The Finite Points Method: I J J I I M L L T T

1
The finite points method

Let ΩI be the interpolation sub-domain, called cloud, and let sj with j =
1, 2, ..., n be a collection of n points with coordinates xj ∈ ΩI . The unknown
function u(x) may be approximated within ΩI by
m
u(x) ∼
X
= û(x) = pl (x)αl = pT (x)T α (1)
l=1
where α = [α1 , α2 , . . . , αn ]T and the vector p(x), called base interpolating

functions, contains typically monomials in the space coordinates ensuring
that the basis is complete. For a 2D problem,
p = [1, x, y]T , m = 3 (2)
and
p = [1, x, y, x2 , xy, y 2 ]T , m = 6 (3)
can be used.
The function u(x) can now be sampled at the n points belonging to ΩI
giving
uh1 ûh1 pT1

     
 uh2  
∼ ûh2   pT2 
uh =  = =  α = Cα (4)
   
.. .. ..
 .   .   . 
uhn ûhn pTn
where uhj = u(xj ) are the unknown values of u at point j, ûj = û(xj ) are
the approximate values, and pj = p(xj ).
In the case of FEM, the number of points is chosen so that m = n. In
this case C is a square matrix.
If n > m, C is not a square matrix and the approximation cannot fit
all the uhj values. This problem can be solved determining the û values by
minimizing with respect to the a parameters the sum of the square of the
error at each point weighted with a function ω(x) as
n
X n
X
J= ω(xj )(uhj − û(xj ))2 = ω(xj )(uhj − pTj α)2 (5)
j=1 j=1
2
This approximation is called weighted least square (WLS) interpolation.

Note that for ω(x) = 1 the standard least square (LSQ) method is repro-
duced.
Function ω(x) is usually built in such way that it takes a unit value
in the vicinity of the point I typically called star node where the function
(or its derivatives) are to be computed and vanishes outside a region ΩI
surrounding the point. The region ΩI can be used to define the number of
sampling points n in the interpolation region. In this work, the normalized
Gaussian weight function ω(x) is used, thus n ≥ m is always required.
Minimization of (5) with respect to a α gives
α = C −1 , C −1 = A−1 B (6)
where
n
X
A= ω(xj )p(xj )pT (xj ) (7)
j=1
and
B = [ω(x1 )p(x1 ), ω(x2 )p(x2 ), . . . , ω(xn )p(xn )] (8)
are the shape functions. It must be noted that because of least square
approximation u(xj ) ∼ = û(xj ) 6= uhj . That is, the local values of the approx-
imating function do not fit the nodal unknown values.
The WLS approximation described above depends strongly on the shape
and the way in which the weighting function is applied. The simplest way is
to define a fixed function ω(x) for each of the ΩI interpolation sub-domains.
Let ω(x) be a weighting function satisfying
ωi (xi ) = 1
ωi (x) 6= 0, x ∈ ΩI (9)
ωi (x) = 0, x 6∈ ΩI
Then the minimization of J in (5) becomes
n
X
JI = ωI (xj )(uhj − û(xj ))2 (10)
j=1
The Gaussian weight function used in this work is given by

3
 dj r
 e− c − e− c

ωI (xj ) = − rc
, dj ≤ r (11)
 1 − e
 0, dj > r
where dj = kxj − xi k, r = qmaxxj ∈ΩI kxj − xI k y c = βr
The support of this function is isotropic, circular and spherical in two
and three-spatial, respectively.
Note that according to Eq. (1), the approximate function û(x) is defined
in each interpolation domain ΩI . In fact, different interpolation sub-domains
can yield different shape functions φIj . As a consequence of this, a point
belonging to two or more overlapping clouds has different values of the shape
functions which means that φij 6= φkj . The interpolation is now multivalued
within ΩI and, therefore for the approximation to be useful, a decision
must be made in order to limit the choice to a single value. Indeed, the
approximate function û(x) will be typically used to provide the value of the
unknown function u(x) and its derivatives only in specific regions within
each interpolation sub-domain. For instance by using point collocation the
validity of the interpolation is limited to a single point xI .

The Finite Points Method: I J J I I M L L T T

Uploaded by

Copyright:

Available Formats

The Finite Points Method: I J J I I M L L T T

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Finite Points Method: I J J I I M L L T T

Uploaded by

Copyright:

Available Formats

1

The finite points method

where α = [α1 , α2 , . . . , αn ]T and the vector p(x), called base interpolating

p = [1, x, y]T , m = 3 (2)

p = [1, x, y, x2 , xy, y 2 ]T , m = 6 (3)

uh1 ûh1 pT1

This approximation is called weighted least square (WLS) interpolation.

B = [ω(x1 )p(x1 ), ω(x2 )p(x2 ), . . . , ω(xn )p(xn )] (8)

Then the minimization of J in (5) becomes

The Gaussian weight function used in this work is given by

You might also like