Parametric surface ($P(u,v)$)

A parametric surface is represented by a map $(u,v)\mapsto P(u,v)$. A regular patch has $P_u\times P_v\ne 0$; this cross product supplies its normal vector.

// Target: analysis.bigb

A general conical surface is swept out by straight generators through one apex $A$. An invertible affine map sends it to another such surface, since it sends each generator to a line through the transformed apex. A projective transformation preserves the projective concurrence of generators, but can send the apex to infinity, producing parallel generators in the affine chart. Therefore finite-apex cones and cylinders are not separately projectively invariant Euclidean classes.

For nonnegative barycentric parameters, these total-degree Bernstein basis functions define a surface over a triangle. The triangular control net has $(n+1)(n+2)/2$ points. Its index constraint couples the three exponents, so it is not a tensor product of two independent univariate bases. The multinomial theorem gives nonnegativity and partition of unity.

Core numerical enquiries return a surface point and its derivatives through second order, together with the parameter domain, patch boundaries and regularity information. At a regular point these determine its tangent plane, normal vector, first and second fundamental forms and curvatures. Subdivision and certified enclosing volumes provide spatial search and intersection support. Singular charts or discontinuity boundaries require one-sided or alternative-chart information; a zero cross product does not define a unique normal vector.

From a point light source $P$, a caster point $C(t)$ projects onto a receiving surface along the displayed ray. On a regular transverse branch, differentiating gives $[S_u,S_v,-(C-P)](u^{\prime },v^{\prime },{\alpha }^{\prime }{)}^T=\alpha C^{\prime }$. Newton method correction and certified patch/curve bounds support continuation and isolation of all visible branches. If the shadow's second derivative is bounded by $M$ on an interval of width $h$, its chord error is at most $M{h}^2/8$. A fixed sample grid can miss entire shadow components; surface boundaries, occlusion changes and grazing rays require explicit treatment.

Products of two univariate bases form a surface basis. Nonnegativity, partition of unity and coordinate-wise linear precision pass to the products by multiplication of the respective univariate identities. If the factors are $C^r$ and $C^s$, each mixed derivative through orders $r,s$ is the product of the corresponding derivatives, giving continuous mixed derivatives and joint $C^{\min (r,s)}$ regularity.

Nonnegative tensor-product surface basis factors yield nonnegative product weights. Their two partition of unity sums multiply to one. If the factors are $C^r$ and $C^s$, mixed derivatives through separate orders $r,s$ are continuous, so the resulting finite or locally finite surface representation is jointly $C^{\min (r,s)}$. Control coefficients can cancel derivative jumps and yield more smoothness; inherited continuity is a guaranteed lower bound, not a claim that every surface has exactly the basis's worst regularity.