1. "any representation of 3D rotations using only three values"
That is not representation, that is parametrization. Euler-angle parametrization sometimes fails because it is not a correct parameterization of SO(3) in general by construction, this is why it sometimes fails (essentially, the three consecutive rotations can sometimes effectively collapse into two for certain set of angles, regardless of how you choose your 3 axes, in which case you can't relate 2 independent parameters back to the 3 independent axis-angle parameters). The correct parametrization of SO(3) is the axis-angle parametrization, which can be represented using quaternions or 3D reals matrices.
The "representation", on the other hand, would typically be unit quaternions or 3D orthogonal matrices.
2."it can be proven that any representation of 3D rotations using only three values must contain discontinuities." where is that proof and what discontinuity are you talking about? It sound like he misunderstood what "SO(3) is not simply connected" means. Lie groups are differentiable.
3 parameters are sufficient to represent any 3D rotation. The natural parametrization of all Lie groups, including SO(3), is the axis-angle parametrization, and their elements have the form exp(i θ n.J) where n is a unit vector defining the axis of rotation, θ determines the amount of rotation, and J is a vector of the generators of the corresponding Lie algebra. The "regular" 3D matrix representation in the axis-angle parameterization is obtained with so(3) generators L_x, L_y, L_z in their fundamental representation.
Basis quaternions i, j, k (which can be represented by Pauli matrices) obey the same Lie algebra as L_x, L_y, L_z, but the group that it corresponds to (which is SU(2)) is a double cover of SO(3) (up to a sign), so they can still be used for implementing 3D rotations once you pick a sign.
1. "any representation of 3D rotations using only three values"
That is not representation, that is parametrization. Euler-angle parametrization sometimes fails because it is not a correct parameterization of SO(3) in general by construction, this is why it sometimes fails (essentially, the three consecutive rotations can sometimes effectively collapse into two for certain set of angles, regardless of how you choose your 3 axes, in which case you can't relate 2 independent parameters back to the 3 independent axis-angle parameters). The correct parametrization of SO(3) is the axis-angle parametrization, which can be represented using quaternions or 3D reals matrices.
The "representation", on the other hand, would typically be unit quaternions or 3D orthogonal matrices.
2."it can be proven that any representation of 3D rotations using only three values must contain discontinuities." where is that proof and what discontinuity are you talking about? It sound like he misunderstood what "SO(3) is not simply connected" means. Lie groups are differentiable.
3 parameters are sufficient to represent any 3D rotation. The natural parametrization of all Lie groups, including SO(3), is the axis-angle parametrization, and their elements have the form exp(i θ n.J) where n is a unit vector defining the axis of rotation, θ determines the amount of rotation, and J is a vector of the generators of the corresponding Lie algebra. The "regular" 3D matrix representation in the axis-angle parameterization is obtained with so(3) generators L_x, L_y, L_z in their fundamental representation. Basis quaternions i, j, k (which can be represented by Pauli matrices) obey the same Lie algebra as L_x, L_y, L_z, but the group that it corresponds to (which is SU(2)) is a double cover of SO(3) (up to a sign), so they can still be used for implementing 3D rotations once you pick a sign.