A formula for Ramanujan's \(\tau\)-function (Q1215660)

Let \(\sigma_k(n) = \sum_{d\mid n} d^k\), \(\sigma(n) = \sigma_1(n)\) and set \(\exp(2\pi iz) = e(z) =x\). Ramanujan's function \(\tau(n)\) is defined as the coefficient of \(x\) in the power series expansion of \[ \Delta(z) =x \prod_{n=1}^\infty (1 -x^n)^{24} = \sum_{n=1}^\infty \tau(n)x^n. \] An explicit formula for \(\tau(n)\), which depends \(\sigma_3(n)\) is known. It also is possible to express \(\sigma_3(n)\) as a function of \(\sigma(n)\), so that the existence of a formula for \(\tau(n)\), depending only on \(\sigma(n)\) is quite plausible. This is in fact the case and the author proves that \[ \tau(n) =n^4 \sigma(n) - 24 \sum_{k=1}^{n-1} (35k^4 - 52k^3n + 18k^2n^2) \sigma(k) \sigma(n-k). \tag{*} \] Sketch of the proof: Set f(z) = -\log \Delta(z)\(; then (*) follows by comparing the coefficients of \)e(nz)\( in \)\( 2^9\cdot 3\pi^6 \Delta(z) =18(f'''(z))^2 + f'(z) f^{(5)}(z) -16 f''(z)f^{(4)}(z) = F(z), \tag{**}\)\( say. One observes that \)F(z)=F(z+1)\(; hence, if one can also show that \)F(-z^{-1}) =z^{12}F(z)\(, then \)F(z)\( is a cusp form of weight 12, just like \)\delta(z)\(, so that (recall that the space of cusp forms of weight 12 is one-dimensional) \)F(z)/\Delta(z) = c\(, a constant. The value of \)c\( is immediately obtained, namely \)c =1\(, because both sides of (**) have as first Fourier coefficient \)2^9\cdot 3\pi^6\(. It only remains to establish that \)F(-z^{-1}) =z^{12}F(z)\(. This is simply checked (in part computationally), by starting from \)f(-z^{-1}) = -12 \log z + f(z)\( (true by the definition of \)f(z))\(. Formula (*) may be quite useful, because \)\sigma(n)\( is more extensively tabulated than \)\sigma_3(n)\(.\)