Switch the two independent variables back to T and P and enter @hpsep for both of them as we know these are constant values. Clear the formula for X and bring up its formula editor (be sure and enter something on the line before hitting the blue icon so the editor comes up rather than the mole fraction entry table).
During the solution, the solver output values may range on either side of 0, but a mole fraction cannot be negative, so our formula must be constructed to ensure only positive values result. The following formula seems to give good results:
1.5 ^ #solver.0 * @feed.l
Any value raised to the power of 0 will result in the value 1, so for the initial trial when all the output values are 0, this formula will simply result in the feed fluid liquid composition being the estimate. Increasingly negative numbers will result in mole fractions smaller than the initial estimate, but they will always remain positive. The 1.5 value is arbitrary, but provides a good balance of sensitivity. The property package will always normalize these values so they sum to 1.0.
That normalization is the reason we used compound flows as our error values in the solver rather than mole fractions and a separate error term for total flow. The normalization means that the mole fraction specification only provides n-1 degrees of freedom, where n is the number of compounds. Thus we need to be able to vary one more variable in the recycle fluid to match the n errors calculated in the solver. This will be the flow rate.
Save the composition formula and open the formula editor for the recycle flow.