physically correct combination of displacement and normals

Consider a texture bundled with both a displacement map and a normal map, as in the output of CrazyBump. It seems to me that the normals should be in tangent space before applying the displacement modifier. The displacement modifier will give rise to altered geometry with different vertex normals and therefore a different tangent space which is not exactly the coordinate system in which the normal map is expressed, right? If I were writing GLSL code, then I could simply leave the vertex normals untouched while displacing the geometry. What is the right thing to do in Blender?

EDIT: i assumed you wanted to make mountains for some reason… if that is not the case then this MIGHT be a bit innapropriate. anyways, give us some more context, what are you actually doing.

you cant use a displacement modifier and a normal map on the same model, unless…

let me explain, you probably know what both maps do so i wont go into detail on that, but a normal map is trying to do what a displacement map does, but without altering the geometry, so if you use both maps on a flat subdivided plane, it will kinda look like you took a normal map and multiplied by two.

if you want to make this work, you have to generate a normal map based off of the displaced model.

the reason you are given a normal map, AND a displacement map from crazybump is not so that you can use both at the same time, but pick and chose which one want to use.

what you want to do depends on what you want to achieve. you most likely want to use a displacement map to generate a base mesh, and if you got time, retopo it… yes that sucks, it is REALLY boring, and takes a LONG time, but it will look a lot better in the end…

then further subdivide this model, and sculpt inn details, and use this hi ress model to bake a normal map into the low ress model.

now mind you, if you have access to Zbrush, you can use their Zremesher to retopo the model. i do have zbrush, so if you want, i can generate the retopod model for you. so you dont have to do it manually.

quick recap, get a basemesh, retopo if you can, duplicate and subdivide this model to sculpt details, then bake this detail into the low ress model.

Now I get it. I had gotten the impression that people commonly use the displacement map for big bumps, together with the normal map for finer detail. I figured that would make sense if the normal mapping were done in the tangent space of the mesh from before the displacement—functionality which I can imagine implementing in GLSL, but is perhaps not available in Blender. Thank you for helping me understand that the idea of the output from CrazyBump is to simply use one or the other: displacement mapping or normal mapping (although as you describe, you could use the displacement modifier to make an extremely detailed mesh from which to compute an object space normal map and then apply that to a mesh that is possibly just a little more detailed than the original).

solved but

i use a lores displacement to make a low poly count mesh , then use a tangent normal map on the mesh

this is a good trade off for making a 20 meg mesh+ textures
a 500+ meg mesh all by its self

an example of this is in this thread on a different site do have a issue with tools like crasybump
if you say start with this shaded image you WILL NOT !!! get a good height map
like this