Line by line carefully. Never skip anything in hopes it will clear up later.
You may allow to yourself not to know with what intention author wrote the line but you should never allow yourself not to know why the author could write this line.
The best way I've found to read math texts is to take it concept by concept. Line by line is too mechanical. Approach each concept and get a "gist" for what is trying to be accomplished. Once you understand where they are starting from for a given concept, where they are going with it, and have a rough idea of how they'll get there, then you read the details. The details are essentially meaningless if you don't know the bigger picture.
You can skip on first reading to get an overview of where the author wants to take you, but don't expect it will clear up later on its own. Read and re-read. Fill in all the little gaps, especially when a sentence begins with "Clearly" or "As we can easily see".
"As we can easily see" == doesn't seem implausible after you read the paper countless times, re-work all the derivations from scratch at least 7 times, read the cited docs, get out your old analysis texts to look up some theorem you'd forgotten existed, and sacrifice a chicken.
I must say, there's nothing like reading a math paper to remind me that there's no shortage of people in the world that are way smarter than me.
You may allow to yourself not to know with what intention author wrote the line but you should never allow yourself not to know why the author could write this line.