) shouldexist for the Mahler measures of polynomials with integral coefficients. Thiswas further studied by Rodriguez-Villegas [71] who interpreted these con-ditions in the context of Bloch’s and Beilinson’s conjectures. Furthermore,Rodriguez-Villegas used modular forms to express the Mahler measures asKronecker-Eisenstein series in more general cases. In turn, this allowed himto prove some equalities such asm(4√2 + x +1x+ y +1y)= L′(E4√2, 0), (3.1.2)m(3√2 + x +1x+ y +1y)=52L′(E3√2, 0), (3.1.3)where Ek denotes the projective closure of k + x+ 1x + y +1y = 0. The firstequality can be proved using the fact that the corresponding elliptic curvehas complex multiplication, and therefore the conjectures are known for thiscase due to Bloch [62]. The second equality depends on the fact that onehas the modular curve X0(24), and the conjectures then follow from a resultof Beilinson.Rodriguez-Villegas [72] subsequently used the relationship between Mahlermeasures and regulators to prove a conjecture of Boyd [63]:m(y2 + 2xy + y − x3 − 2x2 − x) = 57m(y2 + 4xy + y − x3 + x2).43Chapter 3. Functional equations for Mahler measuresIt is important to point out that he proved this identity without actuallyexpressing the Mahler measures in terms of L-series. Bertin [61] has alsoproved similar identities using these ideas.Although the conjecture in Eq. (3.1.1) remains open, we will in factprove two of Boyd’s other conjectures in this paper.Theorem 3.1.1. The following identities are true:m(2 + x +1x+ y +1y)=L′(E3√2, 0), (3.1.4)m(8 + x +1x+ y +1y)=4L′(E3√2, 0). (3.1.5)Our proof of Theorem 3.1.1 follows from combining two interesting “func-tional equations” for the functionm(k) := m(k + x +1x+ y +1y).Kurokawa and Ochiai [67] recently proved the first functional equation. Theyshowed that if k ∈ R\{0}:m(4k2)+ m(4k2)= 2m(2(k +1k)). (3.1.6)In Section 3.3 we use regulators to give a new proof of Eq. (3.1.6). We willalso prove a second functional equation in Section 3.2.1 using q-series. Inparticular, if k is nonzero and |k| < 1:m(2(k +1k))+ m(2(ik +1ik))= m(4k2). (3.1.7)Theorem 3.1.1 follows from setting k = 1/√2 in both identities, and thenshowing that 5m(i√2)= 3m(3√2). We have proved this final equality inSection 3.3.6.This paper is divided into two sections of roughly equal length. In Sec-tion 3.2 we will prove more identities like Eq. (3.1.7), which arise fromexpanding Mahler measures in q-series. In particular, we will look at identi-ties for four special functions defined by the Mahler measures of genus-onecurves (see equations (3.2.1) through (3.2.4) for notation). Equation (3.2.14)undoubtedly constitutes the most important result in this part of the pa-per, since it implies that infinitely many identities like Eq. (3.1.7) exist.44Chapter 3. Functional equations for Mahler measuresSubsections 3.2.1 and 3.2.2 are mostly devoted to transforming special casesof Eq. (3.2.14) into interesting identities between the Mahler measures ofrational polynomials. While the theorems in those subsections rely heav-ily on Ramanujan’s theory of modular equations to alternative bases, wehave attempted to maximize readability by eliminating q-series manipula-tion wherever possible. Finally, we have devoted Subsection 3.2.3 to provingsome useful computational formulas. As a corollary we establish several newtransformations for hypergeometric functions, including:∞∑n=0(k(1− k)2(1 + k)2)n n∑j=0(nj)2(n + jj)=(1 + k)2√(1 + k2)((1− k − k2)2 − 5k2)× 2F1⎛⎜⎝14 , 34; 1; 64k5(1 + k − k2)(1 + k2)2((1− k − k2)2 − 5k2)2⎞⎟⎠ .(3.1.8)We have devoted Section 3.3 to further studying the relationship be-tween Mahler measures and regulators. We show how to recover the Mahlermeasure q-series expansions and the Kronecker-Einsenstein series directlyfrom Bloch’s formula for the regulator. This in turn shows that the Mahlermeasure identities can be viewed as consequences of functional identities forthe elliptic dilogarithm.Many of the identities in this paper can be interpreted from both aregulator perspective, and from a q-series perspective. The advantage of theq-series approach is that it simplifies the process of finding new identities.The fundamental result in Section 3.2, Eq. (3.2.14), follows easily from theMahler measure q-series expansions. Unfortunately the q-series approachdoes not provide an easy way to explain identities like Eq. (3.1.6). Unlikemost of the other formulas in Section 3.2, Kurokawa’s and Ochiai’s resultdoes not follow from Eq. (3.2.14). An advantage of the regulator approach,is that it enables us to construct proofs of both Eq. (3.1.6) and Eq. (3.1.7)from a unified perspective. Additionally, the regulator approach seems toprovide the only way to prove the final step in Theorem 3.1.1, namely toshow that 5m(i√2)= 3m(3√2). Thus, a complete view of this subjectmatter should incorporate both regulator and q-series perspectives.45Chapter 3. Functional equations for Mahler measures3.2 Mahler measures and q-seriesIn this paper we will consider four important functions defined by Mahlermeasures:μ(t) =m(4√t+ x +1x+ y +1y), (3.2.1)n(t) =m(x3 + y3 + 1− 3t1/3xy), (3.2.2)g(t) =m((x + y)(x + 1)(y + 1)− 1txy), (3.2.3)r(t) =m((x + y + 1)(x + 1)(y + 1)− 1txy). (3.2.4)Throughout Section 3.2 will use the notation μ(t) = m(4√t)for conve-nience. Recall from [71] and [76], that each of these functions has a simpleq-series expansion when t is parameterized correctly. To summarize, if welet (x; q)∞ = (1− x)(1− xq)(1− xq2) . . . , andM(q) = 16q(q; q)8∞(q4; q4)16∞(q2; q2)24∞, (3.2.5)N(q) =27q(q3; q3)12∞(q; q)12∞ + 27q (q3; q3)12∞, (3.2.6)G(q) = q1/3(q; q2)∞(q3; q6)3∞, (3.2.7)R(q) = q1/5(q; q5)∞(q4; q5)∞(q2; q5)∞ (q3; q5)∞, (3.2.8)then for |q| sufficiently smallμ (M(q)) =− Re⎡⎣12log(q) + 2∞∑j=1jχ−4(j) log(1− qj)⎤⎦ , (3.2.9)n (N(q)) =− Re⎡⎣13log(q) + 3∞∑j=1jχ−3(j) log(1− qj)⎤⎦ , (3.2.10)g(G3(q))=− Re⎡⎣log(q) + ∞∑j=1(−1)j−1jχ−3(j) log(1− qj)⎤⎦ , (3.2.11)46Chapter 3. Functional equations for Mahler measuresr(R5(q))=− Re⎡⎣log(q) + ∞∑j=1jRe [(2− i)χr(j)] log(1− qj)⎤⎦ . (3.2.12)In particular, χ−3(j) and χ−4(j) are the usual Dirichlet characters, andχr(j) is the character of conductor five with χr(2) = i. We have used thenotation G(q) and R(q), as opposed to something like G˜(q) = G3(q), inorder to preserve Ramanujan’s notation. As usual, G(q) corresponds toRamanujan’s cubic continued fraction, and R(q) corresponds to the Rogers-Ramanujan continued fraction [53].The first important application of the q-series expansions is that they canbe used to calculate the Mahler measures numerically. For example, we cancalculate μ (1/10) with Eq. (3.2.9), provided that we can first determinea value of q for which M(q) = 1/10. Fortunately, the theory of ellipticfunctions shows that if α = M(q), thenq = exp⎛⎝−π 2F1(12, 121; 1− α)2F1(12, 121;α)⎞⎠ . (3.2.13)Using Eq. (3.2.13) we easily compute q = .01975 . . . , and it follows thatμ(1/10) = 2.524718 . . . The function defined in Eq. (3.2.13) is called theelliptic nome, and is sometimes denoted by q2(α). Theorem 3.2.6 providessimilarly explicit inversion formulas for Eqs. (3.2.5) through (3.2.8).The second, and perhaps more significant fact that follows from theseq-series, is that linear dependencies exist between the Mahler measures. Inparticular, iff(q) ∈ {μ (M(q)) , n (N(q)) , g (G3(q)) , r (R5(q))} ,then for an appropriate prime pp−1∑j=0f(e2πij/pq)= (1 + p2χ(p))f (qp)− pχ(p)f(qp2), (3.2.14)where χ(j) is the character from the relevant q-series. The prime p satisfiesthe restriction that p = 2 when f(q) = g (G3(q)), and p ≡ 2, 3 (mod 5) whenf(q) = r(R5(q)). The astute reader will immediately recognize that Eq.(3.2.14) is essentially a Hecke eigenvalue equation. A careful analysis of theexceptional case that occurs when p = 2 and f(q) = g(G3(q))leads to theimportant and surprising inverse relation:3n (N(q)) = g(G3 (q))− 8g (G3 (−q))+ 4g (G3 (q2)) ,3g(G3(q))= n (N(q)) + 4n(N(q2)).(3.2.15)47Chapter 3. Functional equations for Mahler measuresIn the next two subsections we will discuss methods for transforming Eq.(3.2.14) and Eq. (3.2.15) into so-called functional equations.3.2.1 Functional equations from modular equationsSince the primary goal of this paper is to find relations between the Mahlermeasures of rational (or at least algebraic) polynomials, we will require mod-ular equations to simplify our results. For example, consider Eq. (3.2.14)when f(q) = μ (M(q)) and p = 2:μ (M(q)) + μ (M(−q)) = μ (M (q2)) . (3.2.16)For our purposes, Eq. (3.2.16) is only interesting if M(q), M(−q), andM(q2)are all simultaneously algebraic. Fortunately, it turns out that M(q)and M(q2)(hence also M(−q) and M (q2)) satisfy a well known polynomialrelation.Definition 3.2.1. Suppose that F (q) ∈ {M(q), N(q), G(q), R(q)}. An n’thdegree modular equation is an algebraic relation between F (q) and F (qn).We will not need to derive any new modular equations in this paper.Berndt proved virtually all of the necessary modular equations while editingRamanujan’s notebooks, see [53], [56], [57], and [58]. Ramanujan seems tohave arrived at most of his modular equations through complicated q-seriesmanipulations (of course this is speculation since he did not write down anyproofs!). Modular equations involving M(q) correspond to the classical mod-ular equations [57], relations for N(q) correspond to Ramanujan’s signaturethree modular equations [58], and most of the known modular equations forG(q) and R(q) appear in [53].Now we can finish simplifying Eq. (3.2.16). Since the classical second-degree modular equation shows that whenever |q| < 14M(q2)(1 + M (q2))2=(M(q)M(q)− 2)2,we easily obtain the parameterizations: M(q) = 4k2(1+k2)2, M(−q) = −4k2(1−k2)2 ,and M(q2)= k4. Substituting these parametric formulas into Eq. (3.2.16)yields:48Chapter 3. Functional equations for Mahler measuresTheorem 3.2.2. The following identity holds whenever |k| < 1:m(4k2+ x +1x+ y +1y)=m(2(k +1k)+ x +1x+ y +1y)+m(2i(k − 1k)+ x +1x+ y +1y).(3.2.17)We need to make a few remarks about working with modular equationsbefore proving the main theorem in this section. Suppose that for somealgebraic function P (X,Y ):P (F (q), F (qp)) = 0,where F (q) ∈ {M(q), N(q), G(q), R(q)}. Using the elementary change ofvariables, q → e2πij/pq, it follows that P (F (e2πij/pq) , F (qp)) = 0 for everyj ∈ {0, 1, . . . , p − 1}. If P (X,Y ) is symmetric in X and Y , it also followsthat P(F(qp2), F (qp))= 0. Therefore, if P (X,Y ) is sufficiently simple(for example a symmetric genus-zero polynomial), we can find simultaneousparameterizations for F (qp), F(qp2), and F(e2πij/pq)for all j. In suchan instance, Eq. (3.2.14) reduces to an interesting functional equation forone of the four Mahler measures {μ(t), n(t), g(t), r(t)}. Five basic functionalequations follow from applying these ideas to Eq. (3.2.14).Theorem 3.2.3. For |k| < 1 and k = 0, we haveμ(4k2(1 + k2)2)+ μ( −4k2(1− k2)2)= μ(k4). (3.2.18)The following identities hold for |u| sufficiently small but non-zero:n(27u(1 + u)42 (1 + 4u + u2)3)+n(− 27u(1 + u)2 (1− 2u− 2u2)3)= 2n(27u4(1 + u)2 (2 + 2u− u2)3)− 3n(27u2(1 + u)24 (1 + u + u2)3).(3.2.19)If ζ3 = e2πi/3, and Y (t) = 1−(1−t1+2t)3, thenn(u3)=2∑j=0n(Y(ζj3u)). (3.2.20)49Chapter 3. Functional equations for Mahler measuresIf ζ3 = e2πi/3, and Y (t) = t(1−t+t21+2t+4t2), theng(u3) =2∑j=0g(Y(ζj3u)). (3.2.21)If ζ5 = e2πi/5, and Y (t) = t(1−2t+4t2−3t3+t41+3t+4t2+2t3+t4), thenr(u5)=4∑j=0r(Y(ζj5u)). (3.2.22)Proof. We have already sketched a proof of Eq. (3.2.18) in the discussionpreceding Theorem 3.2.2.The proof of Eq. (3.2.19) requires the second-degree modular equa-tion from Ramanujan’s theory of signature three. If β = N(q2), andα ∈ {N(q), N(−q), N (q4)}, then27αβ(1− α)(1− β)− (α + β − 2αβ)3 = 0. (3.2.23)If we choose u so that N(q2)= 27u2(1+u)24(1+u+u2)3, then we can use Eq. (3.2.23)to easily verify that N(q) = 27u(1+u)42(1+4u+u2)3, N(−q) = − 27u(1+u)2(1−2u−2u2)3 , andN(q4)= 27u4(1+u)2(2+2u−u2)3 . The proof of Eq. (3.2.19) follows from applyingthese parameterizations to Eq. (3.2.14) when f(q) = n (N(q)), and p = 2.The proof of Eq. (3.2.20) requires Ramanujan’s third-degree, signaturethree modular equation. In particular, if α = N(q) and β = N(q3), thenα = 1−(1− β1/31 + 2β1/3)3= Y(β1/3). (3.2.24)Since N1/3(q3) = q × {power series in q3}, a short computation shows thatN(ζj3q) = Y(ζj3N1/3(q3))for all j ∈ {0, 1, 2}. Choosing u such thatN(q3) = u3, we must have N(ζj3q)= Y(ζj3u). Eq. (3.2.20) follows fromapplying these parametric formulas to Eq. (3.2.14) when f(q) = n (N(q)),and p = 3.Since the proofs of equations (3.2.21) and (3.2.22) rely on similar argu-ments to the proof of Eq. (3.2.20), we will simply state the prerequisitemodular equations. In particular, Eq. (3.2.21) follows from Ramanujan’s50Chapter 3. Functional equations for Mahler measuresthird-degree modular equation for the cubic continued fraction. If α = G(q)and β = G(q3), thenα3 = β(1− β + β21 + 2β + 4β2). (3.2.25)Similarly, Eq. (3.2.22) follows from the fifth-degree modular equation forthe Rogers-Ramanujan continued fraction. In particular, if α = R(q) andβ = R(q5)α5 = β(1− 2β + 4β2 − 3β3 + β41 + 3β + 4β2 + 2β3 + β4). (3.2.26)The functional equations in Theorem 3.2.3 only hold in restricted subsetsof C. To explain this phenomenon we will go back to Eq. (3.2.14). As ageneral rule, we have to restrict q to values for which none of the Mahlermeasure integrals in Eq. (3.2.14) vanish on the unit torus. In other words,we can only consider the set of q’s for which each term in Eq. (3.2.14)can be calculated from the appropriate q-series. Next, we may need tofurther restrict the domain of q depending on where the relevant parametricformulas hold. For example, parameterizations such as N(q) = 27u(1+u)42(1+4u+u2)3and N(q2) = 27u2(1+u)24(1+u+u2)3hold for |q| sufficiently small, but fail when q isclose to 1. After determining the domain of q, we can calculate the domainof u by solving a parametric equation to express u in terms of a q-series.Theorem 3.2.4. Recall that G(q) is defined in equation (4.2.1). For |p|sufficiently small but non-zero3g(p) = n(27p(1 + 4p)3)+ 4n(27p2(1− 2p)3). (3.2.27)Furthermore, for |u| sufficiently small but non-zero3n(27u(1 + u)42(1 + 4u + u2)3)=g(u2(1 + u)2)− 8g(−u(1 + u)2)+ 4g(u24(1 + u)).(3.2.28)Proof. We will prove Eq. (3.2.28) first. Recall that Eq. (3.2.15) shows that3n (N(q)) = g(G3(q))− 8g (G3(−q))+ 4g (G3 (q2)) .51Chapter 3. Functional equations for Mahler measuresLet us suppose that q = q2(u(2+u)3(1+2u)3), where q2(α) is the elliptic nome. Clas-sical eta function inversion formulas (which we shall omit here) show that for|u| sufficiently small: G3(q) = u2(1+u)2, G3(−q) = −u(1+u)2 , G3(q2)= u24(1+u) ,N(q) = 27u(1+u)42(1+4u+u2)3, and N(q2)= 27u2(1+u)24(1+u+u2)3.To prove Eq. (3.2.27) first recall recall that3g(G3(q))= n (N(q)) + 4n(N(q2)).If we let p = u2(1+u)2, then it follows that G3(q) = p, N(q) = 27p(1+4p)3, andN(q2)= 27p2(1+2p)3.Theorem 3.2.4 shows that g(t) and n(t) are essentially interchangeable.In Section 3.2.3 we will use Eq. (3.2.27) to derive an extremely useful formulafor calculating g(t) numerically.3.2.2 Identities arising from higher modular equationsThe seven functional equations presented in Section 3.2.1 are certainly notthe only interesting formulas that follow from Eq. (3.2.14). Rather thoseresults represent the subset of functional equations in which every Mahlermeasure depends on a rational argument (possibly in a cyclotomic field). Ifwe consider the higher modular equations, then we can establish formulasinvolving the Mahler measures of the modular polynomials themselves. Eq.(3.2.32) is the simplest formula in this class of results.Consider Eq. (3.2.14) when p = 3 and f(q) = μ (M(q)):2∑j=0μ(M(ζj3q))= −8μ (M (q3))+ 3μ (M (q9)) . (3.2.29)The third-degree modular equation shows that ifα ∈ {M (q) ,M (ζ3q) ,M (ζ23q) ,M (q9)}, and β = M (q3), thenG3(α, β) := (α2+β2+6αβ)2−16αβ (4(1 + αβ)− 3(α + β))2 = 0. (3.2.30)Since G3(α, β) = 0 defines a curve with genus greater than zero, it is im-possible to find simultaneous rational parameterizations for all four zeros inα. For example, if we let β = M(q3) = p(2 + p)3/(1 + 2p)3, then we canobtain the rational expression M(q9)= p3(2+p)/(1+2p), and three messy52Chapter 3. Functional equations for Mahler measuresformulas involving radicals for the other zeros. Despite this difficulty, Eq.(3.2.29) still reduces to an interesting formula if we recall the factorizationG3(α,M(q3))=(α−M (q9)) 2∏j=0(α−M(ζj3q)), (3.2.31)and then use the fact that Mahler measure satisfies m(P )+m(Q) = m (PQ).Theorem 3.2.5. If G3(α, β) is defined in Eq. (3.2.30), then for |p| suffi-ciently small but non-zerom(G3((x + x−1)2 (y + y−1)216,1p(1 + 2p2 + p)3))= −16 log(2)− 16μ(p(2 + p1 + 2p)3)+ 8μ(p3(2 + p1 + 2p)).(3.2.32)Proof. First notice that from the elementary properties of Mahler’s measureμ(t) =12m(16(x + x−1)2 (y + y−1)2− t)− 12log |t|.Applying this identity to Eq. (3.2.29), and then appealing to Eq. (3.2.31)yieldsm(G3(16(x + x−1)2 (y + y−1)2,M(q3)))= log∣∣M (q)M (ζ3q)M (ζ23q)M (q9)∣∣− 16μ (M (q3))+ 8μ (M (q9)) .Elementary q-product manipulations show thatM4(q3)= M (q)M (ζ3q)M(ζ23q)M(q9), and since α4β4G3(1α ,1β)= G3(α, β),we obtainm(G3((x + x−1)2 (y + y−1)216,1M (q3)))= −16 log(2)− 16μ (M (q3))+ 8μ (M (q9)) .Finally, if we choose p so that M(q3)= p(2+p1+2p)3, then M(q9)= p3(2+p1+2p),and the theorem follows.53Chapter 3. Functional equations for Mahler measuresAlthough we completely eliminated the q-series expressions from Eq.(3.2.32), this is not necessarily desirable (or even possible) in more com-plicated examples. Consider the identity involving resultants which fol-lows from Eq. (3.2.14) (and some manipulation) when p = 11 and f(q) =r(R5(q)):m(Resz[z5 − xy(x + 1)(y + 1)(x + y + 1), P(z,R5 (q))])= −12m (1 + x + y) + 12 log ∣∣R5 (q)∣∣+ 122r (R5 (q))− 11r (R5 (q11)) .(3.2.33)In this formula P (u, v) is the polynomialP (u, v) = uv(1− 11v5 − v10)(1− 11u5 − u10)− (u− v)12,which also satisfies P(R(q), R(q11))= 0 [73]. Even if rational parameter-izations existed for R(q) and R(q11), substituting such formulas into Eq.(3.2.33) would probably just make the identity prohibitively complicated.3.2.3 Computationally useful formulas, and a few relatedhypergeometric transformationsWhile many methods exist for numerically calculating each of the fourMahler measures {μ(t), n(t), g(t), r(t)}, two simple and efficient methodsare directly related to the material discussed so far.The first computational method relies on the q-series expansions. Forexample, we can calculate μ(α) with Eq. (3.2.9), provided that a value of qexists for which M(q) = α. Amazingly, the elliptic nome function, definedin Eq. (3.2.13), furnishes a value of q whenever |α| < 1. Similar inversionformulas exist for all of the q-products in equations (3.2.5) through (3.2.8).Suppose that for j ∈ {2, 3, 4, 6}qj(α) = exp⎛⎝− πsin(πj) 2F1(1j , 1− 1j ; 1; 1− α)2F1(1j , 1− 1j ; 1;α)⎞⎠ , (3.2.34)then we have the following theorem:Theorem 3.2.6. With α and q appropriately restricted, the following tablegives inversion formulas for equations (3.2.5) through (3.2.8):54Chapter 3. Functional equations for Mahler measuresα qM(q) q2(α)N(q) q3(α)G(q) q2(u(2+u)3(1+2u)3), where α3 = u2(1+u)2R(q) q4(64k(1+k−k2)5(1+k2)2((1+11k−k2)2−125k2)2), where α5 = k(1−k)2(1+k)2For example: If |q| < 1 and α = M(q), then q = q2(α).Proof. The inversion formulas for M(q) and G(q) follow from classical etafunction identities, and the inversion formula for N(q) follows from eta func-tion identities in Ramanujan’s theory of signature three.The inversion formula for R(q) seems to be new, so we will prove it. Letus suppose that α = R(q) and k = R(q)R2(q2), where q is fixed. A formulaof Ramanujan [53] shows that α5 = k(1−k)2(1+k)2, which establishes the secondpart of the formula. Now suppose that q = q2(α2), where α2 = M(q). Aclassical identity shows thatq (−q; q)24∞ =α216(1− α2)2 ,and comparing this to Ramanujan’s identityq (−q; q)24∞ =(k1− k2)(1 + k − k21− 4k − k2)5,we deduce thatα2(1− α2)2 = 16(k1− k2)(1 + k − k21− 4k − k2)5. (3.2.35)Now recall that the theory of the signature 4 elliptic nome shows thatq = q2(α2) = q4(4α2(1 + α2)2)= q4(4α2/(1− α2)21 + 4α2/(1− α2)2).Substituting Eq. (3.2.35) into this final result yieldsq = q4⎛⎜⎝ 64k (1 + k − k2)5(1 + k2)2((1 + 11k − k2)2 − 125k2)2⎞⎟⎠ ,which completes the proof.55Chapter 3. Functional equations for Mahler measuresThe second method for calculating the four Mahler measures,{μ(t), n(t), g(t), r(t)}, depends on reformulating them in terms of hypergeo-metric functions. For example, Rodriguez-Villegas proved [71] the formulaμ(t) = −12Re[log(t/16) +∫ t02F1(12 ,12 ; 1;u)− 1udu].Translated into the language of generalized hypergeometric functions, thisbecomesμ(t) = −Re[t84F3(32, 32,1,12,2,2; t)+12log(t/16)]. (3.2.36)He also proved a formula for n(t) which is equivalent ton(t) = −Re[2t274F3(43, 53,1,12,2,2; t)+13log(t/27)]. (3.2.37)Formulas like Eq. (3.2.36) and Eq. (3.2.37) hold some obvious appeal.From a computational perspective they are useful because most mathematicsprograms have routines for calculating generalized hypergeometric functions.For example, when |t| < 1 the Taylor series for the 4F3 function easily givesbetter numerical accuracy than the Mahler measure integrals. CombiningEq. (3.2.37) with Eq. (3.2.27) also yields a useful formula for calculatingg(t) whenever |t| is sufficiently small:g(t) =− Re[2t(1 + 4t)3 4F3(43, 53,1,12,2,2;27t(1 + 4t)3)+8t2(1− 2t)3 4F3(43, 53,1,12,2,2;27t2(1− 2t)3)+ log(t3(1 + 4t)(1− 2t)4)].(3.2.38)So far we have been unable to to find a similar expression for r(t).Open Problem 2: Express r(t) in terms of generalized hypergeometricfunctions.Besides their computational importance, identities like Eq. (3.2.36) allowfor a reformulation of Boyd’s conjectures in the language of hypergeometricfunctions. For example, the conjecturem(1 + x +1x+ y +1y)?= L′ (E, 0) ,56Chapter 3. Functional equations for Mahler measureswhere E is an elliptic curve with conductor 15, becomesL′(E, 0) ?= −2Re[4F3(32, 32,1,12,2,2; 16)].A proof of this identity would certainly represent an important addition tothe vast literature concerning transformations and evaluations of generalizedhypergeometric functions.In the remainder of this section we will apply our results to deduce afew interesting hypergeometric transformations. For example, differentiat-ing Eq. (3.2.38) leads to an interesting corollary:Corollary 3.2.7. For |t| sufficiently smallω(t) :=∞∑n=0tnn∑k=0(nk)3=11− 2t2F1(13,23; 1;27t2(1− 2t)3), (3.2.39)furthermoreω(p2(1 + p)2)= (1 + p)ω(p24(1 + p)), (3.2.40)whenever |p| is sufficiently small.Proof. We can prove Eq. (3.2.39) by differentiating each side of Eq. (3.2.38),and then by appealing to Stienstra’s formulas [76]. A second possible prooffollows from showing that both sides of Eq. (3.2.39) satisfy the same differ-ential equation.The shortest proof of Eq. (3.2.40) follows from a formula due to Zagier[76]:ω(G3(q))=∞∏n=0(1− q2n) (1− q3n)6(1− qn)2 (1− q6n)3 .First use Zagier’s identity to verify thatG2(q)ω(G3(q))= G(q2)ω(G3(q2)),and then apply the parameterizations for G3(q) and G3(q2)from Theorem3.2.4.We will also make a few remarks about the derivative of r(t). Stienstrahas shown thatr(t) = −Re[log(t) +∫ t0φ(u)− 1udu], (3.2.41)where φ(t) is defined byφ(t) =∞∑n=0tnn∑k=0(nk)2(n + kk). (3.2.42)57Chapter 3. Functional equations for Mahler measuresEven though we have not discovered a formula for r(t) involving hyperge-ometric functions, we can still express φ(t) in terms of the hypergeometricfunction.Theorem 3.2.8. Let φ(t) be defined by Eq. (3.2.42), then for |k| sufficientlysmall:φ(k(1− k1 + k)2)=(1 + k)2√(1 + k2)((1− k − k2)2 − 5k2)× 2F1⎛⎜⎝14 , 34; 1; 64k5(1 + k − k2)(1 + k2)2((1− k − k2)2 − 5k2)2⎞⎟⎠ ,(3.2.43)φ(k2(1 + k1− k))=(1− k)√(1 + k2)((1 + 11k − k2)2 − 125k2)× 2F1⎛⎜⎝14 , 34; 1; 64k(1 + k − k2)5(1 + k2)2((1 + 11k − k2)2 − 125k2)2⎞⎟⎠ .(3.2.44)Furthermore, φ(t) satisfies the functional equation:φ(k2(1 + k1− k))=1− k(1 + k)2φ(k(1− k1 + k)2). (3.2.45)Proof. We will prove Eq. (3.2.45) first. A result of Verrill [77] shows thatφ2(R5(q))=qR5(q)(q5; q5)5∞(q; q)∞. (3.2.46)Combining Eq. (3.2.46) with the trivial formula (q2, q2)∞ = (q; q)∞(−q; q)∞,we haveφ2(R5(q))φ2 (R5(q2))=R5(q2)R5(q){q1/24 (−q; q)∞}{q5/24 (−q5; q5)∞}5 . (3.2.47)We will apply four of Ramanujan’s formulas to finish the proof. If k =R(q)R2(q2), then for |q| sufficiently small [53]:R5(q) =k(1− k1 + k)2, (3.2.48)58Chapter 3. Functional equations for Mahler measuresR5(q2) =k2(1 + k1− k), (3.2.49)q1/24 (−q; q)∞ =(k1− k2)1/24( 1 + k − k21− 4k − k2)5/24, (3.2.50)q5/24(−q5; q5)∞ =( k1− k2)5/24( 1 + k − k21− 4k − k2)1/24. (3.2.51)Eq. (3.2.45) follows immediately from substituting these parametric formu-las into Eq. (3.2.47).Next we will prove Eq. (3.2.43). Combining Eq. (3.2.48) with Entry3.2.15 in [53], we easily obtainq5/24(q5; q5)∞ ={k(1− k2)2(1 + k − k2) (1− 4k − k2)2}1/6q1/24 (q; q)∞ . (3.2.52)Now we will evaluate the eta product q1/24 (q; q)∞. First recall that if q =q4(z), then (see [58], page 148)q1/24(q; q)∞ = 2−1/4z1/24(1− z)1/12√2F1(14,34; 1; z).In Theorem 3.2.6 we showed that if k = R(q)R2(q2)thenq = q4(64k(1+k−k2)5(1+k2)2((1+11k−k2)2−125k2)2), hence it follows thatq1/24 (q; q)∞ =⎛⎜⎝k (1− k2)2 (1 + k − k2)5 (1− 4k − k2)10(1 + k2)6((1 + 11k − k2)2 − 125k2)6⎞⎟⎠1/24×√√√√√√2F1⎛⎜⎝14 , 34; 1; 64k (1 + k − k2)5(1 + k2)2 ((1 + 11k − k2)2 − 125k2)2⎞⎟⎠(3.2.53)Substituting Eq. (3.2.53), Eq. (3.2.52), and Eq. (3.2.48) into Eq. (3.2.46)completes the proof of Eq. (3.2.43). The proof of Eq. (3.2.44) also followsfrom an extremely similar argument.59Chapter 3. Functional equations for Mahler measuresWe will conclude this section by recording a few formulas which do notappear in [53], but which were probably known to Ramanujan. We will pointout that Maier obtained several results along these lines in [69]. Notice thatthe functional equation for φ(t) (after substituting z = k/(1 − k2)) impliesa new hypergeometric transformation:√(1 + 11z)2 − 125z2(1− z)2 − 5z2 2F1⎛⎜⎝14 , 34; 1, 64z5 (1 + z)(1 + 4z2)((1− z)2 − 5z2)2⎞⎟⎠= 2F1⎛⎜⎝14 , 34; 1; 64z (1 + z)5(1 + 4z2)((1 + 11z)2 − 125z2)2⎞⎟⎠(3.2.54)Perhaps not surprisingly, we can also use the arguments in this section todeduce thatq54⎛⎜⎝ 64z (1 + z)5(1 + 4z2)((1 + 11z)2 − 125z2)2⎞⎟⎠ = q4⎛⎜⎝ 64z5 (1 + z)(1 + 4z2)((1− z)2 − 5z2)2⎞⎟⎠ ,(3.2.55)which implies a rational parametrization for the fifth-degree modular equa-tion in Ramanujan’s theory of signature 4.3.3 A regulator explanationNow we will reinterpret our identities in terms of the regulators of ellipticcurves. The elliptic curves in question are defined by the zero varieties ofthe polynomials whose Mahler measure we studied. First we will explain therelationship between Mahler measures and regulators. Then we will use reg-ulators to deduce formulas involving Kronecker-Eisenstein series, includingequations (3.2.9), (3.2.10), (3.2.11), and (3.2.12).We will follow some of the ideas of Rodriguez-Villegas [72].3.3.1 The elliptic regulatorLet F be a field. By Matsumoto’s Theorem, K2(F ) is generated by thesymbols {a, b} for a, b ∈ F , which satisfy the bilinearity relations {a1a2, b} =60Chapter 3. Functional equations for Mahler measures{a1, b}{a2, b} and {a, b1b2} = {a, b1}{a, b2}, and the Steinberg relation {a, 1−a} = 1 for all a = 0.Recall that for a field F , with discrete valuation v, and maximal idealM, the tame symbol is given by(x, y)v ≡ (−1)v(x)v(y)xv(y)yv(x)mod M(see [71]). Note that this symbol is trivial if v(x) = v(y) = 0. In the casewhen F = Q(E) (from now on E denotes an elliptic curve), a valuation isdetermined by the order of the rational functions at each point S ∈ E(Q¯).We will denote the valuation determined by a point S ∈ E(Q¯) by vS .The tame symbol is then a map K2(Q(E))→ Q(S)∗.We have0→ K2(E)⊗Q→ K2(Q(E))⊗Q→∐S∈E(Q¯)QQ(S)∗ ×Q,where the last arrow corresponds to the coproduct of tame symbols.Hence an element {x, y} ∈ K2(Q(E)) ⊗ Q can be seen as an elementin K2(E) ⊗ Q whenever (x, y)vS = 1 for all S ∈ E(Q¯). All of the familiesconsidered in this paper are tempered according to [71], and therefore theysatisfy the triviality of tame symbols.The regulator map (defined by Beilinson after the work of Bloch) maybe defined byr : K2(E)→ H1(E,R){x, y} →{γ →∫γη(x, y)}for γ ∈ H1(E,Z), andη(x, y) := log |x|d arg y − log |y|d arg x.Here we think of H1(E,R) as the dual of H1(E,Z). The regulator is welldefined because η(x, 1− x) = dD(x), whereD(z) = Im (Li2(z)) + arg(1− z) log |z|is the Bloch-Wigner dilogarithm.In terms of the general formulation of Beilinson’s conjectures this defini-tion is not completely correct. One needs to go a step further and considerK2(E), where E is a Ne´ron model of E over Z. In particular, K2(E) is61Chapter 3. Functional equations for Mahler measuresa subgroup of K2(E). It seems [71] that a power of {x, y} always lies inK2(E).Assume that E is defined over R. Because of the way that complex conju-gation acts on η, the regulator map is trivial for the classes in H1(E,Z)+. Inparticular, these cycles remain invariant under complex conjugation. There-fore it suffices to consider the regulator as a function on H1(E,Z)−.We write E(C) ∼= C/Z + τZ, where τ is in the upper half-plane. ThenC/Z+ τZ ∼= C∗/qZ, where z mod Λ = Z+ τZ is identified with e2iπz. Bloch[62] defines the regulator function in terms of a Kronecker-Eisenstein seriesRτ(e2πi(a+bτ))=y2τπ′∑m,n∈Ze2πi(bn−am)(mτ + n)2(mτ¯ + n), (3.3.56)where yτ is the imaginary part of τ .Let J(z) = log |z| log |1− z|, and letD(x) = Im (Li2(x)) + arg(1− x) log |x|be the Bloch-Wigner dilogarithm.Consider the following function on E(C) ∼= C∗/qZ:Jτ (z) =∞∑n=0J(zqn)−∞∑n=1J(z−1qn) +13log2 |q|B3(log |z|log |q|), (3.3.57)where B3(x) = x3− 32x2 + 12x is the third Bernoulli polynomial. If we recallthat the elliptic dilogarithm is defined byDτ (z) :=∑n∈ZD(zqn), (3.3.58)then the regulator function (see [62]) is given byRτ = Dτ − iJτ . (3.3.59)By linearity, Rτ extends to divisors with support in E(C). Let x and ybe non-constant functions on E with divisors(x) =∑mi(ai), (y) =∑nj(bj).Following [62], and the notation in [71], we recall the diamond operationC(E)∗ ⊗ C(E)∗ → Z[E(C)]−(x)  (y) =∑minj(ai − bj).62Chapter 3. Functional equations for Mahler measuresHere Z[E(C)]− means that [−P ] ∼ −[P ].Because Rτ is an odd function, we obtain a mapZ[E(C)]− → R.Theorem 3.3.1. (Beilinson [55]) If E/R is an elliptic curve, x, y are non-constant functions in C(E), and ω ∈ Ω1, then∫E(C)ω¯ ∧ η(x, y) = Ω0Rτ ((x)  (y)),where Ω0 is the real period.Although a more general version of Beilinson’s Theorem exists for ellipticcurves defined over the complex numbers, the above version has a simplerformulation.Corollary 3.3.2. (after an idea of Deninger) If x and y are non-constantfunctions in C(E) with trivial tame symbols, then−∫γη(x, y) = Im(ΩyτΩ0Rτ ((x)  (y)))where Ω =∫γ ω.Proof. Notice that iη(x, y) is an element of the two-dimensional vector spaceH2D(E(C),R(2)) generated by ω and ω¯. Then we may writeiη(x, y) = α[ω] + β[ω¯],from which we obtain ∫γiη(x, y) = αΩ+ βΩ.On the other hand, we have∫E(C)iη(x, y) ∧ ω¯ = α∫E(C)ω ∧ ω¯ = αi2Ω20yτ ,and ∫E(C)iη(x, y) ∧ ω = −βi2Ω20yτ .By Beilinson’s Theorem∫γiη(x, y) = −Rτ ((x)  (y))Ω2Ω0yτ+Rτ ((x)  (y))Ω2Ω0yτ,and the statement follows.63Chapter 3. Functional equations for Mahler measures3.3.2 Regulators and Mahler measureFrom now on, we will set k = 4√tin the first family (3.2.1).Rodriguez-Villegas [71] proved that if Pk(x, y) = k + x+ 1x + y +1y doesnot intersect the torus T2, thenm(k) ∼Z 12π r({x, y})(γ). (3.3.60)Here the ∼Z stands for ”up to an integer number”, and γ is a closed paththat avoids the poles and zeros of x and y. In particular, γ generates thesubgroup H1(E,Z)− of H1(E,Z) where conjugation acts by −1.We would like to use this property, however we need to exercise caution.In particular, Pk(x, y) intersects the torus whenever |k| ≤ 4 and k ∈ R. Letus recall the idea behind the proof of Eq. (3.3.60) for the special case ofPk(x, y). WritingyPk(x, y) = (y − y(1)(x))(y − y(2)(x)),we havem(k) = m(yPk(x, y)) =12πi∫T1(log+ |y(1)(x)|+ log+ |y(2)(x)|)dxx.This last equality follows from applying Jensen’s formula with respect to thevariable y. When the polynomial does not intersect the torus, we may omitthe “+” sign on the logarithm since each y(i)(x) is always inside or outsidethe unit circle. Indeed, there is always a branch inside the unit circle and abranch outside. It follows thatm(k) =12πi∫T1log |y|dxx= − 12π∫T1η(x, y), (3.3.61)where T1 is interpreted as a cycle in the homology of the elliptic curvedefined by Pk(x, y) = 0, namely H1(E,Z).If k ∈ [−4, 4], then we may also assume that k > 0 since this particularMahler measure does not depend on the sign of k. The equationk + x +1x+ y +1y= 0certainly has solutions when (x, y) ∈ T2. However, for |x| = 1 and k real,the number k + x+ 1x is real, and therefore y +1y must be real. This forces64Chapter 3. Functional equations for Mahler measurestwo possibilities: either y is real or |y| = 1. Let x = eiθ, then for −π ≤ θ ≤ πwe have−k − 2 cos θ = y + 1y. (3.3.62)The limiting case occurs when |k + 2 cos θ| = 2. Since we have assumedthat k is positive, this condition becomes k + 2 cos θ = 2, which impliesthat y = −1. When k + 2 cos θ > 2 one solution for y, say, y(1), becomes anegative number less than -1, thus∣∣y(1)∣∣ > 1 (the other solution y(2) is suchthat∣∣y(2)∣∣ < 1). When k+2 cos θ < 2, yi lies inside the unit circle and neverreaches 1. What is important is that∣∣y(1)∣∣ ≥ 1 and ∣∣y(2)∣∣ ≤ 1, so we can stillwrite Eq. (3.3.61) even if there is a nontrivial intersection with the torus.3.3.3 Functional identities for the regulatorFirst recall a result by Bloch [62] which studies the modularity of Rτ :Proposition 3.3.3. Let(α βγ δ)∈ SL2(Z), and let τ ′ = ατ+βγτ+δ . If we let(b′a′)=(δ −γ−β α)(ba),then:Rτ ′(e2πi(a′+b′τ ′))=1γτ¯ + δRτ(e2πi(a+bτ)).We will need to use some functional equations for Jτ . First recall thefollowing trivial property for J(z):J(z) = p∑xp=zJ(x). (3.3.63)Proposition 3.3.4. Let p be an odd prime, let q = e2πiτ , and let qj =e2πi(τ+j)p for j ∈ {0, 1, . . . , p − 1}. Suppose that (N, k) = 1, and p ≡ ±1 or0 (modN). Then(1 + χ−N (p)p2)JNτ (qk) =p−1∑j=0pJN(τ+j)p(qkj ) + χ−N (p)JNpτ (qpk), (3.3.64)and for any z we have(χ−N (p) + p2)JNτ (z) =p−1∑j=0pJN(τ+j)p(z) + χ−N (p)JNpτ (z). (3.3.65)65Chapter 3. Functional equations for Mahler measuresProof. First notice thatp−1∑j=0JN(τ+j)p(qkj ) =∞∑n=0p−1∑j=0J(qNn+kj)−∞∑n=1p−1∑j=0J(qNn−kj)+4π2y2τN23pB3(kN).By Eq. (3.3.63) this becomes=∞∑n=0pNn+k1pJ(qNn+k)−∞∑n=1pNn−k1pJ(qNn−k)+∞∑n=0p|Nn+kpJ(qNn+kp)−∞∑n=1p|Nn−kpJ(qNn−kp)+4π2y2τN23pB3(kN)=∞∑n=01pJ(qNn+k)−∞∑n=11pJ(qNn−k)−∞∑n=0p|Nn+k1pJ(qNn+k)+∞∑n=1p|Nn−k1pJ(qNn−k)+∞∑n=0p|Nn+kpJ(qNn+kp)−∞∑n=1p|Nn−kpJ(qNn−kp)+4π2y2τN23pB3(kN).Rearranging, we find that=1pJNτ(qk)− 4π2y2τN23pB3(kN)−∞∑n=0p|Nn+k1pJ((qp)Nn+kp)+∞∑n=1p|Nn−k1pJ((qp)Nn−kp)+∞∑n=0p|Nn+kpJ(qNn+kp)−∞∑n=1p|Nn−kpJ(qNn−kp)+4π2y2τN23pB3(kN)=1pJNτ(qk)− χ−N (p)pJNpτ (qpk) + χ−N (p)pJNτ (qk),which proves the assertion.The second equality follows in a similar fashion.66Chapter 3. Functional equations for Mahler measuresIt is possible to prove analogous identities for Dτ and Rτ .Proposition 3.3.5.J 2μ+12(eπiμ)= J2μ(eπiμ)− J2μ (−eπiμ) (3.3.66)Proof. Let z = eπiμ, thenJ2μ (z)− J2μ (−z) =J(z)− J(−z)+∞∑n=1(J(zqn)− J(−zqn)− J(z−1qn) + J(−z−1qn))=∞∑n=0(J(eπiμ(4n+1))− J(−eπiμ(4n+1))−J(eπiμ(4n+3))+ J(−eπiμ(4n+3))).On the other hand,J 2μ+12(z) =∞∑n=0(J((−1)neπiμ(2n+1))− J((−1)n+1eπiμ(2n+1))),which proves the equality.3.3.4 The first familyFirst we will write the equationx +1x+ y +1y+ k = 0in Weierstrass form. Consider the rational transformationX =k + x + yx + y= − 1xy, Y =k(y − x)(k + x + y)2(x + y)2=(y − x)(1 + 1xy)2xy,which leads toY 2 = X(X2 +(k24− 2)X + 1).It is useful to state the inverse transformation:x =kX − 2Y2X(X − 1) , y =kX + 2Y2X(X − 1) .67Chapter 3. Functional equations for Mahler measuresNotice that Ek contains a torsion point of order 4 over Q(k), namelyP =(1, k2). Indeed, this family is the modular elliptic surface associated toΓ0(4).We can show that 2P = (0, 0), and 3P =(1,−k2).Now(X) = 2(2P )− 2O,and(x) =(2(P ) + (2P )− 3O)− (2(2P )− 2O)− ((P ) + (3P )− 2O)=(P )− (2P )− (3P ) + O,(y) =(2(3P ) + (2P )− 3O)− (2(2P )− 2O)− ((P ) + (3P )− 2O)=− (P )− (2P ) + (3P ) + O.Computing the diamond operation between the divisors of x and y yields(x)  (y) = 4(P )− 4(−P ) = 8(P ).Now assume that k ∈ R and k > 4. We will choose an orientation forthe curve and compute the real period. Because P is a point of order 4 and∫ 10 ω is real, we may assume that P corresponds to3Ω04 .The next step is to understand the cycle |x| = 1 as an element ofH1(E,Z). We would like to compute the value of Ω =∫γ ω. First recallthatω =dX2Y=dxx(y − y−1) .In the case when k > 4, consider conjugation of ω. This sends x→ x−1, anddxx → −dxx . There is no intersection with the torus, so y remains invariant.Therefore we conclude that Ω is the complex period, and ΩΩ0 = τ , where τis purely imaginary.Therefore for k real and |k| > 4m(k) =4πIm(τyτRτ (−i)).Now take(0 −11 0)∈ SL2(Z). By Proposition 3.3.3Rτ (−i) = Rτ(e−2πi4)= τ¯R− 1τ(e−2πi4τ),thereforem(k) = −4|τ |2πyτJ− 1τ(e−2πi4τ).68Chapter 3. Functional equations for Mahler measuresIf we let μ = − 14τ , then for k ∈ R we obtainm(k) =− 1πyμJ4μ(e2πiμ)= Im(1πyμR4μ(e2πiμ))=Re(16yμπ2′∑m,nχ−4 (m)(m + 4μn)2(m + 4μ¯n)),thus recovering a result of Rodriguez-Villegas. We can extend this result toall k ∈ C, by arguing that both m(k) and − 1πyμJ4μ(e2πiμ)are the real partsof holomorphic functions that coincide at infinitely many points (see [70]).Now we will show how to deduce equations (3.1.7) and (3.1.6). ApplyingEq. (3.3.64) with N = 4, k = 1, and p = 2, we haveJ4μ(q) = 2J2μ(q0) + 2J2(μ+1)(q1),which translates into1y4μJ4μ(e2πiμ)=1y2μJ2μ(eπiμ)+1y2μJ2μ(−eπiμ) .This is the content of Eq. (3.1.7). Setting τ = − 12μ , we may also writeD τ2(−i) = Dτ (−i) + Dτ (−ieπiτ ). (3.3.67)Next we will use Eq. (3.3.66):J 2μ+12(eπiμ)= J2μ(eπiμ)− J2μ (−eπiμ) ,which translates into1y 2μ+12J 2μ+12(eπiμ)=2y2μJ2μ(eπiμ)− 2y2μJ2μ(−eπiμ) .Setting τ = − 12μ , and using(1 0−2 1)∈ SL2(Z) on the left-hand side, wehaveD τ−12(−i) = Dτ (−i)−Dτ(−ieπiτ) . (3.3.68)Combining equations (3.3.67) and (3.3.68), we see that2Dτ (−i) = D τ2(−i) + D τ−12(−i).This is the content of Eq. (3.1.6).Similarly, we may deduce Eq. (3.2.14) from Eq. (3.3.64) when k = 1,N = 4, and p is an odd prime.69Chapter 3. Functional equations for Mahler measures3.3.5 A direct approachIt is also possible to prove equations (3.1.6) and (3.1.7) directly, withoutconsidering the μ-parametrization or the explicit form of the regulator.For those formulas, it is easy to explicitly write the isogenies at the levelof the Weierstrass models. By using the well-known isogeny of degree 2 (seefor example [64]):φ : {E : y2 = x(x2 + ax + b)} → {Ê : yˆ2 = xˆ(xˆ2 − 2axˆ + (a2 − 4b))}given by(x, y)→(y2x2,y(b− x2)x2)(we require that a2 − 4b = 0), we findφ1 : E2(n+ 1n)→ E4n2 , φ2 : E2(n+ 1n) → E 4n2 ,φ1 : (X,Y )→(X(n2X + 1)X + n2,−n3Y(X2 + 2n2X + 1)(X + n2)2),φ2 : (X,Y )→(X(X + n2)n2X + 1,−Y(n2X2 + 2X + n2)n (n2X + 1)2).Let us write x1, y1, X1, Y1 for the rational functions and r1 for theregulator in E4n2 , and x2, y2, X2, Y2, r2 for the corresponding objects inE 4n2.It follows that±m (4n2) = r1 ({x1, y1}) = 12π∫|X1|=1η(x1, y1)=14π∫|X|=1η(x1 ◦ φ1, y1 ◦ φ1)=12r ({x1 ◦ φ1, y1 ◦ φ1}) ,where the factor of 2 follows from the degree of the isogeny. Similarly, wefind that±m(4n2)= r2 ({x2, y2}) = 12 r ({x2 ◦ φ2, y2 ◦ φ2}) .70Chapter 3. Functional equations for Mahler measuresNow we need to compare the values ofr ({x1 ◦ φ1, y1 ◦ φ1}) , r ({x2 ◦ φ2, y2 ◦ φ2}) , and r ({x, y}) .Recall that (x)  (y) = 8(P ), where P = (1, k2 ). When k = 2(n + 1n),we will also consider the point Q =(− 1n2, 0), which has order 2 (thenP + Q =(−1, n− 1n), 2P + Q = (−n2, 0), etc).Let P now denote the point in E2(n+ 1n), and let P1 denote the corre-sponding point in E4n2 . We have the following table:φ1 :3P, P + Q → P12P, Q → 2P1P, 3P + Q → 3P1O0, 2P + Q → O1.Using this table, and the divisors (x1) and (y1) in E4n2 , we can compute(x1 ◦ φ1)  (y1 ◦ φ1). We find that(x1 ◦ φ1)  (y1 ◦ φ1) = −16(P ) + 16(P + Q),and similarly(x2 ◦ φ2)  (y2 ◦ φ2) = −16(P )− 16(P + Q).These computations show that12r0 ({x1 ◦ φ1, y1 ◦ φ1}) + 12 r0 ({x2 ◦ φ2, y2 ◦ φ2}) = 2 r0 ({x0, y0}) ,(3.3.69)and thereforer1 ({x1, y1}) + r2 ({x2, y2}) = 2 r0 ({x0, y0}) . (3.3.70)We can conclude the proof of Eq. (3.1.6) by inspecting signs.To prove Eq. (3.1.7), it is necessary to use the isomorphism φ from Eq.(3.3.71).3.3.6 Relations among m(2), m(8), m(3√2), and m(i√2)Setting n = 1√2in Eq. (3.1.7), we obtainm(3√2)+ m(i√2)= m(8).71Chapter 3. Functional equations for Mahler measuresDoing the same in Eq. (3.1.6), we find thatm(2) + m(8) = 2m(3√2).In this section we will establish the identity3m(3√2) = 5m(i√2),from which we can deduce expressions for m(2) and m(8).Consider the functions f and 1−f , where f =√2Y−X2 ∈ C(E3√2). Theirdivisors are (√2Y −X2)=(2P ) + 2(P + Q)− 3O,(1−√2Y −X2)=(P ) + (Q) + (3P + Q)− 3O.The diamond operation yields(f)  (1− f) = 6(P )− 10(P + Q).But (f)  (1− f) is trivial in K-theory, hence6(P ) ∼ 10(P + Q).Now consider the isomorphism φ:φ : E2(n+ 1n)→ E2(in+ 1in), (X,Y )→ (−X, iY ) (3.3.71)This isomorphism implies thatri√2 ({x, y}) = r3√2 ({x ◦ φ, y ◦ φ}) .But we know that(x ◦ φ)  (y ◦ φ) = 8(P + Q).This implies6 r3√2 ({x, y}) = 10 ri√2({x, y}),and3m(3√2) = 5m(i√2).Consequently, we may conclude thatm(8) =85m(3√2), m(2) =25m(3√2),and finallym(8) = 4m(2).72Chapter 3. Functional equations for Mahler measures3.3.7 The Hesse familyWe will now sketch the case of the Hesse family:x3 + y3 + 1− 3t13xy.This family corresponds to Γ0(3). The diamond operation yields(x)  (y) = 9(P ) + 9(A) + 9(B), (3.3.72)where P is a point of order 3, defined over Q(t1/3), and A,B are points oforder 3 such that A + B + P = O.For 0 < t < 1, we haven(t) =92πIm(τyτ(Rτ(e4πi3)+ Rτ(e4πi(1+τ)3)+ Rτ(e2πi(2+τ)3))).If we let μ = − 1τ , we obtain, after several steps,n(t) = Re⎛⎝27√3yμ4π2′∑k,nχ−3(n)(3μk + n)2(3μ¯k + n)⎞⎠ .Following the previous example, this result may be extended to C \κ bycomparing holomorphic functions.3.3.8 The Γ00(6) exampleWe will now sketch a treatment of Stienstra’s example [76]:(x + 1)(y + 1)(x + y)− 1txy.Applying the diamond operation, we have(x)  (y) = −6(P )− 6(2P ),where P is a point of order 6.For t small, one can writeg(t) =3πIm(τyτRτ (ξ−16 ) + Rτ (ξ−13 )).73Chapter 3. Functional equations for Mahler measuresEventually, one arrives tog(t) =Re(36yμπ2′∑m,nχ−3(m)(m + 6μn)2(m + 6μ¯n))+Re(9yμ2π2′∑m,nχ−3(m)(m + 3μn)2(m + 3μ¯n)),thus recovering a result of [76].3.3.9 The Γ00(5) exampleNow we will consider our final example:(x + y + 1)(x + 1)(y + 1)− 1txy.Applying the diamond operation, we find that(x)  (y) = 10(P ) + 5(2P ),where P is a torsion point of order 5.For t > 0r(t) =52πIm(τyτ(2Rτ(e8πi5)+ Rτ(e6πi5))).Finally,r(t) = −Re(25iyμ4π2′∑m,n2(ζm5 − ζ−m5)+ ζ2m5 − ζ−2m5(m + 5μn)2(m + 5μ¯n)).In conclusion, we see that the modular structure comes from the formof the regulator function, and the functional identities are consequences ofthe functional identities of the elliptic dilogarithm.3.4 ConclusionWe have used both regulator and q-series methods to prove a variety ofidentities between the Mahler measures of genus-one polynomials. We willconclude this paper with a final open problem.Open Problem 3: How do you characterize all the functional equationsof μ(t)?74Chapter 3. Functional equations for Mahler measuresWe have seen that there are identities like Eq. (3.1.6), stating that2m(2(k +1k)+ x +1x+ y +1y)=m(4k2 + x +1x+ y +1y)+m(4k2+ x +1x+ y +1y).While this formula does not follow from Eq. (3.2.14), it can be proved withregulators.Indeed, the last section showed us that we can obtain functional identitiesfor the Mahler measures by looking at functional equations for the ellipticdilogarithm. To have an idea of the dimensions of this problem, let usnote that equation (3.3.64) corresponds to the integration of an identity forthe Hecke operator Tp. This suggests that more identities will follow fromlooking at the general operator Tn. And this is just the beginning of thestory. . .3.5 AcknowledgementsThe authors would like to deeply thank David Boyd and Fernando Rodriguez-Villegas for many helpful discussions, and David Boyd for pointing out thework of Kurokawa and Ochiai [67]. ML extends her gratitude to ChristopherDeninger, Herbert Gangl, and Florian Herzig for enlightening discussions.ML is a postdoctoral fellow at the Pacific Institute for the MathematicalSciences and the University of British Columbia. This research was alsopartially conducted while ML was a member at the Institute for AdvancedStudy, and at the Mathematical Sciences Research Institute, a visitor at theInstitut des Hautes E´tudes Scientifiques, a guest at the Max-Planck-Insitutfu¨r Mathematik, and she was employed by the Clay Mathematics Instituteas a Liftoff Fellow. ML whishes to thank these institutions for their supportand hospitality.This material is partially based upon work supported by the NationalScience Foundation under agreement No. DMS-0111298.75Bibliography[53] G. E. Andrews and B.C. Berndt, Ramanujan’s Lost Notebook, Part I,Springer-Verlag, New York, 2005.[54] A. Beauville, Les familles stables de courbes elliptiques sur P 1 admet-tant quatre fibres singulie`res. (French. English summary) C. R. Acad.Sci. Paris Sr. I Math. 294 (1982), no. 19, 657–660.[55] A. A. Be˘ılinson, Higher regulators and values of L-functions of curves.Funktsional. Anal. i Prilozhen. 14 (1980), no. 2, 46–47.[56] B.C. Berndt, Ramanujan’s Notebooks, Part II, Springer-Verlag, NewYork, 1989.[57] B.C. Berndt, Ramanujan’s Notebooks, Part III, Springer-Verlag, NewYork, 1991.[58] B.C. Berndt, Ramanujan’s Notebooks, Part V, Springer-Verlag, NewYork, 1998.[59] M. J. Bertin, Integral expression of Dirichlet L-series, Slides of a talkgiven at the conference ”Mahler measure in Mobile”, University ofSouth Alabama, January 2006.[60] M. J. Bertin, Mesure de Mahler et re´gulateur elliptique: preuve de deuxrelations ”exotiques”. Number theory , 1–12, CRM Proc. Lecture Notes,36, Amer. Math. Soc., Providence, RI , 2004.[61] M. J. Bertin, Mesure de Mahler d’une famille de polynoˆmes. J. ReineAngew. Math. 569 (2004), 175–188.[62] S. J. Bloch, Higher regulators, algebraic K-theory, and zeta functionsof elliptic curves. CRM Monograph Series, 11. American MathematicalSociety, Providence, RI , 2000. x+97 pp.[63] D. W. Boyd, Mahler’s measure and special values of L-functions, Ex-periment. Math. 7 (1998), 37-82. demic Press, 1994.76Bibliography[64] J. W. S. Cassels, Lectures on elliptic curves. London MathematicalSociety Student Texts, 24. Cambridge University Press, Cambridge,1991. vi+137 pp.[65] C. Deninger, Deligne periods of mixed motives, K-theory and the en-tropy of certain Zn-actions, J. Amer. Math. Soc. 10 (1997), no. 2,259–281.[66] N. Koblitz, Introduction to Elliptic Curves and Modular Forms. Sec-ond edition. Graduate Texts in Mathematics, 97. Springer-Verlag, NewYork, 1993. x+248 pp.[67] N. Kurokawa and H. Ochiai, Mahler measures via crystalization, Com-mentarii Mathematici Universitatis Sancti Pauli, 54 (2005), 121-137.[68] S. Lang, Elliptic Functions. With an appendix by J. Tate. Second edi-tion. Graduate Texts in Mathematics, 112. Springer-Verlag, New York,1987. xii+326 pp.[69] R. S. Maier, Algebraic hypergeometric transformations of modular ori-gin, (preprint 2006).[70] F. Rodriguez-Villegas, Modular Mahler measures (previous version).Unpublished manuscript. See http://www.math.utexas.edu/∼ ville-gas/mahler.dvi.[71] F. Rodriguez-Villegas, Modular Mahler measures I, Topics in numbertheory (University Park, PA, 1997), 17–48, Math. Appl., 467, KluwerAcad. Publ., Dordrecht, 1999.[72] F. Rodriguez-Villegas, Identities between Mahler measures, Numbertheory for the millennium, III (Urbana, IL, 2000), 223–229, A K Peters,Natick, MA, 2002.[73] L. J. Rogers, On a type of modular relation, Proc. London Math. Soc.19 (1920), 387-397.[74] M. D. Rogers, A study of inverse trigonometric integrals associated withthree-variable Mahler measures, and some related identities, J. NumberTheory 121 (2006), no. 2, 265–304.[75] J. H. Silverman, The arithmetic of elliptic curves. Graduate Texts inMathematics, 106. Springer-Verlag, New York, 1992. xii+400 pp.77Bibliography[76] J. Stienstra, Mahler measure variations, eisenstein se-ries and instanton expansions, (preprint 2005). Seehttp://front.math.ucdavis.edu/math.NT/0502193.[77] H. A. Verrill, Picard-Fuchs equations of some families of elliptic curves.Proceedings on Moonshine and related topics (Montrl, QC, 1999), 253–268, CRM Proc. Lecture Notes, 30, Amer. Math. Soc., Providence, RI,2001.[78] D. Zagier, Integral solutions of Ape`ry-like recurrence equations,preprint.Department of Mathematics, University of British Columbia,Vancouver, B.C. V6T 1Z2, Canadamatrogers@math.ubc.ca78Chapter 4New 5F4 transformations andMahler measuresMathew D. Rogers34.1 IntroductionIn this paper we will study the consequences of some recent results of Bertin.Recall that Bertin proved q-series expansions for a pair of three-variableMahler measures in [86]. As usual the Mahler measure of an n-variablepolynomial, P (z1, . . . , zn), is defined bym (P (z1, . . . , zn)) =∫ 10. . .∫ 10log∣∣∣P (e2πiθ1 , . . . , e2πiθn)∣∣∣ dθ1 . . .dθn.We will define g1(u) and g2(u) in terms of the following three-variable Mahlermeasuresg1(u) := m(u + x +1x+ y +1y+ z +1z), (4.1.1)g2(u) := m(−u + 4 + (x + x−1) (y + y−1)+(x + x−1) (z + z−1)+(y + y−1) (z + z−1)).(4.1.2)We can recover Bertin’s original notation by observing that g1(u) = m (Pu),and after substituting (xz, y/z, z/x) → (x, y, z) in Eq. (4.1.2) we see thatg2(u + 4) = m(Qu) [86].In Section 4.2 we will show how to establish a large number of inter-esting relations between g1(u), g2(u), and three more three-variable Mahlermeasures. For example, for |u| sufficiently large Eq. (4.2.21) is equivalent3A version of this chapter has been accepted for publication. Rogers, M. D. (2007)New 5F4 hypergeometric transformations, three-variable Mahler measures, and formulasfor 1/π. Ramanujan Journal.79Chapter 4. New 5F4 transformations and Mahler measurestog1(3(u2 + u−2))=15m(x4 + y4 + z4 + 1 +√3(3 + u4)u3xyz)+35m(x4 + y4 + z4 + 1 +√3(3 + u−4)u−3xyz).(4.1.3)Rodriguez-Villegas briefly mentioned the Mahler measure m(x4 + y4 + z4 + 1 + uxyz)on the last page of [95].We will also show that identities like Eq. (4.1.3) are equivalent to trans-formations for the 5F4 hypergeometric function. Recall that the generalizedhypergeometric function is defined bypFq(a1,a2,...apb1,b2,...bq ;x)=∞∑n=0(a1)n . . . (ap)n(b1)n . . . (bq)nxnn!,where (y)n = Γ(y + n)/Γ(y). We have restated Eq. (4.1.3) as a hypergeo-metric transformation in Eq. (4.2.24). As a special case of Eq. (4.1.3) wecan also deduce that5F4(54, 32, 74,1,12,2,2,2; 1)=2563log(2)− 5120√23π3L(f, 3),where f(q) = q∏∞n=1 (1− qn)2(1− q2n) (1− q4n) (1− q8n)2, and L(f, s) isthe usual L-series of f(q). We will conclude Section 4.2 with a brief dis-cussion of some related, but still unproven, evaluations of the 4F3 and 3F2hypergeometric functions.It turns out that g1(u) and g2(u) are also closely related to Watson’striple integrals. For appropriate values of u, Watson showed that g′1(u) andg′2(u) reduce to products of elliptical integrals (for relevant results see [96],[91], [92], and [90]). In Section 4.3 we will use some related transformationsto prove new formulas for 1/π. For example, we will show that2π=∞∑n=0(−1)n (3n + 1)32nn∑k=0(2n− 2kn− k)(2kk)(nk)2.Notice that this formula for 1/π involves the Domb numbers. Chan, Chanand Liu obtained a similar formula for 1/π involving Domb numbers in [88],we have recovered their result in (4.3.10). Zudilin and Yang also discoveredsome related formulas for 1/π in [98]. All of the 3F2 transformations thatwe will utilize in Section 4.3 follow from differentiating the 5F4 identitiesestablished in Section 4.2.80Chapter 4. New 5F4 transformations and Mahler measures4.2 Identities between Mahler measures andtransformations for the 5F4 functionBertin proved that both g1(u) and g2(u) have convenient q-series expansionswhen u is parameterized correctly. Before stating her theorem, we will definesome notation. As usual let(x, q)∞ = (1− x)(1− xq)(1− xq2) . . . ,and define G(q) byG(q) = Re[− log(q) + 240∞∑n=1n2 log (1− qn)]. (4.2.1)Notice that if q ∈ (0, 1) then G′(q) = −M(q)/q, where M(q) is the Eisensteinseries of weight 4 on the full modular group Γ(1) [81].Theorem 4.2.1. (Bertin) For |q| sufficiently smallg1 (t1(q)) =− 160G(q) +130G(q2)− 120G(q3) +110G(q6), (4.2.2)g2 (t2(q)) =1120G(q)− 115G(q2)− 140G(q3)+15G(q6), (4.2.3)wheret1(q) =v1 +1v1, and v1 = q1/2(q; q2)6∞(q3; q6)6∞,t2(q) =−(v2 − 1v2)2, and v2 = q1/2(q2; q2)6∞(q3; q3)2∞(q12; q12)4∞(q; q)2∞ (q4; q4)4∞ (q6; q6)6∞.In this section we will show that both g1(u) and g2(u) reduce to linearcombinations of 5F4 hypergeometric functions. We will accomplish this goalby first expressing each of the functionsf2(u) :=2m(u1/2 +(x + x−1) (y + y−1) (z + z−1)), (4.2.4)f3(u) :=m(u− (x + x−1)2 (y + y−1)2 (1 + z)3 z−2) , (4.2.5)f4(u) :=4m(x4 + y4 + z4 + 1 + u1/4xyz), (4.2.6)in terms of G(q). We will then exploit those identities to establish linearrelations between functions in the set {f2(u), f3(u), f4(u), g1(u), g2(u)}. This81Chapter 4. New 5F4 transformations and Mahler measuresis significant since f2(u), f3(u), and f4(u) all reduce to 5F4 hypergeometricfunctions. In particular this implies the non-trivial fact that both g1(u) andg2(u) also reduce to linear combinations of 5F4 functions.Proposition 4.2.2. The following identities hold for |u| sufficiently large:f2(u) =Re[log(u)− 8u5F4(32, 32, 32,1,12,2,2,2;64u)], (4.2.7)f3(u) =Re[log(u)− 12u5F4(43, 32, 53,1,12,2,2,2;108u)], (4.2.8)f4(u) =Re[log(u)− 24u5F4(54, 32, 74,1,12,2,2,2;256u)]. (4.2.9)For |u| > 6g1(u) =Re[log(u)−∞∑n=1(1/u)2n2n(2nn) n∑k=0(2kk)(nk)2], (4.2.10)and if |u| > 16g2(u) =Re[log(u)−∞∑n=1(1/u)nnn∑k=0(2n− 2kn− k)(2kk)(nk)2]. (4.2.11)Proof. We can prove each of these identities using a method due to Rodriguez-Villegas [95]. We will illustrate the proof of Eq. (4.2.7) explicitly. Rear-ranging the Mahler measure shows thatf2(u) =Re [log(u)+∫ 10∫ 10∫ 10log(1− 64ucos2(2πt1) cos2(2πt2) cos2(2πt3))dt1dt2dt3].If |u| > 64, then ∣∣64u cos2(2πt1) cos2(2πt2) cos2(2πt3)∣∣ < 1, hence by theTaylor series for the logarithmf2(u) =Re[log(u)−∞∑n=1(64/u)nn(∫ 10cos2n(2πt)dt)3]=Re[log(u)−∞∑n=1(2nn)3 (1/u)nn]=Re[log(u)− 8u5F4(32, 32, 32,1,12,2,2,2;64u)].Notice that Eq. (4.2.7) holds whenever u ∈ [−64, 64], since f2(u) is harmonicin C\[−64, 64].82Chapter 4. New 5F4 transformations and Mahler measuresWhile Proposition 4.2.2 shows that the results in this paper easily trans-late into the language of hypergeometric functions, the relationship to Mahlermeasure is more important than simple pedagogy. Bertin proved that forcertain values of u the zero varieties of the (projectivized) polynomials fromequations (4.1.1) and (4.1.2) define K3 hypersurfaces. She also proved for-mulas relating the L-functions of these K3 surfaces at s = 3 to rationalmultiples of the Mahler measures. Proposition 4.2.2 shows that these re-sults imply explicit 5F4 evaluations (see Corollary 4.2.6 for explicit exam-ples). While it might also be interesting to interpret the polynomials fromequations (4.2.4) through (4.2.6) in terms of K3 hypersurfaces, we will notpursue that direction here.Theorem 4.2.3. For |q| sufficiently smallf2(s2(q)) =− 215G(q)−115G (−q) + 35G(q2), (4.2.12)f3(s3(q)) =− 18G(q) +38G(q3), (4.2.13)f4(s4(q)) =− 13G(q) +23G(q2), (4.2.14)wheres2(q) =q−1(−q; q2)24∞ ,s3(q) =1q(27q(q3; q3)6∞(q; q)6∞+(q; q)6∞(q3; q3)6∞)2,s4(q) =1q(q2; q2)24∞(q; q)24∞(16q(q; q)4∞(q4; q4)8∞(q2; q2)12∞+(q2; q2)12∞(q; q)4∞ (q4; q4)8∞)4.The following inverse relations hold for |q| sufficiently small:G(q) =− 19f2 (s2(q))− 4f2 (s2(−q)) + 24f2(s2(q2))− 12f2(s2(−q2)) , (4.2.15)G(q) =− 192f3 (s3(q))− 32f3(s3(e2πi/3q))− 32f3(s3(e4πi/3q))+92f3(s3(q3)),(4.2.16)G(q) =− 5f4 (s4(q))− 2f4 (s4(−q)) + 4f4(s4(q2)). (4.2.17)83Chapter 4. New 5F4 transformations and Mahler measuresProof. We can use Ramanujan’s theory of elliptic functions to verify thefirst half of this theorem. Recall that the elliptic nome is defined byqj(α) = exp⎛⎝− πsin (π/j)2F1(1j , 1− 1j ; 1; 1− α)2F1(1j , 1− 1j ; 1;α)⎞⎠ .It is a well established fact that sj (qj (α)) is a rational function of α when-ever j ∈ {2, 3, 4}. For example if q = q2(α), then s2 (q) = 16α(1−α) . Thereforewe can verify Eq. (4.2.12) by differentiating with respect to α, and byshowing that the identity holds when q → 0.Observe that when q → 0 both sides of Eq. (4.2.12) approach − log |q|+O(q). Differentiating with respect to α yields− (1− 2α)2α(1− α)3F2(12, 12, 121,1; 4α(1− α))= − 12q(1− 16∞∑n=1n3qn1− qn + 256∞∑n=1n3q4n1− q4n)dqdα.This final identity follows from applying three well known formulas:dqdα=qα(1− α)2F12(12 ,12 ; 1;α) ,3F2(12, 12, 121,1; 4α(1− α))= 2F12(12,12; 1;α),1− 16∞∑n=1n3qn1− qn + 256∞∑n=1n3q4n1− q4n = (1− 2α)2F14(12,12; 1;α).We can verify equations (4.2.13) and (4.2.14) in a similar manner by usingthe fact that s3 (q3(α)) = 27α(1−α) , and s4 (q4(α)) =64α(1−α) .The crucial observation for proving equations (4.2.15) through (4.2.17)is the fact that G(q) satisfies the following functional equation for any primep:p−1∑j=0G(e2πij/pq)=(1 + p3)G (qp)− p2G(qp2). (4.2.18)We will only need the p = 2 case to prove Eq. (4.2.15):G(q) + G(−q) = 9G (q2)− 4G (q4) .84Chapter 4. New 5F4 transformations and Mahler measuresNotice that this last formula always allows us to eliminate G(q4)from anequation. Applying the substitutions q → −q, q → q2, and q → −q2 to Eq.(4.2.12) yields⎛⎝ −2/15 −1/15 3/5−1/15 −2/15 3/5−3/20 −3/20 23/20⎞⎠⎛⎝ G (q)G (−q)G(q2)⎞⎠ =⎛⎝ f2 (s2(q))f2 (s2(−q))2f2(s2(q2))− f2 (s2 (−q2))⎞⎠ .Since this system of equations is non-singular, we can invert the matrix torecover Eq. (4.2.15). We can prove equations (4.2.16) and (4.2.17) in asimilar fashion.If we compare Theorem 4.2.3 with Bertin’s results we can deduce someobvious relationships between the Mahler measures. For example, combiningEq. (4.2.14) with Eq. (4.2.3), and combining Eq. (4.2.13) with Eq. (4.2.2),we find thatg1(t1(q)) =120f4 (s4(q)) +320f4(s4(q3)), (4.2.19)g2 (t2(q)) =− 115f3 (s3(q)) +815f3(s3(q2)). (4.2.20)Notice that many more identities follow from substituting equations (4.2.15)through (4.2.17) into formulas (4.2.2), (4.2.3), (4.2.12), (4.2.13) and (4.2.14).However, for the remainder of this section we will restrict our attention toequations (4.2.19) and (4.2.20). In particular, we will appeal to the theoryof elliptic functions to transform those results into identities which dependon rational arguments.If we let q = q2(α), then it is well known thatqj/24(qj ;qj)∞q1/24(q;q)∞is an algebraicfunction of α for j ∈ {1, 2, 3, . . . } (for example see [82] or [83]). It followsimmediately that s2(q), s3(q), s4(q), t1(q), and t2(q) are also algebraic func-tions of α. The following lemma lists several instances where those functionshave rational parameterizations.Lemma 4.2.4. Suppose that q = q2(α), where α = p(2+p)3/(1+2p)3. Thefollowing identities hold for |p| sufficiently small:s2(q) =16(1 + 2p)6p(1− p)3(1 + p)(2 + p)3 , s2(q3)=16(1 + 2p)2p3(1− p)(1 + p)3(2 + p) ,s2(−q) =− 16(1− p)6(1 + p)2p(2 + p)3(1 + 2p)3, s2(−q3) =− 16(1− p)2(1 + p)6p3(2 + p)(1 + 2p),85Chapter 4. New 5F4 transformations and Mahler measuress2(−q2) =162(1− p)3(1 + p)(1 + 2p)3p2(2 + p)6, s2(−q6) =162(1− p)(1 + p)3(1 + 2p)p6(2 + p)2,s3(q) =4(1 + 4p + p2)6p (1− p2)4 (2 + p)(1 + 2p) , s3(q2)=16(1 + p + p2)6p2 (1− p2)2 (2 + p)2(1 + 2p)2 ,s3(−q) =−4(1− 2p− 2p2)6p(1− p2)(2 + p)(1 + 2p)4 , s3(q4)=4(2 + 2p− p2)6p4 (1− p2) (2 + p)4(1 + 2p) ,s4(q) =16(1 + 14p + 24p2 + 14p3 + p4)4p(1− p)6(1 + p)2(2 + p)3(1 + 2p)3 ,s4(q3)=16(1 + 2p + 2p3 + p4)4p3(1− p)2(1 + p)6(2 + p)(1 + 2p) ,s4(−q) =−16(1− 10p− 12p2 − 4p3 − 2p4)4p(1− p)3(1 + p)(1 + 2p)6(2 + p)3 ,s4(−q3) =− 16 (1 + 2p− 4p3 − 2p4)4p3(1− p)(1 + p)3(1 + 2p)2(2 + p) .Rational formulas also exist for certain values of t21(q) and t2(q):t21(q) =4(1 + p + p2)2 (1 + 4p + p2)2p(1− p2)2(2 + p)(1 + 2p) ,t21(−q) =−4(1 + p + p2)2 (1− 2p− 2p2)2p(1− p2)(2 + p)(1 + 2p)2 ,t2(q) =−4(1− p2)2p(2 + p)(1 + 2p),t2(−q) =− 4(1 + p + p2)2p(1− p2)(2 + p) .The main difficulty with Lemma 4.2.4 is the fact that very few values ofsj(±qn) reduce to rational functions of p. Consider the set{s2(q), s2(−q), s2(−q2) , s2 (q2)} as an example. While Lemma 4.2.4 showsthat s2(q), s2(−q), and s2(−q2) are all rational with respect to p, the for-mula for s2(q2)involves radicals. Recall that if α = p(2 + p)3/(1 + 2p)3,thens2(q2)=4(1 +√1− α)6α2√1− α ,86Chapter 4. New 5F4 transformations and Mahler measureswhere√1− α = 1−p(1+2p)2√(1− p2)(1 + 2p). Since the curve X2 = (1−p2)(1+2p) is elliptic with conductor 24, it follows immediately that rational sub-stitutions for p will never reduce s2(q2)to a rational function. For the sakeof legibility, we will therefore avoid all identities which involve those fourfunctions simultaneously. By avoiding pitfalls of this nature, we can deriveseveral interesting results from Lemma 4.2.4.Theorem 4.2.5. For |z| sufficiently largeg1(3(z + z−1))=120f4(9(3 + z2)4z6)+320f4(9(3 + z−2)4z−6), (4.2.21)g2(z) =− 115f3((16− z)3z2)+815f3(−(4− z)3z). (4.2.22)Proof. These identities follow from applying Lemma 4.2.4 to equations (4.2.19)and (4.2.20). If we consider Eq. (4.2.19), then Lemma 4.2.4 shows that t21(q),s4(q), and s4(q3)are all rational functions of p. Forming a resultant withrespect to p, we obtain0 = Resp[4(1 + p + p2)2 (1 + 4p + p2)2p (1− p2)2 (2 + p)(1 + 2p) − t21(q),16(1 + 14p + 24p2 + 14p3 + p4)4p(1− p)6(1 + p)2(2 + p)3(1 + 2p)3 − s4(q)].Simplifying with the aid of a computer, this becomes0 = s24(q) +(12 + t21(q))4 − s4(q) (−288 + 352t21(q)− 42t41(q) + t61(q)) .If we choose z so that t1(q) = 3(z + z−1), then s4(q) = 9(3 + z2)4z−6, anda formula for s4(q3)follows in a similar fashion.If we let u = 1/z with z ∈ R and sufficiently large, then Eq. (4.2.22)reduces to the following infinite series identity:∞∑n=1unnn∑k=0(2kk)(2n− 2kn− k)(nk)2=15log((1− 16u)(1− 4u)8)+4u5(1− 16u)3 5F4(43, 32, 53,1,12,2,2,2;− 108u(1− 16u)3)+32u25(1− 4u)3 5F4(43, 32, 53,1,12,2,2,2;108u2(1− 4u)3).(4.2.23)87Chapter 4. New 5F4 transformations and Mahler measuresSimilarly, if we let u = 1/z2 then Eq. (4.2.21) is equivalent to:∞∑n=11n(u9(1 + u)2)n(2nn) n∑k=0(2kk)(nk)2=25log(27(1 + u)5(3 + u)3(1 + 3u))+4u35(3 + u)4 5F4(54, 32, 74,1,12,2,2,2;256u39(3 + u)4)+4u15(1 + 3u)4 5F4(54, 32, 74,1,12,2,2,2;256u9(1 + 3u)4).(4.2.24)In Section 4.3 we will differentiate equations (4.2.23) and (4.2.24) to obtainseveral new formulas for 1/π. But first we will conclude this section bydeducing some explicit 5F4 evaluations.Recall that for certain values of u, Bertin evaluated g1(u) and g2(u) interms of the L-series of K3 surfaces. She also proved equivalent formulasinvolving twisted cusp forms. Amazingly, her formulas correspond to caseswhere the right-hand sides of equations (4.2.21) and (4.2.22) collapse to onehypergeometric term. We can combine her results with equations (4.2.23)and (4.2.24) to deduce several new 5F4 evaluations.Corollary 4.2.6. If g(q) = q(q2; q2)3∞(q6; q6)3∞, then5F4(43, 32, 53,1,12,2,2,2; 1)= 18 log(2) + 27 log(3)− 810√3π3L(g, 3). (4.2.25)If f(q) = q (q; q)2∞(q2; q2)∞(q4; q4)∞(q8; q8)2∞, then5F4(54, 32, 74,1,12,2,2,2; 1)=2563log(2)− 5120√23π3L(f, 3). (4.2.26)While many famous 5F4 identities, such as Dougall’s formula [81], reducespecial values of the 5F4 function to gamma functions, equations (4.2.25) and(4.2.26) do not fit into this category. Rather these new formulas are higherdimensional analogues of Boyd’s conjectures. In particular, Boyd has con-jectured large numbers of identities relating two-variable Mahler measures(that mostly reduce to 4F3 functions) to the L-series of elliptic curves [87].The most famous outstanding conjecture of this type asserts thatm(1 + x +1x+ y +1y)= −2Re[4F3(32, 32,1,12,2,2; 16)]?=154π2L(f, 2),88Chapter 4. New 5F4 transformations and Mahler measureswheref(q) = q∞∏n=1(1− qn) (1− q3n) (1− q5n) (1− q15n) ,and “ ?=” indicates numerical equality to at least 50 decimal places. Re-cently, Kurokawa and Ochiai proved a formula [93] which simplifies this lastconjecture to3F2(12, 12, 1232,1;116)?=15π2L(f, 2).Of course it would be highly desirable to rigorously prove Boyd’s conjectures.Failing that, it might be interesting to search for more hypergeometric iden-tities like equations (4.2.25) and (4.2.26). This line of thought suggests thefollowing fundamental problem with which we shall conclude this section:Open Problem: Determine every L-series that can be expressed in termsof generalized hypergeometric functions with algebraic parameters.4.3 New formulas for 1/πIn the previous section we produced several new transformations for the 5F4hypergeometric function. Now we will differentiate those formulas to obtainsome new 3F2 transformations, and several accompanying formulas for 1/π.The following formula is a typical example of the identities in this section:2π=∞∑n=0(−1)n (3n + 1)32nn∑k=0(2n− 2kn− k)(2kk)(nk)2. (4.3.1)Ramanujan first proved identities like Eq. (4.3.1) in his famous paper “Mod-ular equations and approximations to π” [94]. He showed that the followinginfinite series holds for certain constants A, B, and X:1π=∞∑n=0(An + B)(1/2)3nn!3Xn (4.3.2)Ramanujan determined many sets of algebraic values for A, B, and X byexpressing them in terms of the classical singular moduli Gn and gn. Healso stated (but did not prove) several formulas for 1/π where (1/2)3n isreplaced by (1/a)n (1/2)n (1− 1/a)n for a ∈ {3, 4, 6} (for more details see[84] or [89]).89Chapter 4. New 5F4 transformations and Mahler measuresRamanujan’s formulas for 1/π have attracted a great deal of attention be-cause of their intrinsic beauty, and because they converge extremely quickly.For example, Mathematica calculates π using a variant of a Ramanujan-typeformula due to the Chudnovsky brothers [97]:1π= 12∞∑n=0(−1)n(6n)!(13591409 + 54513013n)n!3(3n)! (6403203)n+1/2. (4.3.3)More recent mathematicians including Yang and Zudilin have derived for-mulas for 1/π which are not hypergeometric, but still similar to Eq. (4.3.3).For example, Yang showed that18π√15=∞∑n=0(4n + 1)36nn∑k=0(nk)4,and Zudilin gave many infinite series for 1/π containing nested sums ofbinomial coefficients [98]. All of the formulas that we will prove, includingEq. (4.3.1), are essentially of this type.Theorem 4.3.1. For |u| sufficiently small3F2(13, 12, 231,1;108u2(1− 4u)3)= (1− 4u)∞∑n=0unn∑k=0(2n− 2kn− k)(2kk)(nk)2.(4.3.4)If |u| is sufficiently small3F2(14, 12, 341,1;256u9(1 + 3u)4)=(1 + 3u)(1 + u)∞∑n=0(u9(1 + u)2)n(2nn) n∑k=0(2kk)(nk)2.(4.3.5)Proof. Applying the operator u ddu to Eq. (4.2.23), and then simplifyingyields∞∑n=0unn∑k=0(2kk)(2n− 2kn− k)(nk)2=− (1 + 32u)15(1− 16u)3F2(13, 12, 231,1;− 108u(1− 16u)3)+16(1 + 2u)15(1− 4u)3F2(13, 12, 231,1;108u2(1− 4u)3).Eq. (4.3.4) then follows from applying a standard 3F2 transformation:3F2(13, 12, 231,1;− 108u(1− 16u)3)=(1− 16u)(1− 4u) 3F2(13, 12, 231,1;108u2(1− 4u)3). (4.3.6)90Chapter 4. New 5F4 transformations and Mahler measuresWe can prove Eq. (4.3.5) in a similar manner by differentiating Eq. (4.2.24)and then using3F2(14, 12, 341,1;256u39(3 + u)4)=(3 + u)3(1 + 3u)3F2(14, 12, 341,1;256u9(1 + 3u)4). (4.3.7)While the infinite series in Theorem 4.3.1 are not hypergeometric sincethey involve nested binomial sums, they are still interesting. In particular,those formulas easily translate into unexpected integrals involving powers ofmodified Bessel functions. For |x| sufficiently small, Eq. (4.3.4) is equivalentto∫ ∞0e−3(x+x−1)uI30 (2u) du =x3 (1 + 3x2)3F2(14, 12, 341,1256x29(1 + 3x2)4), (4.3.8)where I0(u) is the modified Bessel function of the first kind. Recall the seriesexpansions for I0(2u) and I20 (2u):I0(2u) =∞∑n=0u2nn!2, I20 (2u) =∞∑n=0(2nn)u2nn!2.While it is not obvious that the Laplace transform of I30 (2u) should equala hypergeometric function, M. Lawrence Glasser has kindly pointed outthat equation (4.3.8) is essentially a well known result. A variety of similarintegrals have also been studied by Joyce [92], Glasser and Montaldi [90],and others.Finally, we will list a few formulas for 1/π. Notice that equation (4.3.10)first appeared in the work of Chan, Chan and Liu [88].Corollary 4.3.2. Let an =∑nk=0(2n−2kn−k)(2kk)(nk)2, then the following for-mulas are true:2π=∞∑n=0(−1)n (3n + 1)32nan, (4.3.9)8√33π=∞∑n=0(5n + 1)64nan, (4.3.10)9 + 5√3π=∞∑n=0(6n + 3−√3)(3√3− 54)nan. (4.3.11)91Chapter 4. New 5F4 transformations and Mahler measuresLet bn =(2nn)∑nk=0(2kk)(nk)2, then the following identity holds:2(64 + 29√3)π=∞∑n=0(520n + 159− 48√3)(80√3− 139484)nbn. (4.3.12)Proof. We can use Eq. (4.3.4) to easily deduce that if∞∑n=0(an + b)(1/3)n(1/2)n(2/3)nn!3(108u2(1− 4u)3)n=∞∑n=0(An + B)unan,then A = a(1− 4u)/(2+ 4u), and B = a(−4u)(1− 4u)/(2+ 4u)+ b(1− 4u).Since the left-hand side of this last formula equals 1/π when(a, b, 108u2(1−4u)3)∈{(6027 , 827 , 227) ,( 2√3 , 13√3 , 12) , (4511 − 533√3, 611 − 1399√3,− 1941331 + 2252662√3)}, it iseasy to verify equations (4.3.9) through (4.3.11) [89].We can verify Eq. (4.3.12) in a similar manner by combining Eq. (4.3.5)with Ramanujan’s formula8π=∞∑n=0(20n + 3)(1/4)n(1/2)n(3/4)nn!3(−14)n.4.4 ConclusionWe will conclude the paper by suggesting two future projects. Firstly, itwould be desirable to determine whether or not a rational series involvingbn exists for 1/π. Secondly, it might be interesting to consider the Mahlermeasuref6(u) = m(u− (z + z−1)6 (y + y−1)2 (1 + x)3x−2) ,since f6(u) arises from Ramanujan’s theory of signature 6.4.5 AcknowledgementsThe author would like to thank David Boyd for the many helpful discussionsand encouragement. The author also thanks Fernando Rodriquez-Villegasand Marie Jose´ Bertin for the useful discussions. The author thanks WadimZudilin for the useful communications and for the reference [98]. The au-thor also thanks Zhiguo Liu for the reference [88]. The author extends hisgratitude to M. Lawrence Glasser for the references [90] and [92]. Finally,thanks to Jianyun Shan from Peking University for pointing out [91].92Bibliography[79] G. E. Andrews and B.C. Berndt, Ramanujan’s Lost Notebook, Part I,Springer-Verlag, New York, 2005.[80] C. M. Bender, D. C. Brody and B. K. Meister, On powers of Besselfunctions, J. Math. Phys. 44(2003), no. 1, 309–314.[81] B.C. Berndt, Ramanujan’s Notebooks, Part II, Springer-Verlag, NewYork, 1989.[82] B.C. Berndt, Ramanujan’s Notebooks, Part III, Springer-Verlag, NewYork, 1991.[83] B.C. Berndt, Ramanujan’s Notebooks, Part V, Springer-Verlag, NewYork, 1998.[84] B.C. Berndt, S. Bhargava and F.G. Garvan, Ramanujan’s theories ofelliptic functions to alternative bases, Trans. Amer. Math. Soc. 347(1995), 4163-4244.[85] M. J. Bertin, Mahler’s measure and L-series of K3 hypersurfaces,(Preprint).[86] M. J. Bertin, Mesure de Mahler D’Hypersurfaces K3, (Preprint 2005).See http://arxiv.org/abs/math/0501153[87] D. W. Boyd, Mahler’s measure and special values of L-functions, Ex-periment. Math. 7 (1998), 37-82. demic Press, 1994.[88] H. H. Chan, S. H. Chan and Z. Liu, Domb’s numbers and RamanujanSato type series for 1/π. Adv. Math. 186 (2004), no. 2, 396–410.[89] H. H. Chan and W. Liaw, Cubic modular equations and newRamanujan-type series for 1/π. Pacific J. Math. 192 (2000), no.2, 219–238.[90] M. L. Glasser and E. Montaldi, Staircase polygons and recurrent latticewalks, Phys. Rev. E, 48, (1993), no. 4.93Bibliography[91] G. S. Joyce and I. J. Zucker, On the evaluation of generalized Watsonintegrals. Proc. Amer. Math. Soc. 133 (2005), no. 1, 71-81 (electronic).[92] G. S. Joyce, On the simple cubic lattice Green function. Philos. Trans.Roy. Soc. London Ser. A 273 (1973), no. 1236, 583–610.[93] N. Kurokawa and H. Ochiai, Mahler measures via crystalization, Com-mentarii Mathematici Universitatis Sancti Pauli, 54 (2005), 121-137.[94] S. Ramanujan, Modular equations and approximations to π, [Quart. J.Math. 45 (1914), 350-372]. Collected papers of Srinivasa Ramanujan,23-29, AMS Chelsea Publ., Providence, RI, 2000.[95] F. Rodriguez-Villegas, Modular Mahler measures I, Topics in numbertheory (University Park, PA, 1997), 17–48, Math. Appl., 467, KluwerAcad. Publ., Dordrecht, 1999.[96] G. N. Watson, Three triple integrals. Quart. J. Math., Oxford Ser. 10,(1939). 266-276.[97] E. W. Weisstein. ”Pi Formulas.” From MathWorld–A Wolfram WebResource. http://mathworld.wolfram.com/PiFormulas.html[98] W. Zudilin, Quadratic transformations and Guillera’s formulae for1/π2, (preprint 2005). See http://wain.mi.ras.ru/publications.htmlDepartment of Mathematics, University of British Columbia,Vancouver, B.C. V6T 1Z2, Canadamatrogers@math.ubc.ca94Chapter 5Trigonometric integrals andMahler measuresMathew D. Rogers45.1 IntroductionIn this paper we will undertake a systematic study of each of the inversetrigonometric integralsT(v, w) =∫ 10tan−1(vx) tan−1(wx)xdx,S(v, w) =∫ 10sin−1(vx) sin−1(wx)xdx,TS(v, w) =∫ 10tan−1(vx) sin−1(wx)xdx.This class of integrals arises when trying to find closed form expressions forthe Mahler measures of certain three-variable polynomials.Recall that the Mahler measure of an n-dimensional polynomial, P (x1, . . . , xn),can be defined bym (P (x1, . . . , xn)) =∫ 10. . .∫ 10log∣∣P (e2πiθ1 , . . . , e2πiθn) ∣∣dθ1 . . .dθn.In the last few years, numerous papers have established explicit formulasrelating multi-variable Mahler measures to special constants. Smyth provedthe first result [102] withm (1 + x + y + z) =72π2ζ(3),4A version of this chapter has been published. Rogers, M. D. (2006) A study of inversetrigonometric integrals associated with three-variable Mahler measures, and some relatedidentities. J. Number Theory 121:265-304.95Chapter 5. Trigonometric integrals and Mahler measureswhere the Riemann zeta function is defined by ζ(s) =∑∞n=11ns .In this paper, we will prove a number of new formulas relating three-variable Mahler measures to the aforementioned trigonometric integrals.Many of our identities generalize previously known results. We will lista few of our main results in this introductory section.For our first example, we can use various properties of T(v, w) to showthatm(1− v4(1− x1 + x)2+(y + v2(1− x1 + x))2z)=4π∫ v0tan−1(u)udu− 8π2T(v,1v)+12m(1− v4(1− x1 + x)2).(5.1.1)This reduces to one of Lalin’s formulas [107] when v = 1:m ((1 + y)(1 + z) + (1− z)(x− y)) = 72π2ζ(3) +log(2)2. (5.1.2)We can use the double arcsine integral, S(v, w), to prove that if v ∈ [0, 1]:m (v(1 + x) + y + z) =2π∫ v0sin−1(u)udu− 4π2S(v, 1)=4π2(Li3(v)− Li3(−v)2).(5.1.3)The second equality has been proved by Vandervelde [115]. Slightly morecomplicated arguments lead to expressions that includem(1− x1/6 + y + z)=2π∫ 120sin−1(u)udu− 12π2S(12,12)(5.1.4)This fractional Mahler measure is defined bym(1− x1/6 + y + z)=∫ 10m(1− e2πiu/6 + y + z)du,notice that m(1− x1/6 + y + z) = m(1 − x + y + z). We can simplify theright-hand side of Eq. (5.1.4) by either expressing S(12 ,12)as a linear com-bination of L-functions, or in terms of a famous binomial sum:S(12,12)=14∞∑n=11n3(2nn) .96Chapter 5. Trigonometric integrals and Mahler measuresCondon [105] proved an identity that Boyd and Rodriguez Villegas con-jectured:m (1 + x + (1− x)(y + z)) = 285π2ζ(3). (5.1.5)His proof also showed (in a slightly disguised form) thatTS(2, 1) =π2∫ 20tan−1(u)udu− 75ζ(3). (5.1.6)We have generalized Condon’s identity to show thatm(1 + x +v2(1− x)(y + z))=2π∫ v0tan−1(u)udu− 4π2TS(v, 1), (5.1.7)where Eq. (5.4.18) expresses TS(v, 1) in terms of polylogarithms. We canuse this result to prove a number of new formulas, including:m(x +v24(1 + x)2 +(y +v2(1 + x))2z)=2π∫ v0tan−1(u)udu− 4π2TS(v, 1) +12m(x +v24(1 + x)2).(5.1.8)When v = 2 this reduces to an interesting identity for ζ(3) and the goldenratio:m(x + (1 + x)2 + (1 + x + y)2z)=285π2ζ(3) + log(1 +√52). (5.1.9)We will show that all of the integrals TS(v, w), T(v, w), and S(v, w)have closed form expressions in terms of polylogarithms. The special case ofTS(v, 1) will warrant extra attention, as it is related to an interesting familyof binomial sums. Our closed forms are all derived through elementarymethods.5.2 Preliminaries: A description of the method,and some two dimensional Mahler measuresAlthough there are many conjectured formulas for multi-variable Mahlermeasures, most are extremely difficult, if not impossible, to prove. Rather97Chapter 5. Trigonometric integrals and Mahler measuresthan attempting to prove any of these conjectures, we will take an easier ap-proach. By investigating promising functions, and rewriting them as Mahlermeasures, we can recover a number of useful formulas.Our first step was to determine a class of functions that we could relateto Mahler’s measure. We chose the three integrals TS(v, w), S(v, w), andT(v, w), based on Condon’s evaluation of TS(2, 1), Eq. (5.1.6). Condon’sformula naturally suggested the existence of a generalized Mahler measureformula involving TS(v, 1). From there, it was a small step to consider thesimilar functions TS(v, w), T(v, w), and S(v, w).We will use the following method to express TS(v, 1), S(v, 1), and T(v, 1/v)as three-variable Mahler measures. First, a simple integration by partschanges each function into a two-dimensional integral, containing either anested arcsine or arctangent integral. Recall that the following integralsdefine the arctangent and arcsine integrals respectively:∫ w0tan−1(u)udu,∫ v0sin−1(u)udu.A typical formula for TS(v, 1), Eq. (5.3.8), can be proved with little trouble:TS(v, 1) =π2∫ v0tan−1(u)udu−∫ π/20∫ v sin(θ)0tan−1(z)zdzdθ.Next, substituting a two-dimensional Mahler measure for the nested arctan-gent or arcsine integral will allow us to obtain a three-dimensional Mahlermeasure evaluation. Theorem 5.3.2, Proposition 5.5.2, and Theorem 5.7.3contain our main results from using this method.Expressing the arcsine and arctangent integrals in terms of Mahler’smeasure represents the main difficulty in this approach. In the remainderof this section we will establish four two-variable Mahler measures for thearctangent integral, and one two-variable Mahler measure for the arcsineintegral.Since many of our results involve polylogarithms, this will be a goodplace to define the polylogarithm.Definition 5.2.1. If |z| < 1, then the polylogarithm of order k is defined byLik(z) =∞∑n=1znnk.We call Li2(z) the dilogarithm, and we call Li3(z) the trilogarithm.98Chapter 5. Trigonometric integrals and Mahler measuresTheorem 5.2.2 requires a formula of Cassaigne and Maillot [111]. Inparticular, Cassaigne and Mallot showed thatπm(a+bx+cy) ={D( |a||b| eiγ)+ α log |a|+ β log |b|+ γ log |c|, if “”π log (max {|a|, |b|, |c|}) , otherwise.The “” condition states that |a|, |b|, and |c| form the sides of a triangle.If “” is true, then α, β, and γ denote the radian measures of the anglesopposite to the sides of length |a|, |b|, and |c| respectively. In this formula,D(z) denotes the Bloch-Wigner dilogarithm. As usual,D(z) = Im (Li2(z)) + log |z| arg(1− z).Now that we have stated Cassaigne and Maillot’s formula, we will proveTheorem 5.2.2.Theorem 5.2.2. If 0 ≤ v ≤ 1 and w ≥ 0, then∫ v0sin−1(u)udu =π2m(2v + y + z) (5.2.1)∫ w0tan−1(u)udu =π2m(1 + w2 + (y + w)2z)− π4log(1 + w2)(5.2.2)Proof. To prove Eq. (5.2.1) first recall the usual formula for this arcsineintegral,∫ v0sin−1(u)udu =12Im(Li2(e2i sin−1(v)))+ sin−1(v) log(2v), (5.2.3)which is valid whenever 0 ≤ v ≤ 1.Now apply Cassaigne and Maillot’s formula to m(2v + y + z); we are inthe “” case since 0 ≤ v ≤ 1. It follows from a little trigonometry thatπm(2v + y + z) = D(e2i sin−1(v))+ 2 sin−1(v) log(2v),Since∣∣∣e2i sin−1(v)∣∣∣ = 1, D (e2i sin−1(v)) = Im (Li2 (e2i sin−1(v))), hence weobtainπm(2v + y + z) = Im(Li2(e2i sin−1(v)))+ 2 sin−1(v) log(2v).Comparing this last formula to Eq. (5.2.3), we haveπ2m(2v + y + z) =∫ v0sin−1(u)udu.99Chapter 5. Trigonometric integrals and Mahler measuresTo prove Eq. (5.2.2) first recall that if 0 ≤ w ≤ 1, then∫ w0tan−1(u)udu = Im (Li2(iw)) .Next observe that by Cassaigne and Maillot’s formulaπm(√1 + w2 + wy + z)= D(eπi/2w)+ tan−1(w) log(w) +π2log(√1 + w2)= Im (Li2(iw)) +π4log(1 + w2).After a change of variables in the Mahler measure, it is clear thatm(√1 + w2 + wy + z)=12{m(√1 + w2 + (1 + wy)iz)+m(√1 + w2 − (1 + wy)iz)}=12m(1 + w2 + (1 + wy)2z2)=12m(1 + w2 + (y + w)2z).It follows that for 0 ≤ w ≤ 1 we have∫ w0tan−1(u)udu =π2m(1 + w2 + (y + w)2z)− π4log(1 + w2).We can extend this formula to the entire positive real line. Suppose thatw = 1/w′ where w′ ≥ 1, then∫ 1/w′0tan−1(u)udu =π2m(1 +1w′2+(y +1w′)2z)− π4log(1 +1w′2)=π2m(1 + w′2 +(y + w′)2z)− π4log(1 + w′2)− π2log(w′).Since the arctangent integral obeys the functional equation [112]∫ w′0tan−1(u)udu =π2log(w′) +∫ 1/w′0tan−1(u)udu, (5.2.4)it follows that∫ w′0tan−1(u)udu =π2m(1 + w′2 +(y + w′)2z)− π4log(1 + w′2).Therefore Eq. (5.2.2) holds for all w ≥ 0.100Chapter 5. Trigonometric integrals and Mahler measuresThe next theorem proves that Eq. (5.2.2) is not unique. Using resultsfrom Theorem 5.6.5, we can derive three more Mahler measures for thearctangent integral.Theorem 5.2.3. Suppose that w ≥ 0, then∫ w0tan−1(u)udu =π4m((1 + w2)(1 + y) + w(1− y)(z + z−1)) , (5.2.5)∫ w0tan−1(u)udu =π2m((y − y−1) + w(z + z−1)) , (5.2.6)∫ w0tan−1(u)udu =π4m⎛⎝(4(1 + y)2 − (z + z−1)2) (1 + w2)2+(z − z−1)2 (1 + y)2(1− w2)2⎞⎠− π4log(2)− π2log(1 + w).(5.2.7)Proof. Since all three of these formulas have similar proofs, we will onlyprove Eq. (5.2.5) and Eq. (5.2.7). It is necessary to remark, that while Eq.(5.2.5) follows from Eq. (5.6.25), and Eq. (5.2.7) follows from Eq. (5.6.19),we must start from Eq. (5.6.16) to prove Eq. (5.2.6).Now we will proceed with the proof of Eq. (5.2.5). From Eq. (5.6.25)we haveπ4klog(1 + k1− k)− 2kIm (Li2(ir)) =∫ 10sin−1(u)1− k2u2du,where k = 2r1+r2, and 0 < k < 1. After an integration by parts this becomesπ4klog(1 + k1− k)−2kIm (Li2(ir))=π4klog(1 + k1− k)− 12k∫ 10log(1 + ku1− ku)du√1− u2 .It follows immediately thatIm (Li2(ir)) =14∫ π/20log(1 + k sin(t)1− k sin(t))dt=18∫ 2π0log+∣∣∣∣1 + k sin(t)1− k sin(t)∣∣∣∣dt.101Chapter 5. Trigonometric integrals and Mahler measuresChanging the “log+ | · |” term into a Mahler measure, which we can do byJensen’s formula, yieldsIm (Li2(ir)) =π4m(y +1 + k z+z−121− k z+z−12).Since k = 2r+r−1 , we haveIm (Li2(ir)) =π4m(y +r + r−1 + (z + z−1)r + r−1 − (z + z−1))=π4m((1 + y)(r + r−1) + (1− y)(z + z−1))− π4m(r + r−1 − (z + z−1))=π4m((1 + y)(r + r−1) + (1− y)(z + z−1))− π4(log+(r) + log+(1r))In order to substitute the arctangent integral for Im (Li2(ir)), we will assumethat 0 < r < 1. With this restriction, the formula becomes∫ r0tan−1(u)udu =π4m((1 + y)(r + r−1) + (1− y)(z + z−1))− π4log(1r)=π4m((1 + y)(1 + r2) + r(1− y)(z + z−1)) (5.2.8)We can manually verify that Eq. (5.2.8) holds when r = 0 and r = 1, andusing Eq. (5.2.4) we can extend Eq. (5.2.8) to all r > 1. Therefore, Eq.(5.2.5) follows immediately.Next we will prove Eq. (5.2.7). Using Eq. (5.6.16), we can show that2Im (Li2(ip)) =π2log(p) +∫ 10sin−1(u)u√(1− u2)(1− k2u2)du,where k = 1−p21+p2, and 0 < k < 1. To satisfy this restriction on k, we willassume that 0 < p < 1. After several elementary simplifications, the right-hand side becomes=π2log(p) +π2log(1 +1√1− k2)+∫ 10log(1 +√1− u21− k2u2)du√(1− u2)102Chapter 5. Trigonometric integrals and Mahler measures=π2log(p) +π2log((1 + p)22p)+∫ π/20log(1 +cos(θ)√1− k2 sin2(θ))dθ=π2log((1 + p)22)+12∫ 2π0log+∣∣∣∣∣1 + cos(θ)√1− k2 sin2(θ)∣∣∣∣∣ dθSince cos(π − θ) = − cos(θ), we have2Im (Li2(ip)) =π2log((1 + p)22)+14∫ 2π0log+∣∣∣∣∣1 + cos(θ)√1− k2 sin2(θ)∣∣∣∣∣ dθ+14∫ 2π0log+∣∣∣∣∣1− cos(θ)√1− k2 sin2(θ)∣∣∣∣∣ dθ.Applying Jensen’s formula yields2Im (Li2(ip)) =π2log((1 + p)22)+14∫ 2π0m((1 + y)2 − cos2(θ)1− k2 sin2(θ))dθ=π2log((1 + p)22)+π2m((1 + y)2 −(z + z−1)24 + k2 (z − z−1)2)=π2log((1 + p)22)− π2m(4 + k2(z − z−1)2)+π2m((4(1 + y)2 − (z + z−1)2)+ k2(1 + y)2 (z − z−1)2)We can simplify the one-dimensional Mahler measure as follows:m(4 + k2(z − z−1)2) = 2m (2 + ik (z − z−1))= 2 log(1 +√1− k2)= 2 log((1 + p)21 + p2).103Chapter 5. Trigonometric integrals and Mahler measuresEliminating k yields2Im (Li2(ip)) =π2log((1 + p)22)− π log((1 + p)21 + p2)+π2m((4(1 + y)2 − (z + z−1)2)+ (1− p21 + p2)2(1 + y)2(z − z−1)2)=− π2log(2)− π log(1 + p)+π2m⎛⎝(4(1 + y)2 − (z + z−1)2) (1 + p2)2+ (1 + y)2(z − z−1)2 (1− p2)2⎞⎠ .Since 0 < p < 1, it follows that2∫ p0tan−1(u)udu =π2m⎛⎝(4(1 + y)2 − (z + z−1)2) (1 + p2)2+ (1 + y)2(z − z−1)2 (1− p2)2⎞⎠− π2log(2)− π log(1 + p).(5.2.9)It is relatively easy to verify that Eq. (5.2.9) holds when p = 0 and p = 1.Using Eq. (5.2.4), we can also extend Eq. (5.2.9) to p > 1, which completesthe proof of Eq. (5.2.7).5.3 Relations between TS(v, 1) and Mahler’smeasure, and a reduction of TS(v, w) tomultiple polylogarithmsThe first goal of this section is to establish five identities relating TS(v, 1) tothree-variable Mahler measures. We will prove these formulas in Theorem5.3.2, using the methods outlined in Section 5.2. Corollary 5.3.3 examinesa few special cases of these results.Theorem 5.3.5 accomplishes the second goal of this section, which isto express TS(v, w) in terms of multiple polylogarithms. This result, whichappears to be new, is stated in Eq. (5.3.14). The importance of Eq. (5.3.14)lies in its easy proof, and more importantly in the fact that it immediatelyreduces TS(v, 1) to multiple polylogarithms. Finally, Proposition 5.3.6 willdemonstrate that the multiple polylogarithms in Eq. (5.3.14) always reduceto standard polylogarithms.We will need the following simple lemma to prove Theorem 5.3.1.104Chapter 5. Trigonometric integrals and Mahler measuresLemma 5.3.1. Assume that v and w are real numbers with v > 0 andw ∈ (0, 1], thenTS(v, w) = tan−1(v)∫ w0sin−1(z)zdz −∫ tan−1(v)0∫ wvtan(θ)0sin−1(z)zdzdθ,(5.3.1)TS(v, w) = sin−1(w)∫ v0tan−1(u)udu−∫ sin−1(w)0∫ vwsin(θ)0tan−1(z)zdzdθ.(5.3.2)Proof. To prove Eq. (5.3.1) first integrate TS(v, w) by parts to obtain:TS(v, w) = tan−1(v)∫ w0sin−1(z)zdz−∫ 10ddu(tan−1(vu)) ∫ wu0sin−1(z)zdzdu.Making the u-substitution θ = tan−1(vu) we have:TS(v, w) = tan−1(v)∫ w0sin−1(z)zdz −∫ tan−1(v)0∫ wvtan(θ)0sin−1(z)zdzdθ,which completes the proof of the identity.The proof of Eq. (5.3.2) follows in a similar manner.The fact that Lemma 5.3.1 expresses TS(v, w) as a double integral intwo different ways, makes TS(v, w) more versatile than either S(v, w) orT(v, w). These two different expansions will allow us to combine TS(v, w)with Mahler measures for both arctangent and arcsine integrals.Theorem 5.3.2. The following Mahler measures hold whenever v ≥ 0:m(1 + x +v2(1− x)(y + z))=2π∫ v0tan−1(u)udu− 4π2TS(v, 1) (5.3.3)m(x +v24(1 + x)2 +(y +v2(1 + x))2z)=2π∫ v0tan−1(u)udu− 4π2TS(v, 1) +12m(x +v24(1 + x)2) (5.3.4)105Chapter 5. Trigonometric integrals and Mahler measuresm((1 + y)(1 +v24(x + x−1)2)+v2(1− y) (x + x−1) (z + z−1))=4π∫ v0tan−1(u)udu− 8π2TS(v, 1)(5.3.5)m((z − z−1) + v2(x + x−1)(y + y−1))=2π∫ v0tan−1(u)udu− 4π2TS(v, 1)(5.3.6)m⎛⎜⎜⎜⎝(4(1 + y)2 − (z + z−1)2)(1 + v24(x + x−1)2)2+(z − z−1)2 (1 + y)2(1− v24(x + x−1)2)2⎞⎟⎟⎟⎠=4π∫ v0tan−1(u)udu− 8π2TS(v, 1) +4π∫ π/20log (1 + v sin(θ)) dθ+ log(2)(5.3.7)Proof. We will prove Eq. (5.3.3) first, since it has the most difficult proof.Letting w = 1 in Eq. (5.3.1) yieldsTS(v, 1) =π2log(2) tan−1(v)−∫ tan−1(v)0∫ tan(θ)/v0sin−1(z)zdzdθ.Since 0 ≤ tan(θ)v ≤ 1, we may substitute Eq. (5.2.1) for the nested arcsineintegral to obtainTS(v, 1) =π2log(2) tan−1(v)− π2∫ tan−1(v)0m(2vtan(θ) + y + z)dθ=π2log(2) tan−1(v)− π2∫ π/20m(2vtan(θ) + y + z)dθ+π2∫ π/2tan−1(v)m(2vtan(θ) + y + z)dθ.In the right-hand integral tan(θ)v ≥ 1, hence by Cassaigne and Maillot’sformulam(2vtan(θ) + y + z)= log(2vtan(θ)).106Chapter 5. Trigonometric integrals and Mahler measuresSubstituting this result yields:TS(v, 1) =π2log(2) tan−1(v) +π2∫ π/2tan−1(v)log(2vtan(θ))dθ− π2∫ π/20m(2vtan(θ) + y + z)dθ=π2∫ v0tan−1(u)udu− π2∫ π/20m(tan(θ) +v2(y + z))dθ=π2∫ v0tan−1(u)udu− π24m(1 + x +v2(1− x)(y + z)).Eq. (5.3.3) follows immediately from rearranging this final identity.The proofs of equations (5.3.4) through (5.3.7) are virtually identical,hence we will only prove Eq. (5.3.5). Letting w = 1 in Eq. (5.3.2), we haveTS(v, 1) =π2∫ v0tan−1(u)udu−∫ π/20∫ v sin(θ)0tan−1(z)zdzdθ. (5.3.8)Substituting Eq. (5.2.5) for the nested arctangent integral yieldsTS(v, 1) =π2∫ v0tan−1(u)udu− π4∫ π/20m((1 + y)(1 + v2 sin2(θ))+ v sin(θ)(1− y)(z + z−1)) dθ=π2∫ v0tan−1(u)udu− π28m((1 + y)(1− v24(x− x−1)2)+v2i(1− y)(x− x−1)(z + z−1)).Letting x→ ix, we obtainTS(v, 1) =π2∫ v0tan−1(u)udu− π28m((1 + y)(1 +v24(x + x−1)2)+v2(1− y)(x + x−1)(z + z−1)).Eq. (5.3.5) follows immediately from rearranging this final equality.Finally, we will remark that the while Eq. (5.3.5) follows from substitut-ing Eq. (5.2.5) into Eq. (5.3.8), we must substitute Eq. (5.2.2) to prove Eq.(5.3.4), Eq. (5.3.6) follows from substituting Eq. (5.2.6), and Eq. (5.3.7)follows from substituting Eq. (5.2.7).107Chapter 5. Trigonometric integrals and Mahler measuresCorollary 5.3.3. The formulas in Theorem 5.3.2 reduce, in order, to thefollowing identities when v = 2:m((1 + x) + (1− x)(y + z)) = 285π2ζ(3), (5.3.9)m(x + (1 + x)2 + (1 + x + y)2z)=285π2ζ(3) + log(1 +√52), (5.3.10)m((1 + x + z)(1 + x−1 + z−1)+ y(1 + x− z) (1 + x−1 − z−1))=565π2ζ(3),(5.3.11)m((z − z−1)+ (x + x−1) (y + y−1)) = 285π2ζ(3), (5.3.12)m((4z(1 + y)2 − (1 + z)2) (1 + 3x + x2)2+ (1− z)2(1 + y)2 (1 + x + x2)2)=565π2ζ(3) +163πG+ log(2).(5.3.13)In Eq. (5.3.13), and throughout the rest of the paper, G denotes Catalan’sconstant. In particular, G = 1− 132+ 152− 172+ . . .Proof. As we have already stated, Condon proved Eq. (5.3.9) in [105]. Hisproof also showed thatTS(2, 1) =π2∫ 20tan−1(u)udu− 75ζ(3).Using this formula, equations (5.3.10) through (5.3.13) follow immediatelyfrom Theorem 5.3.2.Theorem 5.3.2 shows that we can obtain closed forms for several three-variable Mahler measures by reducing TS(v, 1) to polylogarithms. We haveproved a convenient closed form for TS(v, 1) in Eq. (5.4.18). Corollary 5.4.6also shows that this closed form immediately implies Condon’s evaluation ofTS(2, 1). We will postpone further discussion of Eq. (5.4.18) until Section5.4.We will devote the remainder of this section to deriving a closed formfor TS(v, w) in terms of multiple polylogarithms. For convenience, we willuse a slightly non-standard notation for our multiple polylogarithms.108Chapter 5. Trigonometric integrals and Mahler measuresDefinition 5.3.4. Define Fj(x) byFj(x) =∞∑n=0x2n+1(2n + 1)j=Lij(x)− Lij(−x)2,and define Fj,k(x, y) byFj,k(x, y) =∞∑n=0x2n+1(2n + 1)jn∑m=0y2m+1(2m + 1)k.We will employ this notation throughout the rest of the paper.Theorem 5.3.5. If vw ∈ (−i∞,−i] ∪ [i, i∞) and w ∈ [−1, 1], then we canexpress TS(v, w) in terms of multiple polylogarithms. Let R =vw1+√1+( vw )2,and let S = iw +√1− w2, thenTS(v, w) =2F3(R)− F3(RS)− F3(R/S)− 4F1,2(R, 1)+ 2F1,2(R,S) + 2F1,2(R, 1/S)+ i sin−1(w) {F2(RS)− F2(R/S)−2F1,1(R,S) + 2F1,1(R, 1/S)} .(5.3.14)Proof. First note that by u-substitutionTS(v, w) =∫ sin−1(w)0tan−1( vwsin(θ))cot(θ)θdθ. (5.3.15)Since w ∈ [−1, 1], it follows that our path of integration is along the realaxis. Next substitute the Fourier seriestan−1( vwsin(θ))= 2∞∑n=0R2n+12n + 1sin ((2n + 1)θ) , (5.3.16)into Eq. (5.3.15). Swapping the order of summation and integration, wehaveTS(v, w) = 2∞∑n=0R2n+12n + 1∫ sin−1(w)0sin ((2n + 1)θ) cot(θ)θdθ.Uniform convergence justifies this interchange of summation and integration.In particular, Eq. (5.3.16) converges uniformly whenever |R| < 1 and θ ∈ R.It is easy to show that |R| < 1 except when vw ∈ (−i∞,−i]∪[i, i∞), in which109Chapter 5. Trigonometric integrals and Mahler measurescase |R| = 1. If |R| = 1, then Eq. (5.3.16) no longer converges uniformly,and hence the following arguments do not apply.Evaluating the nested integral yieldsTS(v, w) = 4∞∑n=0R2n+12n + 1{sin−1(w)n∑k=0′sin((2k + 1) sin−1(w))2k + 1−n∑k=0′1− cos ((2k + 1) sin−1(w))(2k + 1)2},(5.3.17)wheren∑k=0′ak = a0 + · · ·+an−1 + an2 . Simplifying Eq. (5.3.17) completes ourproof.Eq. (5.3.14) deserves a few remarks, since it is a fairly general result.Firstly, observe that a closer analysis of Eq. (5.3.16) would probably allowus to relax the restriction that w ∈ [−1, 1]. Secondly, Eq. (5.3.14) mostlikely has applications beyond the scope of this paper. For example, we canuse Eq. (5.3.14) to reduce the right-hand side of the following equation∞∑n=1(−1)n(2n + 1)2(2nn)(w2)2n+1 n∑k=1(−1)k+1k= TS(1, w)− π4∫ w0sin−1(t)tdt +log(2)2∫ w0sinh−1(t)tdt,(5.3.18)to multiple polylogarithms.We can use the final result of this section, Proposition 5.3.6, to reduceTS(v, w) to regular polylogarithms. This proposition allows us to equateTS(v, w) with a formula involving around twenty trilogarithms. While aclever usage of trilogarithmic functional equations might simplify this result,it seems more convenient to simply leave Eq. (5.3.14) in its current form.Proposition 5.3.6. The functions F1,1(x, y) and F1,2(x, y) can be expressedin terms of polylogarithms, we have:4F1,1(x, y) =Li2(x(1 + y)1 + x)− Li2(x(1− y)1 + x)− Li2(−x(1 + y)1− x)+ Li2(−x(1− y)1− x).(5.3.19)110Chapter 5. Trigonometric integrals and Mahler measuresTo reduce F1,2(x, y) to polylogarithms, apply Lewin’s formula, Eq. (5.7.5),four times to the following identity:F1,2(x, y) =F3(xy)− 12 log(1− x2)F2(xy)+14∫ x0log(1− u2) log (1+yu1−yu)udu.(5.3.20)Proof. To prove Eq. (5.3.20), first swap the order of summation to obtainF1,2(x, y) = F3(xy) + F1(x)F2(y)−∞∑n=0y2n+1(2n + 1)2n∑k=0x2k+12k + 1.Substituting an integral for the nested sum yieldsF1,2(x, y) = F3(xy) + F1(x)F2(y)−∞∑n=0y2n+1(2n + 1)2∫ x01− u2n+21− u2 du= F3(xy) +∫ x0u1− u2F2(yu)du.Integrating by parts, the identity becomesF1,2(x, y) =F3(xy)− 12 log(1− x2)F2(xy)+14∫ x0log(1− u2) log (1+yu1−yu)udu,which completes the proof of Eq. (5.3.20).We can verify Eq. (5.3.19) by differentiating each side of the equationwith respect to y.Finally, observe that we can obtain simple closed forms for F1,2(x, 1) andF2,1(1, x) from Eq. (5.4.9).5.4 An evaluation of TS(v, 1) using infinite seriesThis evaluation of TS(v, 1) generalizes a theorem due to Condon. Condonproved a formula that Boyd and Rodriguez Villegas conjectured:m(1 + x + (1− x)(y + z)) = 285π2ζ(3).111Chapter 5. Trigonometric integrals and Mahler measuresCondon’s result is equivalent to evaluating TS(2, 1) in closed form. As The-orem 5.3.2 has shown, generalizing this Mahler measure depends on findinga closed form for TS(v, 1). Eq. (5.4.18) accomplishes this goal by expressingTS(v, 1) in terms of polylogarithms.This calculation of TS(v, 1) is based on several series transformations.The first step is to expand TS(v, 1) in a Taylor series; observe that thefollowing formula holds whenever |v| < 1:TS(v, 1) =π2∞∑k=0(−1)k(2k + 1)2v2k+1 − 12∞∑k=0(−1)k(2k + 1)3(2v)2k+1(2kk) . (5.4.1)We can easily prove Eq. (5.4.1) by starting from Eq. (5.3.8). Formula(5.4.1) shows that TS(v, 1) is analytic in the open unit disk. UnfortunatelyEq. (5.4.1) does not converge when v = 2, and hence it can not be used tocalculate TS(2, 1). It will be necessary to find an analytic continuation ofTS(v, 1) in order to carry out any useful computations.The following family of functions will play a crucial role in our calcula-tions.Definition 5.4.1. Define hn(v) by the infinite series,hn(v) =∞∑k=0(−1)k(2k + 1)n(2v)2k+1(2kk) . (5.4.2)Using the definition of h3(v), combined with the identity∞∑k=0(−1)k(2k + 1)2v2k+1 =∫ v0tan−1(u)udu,it follows that Eq. (5.4.1) can be rewritten asTS(v, 1) =π2∫ v0tan−1(u)udu− 12h3(v). (5.4.3)Finding a closed form for TS(v, 1) we will entail finding a closed formfor h3(v). Theorem 5.4.5 accomplishes this goal, however several auxiliarylemmas are needed first. The idea behind our proof is very simple: first finda closed form for h2(v), and then integrate it to find a closed form for h3(v).Batir recently used this method in an interesting paper [99] to obtaina formula that is equivalent to Eq. (5.4.15). Unfortunately Batir seemsto have missed Eq. (5.4.12), so we will provide a full derivation of thisimportant result.112Chapter 5. Trigonometric integrals and Mahler measuresLemma 5.4.2. The function h2(v) is analytic if v ∈ (−i∞,−i] ∪ [i, i∞).Furthermore, we can express h2(v) in terms of the dilogarithm,h2(v) = 4∞∑k=01(2k + 1)2(v1 +√1 + v2)2k+1= 2Li2(v1 +√1 + v2)− 2Li2( −v1 +√1 + v2).(5.4.4)Proof. We use the following elementary identity to prove Eq. (5.4.4),24k(2k + 1)2(2kk) = ∞∑j=0(−1)j2j + 1(2k)!(k + j + 1)!(k − j)! . (5.4.5)Substituting Eq. (5.4.5) into the definition of h2(v), we haveh2(v) =∞∑k=0(−1)k(2k + 1)2(2v)2k+1(2kk)= 4∞∑k=0(−1)k(v2)2k+1 ∞∑j=0(−1)j2j + 1(2k)!(k + j + 1)!(k − j)! .If we assume that |v| < 1, then the series converges uniformly, hence we mayswap the order of summation to obtainh2(v) = 4∞∑j=012j + 1∞∑k=0(−1)k+2j (2k + 2j)!(k + 2j + 1)!k!(v2)2k+2j+1= 4∞∑j=01(2j + 1)2(v2)2j+1 ∞∑k=0(j + 12)k(j + 1)k(2j + 2)k(−v2)kk!,where (x)n =Γ(x+n)Γ(x) . But then we haveh2(v) = 4∞∑j=01(2j + 1)2(v2)2j+12F1[j+ 12,j+12j+2∣∣− v2] ,where 2F1[a,bc∣∣x] is the usual hypergeometric function. A standard hyper-geometric identity [106] shows that2F1[j+ 12,j+12j+2∣∣− v2] = 22j+1(1 +√1 + v2)2j+1,113Chapter 5. Trigonometric integrals and Mahler measuresfrom which we obtainh2(v) = 4∞∑j=01(2j + 1)2(v1 +√1 + v2)2j+1,concluding the proof of the identity.We can use Eq. (5.4.4) to analytically continue h2(v) to a larger domain.Recall that Li2(r)−Li2(−r) is analytic whenever r ∈ (−∞,−1]∪ [1,∞), andv1+√1+v2is analytic whenever v ∈ (−i∞,−i]∪ [i, i∞). Since we have alreadyassumed that v ∈ (−i∞,−i]∪ [i, i∞), we simply have to show that the rangeof r = v1+√1+v2does not intersect the set {(−∞,−1] ∪ [1,∞)}.Some elementary calculus shows that |r| =∣∣∣ v1+√1+v2∣∣∣ ≤ 1 for all v ∈ C,with equality occurring only when v ∈ (−i∞,−i] ∪ [i, i∞). It follows thath2(v) is analytic on C− {(−i∞,−i] ∪ [i, i∞)}.Since we have now expressed h2(v) in terms of dilogarithms, we can finda closed form for h1(v) by differentiating Eq. (5.4.4):h1(v) =2√1 + v2log(v +√1 + v2). (5.4.6)In Theorem 5.4.5, we will integrate Eq. (5.4.4) to find a closed formfor h3(v) involving trilogarithms. To prove this theorem, we first need toestablish two lemmas. Lemma 5.4.3 evaluates a necessary integral, whileLemma 5.4.4 expresses F2,1(1, x) in terms of polylogarithms.Lemma 5.4.3. If j ≥ 0 is an integer, and r = v1+√1+v2, then we have thefollowing identity:∫ v01u(u1 +√1 + u2)2j+1du = log(1 + r1− r)+r2j+12j + 1−2j∑k=0r2k+12k + 1. (5.4.7)Proof. To evaluate the integralwj(v) =∫ v01u(u1 +√1 + u2)2j+1du,first make the substitution z = u1+√1+u2. In particular we can show thatu = 2z1−z2 anddudz = 2(1+z2)(1−z2)2 . Therefore we havewj(v) =∫ r0z2j(1 + z21− z2)dz=∫ r021− z2dz −∫ r01− z2j1− z2 dz −∫ r01− z2j+21− z2 dz.114Chapter 5. Trigonometric integrals and Mahler measuresNext substitute the geometric series 1−z2j1−z2 =∑j−1k=0 z2k into each of the right-hand integrals, and swap the order of summation and integration to obtainwj(v) =∫ r021− z2dz −j−1∑k=0r2k+12k + 1−j∑k=0r2k+12k + 1= log(1 + r1− r)+r2j+12j + 1− 2j∑k=0r2k+12k + 1.Lemma 5.4.4. The following double polylogarithmF2,1(1, x) =∞∑n=01(2n + 1)2n∑k=0x2k+12k + 1(5.4.8)can be evaluated in closed form. If |x| < 1,8F2,1(1, x) =4Li3(x)− Li3(x2)− 4Li3(1− x)− 4Li3(x1 + x)+ 4ζ(3)+ log(1 + x1− x)Li2(x2) +π22log(1 + x) +π26log(1− x)+23log3(1 + x)− 2 log(x) log2(1− x)(5.4.9)Proof. We will verify Eq. (5.4.9) by differentiating each side of the identity.First observe that the infinite series in Eq. (5.4.8) converges uniformlywhenever |x| ≤ 1, hence term by term differentiation is justified at all pointsin the open unit disk. It follows thatddxF2,1(1, x) =∞∑n=01(2n + 1)2(1− x2n+21− x2)=π28(11− x2)− x1− x2(Li2(x)− 14Li2(x2)),(5.4.10)whenever |x| < 1.Let ϕ(x) denote the right-hand side of Eq. (5.4.9). Taking the derivative115Chapter 5. Trigonometric integrals and Mahler measuresof ϕ(x) we obtain:dϕdx=4xLi2(x)− 2xLi2(x2) +41− xLi2(1− x)− 4(1x− 11 + x)Li2(x1 + x)+21− x2Li2(x2)− 2x(log2(1 + x)− log2(1− x))+ π22(11 + x)− π26(11− x)+21 + xlog2(1 + x)− 2xlog2(1− x) + 41− x log(x) log(1− x)(5.4.11)We can simplify Eq. (5.4.11) by eliminating Li2(1− x) and Li2(x1+x)withthe functional equations:Li2(1− x) = π26− log(x) log(1− x)− Li2 (x) ,Li2(x1 + x)= −12log2(1 + x) + Li2(x)− 12Li2(x2).Substituting these identities into Eq. (5.4.11) and simplifying, we are leftwithdϕdx= π2(11− x2)− 8x1− x2(Li2(x)− 14Li2(x2))=ddx{8F2,1(1, x)} .Eq. (5.4.10) justifies this final step. Since the derivatives of 8F2,1(1, x) andϕ(x) are equal on the open unit disk, and since both functions vanish atzero, we may conclude that 8F2,1(1, x) = ϕ(x).The proof of Eq. (5.4.9) requires a remark. Despite the fact that theright-hand side of Eq. (5.4.9) is single valued and analytic whenever |x| < 1,the individual terms involving Li3(1−x) and log(x) are multivalued for x ∈(−1, 0). To avoid all ambiguity, we can simply use F2,1(1, x) = −F2,1(1,−x)to calculate the function at negative real arguments .Theorem 5.4.5. The function h3(v) is analytic on C−{(−i∞,−i] ∪ [i, i∞)}.If v ∈ (−i∞,−i] ∪ [i, i∞), then h3(v) can be expressed in terms of polyloga-116Chapter 5. Trigonometric integrals and Mahler measuresrithms. Let r = v1+√1+v2, thenh3(v) =12Li3(r2) + 4Li3(1− r) + 4Li3(r1 + r)− 4ζ(3)− log(1 + r1− r)Li2(r2)− 2π23log(1− r)− 23log3(1 + r)+ 2 log(r) log2(1− r).(5.4.12)We can recover an equivalent form of Condon’s identity by letting v = 2:h3(2) =145ζ(3). (5.4.13)Proof. This proof is very simple since we have already completed all of thehard computations. Observe from Eq. (5.4.2) that if |v| < 1,h3(v) =∫ v0h2(u)udu. (5.4.14)Lemma 5.4.2 shows that h2(v) is analytic provided that v ∈ (−i∞,−i] ∪[i, i∞). If we assume that the path of integration does not pass througheither of these branch cuts, then it is easy to see that Eq. (5.4.14) providesan analytic continuation of h3(v) to C− {(−i∞,−i] ∪ [i, i∞)}.Next we will prove Eq. (5.4.12). Substituting Eq. (5.4.4) into Eq.(5.4.14) yields an infinite series for h3(v) that is valid whenever v ∈ (−i∞,−i]∪[i, i∞). We haveh3(v) = 4∞∑n=01(2n + 1)2∫ v01u(u1 +√1 + u2)2n+1du.The nested integrals can be evaluated by Lemma 5.4.3. Letting r = v1+√1+v2it is clear thath3(v) =4∞∑n=01(2n + 1)2(log(1 + r1− r)+r2n+12n + 1− 2n∑j=0r2j+12j + 1)=π22log(1 + r1− r)+ 4Li3(r)− 12Li3(r2)− 8F2,1(1, r),(5.4.15)where F2,1(1, r) has a closed form provided by Eq. (5.4.9). Since |r| < 1whenever v ∈ (−i∞,−i] ∪ [i, i∞), we may substitute Eq. (5.4.9) to finishthe calculation.117Chapter 5. Trigonometric integrals and Mahler measuresObserve that when v = 2, we have r =√5−12 . It is easy to verify that3−√52 = r2 = 1− r = r1+r . Using Eq. (5.4.12), it follows thath3(2) =172Li3(3−√52)− 4ζ(3)− 3 log(1 +√52)Li2(3−√52)+4π23log(1 +√52)− 263log3(1 +√52).(5.4.16)Eq. (5.4.13) follows immediately from substituting the classical formulasfor Li3(3−√52)and Li2(3−√52)into Eq. (5.4.16) (see [100], pages 248 and249).Notice that Eq. (5.4.13) is equivalent to a new evaluation of the 4F3hypergeometric function,4F3[1,1, 12, 1232, 32, 32∣∣∣∣− 4] = 710ζ(3). (5.4.17)Corollary 5.4.6. If r = v1+√1+v2and v ∈ (−i∞,−i] ∪ [i, i∞), thenTS(v, 1) =π2∫ v0tan−1(u)udu− 14Li3(r2)− 2Li3(1− r)− 2Li3(r1 + r)+ 2ζ(3) +12log(1 + r1− r)Li2(r2) +π23log(1− r)+13log3(1 + r)− log(r) log2(1− r),(5.4.18)TS(2, 1) =π2∫ 20tan−1(u)udu− 75ζ(3). (5.4.19)Proof. Eq. (5.4.18) follows immediately from substituting Eq. (5.4.12) intoEq. (5.4.3), while Eq. (5.4.19) follows from combining Eq. (5.4.13) withEq. (5.4.3).The fact that we can reduce h1(v), h2(v) and h3(v) to standard poly-logarithms is somewhat miraculous. Integrating Eq. (5.4.15) again, we can118Chapter 5. Trigonometric integrals and Mahler measuresshow thath4(v) =π24(log(1− r2) log(1− r1 + r)+ 2Li2(1− r2)− 2Li2(1 + r2))+ π2F2(r) + 4F3(r)− 8F3,1(1, r)− 8F2,2(1, r)+ 16F2,1,1(1, 1, r).(5.4.20)Considering the complexity of these multiple polylogarithms, it seems un-likely that hn(v) will reduce to standard polylogarithms for n ≥ 4.5.5 Relations between S(v, 1) and Mahler’smeasure, and a closed form for S(v, w).In this section we will study the double arcsine integral, S(v, w). Recall thatwe defined S(v, w) with an integral:S(v, w) =∫ 10sin−1(vx) sin−1(wx)xdx.First, we will show that both S(v, 1) and S(v, v) reduce to standard poly-logarithms. Next, we will discuss several interesting results relating S(v, 1)and S(v, v) to Mahler’s measure and binomial sums. Finally, Theorem 5.5.4concludes this section by expressing S(v, w) in terms of polylogarithms.Theorem 5.5.1. If 0 ≤ v ≤ 1, then S(v, v) and S(v, 1) both have simpleclosed forms:S(v, 1) =π2∫ v0sin−1(x)xdx−(Li3(v)− Li3(−v)2), (5.5.1)S(v, v) =⎛⎝Li3(e2i sin−1(v))+ Li3(e−2i sin−1(v))4⎞⎠− ζ(3)2+ sin−1(v)⎛⎝Li2(e2i sin−1(v))− Li2(e−2i sin−1(v))2i⎞⎠+(sin−1(v))2 log(2v).(5.5.2)Proof. To prove Eq. (5.5.1), we will substitute the Taylor series for sin−1(vx)into the integral S(v, 1) =∫ 10sin−1(vx) sin−1(x)x dx. After swapping the order119Chapter 5. Trigonometric integrals and Mahler measuresof summation and integration, we haveS(v, 1) = 2∞∑n=012n + 1(2nn)(v2)2n+1 ∫ 10sin−1(x)x2ndx= π∞∑n=01(2n + 1)2(2nn)(v2)2n+1 − ∞∑n=0v2n+1(2n + 1)3=π2∫ v0sin−1(x)xdx−(Li3(v)− Li3(−v)2).To prove (5.5.2) make the u-substitution x = sin(t)v , and then integrateby parts as follows:S(v, v) =∫ 10(sin−1(vx))2xdx =∫ sin−1(v)0t2 cot(t)dt=(sin−1(v))2 log(v)− 2∫ sin−1(v)0t log(sin(t))dt.Next substitute the Fourier series for log(sin(t)) into the previous equation.Recall thatlog(sin(t)) = − log(2)−∞∑n=1cos(2nt)nis valid for 0 < t < π. Integrating by parts a second time completes theproof.The function S(v, v) provides a connection to a second family of inter-esting binomial sums. If we recall the formula (see [106], page 61)(sin−1(x))2 =12∞∑n=1(2x)2nn2(2nn) ,then it is immediately obvious that if |v| ≤ 1 we must haveS(v, v) =14∞∑n=1(2v)2nn3(2nn) . (5.5.3)Comparing Eq. (5.5.3) with Eq. (5.5.2) yields a classical formula:S(12,12)=14∞∑n=11n3(2nn) = 12∞∑n=1cos(πn3)n3− ζ(3)2+π6∞∑n=1sin(πn3)n2. (5.5.4)120Chapter 5. Trigonometric integrals and Mahler measuresProposition 5.5.2. If v ∈ [0, 1] and w ∈ (0, 1], we haveS(v, w) = sin−1(w)∫ v0sin−1(u)udu− π2∫ sin−1(w)0m(2vwsin(θ) + y + z)dθ.(5.5.5)Proof. This proof is similar to the proof of Proposition 5.3.1. After anintegration by parts, and the u-substitution u = sin(θ)/w, we obtainS(v, w) = sin−1(w)∫ v0sin−1(u)udu− π2∫ sin−1(w)0∫ vwsin(θ)0sin−1(z)zdzdθ.Since 0 ≤ v ≤ 1 and 0 < w ≤ 1, it follows that 0 ≤ vw sin(θ) ≤ 1. Thereforewe may complete the proof by substituting Eq. (5.2.1) for the nested arcsineintegral.Corollary 5.5.3. We can recover Vandervelde’s formula by letting w = 1in Eq. (5.5.5):m(v(1 + x) + y + z) =2π∫ v0sin−1(u)udu− 4π2S(v, 1)=4π2(Li3(v)− Li3(−v)2) (5.5.6)Notice that if v = w = 12 in Eq. (5.5.5), we haveS(12,12)=π6∫ 1/20sin−1(u)udu− π2∫ π/60m(2 sin(θ) + y + z) dθ=π6∫ 1/20sin−1(u)udu− π212m(1− x1/6 + y + z)(5.5.7)Comparing Eq. (5.5.7) to Eq. (5.5.4) allows us to express a famous binomialsum as the Mahler measure of a three-variable algebraic function.The final result of this section allows us to express S(v, w) in terms ofstandard polylogarithms.121Chapter 5. Trigonometric integrals and Mahler measuresTheorem 5.5.4. Suppose that 0 ≤ v < w ≤ 1, and let θ = sin−1(w) −sin−1(v). Then we have2S(v, w) =S(v, v) + S(w,w)− S (sin(θ), sin(θ))− 2Li3( vw)+ Li3( vweiθ)+ Li3( vwe−iθ)− iθLi2( vweiθ)+ iθLi2( vwe−iθ)+θ22log(1 +v2w2− 2vwcos(θ)).(5.5.8)Notice that Eq. (5.5.2) reduces S(v, v), S(w,w), and S (sin(θ), sin(θ)) tostandard polylogarithms.Proof. The details of this proof are not particularly difficult. First observethe following trivial formula:S(v, v)− 2S(v, w) + S(w,w) =∫ 10(sin−1(wu)− sin−1(vu))2udu.Rearranging, and then applying the arcsine addition formula yields2S(v, w) =S(v, v) + S(w,w)−∫ 10(sin−1(wu√1− v2u2 − vu√1− w2u2))2udu.(5.5.9)This substitution is justified by the monotonicity of the arcsine function. Inparticular, 0 ≤ v < w ≤ 1 implies that 0 ≤ sin−1(wu) − sin−1(vu) ≤ π2 forall u ∈ [0, 1].Next we will make the u-substitution z = wu√1− v2u2− vu√1− w2u2.In particular, we can show thatu2 =z2w2 + v2 − 2vw√1− z2 ,and we can easily verify that1ududz=1z− vwz(v2 + w2 − 2vw√1− z2)√1− z2 .122Chapter 5. Trigonometric integrals and Mahler measuresObserve that the new path of integration will run from z = 0 to z = sin(θ) =w√1− v2 − v√1− w2. Therefore, Eq. (5.5.9) becomes2S(v, w) =S(v, v) + S(w,w)−∫ sin(θ)0(sin−1 (z))2(1z− vwz(v2 + w2 − 2vw√1− z2)√1− z2)dz=S(v, v) + S(w,w)− S (sin(θ), sin(θ))+∫ sin(θ)0(sin−1 (z))2 vwz(v2 + w2 − 2vw√1− z2)√1− z2dz.If we let t = sin−1(z), then this last integral becomes2S(v, w) =S(v, v) + S(w,w)− S (sin(θ), sin(θ))+∫ θ0t2vw sin(t)v2 + w2 − 2vw cos(t)dt.(5.5.10)Since 0 ≤ v < w ≤ 1, a formula from [106], page 48, shows thatvw sin(t)v2 + w2 − 2vw cos(t) =∞∑n=1( vw)nsin(nt). (5.5.11)The Fourier series in Eq. (5.5.11) converges uniformly since v < w. It followsthat we may substitute Eq. (5.5.11) into Eq. (5.5.10), and then swap theorder of summation and integration to obtain:2S(v, w) =S(v, v) + S(w,w)− S (sin(θ), sin(θ))+∞∑n=1( vw)n ∫ θ0t2 sin(nt)dt.(5.5.12)Simplifying Eq. (5.5.12) completes the proof of Eq. (5.5.8).5.6 q-series for the dilogarithm, and someassociated trigonometric integralsIn this section we will prove several double q-series expansions for the dilog-arithm. While these formulas are relatively simple, it appears that they arenew. The first of these formulas, Eq. (5.6.8), follows from a few simplemanipulations of Eq. (5.5.1). The remaining formulas follow from integrals123Chapter 5. Trigonometric integrals and Mahler measuresthat we have evaluated in Theorem 5.6.5. Recall that Theorem 5.6.5 figuredprominently in the proof of Theorem 5.2.3.In this section, the twelve Jacobian elliptic functions will play an im-portant role our calculations. Recall that the Jacobian elliptic functions aredoubly periodic and meromorphic on C. The Jacobian sine function, sn(u),inverts the incomplete elliptical integral of the first kind. If u ∈ C is anarbitrary number, then under a suitable path of integration:u =∫ sn(u)0dz√(1− z2)(1− k2z2) .The Jacobian amplitude can be defined by the equation sn(u) = sin(am(u)),and the Jacobian cosine function is defined by cn(u) = cos(am(u)). As usualthe complementary sine function is given by dn(u) =√1− k2sn2(u). Noticethat every Jacobian elliptic function implicitly depends on k; this parameterk is called the elliptic modulus.Following standard notation, we will denote the real one-quarter periodof sn(u) by K. Since sn(K) = 1, we may compute K from the usual formulaK := K(k) =∫ 10dz√(1− z2)(1− k2z2)=π2 2F1[12, 121∣∣k2].Let K ′ = K(√1− k2), and finally define the elliptic nome by q = e−π K′K .Proposition 5.6.1. If k ∈ (0, 1), then we have the following integral:∫ K0am(u)cn(u)du =π2sin−1(k)k−(Li2(k)− Li2(−k)2k). (5.6.1)Proof. Taking the derivative of each side of Eq. (5.5.1), we obtain:ddkS(k, 1) =∫ 10sin−1(x)√1− k2x2dx =π2sin−1(k)k−(Li2(k)− Li2(−k)2k).(5.6.2)Making the u-substitution x = sn(u) completes the proof.We will need the following two inversion formulas for the elliptic nome.124Chapter 5. Trigonometric integrals and Mahler measuresLemma 5.6.2. Let q be the usual elliptic nome. Suppose that q ∈ (0, 1),then q is invertible using either of the formulas:k = sin(4∞∑n=0(−1)n2n + 1qn+1/2(1 + q2n+1)), (5.6.3)k = tanh(4∞∑n=012n + 1qn+1/2(1− q2n+1)). (5.6.4)Proof. To prove Eq. (5.6.3) observe thatsin−1(k) = k∫ 10dx√1− k2x2= k∫ K0cn(u)du (5.6.5)Recall the Fourier series expansion for cn(u) (see [106], page 916):cn(u) =2πkK∞∑n=0qn+1/21 + q2n+1cos(π(2n + 1)2Ku). (5.6.6)Since 0 < q < 1, this Fourier series converges uniformly. It follows thatwe may substitute Eq. (5.6.6) into Eq. (5.6.5), and then swap the order ofsummation and integration to obtain:sin−1(k) =2πK∞∑n=0qn+1/21 + q2n+1∫ K0cos(π(2n + 1)2Ku)du= 4∞∑n=0(−1)n2n + 1qn+1/2(1 + q2n+1). (5.6.7)Eq. (5.6.3) follows immediately from taking the sine of both sides of theequation.Eq. (5.6.4) can be proved in a similar manner when starting from theintegraltanh−1(k) = k∫ 10dx1− k2x2 .Next we will utilize the Fourier-series expansions for the Jacobian ellipticfunctions to prove the following theorem:125Chapter 5. Trigonometric integrals and Mahler measuresTheorem 5.6.3. If q is the usual elliptic nome, then the following formulaholds for the dilogarithm:Li2(k)− Li2(−k)8=∞∑n=01(2n + 1)2qn+1/2(1 + q2n+1)+ 4∞∑n=0m=11(2n + 1)2 − (2m)2qn+m+1/2(1 + q2m)(1 + q2n+1)(5.6.8)Proof. We have already stated the Fourier series expansion for cn(u) in Eq.(5.6.6). We will also require the Fourier series [106] for am(u):am(u) =π2Ku + 2∞∑n=11nqn1 + q2nsin(πnKu). (5.6.9)Substituting Eq. (5.6.6) and Eq. (5.6.9) into the integral in Eq. (5.6.1), andthen simplifying yields:Li2(k)− Li2(−k)8− π8sin−1(k)= −π2∞∑n=0(−1)n2n + 1qn+1/2(1 + q2n+1)+∞∑n=01(2n + 1)2qn+1/2(1 + q2n+1)+ 4∞∑n=0m=11(2n + 1)2 − (2m)2qn+m+1/2(1 + q2m)(1 + q2n+1).(5.6.10)This proof is nearly complete, the final step is to substitute the identitysin−1(k) = 4∞∑n=0(−1)n2n + 1qn+1/2(1 + q2n+1)into Eq. (5.6.10). This formula for sin−1(k) follows immediately fromLemma 5.6.2.The fact that Eq. (5.6.8) follows easily from an integral of the form∫ K0am(u)ϕ(u)du,126Chapter 5. Trigonometric integrals and Mahler measuressuggests that we should try to generalize Eq. (5.6.8) by allowing ϕ(u) toequal one of the other eleven Jacobian elliptic functions. Theorem 5.6.5proves that ten of these eleven integrals reduce to dilogarithms and elemen-tary functions. First, Theorem 5.6.4 will prove that the one exceptionalintegral can be expressed as the Mahler measure of an elliptic curve.Theorem 5.6.4. The following formulas hold whenever k ∈ (0, 1]:m(4k+ x +1x+y +1y)= − log(k1 +√1− k2)+2π∫ 10sin−1(x)x√1− k2x2dx (5.6.11)= − log(k1 +√1− k2)+2π∫ K0am(u)cn(u)sn(u)du. (5.6.12)Proof. First observe that if k ∈ R and 0 < k ≤ 1, thenm(4k+ x +1x+ y +1y)= − log(k4)+m(1 +k4(x +1x+ y +1y)).For brevity let ϕ(k) = m(1 + k4(x + 1x + y +1y)). Making the change ofvariables (x, y)→ (x/y, yx), we have:ϕ(k) =m(1 +k4(x + x−1) (y + y−1))=m(k4(y + y−1))+m(x2 +4k(1y + y−1)x + 1)= log(k4)+m(x2 +4k(1y + y−1)x + 1).Applying Jensen’s formula with respect to x reduces ϕ(k) to a pair of one-dimensional integrals:ϕ(k) = log(k4)+12π∫ 2π0log+∣∣∣∣∣1 +√1− k2 cos2(θ)k cos(θ)∣∣∣∣∣ dθ+12π∫ 2π0log+∣∣∣∣∣1−√1− k2 cos2(θ)k cos(θ)∣∣∣∣∣ dθ.(5.6.13)The right-hand integral vanishes under the assumption that 0 < k ≤ 1.Therefore, it follows that Eq. (5.6.13) reduces toϕ(k) = log(k4)+2π∫ π/20log(1 +√1− k2 cos2(θ)k cos(θ))dθ.127Chapter 5. Trigonometric integrals and Mahler measuresWith the observation that∫ π/20 log (cos(θ)) dθ = −π2 log(2), this formulabecomes:ϕ(k) =2π∫ π/20log(1 +√1− k2 cos2(θ)2)dθ. (5.6.14)Making the u-substitution of x = cos(θ), we obtainϕ(k) =2π∫ 10log(1 +√1− k2x22)1√1− x2dx.Integrating by parts to eliminate the logarithmic term yields:ϕ(k) = log(1 +√1− k22)+2π∫ 10sin−1(x)x(1−√1− k2x2√1− k2x2)dx= log(1 +√1− k22)+2π∫ 10sin−1(x)x√1− k2x2dx−2π∫ 10sin−1(x)xdx.Since∫ 10sin−1(x)x dx =π2 log(2), it follows thatϕ(k) = log(1 +√1− k24)+2π∫ 10sin−1(x)x√1− k2x2dx,from which we obtainm(4k+ x +1x+ y +1y)= − log(k1 +√1− k2)+2π∫ 10sin−1(x)x√1− k2x2dx.To prove Eq. (5.6.12) simply make the u-substitution x = sn(u).The elliptic curve defined by the equation 4/k+x+1/x+y+1/y = 0 wasone of the simplest curves that Boyd studied in [103]. Rodriguez Villegasderived q-series expansions for a wide class of functions defined by the Mahlermeasures of elliptic curves in [113]. We can recover one of his results bysubstituting the Fourier series expansions for am(u) and cn(u)/sn(u) intoEq. (5.6.12).If we let k = sin(θ), and then integrate Eq. (5.6.11) from θ = 0 to θ = π2 ,we can prove thatm(8 +(z +1z)(x +1x+ y +1y))=4πG +4π2∫ 10sin−1(x)xK(x)dx.(5.6.15)128Chapter 5. Trigonometric integrals and Mahler measuresUsing Mathematica, we can reduce the right-hand integral to a rather com-plicated expression involving balanced hypergeometric functions evaluatedat one.Theorem 5.6.5. We will assume that 0 < k < 1 and that each Jacobianelliptic function has modulus k. Let p =√1−k1+k , r =k1+√1−k2 , and s =k√1−k2 , then∫ K0am(u)sn(u)du =∫ 10u sin−1(u)√(1− u2)(1− k2u2)du=Li2(is)− Li2(−is)2ki(5.6.16)∫ K0am(u)cn(u)du =∫ 10sin−1(u)√1− k2u2du=π2sin−1(k)k− Li2(k)− Li2(−k)2k(5.6.17)∫ K0am(u)dn(u)du =π28(5.6.18)∫ K0am(u)1sn(u)du =∫ 10sin−1(u)u√(1− u2)(1− k2u2)du= −π2log (p) +Li2(ip)− Li2(−ip)i(5.6.19)∫ K0am(u)1cn(u)du =∞ (5.6.20)∫ K0am(u)1dn(u)du =∫ 10sin−1(u)(1− k2u2)√1− u2du=1√1− k2(π28+Li2(r2)− Li2(−r2)2) (5.6.21)∫ K0am(u)sn(u)cn(u)du =∞ (5.6.22)∫ K0am(u)sn(u)dn(u)du =∫ 10u sin−1(u)(1− k2u2)√1− u2du=Li2(r)− Li2(−r)k√1− k2(5.6.23)129Chapter 5. Trigonometric integrals and Mahler measures∫ K0am(u)cn(u)sn(u)du =∫ 10sin−1(u)u√1− k2u2du=π2log (r) +π2m(4k+ x +1x+ y +1y) (5.6.24)∫ K0am(u)cn(u)dn(u)du =∫ 10sin−1(u)1− k2u2du= − π2klog(p)− Li2(ir)− Li2(−ir)ki(5.6.25)∫ K0am(u)dn(u)sn(u)du = 2G (5.6.26)∫ K0am(u)dn(u)cn(u)du =∞ (5.6.27)Proof. First observe that Eq. (5.6.20), Eq. (5.6.22), and Eq. (5.6.27) allfollow from the fact that cn(K) = 0. Similarly, Eq. (5.6.18) and Eq. (5.6.26)both follow from the formula dduam(u) = dn(u).We already proved Eq. (5.6.17) in Proposition 5.6.1, and Eq. (5.6.24)was proved in Theorem 5.6.4. This leaves a total of five formulas to prove.To prove Eq. (5.6.16), observe that after letting u =√1− z2, we have∫ 10u sin−1(u)√(1− u2)(1− k2u2)du =iski∫ 10sin−1(√1− z2)√1− (is)2z2 dz.If 0 < k ≤ 1/√2, then |s| ≤ 1. With this restriction on k, we may expandthe square root in a Taylor series to obtain:=1ki∞∑m=0(−1)m(−1/2m)(is)2m+1∫ 10sin−1(√1− z2)z2mdz=1ki∞∑m=0(is)2m+1(2m + 1)2=Li2(is)− Li2(−is)2ki. (5.6.28)Notice that Eq. (5.6.28) extends to 0 < k < 1, since both sides of theequation are analytic in this interval. Therefore, Eq. (5.6.16) follows imme-diately.To prove Eq. (5.6.19) make the u-substitution u = z√1−k2+z2 . Recalling130Chapter 5. Trigonometric integrals and Mahler measuresthat sin−1(z√1−k2+z2)= tan−1(z√1−k2), we obtain∫ 10sin−1(u)u√(1− u2)(1− k2u2)du =∫ ∞0tan−1(z√1−k2)z√1 + z2dz.Using Mathematica to evaluate this last integral yields:= −π2log(p) +√1− k23F2[12,1,132, 32∣∣∣∣1− k2]= −π2log(p) +12∞∑n=0(2√1− k2)2n+1(2n + 1)2(2nn)= −π2log(p) + 2(Li2(ip)− Li2(−ip)2i),where Eq. (5.4.4) justifies the final step.To prove Eq. (5.6.21) observe that after the u-substitution u = sin(θ)we have ∫ 10sin−1(u)(1− k2u2)√1− u2du =∫ π/20θ1− k2 sin2(θ)dθ.Now substitute the Fourier series√1− k21− k2 sin2(θ) = 1 + 2∞∑m=1(−1)m(k1 +√1− k2)2mcos(2mθ) (5.6.29)into the integral, and simplify to complete the proof.The proof of Eq. (5.6.23) follows the same lines as the derivation of Eq.(5.6.21). Observe that∫ 10u sin−1(u)(1− k2u2)√1− u2du =∫ π/20θ sin(θ)1− k2 sin2(θ)dθ.Now substitute the Fourier seriesk√1− k2 sin(θ)1− k2 sin2(θ) = 2∞∑m=0(−1)m(k1 +√1− k2)2m+1sin ((2m + 1)θ)(5.6.30)into the integral, and simplify to complete the proof.131Chapter 5. Trigonometric integrals and Mahler measuresFinally, we are left with Eq. (5.6.25). Expanding 1/(1 − k2u2) in ageometric series yields:∫ 10sin−1(u)1− k2u2du =∞∑n=0k2n∫ 10sin−1(u)u2ndu=∞∑n=0k2n(π/22n + 1− 22n(2n + 1)2(2nn))= − π4klog(1− k1 + k)− h2(ik)2ik.Substituting the closed form for h2(ik) provided by Eq. (5.4.4) completesthe proof.We can obtain each of the following q-series by applying the methodfrom Theorem 5.6.3 to the formulas in Theorem 5.6.5.Corollary 5.6.6. Let p =√1−k1+k , and let r =k1+√1−k2 . The followingformulas hold for the dilogarithm:Li2(k)− Li2(−k)8=∞∑n=01(2n + 1)2qn+1/2(1 + q2n+1)+ 4∞∑n=0m=11(2n + 1)2 − (2m)2qm+n+1/2(1 + q2m)(1 + q2n+1),(5.6.31)Li2(r)− Li2(−r)4=∞∑n=01(2n + 1)2qn+1/2(1 + q2n+1)+ 4∞∑n=0m=1(−1)m(2n + 1)2 − (2m)2qm+n+1/2(1 + q2m)(1 + q2n+1),(5.6.32)Li2(ip)− Li2(−ip)8i=G4+π16log(p) +∞∑n=0(−1)n(2n + 1)2q2n+1(1− q4n+2)+ 4∞∑n=0m=1(−1)n+m(2n + 1)2 − (2m)2qm+2n+1(1 + q2m)(1− q4n+2) .(5.6.33)132Chapter 5. Trigonometric integrals and Mahler measuresProof. As we have already stated, each of these formulas can be proved bysubstituting Fourier series expansions for the Jacobian elliptic functions intoTheorem 5.6.5.Using the method described, we have already proved Eq. (5.6.31) inTheorem 5.6.3. Eq. (5.6.32) follows in a similar manner from Eq. (5.6.23).Eq. (5.6.33) is a little trickier to prove. Expanding Eq. (5.6.26) in aq-series yields the identity∞∑n=11nqn1 + q2nn−1∑j=0(−1)j2j + 1=∞∑n=0(−1)n(2n + 1)2q2n+11 + q2n+1+ 4∞∑n=0m=1(−1)n+m(2n + 1)2 − (2m)2qm+2n+1(1 + q2m)(1 + q2n+1).(5.6.34)Next expand Eq. (5.6.19) in the q-seriesLi2(ip)− Li2(−ip)4i=G2+π8log(p)+∞∑n=11nqn1 + q2nn−1∑j=0(−1)j2j + 1+∞∑n=0(−1)n(2n + 1)2q2n+11− q2n+1+ 4∞∑n=0m=1(−1)n+m(2n + 1)2 − (2m)2qm+2n+1(1 + q2m)(1− q2n+1) ,and then combine it with Eq. (5.6.34) to complete the proof of Eq. (5.6.33).It is important to notice that the nine convergent integrals in Theorem5.6.5 only produce three interesting q-series for the dilogarithm. The otherq-series we may obtain from Theorem 5.6.5 really just restate known factsabout the elliptic nome. For example, if we expand Eq. (5.6.21) in a q-series, we will obtain Eq. (5.6.31) with q replaced by q2 and k replacedby r2. This is equivalent to the fact that q((k1+√1−k2)2)= q2(k). If welet  =(k1+√1−k2)2, then clearly k and  satisfy a second degree modularequation [101].133Chapter 5. Trigonometric integrals and Mahler measures5.7 A closed form for T(v, w), and Mahlermeasures for T(v, 1v)Recall that we defined T(v, w) using the following integral:T(v, w) =∫ 10tan−1(vx) tan−1(wx)xdx. (5.7.1)Since this integral involves two arctangents, rather than one or two arcsines,T(v, w) possesses a number of useful properties that S(v, w) and TS(v, w)appear to lack.First observe that T(v, w) obeys an eight term functional equation. Ifwe let T(v) =∫ v0tan−1(x)x dx, then we can use properties of the arctangentfunction to prove the following formula:T(v, w) + T(1v,1w)− T(wv, 1)− T( vw, 1)=π2(T(v) + T(1w)− T( vw)− T (1)).(5.7.2)If |v| < 1 and |w| < 1, we can substitute arctangent Taylor series expan-sions into Eq. (5.7.1) to obtain:T(v, w) =∞∑n=0(−1)nw2n+2(2n + 2)2n∑m=0(v/w)2m+12m + 1+∞∑n=0(−1)nv2n+2(2n + 2)2n∑m=0(w/v)2m+12m + 1.(5.7.3)Eq. (5.7.3) immediately reduces T(v, w) to multiple polylogarithms. Theo-rem 5.7.1 improves upon this result by expressing T(v, w) in terms of stan-dard polylogarithms.134Chapter 5. Trigonometric integrals and Mahler measuresTheorem 5.7.1. If v and w are real numbers such that |w/v| ≤ 1, then−4T(v, w) =2Li3(wv)− 2Li3(−wv)+ Li3(1− vi1− wi)+ Li3(1 + vi1 + wi)− Li3(1 + vi1− wi)− Li3(1− vi1 + wi)− Li3(w(1− vi)v(1− wi))− Li3(w(1 + vi)v(1 + wi))+ Li3(−w(1 + vi)v(1− wi))+ Li3(−w(1− vi)v(1 + wi))+ log(1 + v21 + w2)(Li2(wv)− Li2(−wv))− 4 tan−1(v)(Li2(wi)− Li2(−wi)2i)− 4 tan−1(w)(Li2(vi)− Li2(−vi)2i)− π log(1 + v21 + w2)tan−1(w) + 4 log(v) tan−1(v) tan−1(w).(5.7.4)Proof. Substituting logarithms for the inverse tangents, we obtain−4T(v, w) =∫ 10log(1 + ivu1− ivu)log(1 + iwu1− iwu)duu=∫ iw0log(1 + vwu1− vwu)log(1 + u1− u)duu.The identity then follows (more or less) immediately from four applications135Chapter 5. Trigonometric integrals and Mahler measuresof Lewin’s formula∫ x0log (1− z) log (1− cz) dzz=Li3(1− cx1− x)+ Li3(1c)+ Li3(1)− Li3(1− cx)− Li3(1− x)− Li3(1− cxc(1− x))+ log(1− cx)[Li2(1c)− Li2(x)]+ log(1− x)[Li2(1− cx)− Li2(1c)+π26]+12log(c) log2(1− x),(5.7.5)which was proved in [110], page 270. Condon has discussed the intricaciesof applying this equation in [105].This closed form for T(v, w) is quite complicated. Notice that a slightchange in the integrand in Eq. (5.7.1) produces a remarkably simplifiedformula:∫ 10tan−1(vx) tan−1(wx)√1− x2 dx = π∞∑n=0(v1+√1+v2w1+√1+w2)2n+1(2n + 1)2. (5.7.6)To prove Eq. (5.7.6), make the u-substitution x = sin(θ), and then applyEq. (5.3.16) twice.There are two special cases of Eq. (5.7.4) worth mentioning. Firstobserve that T(v, 1v)reduces to a very simple expression. If we let w → 1/vin Eq. (5.7.4), and perform a few torturous manipulations, we can showthatT(v,1v)=π2Im [Li2(iv)]− 12(Li3(v2)− Li3(−v2))+log(v)2(Li2(v2)− Li2(−v2)).(5.7.7)Lal´ın obtained an equivalent form of Eq. (5.7.7) using a different method.(See Appendix 2 in [107]. Lal´ın’s formula for T(v, 1) + T(1/v, 1) reduces toEq. (5.7.7) after applying Eq. (5.7.2) with w = 1/v). Observe that when136Chapter 5. Trigonometric integrals and Mahler measuresw = v in Eq. (5.7.4), we haveT(v, v) =12Re[Li3(1 + vi1− vi)− Li3(−1 + vi1− vi)]− 78ζ(3)+ 2 tan−1(v)Im [Li2(iv)]− log(v)(tan−1(v))2.(5.7.8)Finally, it appears that T(v, 1) does not reduce to any particularly simpleexpression. Letting w → 1 fails to simplify Eq. (5.7.4) in any appreciableway. Expanding T(v, 1) in a Taylor series results in an equally complicatedexpression:T(v, 1) =12∞∑n=0v2n+1(2n + 1)2n∑k=1(−1)k+1k+π4∫ v0tan−1(x)xdx− log(2)4(Li2(v)− Li2(−v)) .(5.7.9)Theorem 5.7.3 relates T(v, w) to three-variable Mahler measures, andgeneralizes one of Lal´ın’s formulas. Once again, we will need a simple lemmabefore we prove our theorem.Lemma 5.7.2. If v and w are positive real numbers, thenT(v, w) = tan−1(v)∫ w0tan−1(u)udu−∫ tan−1(v)0∫ wvtan(θ)0tan−1(z)zdzdθ,(5.7.10)T(v,1v)=π2∫ v0tan−1(u)udu− 12∫ π/20∫ v2 tan(θ)0tan−1(z)zdzdθ. (5.7.11)Proof. While we can verify Eq. (5.7.10) with a trivial integration by parts,the proof of Eq. (5.7.11) is slightly more involved.To prove Eq. (5.7.11), first let w = 1v in Eq. (5.7.10). This producesT(v,1v)=tan−1(v)∫ 1/v0tan−1(u)udu−∫ tan−1(v)0∫ tan(θ)v20tan−1(z)zdzdθ.(5.7.12)Letting v → 1/v in Eq. (5.7.12) givesT(1v, v)=tan−1(1v)∫ v0tan−1(u)udu−∫ tan−1( 1v )0∫ v2 tan(θ)0tan−1(z)zdzdθ137Chapter 5. Trigonometric integrals and Mahler measures=(π2− tan−1(v))∫ v0tan−1(u)udu−∫ π/2tan−1(v)∫ v2tan(θ)0tan−1(z)zdzdθNow apply Eq. (5.2.4) twice, which transforms this last identity toT(1v, v)=(π2− tan−1(v))(∫ 1v0tan−1(u)udu +π2log(v))−∫ π/2tan−1(v)(∫ tan(θ)v20tan−1(z)zdz − π2log(1v2tan(θ)))dθ(5.7.13)To complete the proof, simply add equations (5.7.12) and (5.7.13) together,and then simplify the resulting sum.Theorem 5.7.3. If v > 0, then the following Mahler measures hold:m(1− v4(1− x1 + x)2+(y + v2(1− x1 + x))2z)=4π∫ v0tan−1(u)udu− 8π2T(v,1v)+12m(1− v4(1− x1 + x)2),(5.7.14)m(1− v4(1− x1 + x)2+ v2(1− x1 + x)(1− y1 + y)(z − z−1))=8π∫ v0tan−1(u)udu− 16π2T(v,1v),(5.7.15)m((y − y−1)+ v2(1− x1 + x)(z − z−1))=4π∫ v0tan−1(u)udu− 8π2T(v,1v),(5.7.16)138Chapter 5. Trigonometric integrals and Mahler measuresm⎛⎜⎜⎜⎜⎜⎝(4(1 + y)2 − (z + z−1)2)(1− v4(1− x1 + x)2)2+(z − z−1)2 (1 + y)2(1 + v4(1− x1 + x)2)2⎞⎟⎟⎟⎟⎟⎠=8π∫ v0tan−1(u)udu− 16π2T(v,1v)+4π∫ π/20log(1 + v2 tan(θ))dθ+ log(2).(5.7.17)Proof. Each of these results follows, in order, from substituting Eq. (5.2.2),Eq. (5.2.5), Eq. (5.2.6), and Eq. (5.2.7), into Eq. (5.7.11).Corollary 5.7.4. The following identities are true:m((1 + z)(1 + y) + (1− z)(x− y)) = 72π2ζ(3) +log(2)2, (5.7.18)m(4(1 + y) + (1− y) (x− x−1) (z − z−1)) = 14π2ζ(3) (5.7.19)m((1 + x)(y − y−1)+ (1− x) (z − z−1)) = 7π2ζ(3) (5.7.20)m(16(1 + y)2 − 4 (z + z−1)2 + (1 + y)2 (z − z−1)2 (x + x−1)2)=14π2ζ(3) +4πG(5.7.21)Proof. To prove Eq. (5.7.18), let v = 1 in Eq. (5.7.14). From Eq. (5.7.7)we know that T(1, 1) = π2G− 78ζ(3), hence7π2ζ(3) + log(2) = m(1−(1− x1 + x)2+(y +1− x1 + x)2z)= m(4x + ((1 + x)y + (1− x))2 z).Now let (x, y, z)→ (x, yz ,−xz2) to obtain7π2ζ(3) + log(2) = m(4x− ((1 + x)y + (1− x)z)2 x)= m(4− ((1 + x)y + (1− x)z)2)139Chapter 5. Trigonometric integrals and Mahler measures= 2m(2 + (1 + x)y + (1− x)z) .With the final change of variables (x, y, z)→(z, 1yz ,xyz), we have7π2ζ(3) + log(2) = 2m(2 +(1 + z)yz+(1− z)xyz)= 2m((1 + z)(1 + y) + (1− z)(x− y)) ,completing the proof of Eq. (5.7.18).The proof of Eq. (5.7.19) through Eq. (5.7.21) follows almost immedi-ately from our evaluation of T(1, 1). The proof Eq. (5.7.21) also requiresthe fairly easy fact that∫ π/20 log (1 + tan(θ)) dθ = G +π4 log(2)5.8 ConclusionIn principle, we should be able to apply the techniques in this paper toprove formulas for infinitely many three-variable Mahler measures. Themain difficulty, which is significant, lies in the challenge of finding infinitelymany Mahler measures for the arctangent and arcsine integrals. In Section5.2 we proved one such formula for the arcsine integral, and four formulasfor the arctangent integral.5.9 AcknowledgementsI would like to thank my advisor, David Boyd, for bringing Condon’s paperto my attention. He also made several useful suggestions on evaluating theintegrals in Theorem 5.6.5. I truly appreciate his support.Finally, I would like the thank the Referee for helpful suggestions.140Bibliography[99] N. Batir, Integral representations of some series involving(2kk)−1k−nand some related series, Applied Math. and Comp. 147 (2004), 645-667.[100] B.C. Berndt, Ramanujan’s Notebooks part I, Springer-Verlag, NewYork, 1985.[101] B.C. Berndt, Ramanujan’s Notebooks part IV, Springer-Verlag, NewYork, 1994.[102] D.W. Boyd, Speculations concerning the range of Mahler’s measure,Canad. Math. Bull. 24 (1981), 453-469.[103] D.W. Boyd, Mahler’s measure and special values of L-functions, Ex-periment. Math. 7 (1998), 37-82.[104] D.W. Boyd, F. Rodriguez Villegas, Mahler’s measure and the diloga-rithm (I), Canad. J. Math. 54 (2002), 468-492.[105] J. Condon, Calculation of the Mahler measure of a three variable poly-nomial, (preprint, October 2003).[106] I.S. Gradshteyn, I.M. Ryzhik, Table of Integrals, Series and Products,Academic Press, 1994.[107] M. N. Lal´ın, Some examples of Mahler measures as multiple polylog-arithms, J. Number Theory. 103 2003, 85-108.[108] M. N. Lal´ın, Mahler measure of some n-variable polynomial families,(preprint 2004, to appear in J. Number Theory).[109] M. N. Lal´ın, Some relations of Mahler measure with hyperbolic vol-umes and special values of L-functions, Doctoral Dissertation, The Uni-versity of Texas at Austin, 2005.141Bibliography[110] L. Lewin, Polylogarithms and Associated Functions, Elsevier NorthHolland, New York, 1981.[111] V. Maillot, Ge´ome´trie d’Arakelov des varie´te´s toriques et fibre´s endroites inte´grables. Me´m. Soc. Math. Fr. (N.S) 80 (2000), 129pp.[112] S. Ramanujan, On the integral∫ x0tan−1(t)t dt, J. Ind. Math. Soc. 7(1915), 93-96.[113] F. Rodriguez Villegas, Modular Mahler measures I, Topics in numbertheory (University Park, PA, 1997), 17–48, Math. Appl., 467, KluwerAcad. Publ., Dordrecht, 1999.[114] C.J. Smyth, An explicit formula for the Mahler measure of a family of3-variable polynomials, J. Th. Nombres Bordeaux, 14 (2002), 683-700.[115] S. Vandervelde, A formula for the Mahler measure of axy+bx+cy+d,J. Number Theory, 100 (2003), 184-202.Department of Mathematics, University of British Columbia,Vancouver, B.C. V6T 1Z2, Canadamatrogers@math.ubc.ca142Chapter 6Conclusion6.1 Computational proofs?We will conclude this thesis by raising the question as to whether or notcomputational proofs exist for formulas like (1.1.1) and (1.1.2). Indeed,algorithms such as the PSLQ algorithm and the WZ algorithm now makeit possible for computers to both discover and prove interesting formulaswith relatively little human assistance (see [119] and [116]). For example,Guillera recently proved the following hypergeometric series for 1/π2 usingpurely computational methods [118]:128π2=∞∑n=0(−1)n220n(2nn)5 (820n2 + 180n + 13). (6.1.1)The novelty of Guillera’s approach is that it applies to both Ramanujan’soriginal formulas for 1/π [120], and to computationally discovered examplessuch as (6.1.1). At the very least, it seems extremely likely that a WZproof should exist for equation (1.1.1), since that identity closely resemblesformulas in [118]. Such an approach could also be used to eliminate thealgebraic K theory from proofs of similar formulas in [121] and [117].It seems more difficult to speculate on whether or not computationalproofs should exist for identities such as (1.1.2). In Chapter 2 we reducedthat identity to an equivalent relation between a lattice sum and a hyper-geometric function:∞∑n=0(2nn)2 (1/16)2n+12n + 1?=540π2∞∑ni=−∞i∈{1,2,3,4}(−1)n1+n2+n3+n4((6n1 − 1)2 + 3(6n2 − 1)2 + 5(6n3 − 1)2 + 15(6n4 − 1)2)2.(6.1.2)For reasons outlined in Section 2.4, it seems likely that equation (6.1.2) is143Chapter 6. Conclusionreally a special case of a more general formula involving Meijer G-functions.So far, we have been unable to prove or disprove this last statement.144Bibliography[116] D. H. Bailey, J. M. Borwein, Mathematics by experiment. Plausiblereasoning in the 21st century. A K Peters, Ltd., Natick, MA, 2004.x+288 pp.[117] M. J. Bertin, Mesure de Mahler d’une famille de polynoˆmes. J. ReineAngew. Math. 569 (2004), 175–188.[118] J. Guillera, Hypergeometric identities for 10 extended Ramanujan-type series, to appear in the Ramanujan Journal.[119] M. Petkovsˇek, H. S. Wilf, D. Zeilberger, A = B. With a foreword byDonald E. Knuth. A K Peters, Ltd., Wellesley, MA, 1996.[120] S. Ramanujan, Modular equations and approximations to π, [Quart.J. Math. 45 (1914), 350-372]. Collected papers of Srinivasa Ramanujan,23-29, AMS Chelsea Publ., Providence, RI, 2000.[121] F. Rodriguez-Villegas, Identities between Mahler measures, Numbertheory for the millennium, III (Urbana, IL, 2000), 223–229, A K Peters,Natick, MA, 2002.Department of Mathematics, University of British Columbia,Vancouver, B.C. V6T 1Z2, Canadamatrogers@math.ubc.ca145

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0066504/manifest

Comment

Related Items