The
ray epsilon exists to avoid "self shadowing" due to limited numerical precision during path tracing, i.e. surfaces shadowing themselves because a reflection
ray intersects immediately with the surface it reflected off. To avoid this problem a reflected
ray doesn't start exactly at the point it bounced off, but with some distance from the surface. The
ray epsilon is specifying that distance and is in meters.
E.g. if the
ray epsilon is 0.1, then the offset is 0.1m (= 10cm). If your scene is modeled to scale, the shadow
ray would start very close to the bottom of the chair causing these strong shadows, because the bottom of the chair is maybe just 15-20cm above the ground. During rendering it appears as if the chair would only sit 5-10cm above the floor - making those shadows quite strong. -> In general you want to make the
ray epsilon so small that the offset isn't relevant anymore. As a rule of thumb, you can make it maybe ~1/100000th - 1/1000th of the camera <-> object distance, but is of course scene dependent. In most scenes that are in the "human" scale range (with objects of size 0.1m - 100m), the default (0.0001) should work just fine.
As a side effect, the
ray epsilon actually behaves like the shadow bias in VRay, if I understand its function correctly, but currently it can't be controlled per light source, which makes it hard to misuse it to avoid "terminator acne".
I hope that makes some sense
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra