Deterministic cosine-weighted unit directions over the hemisphere around
normal, count of them in index order.
The set is a Hammersley sequence mapped by Malley's method, so it is a
property of the normal and sample count: no random state, no host entropy, the
same directions in the same order on Windows and POSIX. Cosine weighting is
what makes the estimator exact for the case that matters most: a plane facing
the ground's own up with nothing above it sees every one of its directions,
so the estimate of an isotropic sky's contribution collapses to the declared
horizontal illuminance with no sampling error in ideal arithmetic. A tilted
plane loses the directions that point below the horizon, and the fraction it
loses does carry sampling error.
normal need not be unit length; it is normalized first. Throws when count
is not a positive safe integer, or when normal has a non-finite component
or a measured norm that is nonfinite or 1e-12 or less. Memory grows with count, which has no upper
bound here.
Deterministic cosine-weighted unit directions over the hemisphere around
normal,countof them in index order.The set is a Hammersley sequence mapped by Malley's method, so it is a property of the normal and sample count: no random state, no host entropy, the same directions in the same order on Windows and POSIX. Cosine weighting is what makes the estimator exact for the case that matters most: a plane facing the ground's own up with nothing above it sees every one of its directions, so the estimate of an isotropic sky's contribution collapses to the declared horizontal illuminance with no sampling error in ideal arithmetic. A tilted plane loses the directions that point below the horizon, and the fraction it loses does carry sampling error.
normalneed not be unit length; it is normalized first. Throws whencountis not a positive safe integer, or whennormalhas a non-finite component or a measured norm that is nonfinite or1e-12or less. Memory grows withcount, which has no upper bound here.