Shadow Map Projection aliasing?

This is a common issue with shadow maps. You need to use a bias value (ie mulitply one of the values in the shadow comparison by a like 1.001)... also use a higher precision render target... most shadow map implementations use floating point textures, not integers.

Dependent on which API and hardware you are using you could add the bias implicitly with render states.

perspective shadow map approaches increase the depth aliasing errors. You want to use an orthographic projection and use cascaded shadow maps or any other multi-frustum approach.
The amount of depth aliasing errors and the very view dependent quality of perspective shadow maps make them unusable in games with a time of day feature where you need full-screenshadows.

Yes, I suppose you're right, my experiments demonstrate it clearly.
I will probably try to use cascaded shadow maps, but I did not understand exactly how they work, please correct me if I miss something:

first of all: do I have to divide the depth by the homogeneous coordinate? always? with lispsm too?

now about the cascaded shadow:

1. you divide the frustum in some slices ( e.g. 4 )
2. calculate the light space bounding box of each frustum;
3. calculate 4 view projection matrices for the 4 frusta; the projection is orthographic
4. now you should draw 4 shadow maps, correct? can't you draw a single shadow map writing the slice depth in the color channels? e.g. the first depth in red, the second slice in green etc... ?
5. reuse the 4 shadow maps to estimate the shadowing coefficient... but how? what is the depth to be compared with? here is my confusion...

Thanks
Emanuele

there is a nice example implementation available for Parallel-Split shadow maps ... those are cascaded shadow maps with an algorithm that picks the split distance. Search for this term on Google. They have a OpenGL and a D3D implementation.