This is likely a very poor intepretation, and probably patronising, but I think about it like adding extra requirements to a person hanging on for dear life suspended between two sides of a cliff.
They are hanging on with thier hands, so thier forearms, upper arms, and body are otherwise freely hanging (G0).
You then demand that thier forearms must meet the cliff edges at a tangent. To achieve this, more energy is expended towards the other components; the biceps and body, but they manage (G1).
You then state that both thier forearms and thier upper arms must achieve tangency, so the cliff top is level with thier shoulders. More energy is then placed towards the body (abs, chest), but again, they manage (G2).
The problem now is that thier elbow (first CV) and shoulder (second CV) is level with the two cliff tops, but the body (and hence the other internal joints) is doing a lot of work to maintain that position. Accordingly, the muscles on the body become stressed. The problem then becomes optimising (minimising) the energy that the body is using to maintain the requirement, and releasing joints that need not be involved (span/CV, and surface tension/energy reduction; I donāt need this pair of abs for this activity).
The problem is, they usually rely on numerical approaches which can only ever return an approximation to often non-linear problems. In some sense, itās always the same problem; that you are trying to fit a model to the data, and while in areas that model becomes acceptable (you gain continuity, your model reproduces the data exactly at the edges), this comes at the expense of error in other locations where the solver cannot reconcile your need for continuity, with how the parameters required act somewhere else, without looking ugly at certain spatial scales.
So at the end, for surfaces with few relatable parameters to eachother in 3D space, you get a choice, you can have your pretty edge continuity, but you may accept some level of internal problems (Bobi demonstrated Xnurbs doing this). Alternatively, you can get highly ideal scenarios where the solver has an easy time finding solutions where all the joints are quite relaxed (you have flat, level cliff tops all around, so you can place your hands wherever you want, your wrists and joints are not twisting down non-level cliff tops as you rotate, and the right-hand set of abs isnāt contradicting what the left is trying to do, no twisting, for example).
If this way of thinking is very wrong, then I will delete later. But it tends to be how I think of it anyway.
