5
$\begingroup$

A145722 is Expansion of f(q) * f(q^5) / phi(-q^2)^2 in powers of q where f(), phi() are Ramanujan theta functions.

Using the pari program and offset 0, up to $2000$, $$A145722(n-1) \equiv \sigma(4n-3) \pmod{5}$$

Q1 Is this congruence true?

@Gjergji Zaimi proved similar congruence in another answer.

Since sigma is multiplicative, the offsets bug me a bit.

I would expect if $a(f(n)) \equiv \sigma(g(n))$ and $a$ is defined in a "natural" way, then either $f(n) \mid g(n)$ or $g(n) \mid f(n)$.

Q2 In case the congruence is true, where $n-1$ and $4n-3$ come from?

Certain sequences (like modular forms, sigma, totient) can be computed significantly faster if the index is factored.

Q3 Does factoring $n-1$ helps in computing $A145722(n-1)$? (I expect no).

$\endgroup$
4
  • $\begingroup$ You may want to add the tag [modular-forms]. $\endgroup$
    – Wolfgang
    Feb 10, 2016 at 9:54
  • $\begingroup$ @Wolfgang I don't understand modular forms and don't have reference for relation. Feel free to add the tag. btw, can you reproduce my claim up to 2000? $\endgroup$
    – joro
    Feb 10, 2016 at 11:11
  • $\begingroup$ Those q-expansions are closely related to modular forms. I haven't programmed yours. Can you please add the link for "Gjergji Zaimi proved similar congruence in another answer."? There is a similar one mathoverflow.net/questions/84443/…. $\endgroup$
    – Wolfgang
    Feb 10, 2016 at 14:36
  • $\begingroup$ @Wolfgang I mean this proof: mathoverflow.net/questions/59178/… $\endgroup$
    – joro
    Feb 10, 2016 at 14:43

1 Answer 1

9
$\begingroup$

Yes, the congruence is true. Here's a modular forms proof. Let $\eta(z) = q^{1/24} \prod_{n=1}^{\infty} (1-q^{n})$, where $q = e^{2 \pi i z}$. Define $$ g(z) = \frac{\eta(16z) \eta(40z)^{3}}{\eta(4z) \eta(8z) \eta(20z) \eta(80z)} = \sum_{n=0}^{\infty} {\rm A145722}(n) q^{4n+1}. $$ This is a modular form of weight zero for $\Gamma_{0}(80)$ and character $\chi_{5}$. Fermat's little theorem shows that $\eta^{5}(z) \equiv \eta(5z) \pmod{5}$ (in the sense that their Fourier coefficients are congruent modulo $5$). Applying $\eta(40z) \equiv \eta(8z)^{5}$, $\eta(20z) \equiv \eta(4z)^{5}$ and $\eta(80z) \equiv \eta(16z)^{5}$ shows that $g(z) \equiv h(z) \pmod{5}$ where $$ h(z) = \frac{\eta(8z)^{14}}{\eta(4z)^{6} \eta(16z)^{4}}. $$ This is a holomorphic modular form of weight $2$ for $\Gamma_{0}(16)$.

It is reasonably well-known that $F(z) = \frac{\eta(4z)^{8}}{\eta(2z)^{4}} = \sum_{n=0}^{\infty} \sigma(2n+1) q^{2n+1}$ is a modular form of weight $2$ for $\Gamma_{0}(4)$. Thus $F \otimes \chi_{-1} = \sum_{n=1}^{\infty} \chi_{-1}(2n+1) \sigma(2n+1) q^{2n+1}$ is a modular form of weight $2$ and level $16$. The space of modular forms of weight $2$ and level $16$ is $5$-dimensional and by checking the coefficients of $q^{0}$ through $q^{4}$, we can see that $$ h(z) = \frac{F + F \otimes \chi_{-1}}{2} = \sum_{n=0}^{\infty} \sigma(4n+1) q^{4n+1}. $$ This proves the congruence. Hopefully this explains naturally why the shift in offset occurs.

The most efficient way to compute ${\rm A145722}(n)$ is probably using some analogue of the Rademacher formula for $p(n)$, and in no instance can this be computed more quickly by factoring $n$. In fact, since $\log({\rm A145722}(n)) \sim \frac{2 \pi \sqrt{n}}{\sqrt{5}}$ computing ${\rm A145722}(n)$ must take $\Omega(n^{1/2})$ time. Factoring $n$ is much faster than that.

$\endgroup$
1
  • $\begingroup$ Thank you. I meant to compute it modulo $5$. Factoring $4n+1$ computes it via the RHS modulo $5$. $\endgroup$
    – joro
    Feb 11, 2016 at 5:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.