Projecting a polynomial curve
A polynomial Bézier curve is formed by affine combinations of control points. Projection by a homogeneous coordinate divides by a varying denominator, so the projected curve is generally rational rather than a polynomial Bézier curve with unchanged weights. The existing curve lesson identifies rational curves as a further topic; here we derive one finite, positive-weight case.
Definition 8.1.1
(Rational Bézier curve)
For an integer and , let for . Take planar control points and strictly positive weights . Form homogeneous controls
Project a homogeneous point with to . The projected curve is
This is the rational Bézier curve with controls and weights .
Proposition 8.1.2
(Positive weights give a well-defined curve in the convex hull)
Let all . Then the denominator of is strictly positive on , so is defined there, and lies in the convex hull of . Moreover and , and multiplying all weights by the same positive constant does not change .
Proof
(Nonnegative coefficients that sum to one)
Every on and . Hence the denominator is a convex combination of the , so it is at least . The coefficients are nonnegative and sum to one, and , which is a point of the convex hull. At only is nonzero, equal to , so ; likewise . Multiplying every by multiplies numerator and denominator by .
Thus this construction never divides by zero on . Changing just one weight, unlike multiplying all of them, can change the curve. Positivity is a sufficient condition for this argument; it is not asserted to be necessary for every possible rational curve.
A quadratic quarter circle
Choose
Writing and , the denominator and coordinate numerators are
Thus .
Theorem 8.1.3
(The rational quadratic traces the quarter circle)
For these controls and weights, lies on the unit circle in the first quadrant for every , runs from to , and takes every point of that quarter circle. At it passes through . The parameter is not proportional to the angle.
Proof
(Expanding the squares)
Expand the squares. The shared cross terms cancel and
As , the projected point satisfies . Both numerators are nonnegative on , so the curve stays in the first quadrant. Its endpoints are and . By the intermediate value theorem, its continuous -coordinate takes every value in ; the nonnegative -coordinate is then uniquely determined by the circle equation. Thus the whole quarter circle is traced. At , and , so , and by symmetry . The value in the next section shows that the parameter is not proportional to the angle.
Proof X-ray
Where the middle weight and the positivity are used
The whole proof rests on one cancellation. Keep but give the middle weight a general value . Then , , , and expanding gives
This vanishes for every exactly when , so among positive weights is forced. The cross terms in cancel for every ; only the term decides.
Positivity enters twice: is what allows dividing by , and nonnegative numerators keep the curve in the first quadrant, so that the circle equation fixes from . The argument closes through the intermediate value theorem, which needs continuity of and nothing about its speed. It says nothing about how fast the point moves along the arc.
Counterexample mode
Projecting controls and discarding weights changes the curve
If the same three planar controls are instead used with all weights equal to one, the ordinary quadratic at is , whose squared distance from the origin is , not . Projecting the controls and then forgetting their relative homogeneous weights is not generally an exact projection of the original curve.
Worked example 8.1.4
(A point of the quarter circle at one quarter)
At we have , , so , and . Hence . Check: , so the point is on the unit circle. Its polar angle is about , whereas a parameter proportional to the angle would give .
Worked example 8.1.5
(Raising the middle weight pulls the curve toward its control point)
Keep and take weights . At we have , and the numerator is , so . Its squared distance from the origin is . Comparing the three midpoints, the coordinate is for , for and for : the larger the middle weight, the closer the midpoint is to .
Worked example 8.1.6
(Rescaling all weights leaves the arc unchanged)
Multiply the quarter-circle weights by to get . At the denominator is and the -numerator is . Therefore , the same value as before.
Common mistake
(Reading the parameter as the angle)
The rational quadratic traces the arc exactly, but the parameter is not proportional to the angle: at the angle is about , not . Equal steps in do not give equal arcs, so the formula should not be used to place points at equal angles.
Common mistake
(Changing a single weight as if it were a rescaling)
Multiplying all weights by one constant leaves the curve unchanged, but changing one weight moves it. With weights instead of the midpoint leaves the circle, as the worked example shows.
Denominators and clipping
The positivity argument was needed. With quadratic weights , the denominator is , which vanishes at ; the displayed projection formula is not defined there. A particular application must handle its homogeneous denominator and clipping rather than assuming the positive-weight case automatically. This lesson does not give a clipping algorithm or claim every rational quadratic is a circular arc.
Exercises
Quick check
(Under the positive-weight hypotheses, what are and ?)
The endpoint Bernstein weights are or .
Solution · Answer
and , because the endpoint Bernstein weights select those controls.
Quick check
(Does multiplying all by change ?)
Look at the numerator and the denominator.
Solution · Answer
No. Numerator and denominator are both multiplied by .
Quick check
(What is the quarter-circle example's point at ?)
Use .
Solution · Answer
.
Quick check
(Why is the weight list not covered by the positive-weight conclusion?)
Compute the denominator.
Solution · Answer
One weight is negative and the denominator vanishes at .