Let $x_{1},x_{2},y_{1}$ and $y_{2}$ be distinct positive real numbers. I would like to upper bound the following quantity: \begin{equation} \log\frac{x_{2}y_{2}}{x_{1}y_{1}}+\frac{1}{2}(x_{1}+y_{1})\bigg(\frac{1}{x_{2}}+\frac{1}{y_{2}}\bigg)-2. \end{equation} Basically, I would like to get rid of the log term and get an overall $\frac{f(x_{1},y_{1},x_{2},y_{2})}{x_{1}y_{1}x_{2}y_{2}}$ type dependence if possible, where $f$ is allowed to depend on the differences $(x_{1}-x_{2})$ and $(y_{1}-y_{2})$.
Any suggestions appreciated.
EDIT: The problem has some more structure that I had abstracted out as I thought it wouldn't be required, but looks like as stated, the expression cannot be bounded above.
Each $x_i$ and $y_i$ is a sum of $n$ positive numbers and I am trying to get a $1/n^4$ dependence. The differences $x_1-x_2$ and $y_1-y_2$ can be bounded above by constants.
Let $x_1=y_1=kx_2=ky_2.$
Thus, for $k\rightarrow+\infty$ we have:$$\log\frac{x_{2}y_{2}}{x_{1}y_{1}}+\frac{1}{2}(x_{1}+y_{1})\bigg(\frac{1}{x_{2}}+\frac{1}{y_{2}}\bigg)-2=2(k-\log{k}-1)\rightarrow+\infty,$$ which says that a maximal value does not exist.