If $d(n) = \sum_{d|n}1$ is the divisor function, it is a fact that $d$ has summatory behaviour \begin{equation} \sum_{n \leq x} d(n) = x \log x + (2 \gamma - 1)x + O(\sqrt{x}). \end{equation}
I'm interested in the behaviour of $\sum_{n \leq x} \frac{d(n)}{n}$ for large $x$. Now, this answer uses the Dirichlet Hyperbola Method, which I understand, to get the result \begin{equation} \sum_{n \leq x} \frac{d(n)}{n} = \frac{1}{2} \log^2 x + 2 \gamma \log x + \gamma^2 -2\gamma_1 + O\left(\frac{1}{\sqrt x}\right). \end{equation} However, if I try using Abel summation (motivated by this similar question), I get the bound \begin{equation} \sum_{n \leq x} \frac{d(n)}{n} = \frac{1}{x}\left(x \log x + (2 \gamma - 1) x + O(\sqrt x)\right) -1 + \int_1^x\frac{t \log t + (2 \gamma - 1) t + O(\sqrt t)}{t^2}dt \end{equation} \begin{equation} =\frac{1}{2}\log^2 x + 2 \gamma\log x + 2(\gamma - 1) + O\left(\frac{1}{\sqrt x} \right). \end{equation}
Now the numerical constants $\gamma^2-2\gamma_1$ and $2(\gamma-1)$ are not equal; more worryingly I see no way that my approach could have yielded $\gamma_1$ leading me to believe the idea was fundamentally flawed. Would appreciate if someone could aide my understanding.