T.R | Title | User | Personal Name | Date | Lines |
---|
1809.1 | | HANNAH::OSMAN | see HANNAH::IGLOO$:[OSMAN]ERIC.VT240 | Fri Oct 22 1993 12:38 | 23 |
|
I was always taught that 2 points determines a line, 3 points determines a
parabola, 4 points determines a a cubic etc.
So, I don't get what you're saying either.
For example, suppose we want to determine the equation
Ax^3 + Bx^2 + Cx + D = y
that passes through the points
(1,1) and (2,7) and (3,3) and (4,0) [I picked these arbitrarily]
We've got 4 unknowns A,B,C,D and we can make 4 equations from the 4 points,
and hence we have a regular simultaneous system of equations.
My understanding has always been that if we add, say, another point, now we'll
have 5 equations, so no longer is there guarantee of a solution.
/Eric
|
1809.2 | | CSC32::D_DERAMO | Dan D'Eramo, Customer Support Center | Fri Oct 22 1993 13:31 | 14 |
| If the x coordinates of the points are all distinct, then n+1
points will determine a unique y = a(n) x^n + a(n-1) x^(n-1) +
... + a(1) x + a(0).
Whether or not the x coordinates are not distinct, you can try
to fit a curve f(x,y) = 0 through all of the points. For example,
a circle x^2 + y^2 - 2 = 0 can be fit through the four points
(+/- 1, +/- 1).
Perhaps it was talking about the degree of a polynomial in x
and y together.
Dan
|
1809.3 | Would still like to see some examples | WIBBIN::NOYCE | It's the memory interface, stupid! | Fri Oct 22 1993 13:59 | 14 |
| >> Perhaps it was talking about the degree of a polynomial in x
>> and y together.
That seems likely, since the number of terms in such a polynomial of degree
n is 1+n(n+3)/2.
n=0 a 1 coeff
n=1 a + bx + cy 3 coeffs
n=2 a + bx + cy + dx^2 + exy + fy^2 6 coeffs
n=3 above + gx^3 + hx^2y + ixy^2 + j y^3 10 coeffs
But if the equation is of the form F(x,y)=0, then you can arbitrarily
set one nonzero coefficient to 1, so there are really only about n(n+3)/2
unknowns.
|
1809.4 | perhaps degenerate cases | TROOA::RITCHE | From the desk of Allen Ritche... | Fri Oct 22 1993 15:35 | 60 |
| >
> "a curve of the nth order is not always determined by 1/2 n(n+3)
> points. so that 9 points may not uniquely determine a cubic while
> 10 would be too many."
>
Here is one interpretation (I have not heard of this paradox before).
Assuming we a dealing with nth order curves in 2 dimensions that are
represented by the general form f(x,y)=0 where
f(x,y) = SIGMA [C(i)* x^j * y^k] = 0 (j+k not greater than n)
Taking the first three orders, for...
n=1, Ax + By + C = 0 (P=2)
n=2, Ax^2 + Bxy + Cy^2 + Dx + Ey + F = 0 (P=5)
n=3, Ax^3 + Bx^2y + Cxy^2 + Dy^3 + Ex^2 + Fy^2 + Gxy + Hx + Iy + J = 0
(P=9)
The number of 0-order terms in each polynomial is 1
1-order terms ................. is 2
2-order terms ........ is 3
3-order terms ........ is 4
n-order terms in each polynomial is n+1,
Total number of terms is (n+1)(n+2)/2 = n(n+3)/2 + 1
One of these parameters is redundant because f(x,y) could be divided by a
non-zero co-efficient. Thus P=n(n+3)/2 points would generate P simultaneous
linear equations in P unknowns to solve for all the co-efficients for a unqiue
curve. I think this part is obvious to all.
To understand the paradox, I looked at the quadratic as a good example since
I can't count much over order 2.
5 points in the plane will uniquely determine a quadratic curve (taking into
consideration transformations like rotated parabolas, etc.) from the set of
all quadratic curves. So the paradox to analyze is
"A 2-order curve is not always determined by 5
points. So that 5 points may not uniquely determine a quadratic while
6 would be too many."
Perhaps the paradox occurs in degenerate cases. Take f(x,y)=x^2-y^2=0
which plots as two intersecting lines y=x and y=-x.
The *4* points (1,1), (-1,1), (-1,-1), (1,-1) clearly do not uniquely determine
the "curve". (could be a circle) but adding point (0,0) the *5* points _do_
uniquely define the curve. If the fifth point were (sqrt(2),0), a circle is
forced. A sixth point is easy to find so that no quadratic curve is possible.
The challenge to illustrate this paradox is to find 5 points that yield an
ambiguous quadratic. !! ??? Is it possible. Or to find 5 points that
will not produce any quadratic curve. Is that possible??
Allen
|
1809.5 | example | TROOA::RITCHE | From the desk of Allen Ritche... | Fri Oct 22 1993 15:53 | 22 |
| >
>5 points in the plane will uniquely determine a quadratic curve (taking into
>consideration transformations like rotated parabolas, etc.) from the set of
>all quadratic curves. So the paradox to analyze is
>
> "A 2-order curve is not always determined by 5
> points. So that 5 points may not uniquely determine a quadratic while
> 6 would be too many."
>
...
>The challenge to illustrate this paradox is to find 5 points that yield an
>ambiguous quadratic. !! ??? Is it possible. Or to find 5 points that
>will not produce any quadratic curve. Is that possible??
Perhaps there is one more interpretation...that 5 points may not determine
a quadratic.
So take any 5 points on a straight line... NO quadratic (i.e. its order-1)
Take any other 6th point not on the line....TOO MANY, no quadratic.
Allen
|
1809.6 | Maybe not polynomials over reals? | CADSYS::COOPER | Topher Cooper | Fri Oct 22 1993 17:21 | 9
|