SLATEC Routines --- DPOLFT ---


*DECK DPOLFT
      SUBROUTINE DPOLFT (N, X, Y, W, MAXDEG, NDEG, EPS, R, IERR, A)
C***BEGIN PROLOGUE  DPOLFT
C***PURPOSE  Fit discrete data in a least squares sense by polynomials
C            in one variable.
C***LIBRARY   SLATEC
C***CATEGORY  K1A1A2
C***TYPE      DOUBLE PRECISION (POLFIT-S, DPOLFT-D)
C***KEYWORDS  CURVE FITTING, DATA FITTING, LEAST SQUARES, POLYNOMIAL FIT
C***AUTHOR  Shampine, L. F., (SNLA)
C           Davenport, S. M., (SNLA)
C           Huddleston, R. E., (SNLL)
C***DESCRIPTION
C
C     Abstract
C
C     Given a collection of points X(I) and a set of values Y(I) which
C     correspond to some function or measurement at each of the X(I),
C     subroutine  DPOLFT  computes the weighted least-squares polynomial
C     fits of all degrees up to some degree either specified by the user
C     or determined by the routine.  The fits thus obtained are in
C     orthogonal polynomial form.  Subroutine  DP1VLU  may then be
C     called to evaluate the fitted polynomials and any of their
C     derivatives at any point.  The subroutine  DPCOEF  may be used to
C     express the polynomial fits as powers of (X-C) for any specified
C     point C.
C
C     The parameters for  DPOLFT  are
C
C     Input -- All TYPE REAL variables are DOUBLE PRECISION
C         N -      the number of data points.  The arrays X, Y and W
C                  must be dimensioned at least  N  (N .GE. 1).
C         X -      array of values of the independent variable.  These
C                  values may appear in any order and need not all be
C                  distinct.
C         Y -      array of corresponding function values.
C         W -      array of positive values to be used as weights.  If
C                  W(1) is negative,  DPOLFT  will set all the weights
C                  to 1.0, which means unweighted least squares error
C                  will be minimized.  To minimize relative error, the
C                  user should set the weights to:  W(I) = 1.0/Y(I)**2,
C                  I = 1,...,N .
C         MAXDEG - maximum degree to be allowed for polynomial fit.
C                  MAXDEG  may be any non-negative integer less than  N.
C                  Note -- MAXDEG  cannot be equal to  N-1  when a
C                  statistical test is to be used for degree selection,
C                  i.e., when input value of  EPS  is negative.
C         EPS -    specifies the criterion to be used in determining
C                  the degree of fit to be computed.
C                  (1)  If  EPS  is input negative,  DPOLFT  chooses the
C                       degree based on a statistical F test of
C                       significance.  One of three possible
C                       significance levels will be used:  .01, .05 or
C                       .10.  If  EPS=-1.0 , the routine will
C                       automatically select one of these levels based
C                       on the number of data points and the maximum
C                       degree to be considered.  If  EPS  is input as
C                       -.01, -.05, or -.10, a significance level of
C                       .01, .05, or .10, respectively, will be used.
C                  (2)  If  EPS  is set to 0.,  DPOLFT  computes the
C                       polynomials of degrees 0 through  MAXDEG .
C                  (3)  If  EPS  is input positive,  EPS  is the RMS
C                       error tolerance which must be satisfied by the
C                       fitted polynomial.  DPOLFT  will increase the
C                       degree of fit until this criterion is met or
C                       until the maximum degree is reached.
C
C     Output -- All TYPE REAL variables are DOUBLE PRECISION
C         NDEG -   degree of the highest degree fit computed.
C         EPS -    RMS error of the polynomial of degree  NDEG .
C         R -      vector of dimension at least NDEG containing values
C                  of the fit of degree  NDEG  at each of the  X(I) .
C                  Except when the statistical test is used, these
C                  values are more accurate than results from subroutine
C                  DP1VLU  normally are.
C         IERR -   error flag with the following possible values.
C             1 -- indicates normal execution, i.e., either
C                  (1)  the input value of  EPS  was negative, and the
C                       computed polynomial fit of degree  NDEG
C                       satisfies the specified F test, or
C                  (2)  the input value of  EPS  was 0., and the fits of
C                       all degrees up to  MAXDEG  are complete, or
C                  (3)  the input value of  EPS  was positive, and the
C                       polynomial of degree  NDEG  satisfies the RMS
C                       error requirement.
C             2 -- invalid input parameter.  At least one of the input
C                  parameters has an illegal value and must be corrected
C                  before  DPOLFT  can proceed.  Valid input results
C                  when the following restrictions are observed
C                       N .GE. 1
C                       0 .LE. MAXDEG .LE. N-1  for  EPS .GE. 0.
C                       0 .LE. MAXDEG .LE. N-2  for  EPS .LT. 0.
C                       W(1)=-1.0  or  W(I) .GT. 0., I=1,...,N .
C             3 -- cannot satisfy the RMS error requirement with a
C                  polynomial of degree no greater than  MAXDEG .  Best
C                  fit found is of degree  MAXDEG .
C             4 -- cannot satisfy the test for significance using
C                  current value of  MAXDEG .  Statistically, the
C                  best fit found is of order  NORD .  (In this case,
C                  NDEG will have one of the values:  MAXDEG-2,
C                  MAXDEG-1, or MAXDEG).  Using a higher value of
C                  MAXDEG  may result in passing the test.
C         A -      work and output array having at least 3N+3MAXDEG+3
C                  locations
C
C     Note - DPOLFT  calculates all fits of degrees up to and including
C            NDEG .  Any or all of these fits can be evaluated or
C            expressed as powers of (X-C) using  DP1VLU  and  DPCOEF
C            after just one call to  DPOLFT .
C
C***REFERENCES  L. F. Shampine, S. M. Davenport and R. E. Huddleston,
C                 Curve fitting by polynomials in one variable, Report
C                 SLA-74-0270, Sandia Laboratories, June 1974.
C***ROUTINES CALLED  DP1VLU, XERMSG
C***REVISION HISTORY  (YYMMDD)
C   740601  DATE WRITTEN
C   890531  Changed all specific intrinsics to generic.  (WRB)
C   891006  Cosmetic changes to prologue.  (WRB)
C   891006  REVISION DATE from Version 3.2
C   891214  Prologue converted to Version 4.0 format.  (BAB)
C   900315  CALLs to XERROR changed to CALLs to XERMSG.  (THJ)
C   900911  Added variable YP to DOUBLE PRECISION declaration.  (WRB)
C   920501  Reformatted the REFERENCES section.  (WRB)
C   920527  Corrected erroneous statements in DESCRIPTION.  (WRB)
C***END PROLOGUE  DPOLFT