I guess I could've said that better. Yes, the proper length of pipe (the virtual anchor length, VAL) will disconnect the system into an upstream leg and a downstream leg. But you want to know how much of that upstream or downstream leg must be modeled and whether or not an anchor be placed in yoru CAESAR II model.
Maybe an example will help...
Let's say you have a buried run of pipe between two 90 degree elbows and a given VAL of 1000ft. If that run between elbows is less than 1000ft, then there is no separation and the model should continue beyond that second elbow. If that run between elbows is 5000ft, then a VAL exists and the two elbows can be evaluated in separate models. How do you end these two models? I suggest you run at least VAL beyond the upstream elbow to establish enough axial resrtaint through friction to get a sensible response for the elbow. Focus on the elbow, not on the pipe on the other side of the VAL. Since that elbow may get pushed back (according to the bearing capacity of the soil on the incoming leg of this elbow), the thermal strain and therefore axial load is reduced. With a reduced axial load, a shorter VAL would exist.
So one difficulty is what's going on on the other sides of these bends - that bearing stiffness of the soil and the pipe cantilever stiffness. Another confusing point is how CAESAR II treats this soil friction. We model this friction with springs. These springs have to be moved to generate load. This necessary motion makes the VAL in the CAESAR II model greater than the VAL in theory. (Note that the newer ALA soil model gives you more control over this axial soil stiffness.)
Back to the example...
If I have 5000ft between these elbows, I would probably run through both those elbows. It's just two more pieces of pipe. CAESAR II will do the work of "burying" it. You would then look for motion of each elbow away from the 5000ft run and a section of the 5000ft run that has no axial deflection. The length of pipe between the elbow and the "inert" pipe is the effective VAL for this CAESAR II analysis (and, again, it is also influenced by what's around the bend). If I have a 10,000ft straight run between elbows, I would probably run at least 2000ft beyond the first elbow and stop there. I would let the axial friction balance the thermal strain and I would not add an anchor. If you have the time, you can check the maximum axial load in the run headed to the VAL. Then run the job again with a longer run. If the maximum axial load stays the same, the run is long enough.
That's my opinion, I invite others.