NAG Library Routine Document
d02saf (bvp_shoot_genpar_algeq)
1
Purpose
d02saf solves a two-point boundary value problem for a system of first-order ordinary differential equations with boundary conditions, combined with additional algebraic equations. It uses initial value techniques and a modified Newton iteration in a shooting and matching method.
2
Specification
Fortran Interface
Subroutine d02saf ( |
p, m, n, n1, pe, pf, e, dp, npoint, swp, ldswp, icount, range, bc, fcn, eqn, constr, ymax, monit, prsol, w, ldw, sdw, ifail) |
Integer, Intent (In) | :: | m, n, n1, npoint, ldswp, icount, ldw, sdw | Integer, Intent (Inout) | :: | ifail | Real (Kind=nag_wp), Intent (In) | :: | pe(m), e(n), ymax | Real (Kind=nag_wp), Intent (Inout) | :: | p(m), pf(m), dp(m), swp(ldswp,6), w(ldw,sdw) | Logical, External | :: | constr | External | :: | range, bc, fcn, eqn, monit, prsol |
|
C Header Interface
#include <nagmk26.h>
void |
d02saf_ (double p[], const Integer *m, const Integer *n, const Integer *n1, const double pe[], double pf[], const double e[], double dp[], const Integer *npoint, double swp[], const Integer *ldswp, const Integer *icount, void (NAG_CALL *range)(double x[], const Integer *npoint, const double p[], const Integer *m), void (NAG_CALL *bc)(double g1[], double g2[], const double p[], const Integer *m, const Integer *n), void (NAG_CALL *fcn)(const double *x, const double y[], double f[], const Integer *n, const double p[], const Integer *m, const Integer *i), void (NAG_CALL *eqn)(double e[], const Integer *q, const double p[], const Integer *m), logical (NAG_CALL *constr)(const double p[], const Integer *m), const double *ymax, void (NAG_CALL *monit)(const Integer *istate, const Integer *iflag, const Integer *ifail1, const double p[], const Integer *m, const double f[], const double *pnorm, const double *pnorm1, const double *eps, const double d[]), void (NAG_CALL *prsol)(double *z, const double y[], const Integer *n), double w[], const Integer *ldw, const Integer *sdw, Integer *ifail) |
|
3
Description
d02saf solves a two-point boundary value problem for a system of
first-order ordinary differential equations with separated boundary conditions by determining certain unknown arguments
. (There may also be additional algebraic equations to be solved in the determination of the arguments and, if so, these equations are defined by
eqn.) The arguments may be, but need not be, boundary values; they may include eigenvalues, arguments in the coefficients of the differential equations, coefficients in series expansions or asymptotic expansions for boundary values, the length of the range of definition of the system of differential equations, etc.
It is assumed that we have a system of
differential equations of the form
where
is the vector of arguments, and that the derivative
is evaluated by
fcn. Also,
of the equations are assumed to depend on
. For
the
equations of the system are not involved in the matching process. These are the driving equations; they should be independent of
and of the solution of the other
equations. In numbering the equations in
fcn and
bc the driving equations must be put
first (as they naturally occur in most applications). The range of definition [
] of the differential equations is defined by
range and may depend on the arguments
(that is, on
).
range must define the points
,
, which must satisfy
(or a similar relationship with all the inequalities reversed).
If
the points
can be used to break up the range of definition. Integration is restarted at each of these points. This means that the differential equations
(1) can be defined differently in each sub-interval
, for
. Also, since initial and maximum integration step sizes can be supplied on each sub-interval (via the array
swp), you can indicate parts of the range
where the solution
may be difficult to obtain accurately and can take appropriate action.
The boundary conditions may also depend on the arguments and are applied at
and
. They are defined (in
bc) in the form
The boundary value problem is solved by determining the unknown arguments
by a shooting and matching technique. The differential equations are always integrated from
to
with initial values
. The solution vector thus obtained at
is subtracted from the vector
to give the
residuals
, ignoring the first
, driving equations. Because the direction of integration is always from
to
, it is unnecessary, in
bc, to supply values for the first
boundary values at
, that is the first
components of
in
(3). For
then
. Together with the
equations defined by
eqn,
these give a vector of residuals
, which at the solution,
, must satisfy
These equations are solved by a pseudo-Newton iteration which uses a modified singular value decomposition of
when solving the linear equations which arise. The Jacobian
used in Newton's method is obtained by numerical differentiation. The arguments at each Newton iteration are accepted only if the norm
is much reduced from its previous value. Here
is the pseudo-inverse, calculated from the singular value decomposition, of a modified version of the Jacobian
(
is actually the inverse of the Jacobian in well-conditioned cases).
is a diagonal matrix with
where
pf is an array of floor values.
See
Deuflhard (1974) for further details of the variants of Newton's method used,
Gay (1976) for the modification of the singular value decomposition and
Gladwell (1979) for an overview of the method used.
Two facilities are provided to prevent the pseudo-Newton iteration running into difficulty. First, you are permitted to specify constraints on the values of the arguments
via a
constr. These constraints are only used to prevent the Newton iteration using values for
which would violate them; that is, they are not used to determine the values of
. Secondly, you are permitted to specify a maximum value
for
at all points in the range
. It is intended that this facility be used to prevent machine ‘overflow’ in the integrations of equation
(1) due to poor choices of the arguments
which might arise during the Newton iteration. When using this facility, it is presumed that you have an estimate of the likely size of
at all points
.
should then be chosen rather larger (say by a factor of
) than this estimate.
You are strongly advised to supply a
monit (or to call the ‘default’ routine d02hbx, see
monit) to monitor the progress of the pseudo-Newton iteration. You can output the solution of the problem
by supplying a suitable
prsol (an example is given in
Section 10 of a routine designed to output the solution at equally spaced points).
d02saf is designed to try all possible options before admitting failure and returning to you. Provided the routine can start the Newton iteration from the initial point it will exhaust all the options available to it (though you can override this by specifying a maximum number of iterations to be taken). The fact that all its options have been exhausted is the only error exit from the iteration. Other error exits are possible, however, whilst setting up the Newton iteration and when computing the final solution.
If you require more background information about the solution of boundary value problems by shooting methods you are recommended to read the appropriate chapters of
Hall and Watt (1976), and for a detailed description of
d02saf
Gladwell (1979) is recommended.
4
References
Deuflhard P (1974) A modified Newton method for the solution of ill-conditioned systems of nonlinear equations with application to multiple shooting Numer. Math. 22 289–315
Gay D (1976) On modifying singular values to solve possibly singular systems of nonlinear equations Working Paper 125 Computer Research Centre, National Bureau for Economics and Management Science, Cambridge, MA
Gladwell I (1979) The development of the boundary value codes in the ordinary differential equations chapter of the NAG Library Codes for Boundary Value Problems in Ordinary Differential Equations. Lecture Notes in Computer Science (eds B Childs, M Scott, J W Daniel, E Denman and P Nelson) 76 Springer–Verlag
Hall G and Watt J M (ed.) (1976) Modern Numerical Methods for Ordinary Differential Equations Clarendon Press, Oxford
5
Arguments
- 1: – Real (Kind=nag_wp) arrayInput/Output
-
On entry: must be set to an estimate of the th argument, , for .
On exit: the corrected value for the th argument, unless an error has occurred, when it contains the last calculated value of the argument.
- 2: – IntegerInput
-
On entry: , the number of arguments.
Constraint:
.
- 3: – IntegerInput
-
On entry: , the total number of differential equations.
Constraint:
.
- 4: – IntegerInput
-
On entry:
, the number of differential equations active in the matching process. The active equations must be placed last in the numbering in
fcn and
bc. The
first equations are used as the driving equations.
Constraint:
, and .
- 5: – Real (Kind=nag_wp) arrayInput
-
On entry:
, for
, must be set to a positive value for use in the convergence test in the
th argument
. See the description of
pf for further details.
Constraint:
, for .
- 6: – Real (Kind=nag_wp) arrayInput/Output
-
On entry:
, for
, should be set to a ‘floor’ value in the convergence test on the
th argument
. If
on entry then it is set to the small positive value
(where
may in most cases be considered to be
machine precision); otherwise it is used unchanged.
The Newton iteration is presumed to have converged if a full Newton step is taken (
in the specification of
monit), the singular values of the Jacobian are not being significantly perturbed (also see
monit) and if the Newton correction
satisfies
where
is the current value of the
th argument. The values
are also used in determining the Newton iterates as discussed in
Section 3, see equation
(6).
On exit: the values actually used.
- 7: – Real (Kind=nag_wp) arrayInput
-
On entry: values for use in controlling the local error in the integration of the differential equations. If
is an estimate of the local error in
, for
,
where
may in most cases be considered to be
machine precision.
Suggested value:
.
Constraint:
, for .
- 8: – Real (Kind=nag_wp) arrayInput/Output
-
On entry: a value to be used in perturbing the argument
in the numerical differentiation to estimate the Jacobian used in Newton's method. If
on entry, an estimate is made internally by setting
where
is the initial value of the argument supplied by you and
may in most cases be considered to be
machine precision. The estimate of the Jacobian,
, is made using forward differences, that is for each
, for
,
is perturbed to
and the
th column of
is estimated as
where the other components of
are unchanged (see
(3) for the notation used). If this fails to produce a Jacobian with significant columns, backward differences are tried by perturbing
to
and if this also fails then central differences are used with
perturbed to
. If this also fails then the calculation of the Jacobian is abandoned. If the Jacobian has not previously been calculated then an error exit is taken. If an earlier estimate of the Jacobian is available then the current argument set,
, for
, is abandoned in favour of the last argument set from which useful progress was made and the singular values of the Jacobian used at the point are modified before proceeding with the Newton iteration. You are recommended to use the default value
unless you have prior knowledge of a better choice. If any of the perturbations described are likely to lead to an unfortunate set of argument values then you should use the LOGICAL FUNCTION
constr to prevent such perturbations (all changes of arguments are checked by a call to
constr).
On exit: the values actually used.
- 9: – IntegerInput
-
On entry: 2 plus the number of break-points in the range of definition of the system of differential equations
(1).
Constraint:
.
- 10: – Real (Kind=nag_wp) arrayInput/Output
-
On entry:
must contain an estimate for an initial step size for integration across the
th sub-interval
, for
, (see
range).
should have the same sign as
if it is nonzero. If
, on entry, a default value for the initial step size is calculated internally. This is the recommended mode of entry.
must contain a lower bound for the modulus of the step size on the
th sub-interval , for . If on entry, a very small default value is used. By setting but smaller than the expected step sizes (assuming you have some insight into the likely step sizes) expensive integrations with arguments far from the solution can be avoided.
must contain an upper bound on the modulus of the step size to be used in the integration on , for . If on entry no bound is assumed. This is the recommended mode of entry unless the solution is expected to have important features which might be ‘missed’ in the integration if the step size were permitted to be chosen freely.
On exit:
contains the initial step size used on the last integration on
, for
, (excluding integrations during the calculation of the Jacobian).
, for , is usually unchanged. If the maximum step size is so small or the length of the range is so short that on the last integration the step size was not controlled in the main by the size of the error tolerances but by these other factors, is set to the floating-point value of if the problem last occurred in . Any results obtained when this value is returned as nonzero should be viewed with caution.
, for , are unchanged.
If an error exit with
,
or
(see
Section 6) occurs on the integration made from
to
the floating-point value of
is returned in
. The actual point
where the error occurred is returned in
(see also the specification of
w). The floating-point value of
npoint is returned in
if the error exit is caused by a call to
bc.
If an error exit occurs when estimating the Jacobian matrix (
,
,
,
,
or
, see
Section 6) and if argument
was the cause of the failure then on exit
contains the floating-point value of
.
contains the point , for , used at the solution or at the final values of if an error occurred.
swp is also partly used as workspace.
- 11: – IntegerInput
-
On entry: the first dimension of the array
swp as declared in the (sub)program from which
d02saf is called.
Constraint:
.
- 12: – IntegerInput
-
On entry: an upper bound on the number of Newton iterations. If on entry, no check on the number of iterations is made (this is the recommended mode of entry).
Constraint:
.
- 13: – Subroutine, supplied by the user.External Procedure
-
range must specify the break-points
, for
, which may depend on the arguments
, for
.
The specification of
range is:
Fortran Interface
Integer, Intent (In) | :: | npoint, m | Real (Kind=nag_wp), Intent (In) | :: | p(m) | Real (Kind=nag_wp), Intent (Out) | :: | x(npoint) |
|
C Header Interface
#include <nagmk26.h>
void |
range (double x[], const Integer *npoint, const double p[], const Integer *m) |
|
- 1: – Real (Kind=nag_wp) arrayOutput
-
On exit: the
th break-point, for
. The sequence
must be strictly monotonic, that is either
or
- 2: – IntegerInput
-
On entry: plus the number of break-points in .
- 3: – Real (Kind=nag_wp) arrayInput
-
On entry: the current estimate of the
th argument, for .
- 4: – IntegerInput
-
On entry: , the number of arguments.
range must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
Note: range should not return floating-point NaN (Not a Number) or infinity values, since these are not handled by
d02saf. If your code inadvertently
does return any NaNs or infinities,
d02saf is likely to produce unexpected results.
- 14: – Subroutine, supplied by the user.External Procedure
-
bc must place in
g1 and
g2 the boundary conditions at
and
respectively.
The specification of
bc is:
Fortran Interface
Integer, Intent (In) | :: | m, n | Real (Kind=nag_wp), Intent (In) | :: | p(m) | Real (Kind=nag_wp), Intent (Out) | :: | g1(n), g2(n) |
|
C Header Interface
#include <nagmk26.h>
void |
bc (double g1[], double g2[], const double p[], const Integer *m, const Integer *n) |
|
- 1: – Real (Kind=nag_wp) arrayOutput
-
On exit: the value of
, for , (where this may be a known value or a function of the parameters
, for ).
- 2: – Real (Kind=nag_wp) arrayOutput
-
On exit: the value of
, for
, (where these may be known values or functions of the parameters
, for
). If
, so that there are some driving equations, the first
values of
g2 need not be set since they are never used.
- 3: – Real (Kind=nag_wp) arrayInput
-
On entry: an estimate of the
th argument, , for .
- 4: – IntegerInput
-
On entry: , the number of arguments.
- 5: – IntegerInput
-
On entry: , the number of differential equations.
bc must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
Note: bc should not return floating-point NaN (Not a Number) or infinity values, since these are not handled by
d02saf. If your code inadvertently
does return any NaNs or infinities,
d02saf is likely to produce unexpected results.
- 15: – Subroutine, supplied by the user.External Procedure
-
fcn must evaluate the functions
(i.e., the derivatives
), for
.
The specification of
fcn is:
Fortran Interface
Subroutine fcn ( |
x, y, f, n, p, m, i) |
Integer, Intent (In) | :: | n, m, i | Real (Kind=nag_wp), Intent (In) | :: | x, y(n), p(m) | Real (Kind=nag_wp), Intent (Out) | :: | f(n) |
|
C Header Interface
#include <nagmk26.h>
void |
fcn (const double *x, const double y[], double f[], const Integer *n, const double p[], const Integer *m, const Integer *i) |
|
- 1: – Real (Kind=nag_wp)Input
-
On entry: , the value of the argument.
- 2: – Real (Kind=nag_wp) arrayInput
-
On entry: , for , the value of the argument.
- 3: – Real (Kind=nag_wp) arrayOutput
-
On exit: the derivative of
, for
, evaluated at
.
may depend upon the parameters
, for
. If there are any driving equations (see
Section 3) then these must be numbered first in the ordering of the components of
f.
- 4: – IntegerInput
-
On entry: , the number of equations.
- 5: – Real (Kind=nag_wp) arrayInput
-
On entry: the current estimate of the
th argument , for .
- 6: – IntegerInput
-
On entry: , the number of arguments.
- 7: – IntegerInput
-
On entry: specifies the sub-interval on which the derivatives are to be evaluated.
fcn must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
Note: fcn should not return floating-point NaN (Not a Number) or infinity values, since these are not handled by
d02saf. If your code inadvertently
does return any NaNs or infinities,
d02saf is likely to produce unexpected results.
- 16: – Subroutine, supplied by the NAG Library or the user.External Procedure
-
eqn is used to describe the additional algebraic equations to be solved in the determination of the parameters,
, for
. If there are no additional algebraic equations (i.e.,
) then
eqn is never called and the dummy routine d02hbz should be used as the actual argument.
The specification of
eqn is:
Fortran Interface
Subroutine eqn ( |
e, q, p, m) |
Integer, Intent (In) | :: | q, m | Real (Kind=nag_wp), Intent (In) | :: | p(m) | Real (Kind=nag_wp), Intent (Out) | :: | e(q) |
|
C Header Interface
#include <nagmk26.h>
void |
eqn (double e[], const Integer *q, const double p[], const Integer *m) |
|
- 1: – Real (Kind=nag_wp) arrayOutput
-
On exit: the vector of residuals, , that is the amount by which the current estimates of the arguments fail to satisfy the algebraic equations.
- 2: – IntegerInput
-
On entry: the number of algebraic equations, .
- 3: – Real (Kind=nag_wp) arrayInput
-
On entry: the current estimate of the
th argument , for .
- 4: – IntegerInput
-
On entry: , the number of arguments.
eqn must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
Note: eqn should not return floating-point NaN (Not a Number) or infinity values, since these are not handled by
d02saf. If your code inadvertently
does return any NaNs or infinities,
d02saf is likely to produce unexpected results.
- 17: – Logical Function, supplied by the user.External Procedure
-
constr is used to prevent the pseudo-Newton iteration running into difficulty.
constr should return the value .TRUE. if the constraints are satisfied by the parameters
. Otherwise
constr should return the value .FALSE.. Usually the dummy function d02hby, which returns the value .TRUE. at all times, will suffice and in the first instance this is recommended as the actual argument.
The specification of
constr is:
Fortran Interface
Logical | :: | constr | Integer, Intent (In) | :: | m | Real (Kind=nag_wp), Intent (In) | :: | p(m) |
|
C Header Interface
#include <nagmk26.h>
Nag_Boolean |
constr (const double p[], const Integer *m) |
|
- 1: – Real (Kind=nag_wp) arrayInput
-
On entry: an estimate of the
th argument, , for .
- 2: – IntegerInput
-
On entry: , the number of arguments.
constr must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
- 18: – Real (Kind=nag_wp)Input
-
On entry: a non-negative value which is used as a bound on all values where is the solution at any point between and for the current arguments . If this bound is exceeded the integration is terminated and the current arguments are rejected. Such a rejection will result in an error exit if it prevents the initial residual or Jacobian, or the final solution, being calculated. If on entry, no bound on the solution is used; that is the integrations proceed without any checking on the size of .
- 19: – Subroutine, supplied by the NAG Library or the user.External Procedure
-
monit enables you to monitor the values of various quantities during the calculation.
It
is called by
d02saf after every calculation of the norm
which determines the strategy of the Newton method, every time there is an internal error exit leading to a change of strategy, and before an error exit when calculating the initial Jacobian.
Usually the routine d02hbx will be adequate and you are advised to use this as the actual argument for
monit in the first instance. (In this case a call to
x04abf must be made before the call of
d02saf.) If no monitoring is required, the dummy routine d02sas may be used.
The specification of
monit is:
Fortran Interface
Integer, Intent (In) | :: | istate, iflag, ifail1, m | Real (Kind=nag_wp), Intent (In) | :: | p(m), f(m), pnorm, pnorm1, eps, d(m) |
|
C Header Interface
#include <nagmk26.h>
void |
monit (const Integer *istate, const Integer *iflag, const Integer *ifail1, const double p[], const Integer *m, const double f[], const double *pnorm, const double *pnorm1, const double *eps, const double d[]) |
|
- 1: – IntegerInput
-
On entry: the state of the Newton iteration.
- The calculation of the residual, Jacobian and are taking place.
- to
- During the Newton iteration a factor of of the Newton step is being used to try to reduce the norm.
- The current Newton step has been rejected and the Jacobian is being re-calculated.
- to
- An internal error exit has caused the rejection of the current set of argument values, . is the value which istate would have taken if the error had not occurred.
- An internal error exit has occurred when calculating the initial Jacobian.
- 2: – IntegerInput
-
On entry: whether or not the Jacobian being used has been calculated at the beginning of the current iteration. If the Jacobian has been updated then ; otherwise . The Jacobian is only calculated when convergence to the current argument values has been slow.
- 3: – IntegerInput
-
On entry: if
,
ifail1 specifies the
ifail error number that would be produced were control returned to you.
ifail1 is unspecified for values of
istate outside this range.
- 4: – Real (Kind=nag_wp) arrayInput
-
On entry: the current estimate of the
th argument , for .
- 5: – IntegerInput
-
On entry: , the number of arguments.
- 6: – Real (Kind=nag_wp) arrayInput
-
On entry:
, the residual corresponding to the current argument values, provided
or
.
f is unspecified for other values of
istate.
- 7: – Real (Kind=nag_wp)Input
-
On entry: a quantity against which all reductions in norm are currently measured.
- 8: – Real (Kind=nag_wp)Input
-
On entry:
, the norm of the current arguments. It is set for
and is undefined for other values of
istate.
- 9: – Real (Kind=nag_wp)Input
-
On entry: gives some indication of the convergence rate. It is the current singular value modification factor (see
Gay (1976)). It is zero initially and whenever convergence is proceeding steadily.
eps is
or greater (where
may in most cases be considered
machine precision) when the singular values of
are approximately zero or when convergence is not being achieved. The larger the value of
eps the worse the convergence rate. When
eps becomes too large the Newton iteration is terminated.
- 10: – Real (Kind=nag_wp) arrayInput
-
On entry:
, the singular values of the current modified Jacobian matrix. If
is small relative to
for a number of Jacobians corresponding to different argument values then the computed results should be viewed with suspicion. It could be that the matching equations do not depend significantly on some argument (which could be due to a programming error in
fcn,
bc,
range or
eqn). Alternatively, the system of differential equations may be very ill-conditioned when viewed as an initial value problem, in which case
d02saf is unsuitable. This may also be indicated by some singular values being very large. These values of
, for
, should not be changed.
monit must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
Note: monit should not return floating-point NaN (Not a Number) or infinity values, since these are not handled by
d02saf. If your code inadvertently
does return any NaNs or infinities,
d02saf is likely to produce unexpected results.
- 20: – Subroutine, supplied by the NAG Library or the user.External Procedure
-
prsol can be used to obtain values of the solution
at a selected point
by integration across the final range
. If no output is required d02hbw can be used as the actual argument.
The specification of
prsol is:
Fortran Interface
Subroutine prsol ( |
z, y, n) |
Integer, Intent (In) | :: | n | Real (Kind=nag_wp), Intent (In) | :: | y(n) | Real (Kind=nag_wp), Intent (Inout) | :: | z |
|
C Header Interface
#include <nagmk26.h>
void |
prsol (double *z, const double y[], const Integer *n) |
|
- 1: – Real (Kind=nag_wp)Input/Output
-
On entry: contains
on the first call. On subsequent calls
z contains its previous output value.
On exit: the next point at which output is required. The new point must be nearer
than the old.
If
z is set to a point outside
the process stops and control returns from
d02saf to the (sub)program from which
d02saf is called. Otherwise the next call to
prsol is made by
d02saf at the point
z, with solution values
at
z contained in
y. If
z is set to
exactly, the final call to
prsol is made with
as values of the solution at
produced by the integration. In general the solution values obtained at
from
prsol will differ from the values obtained at this point by a call to
bc. The difference between the two solutions is the residual
. You are reminded that the points
are available in the locations
at all times.
- 2: – Real (Kind=nag_wp) arrayInput
-
On entry: the solution value
, for , at .
- 3: – IntegerInput
-
On entry: , the total number of differential equations.
prsol must either be a module subprogram USEd by, or declared as EXTERNAL in, the (sub)program from which
d02saf is called. Arguments denoted as
Input must
not be changed by this procedure.
Note: prsol should not return floating-point NaN (Not a Number) or infinity values, since these are not handled by
d02saf. If your code inadvertently
does return any NaNs or infinities,
d02saf is likely to produce unexpected results.
- 21: – Real (Kind=nag_wp) arrayOutput
-
On exit: in the case of an error exit of the type where the point of failure is returned in
, the solution at this point of failure is returned in
, for
.
Otherwise
w is used for workspace.
- 22: – IntegerInput
-
On entry: the first dimension of the array
w as declared in the (sub)program from which
d02saf is called.
Constraint:
.
- 23: – IntegerInput
-
On entry: the second dimension of the array
w as declared in the (sub)program from which
d02saf is called.
Constraint:
.
- 24: – IntegerInput/Output
-
On entry:
ifail must be set to
,
. If you are unfamiliar with this argument you should refer to
Section 3.4 in How to Use the NAG Library and its Documentation for details.
For environments where it might be inappropriate to halt program execution when an error is detected, the value
is recommended. If the output of error messages is undesirable, then the value
is recommended. Otherwise, if you are not familiar with this argument, the recommended value is
.
When the value is used it is essential to test the value of ifail on exit.
On exit:
unless the routine detects an error or a warning has been flagged (see
Section 6).
6
Error Indicators and Warnings
If on entry
or
, explanatory error messages are output on the current error message unit (as defined by
x04aaf).
Errors or warnings detected by the routine:
-
On entry, an element of the parameter convergence control array is zero or negative.
On entry, .
Constraint: .
On entry, and .
Constraint: .
On entry, and .
Constraint: .
On entry, and .
Constraint: .
On entry, and .
Constraint: .
On entry, .
Constraint: .
On entry, and .
Constraint: .
On entry, .
Constraint: .
On entry, and .
Constraint: .
On entry, and .
Constraint: .
On entry, .
Constraint: .
On entry an element of the local error control array is zero or negative.
-
The constraints have been violated by the initial parameters.
-
The sequence of break-points is not strictly monotonic in their initial specification.
-
In the integration from first to last break-point with initial or final parameters, the step size was reduced too far for the integration to proceed. Consider reversing the order of break-points or relaxing the local error control. If this error exit still results, either this routine is not a suitable method for solving the problem, or the initial choice of parameters is very poor.
-
In the integration from first to last break-point with initial or final parameters, a suitable initial step could not be found on one of the intervals. Consider reversing the order of break-points or relaxing the local error control. If this error exit still results, either this routine is not suitable for solving the problem, or the initial choice of parameters is very poor.
-
During integration the solution exceeded
ymax in magnitude, where, on entry,
.
-
On calculating the initial approximation to the Jacobian, the constraints were violated.
-
On perturbing the parameters when calculating the initial approximation to the Jacobian, the monotonicity condition on break-points is violated.
-
An initial step-length could be found for integration to proceed with the current parameters.
-
The step-length required to calculate the Jacobian to sufficient accuracy is too small.
-
On calculating the initial approximation to the Jacobian, the solution to the system of differential equations exceeded
ymax in magnitude, where, on entry,
.
-
The Jacobian has an insignificant column. Make sure that the solution vector depends on all the parameters.
-
An internal singular value decomposition has failed. This error can be avoided by changing the initial parameter estimates.
-
The Newton iteration has failed to converge. This can indicate a poor initial choice of parameters or a very difficult problem. Consider varying elements of the parameter convergence control if the residuals are small; otherwise vary initial parameter estimates.
-
The number of iterations permitted by
icount has been exceeded where this was set to a positive value on entry.
.
-
Internal error in calculating Jacobian. Please contact
NAG.
-
Internal error in calculating residual. Please contact
NAG.
-
Internal error in integration. Please contact
NAG.
-
Internal error in Newton method. Please contact
NAG.
An unexpected error has been triggered by this routine. Please
contact
NAG.
See
Section 3.9 in How to Use the NAG Library and its Documentation for further information.
Your licence key may have expired or may not have been installed correctly.
See
Section 3.8 in How to Use the NAG Library and its Documentation for further information.
Dynamic memory allocation failed.
See
Section 3.7 in How to Use the NAG Library and its Documentation for further information.
7
Accuracy
If the iteration converges, the accuracy to which the unknown arguments are determined is usually close to that specified by you. The accuracy of the solution (output via
prsol) depends on the error tolerances
, for
. You are strongly recommended to vary all tolerances to check the accuracy of the arguments
and the solution
.
8
Parallelism and Performance
d02saf is not thread safe and should not be called from a multithreaded user program. Please see
Section 3.12.1 in How to Use the NAG Library and its Documentation for more information on thread safety.
d02saf makes calls to BLAS and/or LAPACK routines, which may be threaded within the vendor library used by this implementation. Consult the documentation for the vendor library for further information.
Please consult the
X06 Chapter Introduction for information on how to control and interrogate the OpenMP environment used within this routine. Please also consult the
Users' Note for your implementation for any additional implementation-specific information.
The time taken by
d02saf depends on the complexity of the system of differential equations and on the number of iterations required. In practice, the integration of the differential system
(1) is usually by far the most costly process involved. The computing time for integrating the differential equations can sometimes depend critically on the quality of the initial estimates for the arguments
. If it seems that too much computing time is required and, in particular, if the values of the residuals (output in
monit) are much larger than expected given your knowledge of the expected solution, then the coding of
fcn,
eqn,
range and
bc should be checked for errors. If no errors can be found then an independent attempt should be made to improve the initial estimates
.
In the case of an error exit in the integration of the differential system indicated by
,
,
or
you are strongly recommended to perform trial integrations with
d02pff to determine the effects of changes of the local error tolerances and of changes to the initial choice of the arguments
, for
, (that is the initial choice of
).
It is possible that by following the advice given in
Section 6 an error exit with
,
,
,
or
might be followed by one with
(or vice-versa) where the advice given is the opposite. If you are unable to refine the choice of
, for
, such that both these types of exits are avoided then the problem should be rescaled if possible or the method must be abandoned.
The choice of the ‘floor’ values
, for , may be critical in the convergence of the Newton iteration. For each value , the initial choice of and the choice of should not both be very small unless it is expected that the final argument will be very small and that it should be determined accurately in a relative sense.
For many problems it is critical that a good initial estimate be found for the arguments
or the iteration will not converge or may even break down with an error exit. There are many mathematical techniques which obtain good initial estimates for
in simple cases but which may fail to produce useful estimates in harder cases. If no such technique is available it is recommended that you try a continuation (homotopy) technique preferably based on a physical argument (e.g., the Reynolds or Prandtl number is often a suitable continuation argument). In a continuation method a sequence of problems is solved, one for each choice of the continuation argument, starting with the problem of interest. At each stage the arguments
calculated at earlier stages are used to compute a good initial estimate for the arguments at the current stage (see
Hall and Watt (1976) for more details).
10
Example
This example intends to illustrate the use of the break-point and equation solving facilities of
d02saf. Most of the facilities which are common to
d02saf and
d02hbf are illustrated in the example in the specification of
d02hbf (which should also be consulted).
The program solves a projectile problem in two media determining the position of change of media, , and the gravity and viscosity in the second medium ( represents gravity and represents viscosity).
10.1
Program Text
Program Text (d02safe.f90)
10.2
Program Data
Program Data (d02safe.d)
10.3
Program Results
Program Results (d02safe.r)