That’s a good question.  Here’s a simple test model that, I think, can help provide an answer.  It’s based on one of the sample Grid Sag surfaces provided with OpticStudio.

In this case, the thickness of the first Grid Sag surface is zero, and the material is air.  On the output side of the Gradient4 surface there is a dummy surface with zero thickness and the material is air.  So, the question is, do either or both of the Grid Sag surfaces refract rays into or out of the intermediate Gradient4 surface?  Well, we can just look at the ray trace in a Layout window and see that, for this particular configuration, the answer is no -- there is no refraction associated with either one of the Grid Sag surfaces:

So, it’s clear that if the material on both sides of a Grid Sag surface is air (or in general the same material), then there is no refractive ray bending at the surface.  Of course here the input-side material is designated by the surface prior to the Grid Sag surface, and the output-side material is designated by the material of the Grid Sag surface itself.  This is true independent of the Grid Sag surface thickness, even for zero thickness.

Now let’s set the refractive index for the first Grid Sag surface equal to the nominal value (n0) of the Gradient4 surface:

which yields refraction by the first Grid Sag surface:

However, the second Grid Sag surface is still not refracting rays.  So, let’s ignore the dummy surface:

and we see that the second Grid Sag surface now causes refraction:

So, for your problem, I would suggest setting the refractive index of the first Grid Sag surface equal to n0 (i.e., the average index of the Gradient4 surface) in order to make this Grid Sag surface “active.”  Assuming that its sag is small, then using a constant refractive index should be a very reasonable approximation compared to having the material be GRIN (which doesn’t appear to be possible).  Also, there should be no dummy surfaces.

I think in your current setup, the second Grid Sag surface is probably operating correctly, which is why you see a change when switching to a flat surface.

Regards,

Jeff