Values of the Ramanujan -function

Ramanujan’sτ -function

MichaelBennett

Values of the Ramanujan τ -function

Michael Bennett (joint with Adela Gherga, Vandita Pateland Samir Siksek)

University of British Columbia (and University of Manchester, WarwickUniversity)

Fields Institute : November 2020

MichaelBennett The Ramanujan τ -function

The Ramanujan τ -function τ(n) is defined via the expansion

∆(z) = q

∞∏n=1

(1− qn)24 =

∞∑n=1

τ(n)qn, q = e2πiz.

Explicitly, we have

∆(z) = q − 24q2 + 252q3 − 1472q4 + 4830q5 − 6048q6 + · · ·

MichaelBennett

The Ramanujan τ -function : multiplicativity

It was conjectured by Ramanujan and proved by Mordell thatτ(n) is a multiplicative function, i.e. that

τ(n1n2) = τ(n1)τ(n2),

for all coprime pairs of positive integers n1 and n2.

MichaelBennett

The Ramanujan τ -function : parity

Further, we have

∞∑n=1

τ(n)qn ≡ q∞∏n=1

(1 + q8n)3 ≡ q∞∏n=1

(1− q8n)(1 + q8n)2

and so∞∑n=1

τ(n)qn ≡∞∑m=0

q(2m+1)2 mod 2,

via Jacobi’s triple product formula, whence τ(n) is oddprecisely when n is an odd square and, in particular, τ(p) iseven for every prime p.

MichaelBennett

The Ramanujan τ -function : arithmetic

Amongst the many open questions about the possible values ofτ(n), the most notorious is an old conjecture of Lehmer to theeffect that τ(n) never vanishes.

There are many papers in the literature proving results aboutthe arithmetic nature of τ(n) under various hypotheses : onaverage, for most n, subject to GRH, etc. See e.g. work ofMurty-Murty, Garavev, Shparlinski, Konyagin, etc.

MichaelBennett

The Ramanujan τ -function : upper bounds

In terms of the size of values of τ , one has the upper bound ofDeligne (originally conjectured by Ramanujan) :

|τ(p)| ≤ 2 · p11/2,

valid for prime p.

MichaelBennett The Ramanujan τ -function : lower bounds

In the other direction, Atkin and Serre conjectured (as astrengthening of Lehmer’s conjecture) that, for ε > 0,

|τ(p)| �ε p9/2−ε,

so that, in particular, given a fixed integer a, there are at mostfinitely many primes p for which

τ(p) = a.

MichaelBennett

The Ramanujan τ -function : odd values

While this problem remains open, in the special case where theinteger a is odd, Murty, Murty and Shorey (1987) proved thatthe equation

τ(n) = a,

has at most finitely many solutions in integers n (note that, inthis case, n is necessarily an odd square). More precisely, theydemonstrated the existence of an effectively computablepositive constant c such that if τ(n) is odd, then

|τ(n)| > (log(n))c.

MichaelBennett

The Ramanujan τ -function : odd values continued

A number of recent papers have treated the problem ofexplicitly demonstrating that the equation

τ(n) = a

has, in fact, no solutions, for various odd values of a, including

a = ±1 (Lygeros and Rozier)

|a| < 100 an odd prime (Balakrishnan, Craig and Ono,Balakrishnan, Craig, Ono and Tsai, Dembner and Jain)

|a| < 100 an odd integer (Hanada and Madhukara)

MichaelBennett

τ(n) = a

MichaelBennett

τ(n) = a

MichaelBennett

τ(n) = a

MichaelBennett

A non-Archimidean analogue of Murty-Murty-Shorey

Theorem (B., Gherga, Patel, Siksek 2020)

There exists an effectively computable constant κ > 0 suchthat if τ(n) is odd, with n ≥ 25, then either

P (τ(n)) > κlog log log n

log log log log n,

or there exists a prime p | n for which τ(p) = 0.

MichaelBennett

Recall that a powerful number (also known as squarefull or2-full) is defined to be an integer n with the property that if aprime p | n, then necessarily p2 | n.

Equivalently, we can write such an integer as n = a2b3, wherea and b are integers. Our techniques actually show thefollowing (from which the preceding Theorem is an immediateconsequence).

MichaelBennett

We havelimn→∞

P (τ(n)) =∞,

where the limit is taken over powerful numbers n for whichτ(p) 6= 0 for each p | n. More precisely, there exists aneffectively computable constant κ > 0 such that if n ≥ 25 ispowerful, either

log log log log n,

or there exists a prime p | n for which τ(p) = 0.

MichaelBennett

If Lehmer’s Conjecture is false.....

The restriction that n have no prime divisors p for whichτ(p) = 0 is, in fact, necessary if one wishes to obtain a lowerbound upon P (τ(n)) that tends to ∞ with n. Indeed, one mayobserve that, if τ(p) = 0, then

P(τ(p2k)

((−1)kp11k

is bounded, independent of k. While Lehmer’s conjectureremains unproven, we do know that if there is a prime p forwhich τ(p) = 0, then

p > 816212624008487344127999,

by work of Derickx, van Hoeij and Zeng.

MichaelBennett

What we really prove....

We will actually show that, given m ≥ 2 and prime p for whichτ(p) 6= 0,

P (τ(pm))� log log(pm), for m ∈ {2, 3, 4, 5}, say,

P (τ(pm))� log log(pm)

log log log(pm),

more generally, where the implied constants are absolute.

MichaelBennett

Explicit results

To demonstrate that these results and the techniquesunderlying them are somewhat practical, we prove the followingcomputational “coda” :

If n is a powerful positive integer, then either n = 8, where wehave

τ(8) = 29 · 3 · 5 · 11,

orP (τ(n)) ≥ 13.

MichaelBennett

Explicit results

Corollary (B., Gherga, Patel, Siksek 2020)

If n is a positive integer for which τ(n) is odd, then

P (τ(n)) ≥ 13.

In other words, the equation

τ(n) = ±3α5β7γ11δ

has no solutions in integers n ≥ 2 and α, β, γ, δ ≥ 0.

MichaelBennett

The Ramanujan τ -function : prime values

It is conjectured that |τ(n)| takes on infinitely many primevalues, the smallest of which corresponds to

τ(2512) = −80561663527802406257321747.

Our arguments enable us to eliminate the possibility of powersof small primes arising as values of τ . By way of example, wehave the following.

The equationτ(n) = ±qα

has no solutions in prime q with 3 ≤ q < 100 and α, n ∈ Z.

MichaelBennett

Our techniques

Our proofs rely upon two quite different approaches. The first,to derive inequalities of the shape

log log log log n,

proceeds in the same way as the original work ofMurty-Murty-Shorey, appealing to bounds for linear forms inlogarithms (though we necessarily must use both bounds forcomplex and p-adic logarithms). The second (for our explicitresults), uses Frey curves and the modularity of associatedGalois representations.

MichaelBennett

Generalities: coefficients of cuspidal newforms

It is worth observing that the techniques we employ are readilyextended to treat more generally coefficients λf (n) of cuspidalnewforms of (even) weight k ≥ 4 for Γ0(N), with trivialcharacter and λf (p) even for suitably large prime p; our resultscorrespond to the case of ∆(z), where k = 12 and N = 1. Werestrict our attention to this latter case for simplicity.

MichaelBennett

Lucas sequences

A Lucas pair is a pair (α, β) of algebraic numbers such thatα+ β and αβ are non-zero coprime rational integers, and α/βis not a root of unity. In particular, associated to the Lucas pair(α, β) is a characteristic polynomial

X2 − (α+ β)X + αβ ∈ Z[X].

This polynomial has discriminant D = (α− β)2 ∈ Z \ {0}.Given a Lucas pair (α, β), the corresponding Lucas sequenceis given by

un =αn − βn

α− β, n = 0, 1, 2, . . . .

MichaelBennett

Lucas sequences : primitive divisors

Let (α, β) be a Lucas pair. A prime ` is a primitive divisor ofthe n-th term of the corresponding Lucas sequence if ` dividesun but ` fails to divide (α− β)2 · u1u2 . . . un−1.

We shall make essential use of the celebrated Primitive DivisorTheorem of Bilu, Hanrot and Voutier.

Theorem (Bilu, Hanrot and Voutier)

Let (α, β) be a Lucas pair. If n ≥ 5 and n 6= 6 then un has aprimitive divisor.

MichaelBennett

Lucas sequences : rank of apparition

Let ` be a prime. The smallest positive integer m such that` | um is called the rank of apparition of `; we denote this bym`.

The important result is that, if m` is finite, then either m` = `,or m` | `± 1.

MichaelBennett From Ramanujan τ to Lucas sequences

Let us fix a prime p and consider the sequence{1, τ(p), τ(p2), τ(p3), . . .

We will associate to this a Lucas pair and a correspondingLucas sequence. Our starting point is the identity

τ(pm) = τ(p)τ(pm−1)− p11τ(pm−2),

valid for all integer m ≥ 2. Once again, this was conjectured byRamanujan and proved by Mordell.

MichaelBennett

The degenerate case

If Lehmer’s conjecture fails and there exists a prime p for whichτ(p) = 0, then our recursion implies that

τ(pn) = 0 for all odd n, and

τ(pn) = (−1)n/2p11n/2 for even n.

In particular, this does not correspond to a Lucas sequence.

MichaelBennett

The degenerate case

MichaelBennett

The degenerate case

MichaelBennett

Otherwise

If we have that pk | τ(p), we may conclude that k ≤ 5, viaDeligne’s bounds. Supposing that pr ‖ τ(p), we write

un =τ(pn−1)

pr(n−1), n ≥ 1.

We may show that {un} is a Lucas sequence with

α =α′

pr, β =

for α′ and β′ the roots of the quadratic equation

X2 − τ(p)X + p11 = 0.

MichaelBennett

An aside : p | τ(p)

We note that p | τ(p) for

p = 2, 3, 5, 7, 2411, 7758337633, . . .

We expect that p | τ(p) for infinitely many primes p; see workof Lygeros and Rozier for a discussion of this problem andrelated computations.

MichaelBennett Towards the proofs

We haveτ(p2) = τ2(p)− p11,

τ(p3) = τ(p)(τ2(p)− 2p11)

andτ(p4) = p22 − 3p11τ2(p) + τ4(p).

We can rewrite this last equation as

4τ(p4) =(2τ2(p)− 3p11

)2 − 5p22.

MichaelBennett

τ(pm),m ∈ {2, 3, 4, 5}

In each case, we have that

P (τ(pm)) ≥ P (Y 2 −DX11),

for D ≤ 5, X and Y integers.

Appealing to bounds for linear forms in complex and q-adiclogarithms (say due to Bugeaud and Gyory), we find that

P (τ(pm)) ≥ c1 log log(pm), for m ∈ {2, 3, 4, 5},

for some positive constant c1.

MichaelBennett

τ(pm),m ∈ {2, 3, 4, 5}

In each case, we have that

P (τ(pm)) ≥ P (Y 2 −DX11),

for D ≤ 5, X and Y integers.

Appealing to bounds for linear forms in complex and q-adiclogarithms (say due to Bugeaud and Gyory), we find that

P (τ(pm)) ≥ c1 log log(pm), for m ∈ {2, 3, 4, 5},

for some positive constant c1.

MichaelBennett

τ(pm),m ≥ 6

More generally, for m ≥ 6, we can write

τ(pm) = τ δ(p)Fm(p11, τ2(p)),

where δ ∈ {0, 1}, δ ≡ m mod 2 and Fm(X,Y ) is a binaryform with integer coefficients and degree [m/2].Explicitly, expanding

1−√Y T +XT 2

∞∑m=0

Y δ/2 Fm(X,Y )Tm,

we find, for example, that

F6(X,Y ) = −X3 + 6X2Y − 5XY 2 + Y 3.

MichaelBennett

τ(pm),m ≥ 6

Note that from the fact that the Lucas sequences {un} aredivisibility sequences, we have

Fm(X,Y ) | Fn(X,Y ) if n ≡ −1 mod m+ 1,

whence many of these forms are, in fact, reducible. Explicitly,we can write

Fm(X,Y ) =

[m/2]∑j=1

(−1)j(m− jj

)Y [m/2]−jXj

and so

Fm(X,Y ) =

[m/2]∏j=1

(Y − 4X cos2

MichaelBennett τ(pm),m ≥ 6

We observe that

Fm(X,Y ) = Gm(X,Y − 2X), (1)

Gm(X,Y ) =

[m/2]∏j=1

(Y − 2X cos

)). (2)

MichaelBennett

τ(pm),m ≥ 6 : Theorem 1

We write

τ(pm) = prmum+1 = prm · um+1

uqα· uq

uqα−1

· uqα−1 ,

where qα is the largest prime power divisor of m+ 1.

After ruling out some small cases, we have that the termuqα

uqα−1

is an irreducible binary form F (X,Y ) in X = p11−2r andY = p−2rτ2(p), of degree φ(qα)/2 ≥ 3, where φ denotes theEuler phi-function. Explicitly, we have

F (X,Y ) =

φ(qα)/2∏j=1

gcd(j,q)=1

(Y − 4X cos2

MichaelBennett

τ(pm),m ≥ 6 : Theorem 1

We write

τ(pm) = prmum+1 = prm · um+1

uqα· uq

uqα−1

· uqα−1 ,

where qα is the largest prime power divisor of m+ 1.

After ruling out some small cases, we have that the termuqα

uqα−1

is an irreducible binary form F (X,Y ) in X = p11−2r andY = p−2rτ2(p), of degree φ(qα)/2 ≥ 3, where φ denotes theEuler phi-function. Explicitly, we have

F (X,Y ) =

φ(qα)/2∏j=1

gcd(j,q)=1

(Y − 4X cos2

MichaelBennett

Thue-Mahler equations

We haveP (F (X,Y )) ≤ P (τ(pm)).

Putting this all together (using bounds for solutions ofThue-Mahler equations due to Bugeaud and Gyory, we find that

P (τ(pm))� log log(pm)

log log log(pm).

If we choose pm to be the largest prime power divisor of ourpowerful number n, then pm � log n, whereby

P (τ(n))� log log log n

log log log log n,

as claimed.

MichaelBennett

P (τ(pm)) : explicit results

To deal with, for example, the equation

τ(pm) = ±2α3β5γ7δ11ω,

our approach is as follows

Use Ramanujan’s congruences for τ(n) modulo powers of2, 3, 5, 7 and 23 to reduce the possibilities for theexponents

Appeal to the Primitive Divisor Theorem to restrict m

Solve the Thue-Mahler equations associated with m ≥ 6

Treat the equations corresponding to τ(pm), m ≤ 5.

MichaelBennett

τ(pm) = ±2α3β5γ7δ11ω,

MichaelBennett

τ(pm) = ±2α3β5γ7δ11ω,

MichaelBennett

τ(pm) = ±2α3β5γ7δ11ω,

MichaelBennett P (τ(p2)) :Frey curves

If we suppose that

τ(p2) = ±2α3β5γ7δ11ω,

for p odd, then we necessarily have α = γ = δ = 0 andβ ∈ {0, 1}.

We consider the Frey curves{Ep : Y 2 = X(X2 + 2τ(p)X + τ(p)2 − p11) if p ≡ 1 mod 4,

Ep : Y 2 = X(X2 + 2τ(p)X + p11) if p ≡ 3 mod 4.

If we suppose that

τ(p2) = ±2α3β5γ7δ11ω,

for p odd, then we necessarily have α = γ = δ = 0 andβ ∈ {0, 1}. We consider the Frey curves{Ep : Y 2 = X(X2 + 2τ(p)X + τ(p)2 − p11) if p ≡ 1 mod 4,

Ep : Y 2 = X(X2 + 2τ(p)X + p11) if p ≡ 3 mod 4.

MichaelBennett

P (τ(p2)) :Frey curves

It follows from the modularity theorem and Ribet’s levellowering theorem that there is a normalized newform

f = q +

∞∑n=1

of weight 2 and level 25 · 3δ111δ2 and a prime $ | 11 in theintegers of K = Q(c1, c2, . . . ) so that

ρEp,11 ∼ ρf,$.

In particular, we have

NormK/Q(a`(Ep)− c`) ≡ 0 mod 11.

MichaelBennett

P (τ(p2)) :Frey curves

Recall that we had

τ2(p)− p11 = ±3β11ω, β ∈ {0, 1}.

Sieving using

NormK/Q(a`(Ep)− c`) ≡ 0 mod 11,

for ` - 66, ` < 200, we find that either

β = 1, ω = 0 or β = 0, 11 | ω.

MichaelBennett P (τ(p2)) :Frey curves, final steps

We eliminate the first of these cases by reducing it to the Thueequation

U11 + 22U10V + 165U9V 2 + 990U8V 3 + 2970U7V 4

+ 8316U6V 5 + 12474U5V 6 + 17820U4V 7

+ 13365U3V 8 + 8910U2V 9 + 2673UV 10 + 486V 11 = 1.

The Magma Thue equation solver gives that the only solution is(U, V ) = (1, 0), corresponding to p = U2 − 3V 2 = 1, acontradiction.

MichaelBennett

P (τ(p2)) :Frey curves, final steps

Writing ω = 11t, The second case reduces to

p11 ± (11t)11 = τ2(p),

which, via work of Darmon and Merel, has only trivial solutions.

We have Ep of the form Y 2 = F (X) for F (X) one of

X(X2 + (3p11 − 2τ(p)2)X + τ(p)4 − 3p11τ(p)2 + p22)

X(X2 + (2τ(p)2 − 3p11)X + τ(p)4 − 3p11τ(p)2 + p22).

The proof is similar, if somewhat more involved.

We have Ep of the form Y 2 = F (X) for F (X) one of

X(X2 + (3p11 − 2τ(p)2)X + τ(p)4 − 3p11τ(p)2 + p22)

X(X2 + (2τ(p)2 − 3p11)X + τ(p)4 − 3p11τ(p)2 + p22).

The proof is similar, if somewhat more involved.

MichaelBennett

Some references : page 1

J. S. Balakrishnan, W. Craig and K. Ono,Variations of Lehmer’s conjecture for Ramanujan’stau-function,J. Number Theory, to appear.

J. S. Balakrishnan, W. Craig, K. Ono and W.-L. Tsai,Variants of Lehmer’s speculation for newforms,submitted for publication.

Y. Bilu, G. Hanrot and P. Voutier,Existence of primitive divisors of Luca and Lehmernumbers,J. Reine Angew. Math. 539 (2001), 75–122.

MichaelBennett

P. Deligne,La conjecture de Weil I,Inst. Hautes Etudes Sci. Publ. Math. 43 (1974), 273–307.

S. Dembner and V. Jain,Hyperelliptic curves and newform coefficients,submitted for publication.

M. Derickx, M. Van Hoeij and Jinxiang Zeng,Computing Galois representations and equations formodular curves XH(`),arXiv:1312.6819v2.

MichaelBennett

M. Hanada and R. Madhukara,Fourier coefficients of level 1 Hecke eigenforms,submitted for publication.

D. H. Lehmer,The vanishing of Ramaujan’s function τ(n),Duke Math. J. 14 (1947), 429–433.

F. Luca, S. Mabaso and P. Stanica,On the prime factors of the iterates of the Ramanujanτ -function,Proc. Edinburgh Math. Soc., to appear.

MichaelBennett Some references : page 4

F. Luca and I. E. Shparlinski,Arithmetic properties of the Ramanujan function,Proc. Indian Acad. Sci. (Math. Sci.) 116 (2006), 1–8.

N. Lygeros and O. Rozier,A new solution to the equation τ(p) ≡ 0 mod p,J. Integer Sequences 13 (2010), Article 10.7.4.

N. Lygeros and O. Rozier,Odd prime values of the Ramanujan tau function,Ramanujan Journal 32 (2013), 269–280.

MichaelBennett

L. J. Mordell,On Mr. Ramanujan’s empirical expansions of modularfunctions,Proc. Cambridge Phil. Soc. 19 (1917), 117–124.

M. Ram Murty and V. Kumar Murty,Odd values of Fourier coefficients of certain modular forms,

Int. J. Number Theory 3 (2007), 455-470.

MichaelBennett

M. Ram Murty, V. Kumar Murty and T. N. Shorey,Odd values of the Ramanujan τ -function,Bulletin Soc. Math. France 115 (1987), 391–395.

S. Ramanujan,On certain arithmetical functions,Trans. Camb. Philos. Soc. 22 (1916), 159–184.

Values of the Ramanujan -function

Documents

Transcript of Values of the Ramanujan -function

Growth of function 2008. 1. 28 : 1 Chapter 3. Growth of function.

neuron function

Local function vs. local closure function · Local function vs. local closure function ... Let ˝be a topology on X. Then Cl (A) ... [Kuratowski 1933]. Local closure function

Values & means: summary (Falconer & Mackay: chapter 7)

Chapter 10 Eigenvalues and Singular Values 10 Eigenvalues and Singular Values This chapter is about eigenvalues and singular values of matrices. Computational algorithms and sensitivity

TRIGONOMETRIC FUNCTIONS trigonometry solutions.pdf · 2013. 7. 18. · Find the value of the trigonometric function cosec (–1410°) It is known that the values of cosec x repeat

Testing for Serial Correlation by means of Extreme Values

Core Values for Medical Doctor

Discrete Calculus A - rd.springer.com978-3-642-02475-7/1.pdfDiscrete Calculus A ... where h = π/nand n 1 is an integer. The associated function values are deﬁned as s ... 578 A

Dirac Delta Function

Arithmetic of linear forms involving odd zeta values

Ramanujan Type Trigonometric Formulas: 2π The …...23 11 Article 09.8.5 2 Journal of Integer Sequences, Vol. 12 (2009), 3 6 1 47 Ramanujan Type Trigonometric Formulas: The General

X-Ramanujan graphssidhanthm.com/pdf/X-Ramanujan-Graphs.pdf · This special quantity, 2 2, is precisely the spectral radius of the inﬁnite graph T 3. (In general, the spectral radius

NTC Characteristic Values - Vincotech

Reference Guide to Turbo Pascal ProgramsTurbo Pascal Programs ArFcnTab Function Constructs a TABle of values of the six ARithmetic FunCtioNsP !(n) = pjn 1, (n) = P pakn a, (n), d(n)

Chapter 10 Eigenvalues and Singular Values - MathWorks · Chapter 10 Eigenvalues and Singular Values This chapter is about eigenvalues and singular values of matrices. Computational

A FRAMEWORK OF ROGERS{RAMANUJAN IDENTITIES AND …

AVDP8 - sinou-hk.com file・ Format conversion function ・ α-blending function at the rendering (α value can be designated in dot.) ・ Color Mask function ・ Antialias function

Absolute Value of Dirichlet Eta Function - Fractional Calculus · 04 Absolute Value of Dirichlet Eta Function 4.1 Dirichlet Eta Function 4.1.1 Definition Dirichlet Eta Function ()z

Restricted Sine Function.