NAG Library Routine Document
d02pgf (ivp_rk_step_revcomm)
1
Purpose
d02pgf is a reverse communication onestep routine for solving an initial value problem for a firstorder system of ordinary differential equations using Runge–Kutta methods. The direct communication version of this routine is
d02pff. See
Section 3.3.3 in How to Use the NAG Library and its Documentation for the difference between forward and reverse communication.
2
Specification
Fortran Interface
Integer, Intent (In)  ::  n  Integer, Intent (Inout)  ::  irevcm, iwsav(130), ifail  Real (Kind=nag_wp), Intent (In)  ::  yp(n)  Real (Kind=nag_wp), Intent (Inout)  ::  rwsav(32*n+350)  Real (Kind=nag_wp), Intent (Out)  ::  t, y(n) 

C Header Interface
#include nagmk26.h
void 
d02pgf_ (Integer *irevcm, const Integer *n, double *t, double y[], const double yp[], Integer iwsav[], double rwsav[], Integer *ifail) 

3
Description
d02pgf and its associated routines
(
d02phf,
d02pjf,
d02pqf,
d02prf,
d02ptf and
d02puf) solve an initial value problem for a firstorder system of ordinary differential equations. The routines, based on Runge–Kutta methods and derived from RKSUITE (see
Brankin et al. (1991)), integrate
where
$y$ is the vector of
$\mathit{n}$ solution components and
$t$ is the independent variable.
d02pgf is designed to be used in complicated tasks when solving systems of ordinary differential equations. You must first call
d02pqf to specify the problem and how it is to be solved. Thereafter you (repeatedly) call
d02pgf in reverse communication loops to take one integration step at a time from
tstart in the direction of
tend (as specified in
d02pqf). In this manner
d02pgf returns an approximation to the solution
y and its derivative
yp at successive points
t. If
d02pgf encounters some difficulty in taking a step, the integration is not advanced and the routine returns with the same values of
t,
y and
yp as returned on the previous successful step.
d02pgf tries to advance the integration as far as possible subject to passing the test on the local error and not going past
tend.
In the call to
d02pqf you can specify either the first step size for
d02pgf to attempt or it computes automatically an appropriate value. Thereafter
d02pgf estimates an appropriate step size for its next step. This value and other details of the integration can be obtained after a completed step by
d02pgf by a call to
d02ptf. The local error is controlled at every step as specified in
d02pqf. If you wish to assess the true error, you must set
method to a positive value
in the call to
d02pqf. This assessment can be obtained after any call to
d02pgf by a call to
d02puf.
If you want answers at specific points there are two ways to proceed:
(i) 
The more efficient way is to step past the point where a solution is desired, and then call d02phf and d02pjf to get an answer there. Within the span of the current step, you can get all the answers you want at very little cost by repeated calls to d02pjf. This is very valuable when you want to find where something happens, e.g., where a particular solution component vanishes. 
(ii) 
Alternatively, set tend to the desired value and integrate to tend. d02pgf will not step past tend, so when a step would carry it past, it will reduce the step size so as to produce an answer at tend exactly. After getting an answer there (${\mathbf{t}}={\mathbf{tend}}$), you can reset tend to the next point where you want an answer, and repeat. tend could be reset by a call to d02pqf, but you should not do this. You should use d02prf instead because it is both easier to use and much more efficient. This way of getting answers at specific points can be used with any of the available methods, but it can be inefficient. Should this be the case, the code will bring the matter to your attention. 
4
References
Brankin R W, Gladwell I and Shampine L F (1991) RKSUITE: A suite of Runge–Kutta codes for the initial value problems for ODEs SoftReport 91S1 Southern Methodist University
5
Arguments
Note: this routine uses
reverse communication. Its use involves an initial entry, intermediate exits and reentries, and a final exit, as indicated by the argument
irevcm. Between intermediate exits and reentries,
all arguments other than those specified by the value of irevcm must remain unchanged.
 1: $\mathbf{irevcm}$ – IntegerInput/Output

On initial entry:
irevcm must be set to zero to indicate that a new step is being taken.
On intermediate reentry:
irevcm should remain unchanged.
On intermediate exit:
irevcm returns a value
$\text{}>0$ to indicate that a function evaluation is required prior to reentry; the value of the derivatives
${y}^{\prime}=f\left(t,y\right)$ must be returned in
yp where the value of
$t$ is supplied in
t and the values
$y\left(t\right)$ are supplied in the array
y. The value of
irevcm indicates the reason for the function evaluation as follows:
 ${\mathbf{irevcm}}=1$
 For initial entry values of t and y.
 ${\mathbf{irevcm}}=2$
 To determine stiffness of system.
 ${\mathbf{irevcm}}=3$
 For the stages of the primary step.
 ${\mathbf{irevcm}}=4$
 A final stage of the primary step.
 ${\mathbf{irevcm}}=5$
 For the stages of a secondary step (if global error assessment is required).
On final exit:
 ${\mathbf{irevcm}}=1$
 Successful exit; t, y and yp contain the solution at the end of a successful integration step.
 ${\mathbf{irevcm}}=2$
 Error exit; ifail should be interrogated to determine the nature of the error.
Note: any values you return to d02pgf as part of the reverse communication procedure should not include floatingpoint NaN (Not a Number) or infinity values, since these are not handled by d02pgf. If your code inadvertently does return any NaNs or infinities, d02pgf is likely to produce unexpected results.
 2: $\mathbf{n}$ – IntegerInput

On entry: $n$, the number of ordinary differential equations in the system to be solved.
Constraint:
${\mathbf{n}}\ge 1$. This must be the same value as supplied in a previous call to
d02pqf.
 3: $\mathbf{t}$ – Real (Kind=nag_wp)Output

On intermediate exit:
t contains the value of the independent variable
$t$ at which the derivatives
${y}^{\prime}$ are to be evaluated.
On final exit: the value of $t$ at which a solution has been computed following a successful step.
 4: $\mathbf{y}\left({\mathbf{n}}\right)$ – Real (Kind=nag_wp) arrayOutput

On intermediate exit:
y contains the value of the solution
$y$ at which the derivatives
${y}^{\prime}$ are to be evaluated.
On final exit: the approximation to the solution computed following a successful step.
 5: $\mathbf{yp}\left({\mathbf{n}}\right)$ – Real (Kind=nag_wp) arrayInput

On initial entry:
yp need not be set.
On intermediate reentry:
yp must contain the value of the derivatives
${y}^{\prime}=f\left(t,y\right)$ where
$t$ is supplied in
t and
$y$ is supplied in the array
y.
 6: $\mathbf{iwsav}\left(130\right)$ – Integer arrayCommunication Array
 7: $\mathbf{rwsav}\left(32\times {\mathbf{n}}+350\right)$ – Real (Kind=nag_wp) arrayCommunication Array

On entry: these must be the same arrays supplied in a previous call to
d02pqf. They must remain unchanged between calls.
On exit: information about the integration for use on subsequent calls to d02pgf or other associated routines.
 8: $\mathbf{ifail}$ – IntegerInput/Output

On initial entry:
ifail must be set to
$0$,
$1\text{ or}1$. 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
$1\text{ or}1$ is recommended. If the output of error messages is undesirable, then the value
$1$ is recommended. Otherwise, because for this routine the values of the output arguments may be useful even if
${\mathbf{ifail}}\ne {\mathbf{0}}$ on exit, the recommended value is
$1$.
When the value $\mathbf{1}\text{ or}1$ is used it is essential to test the value of ifail on exit.
On final exit:
${\mathbf{ifail}}={\mathbf{0}}$ unless the routine detects an error or a warning has been flagged (see
Section 6).
6
Error Indicators and Warnings
If on entry
${\mathbf{ifail}}=0$ or
$1$, explanatory error messages are output on the current error message unit (as defined by
x04aaf).
Errors or warnings detected by the routine:
 ${\mathbf{ifail}}=1$

A call to this routine cannot be made after it has returned an error.
The setup routine must be called to start another problem.
${\mathbf{irevcm}}<0$ on entry.
On entry, a previous call to the setup routine has not been made or the communication arrays have become corrupted.
On entry, ${\mathbf{n}}=\u2329\mathit{\text{value}}\u232a$, but the value passed to the setup routine was ${\mathbf{n}}=\u2329\mathit{\text{value}}\u232a$.
On entry, the communication arrays have become corrupted, or a catastrophic error has already been detected elsewhere. You cannot continue integrating the problem.
tend, as specified in the setup routine, has already been reached. To start a new problem, you will need to call the setup routine. To continue integration beyond
tend then
d02prf must first be called to reset
tend to a new end value.
 ${\mathbf{ifail}}=2$

More than
$100$ output points have been obtained by integrating to
tend (as specified in the setup routine). They have been so clustered that it would probably be (much) more efficient to use the interpolation routine. However, you can continue integrating the problem.
 ${\mathbf{ifail}}=3$

Approximately $\u2329\mathit{\text{value}}\u232a$ function evaluations have been used to compute the solution since the integration started or since this message was last printed. However, you can continue integrating the problem.
 ${\mathbf{ifail}}=4$

Approximately
$\u2329\mathit{\text{value}}\u232a$ function evaluations have been used to compute the solution since the integration started or since this message was last printed. Your problem has been diagnosed as stiff. If the situation persists, it will cost roughly
$\u2329\mathit{\text{value}}\u232a$ times as much to reach
tend (setup) as it has cost to reach the current time. You should probably call routines intended for stiff problems. However, you can continue integrating the problem.
 ${\mathbf{ifail}}=5$

In order to satisfy your error requirements the solver has to use a step size of $\u2329\mathit{\text{value}}\u232a$ at the current time, $\u2329\mathit{\text{value}}\u232a$. This step size is too small for the machine precision, and is smaller than $\u2329\mathit{\text{value}}\u232a$.
 ${\mathbf{ifail}}=6$

The global error assessment algorithm failed at start of integration.
The integration is being terminated.
The global error assessment may not be reliable for times beyond $\u2329\mathit{\text{value}}\u232a$.
The integration is being terminated.
 ${\mathbf{ifail}}=99$
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.
 ${\mathbf{ifail}}=399$
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.
 ${\mathbf{ifail}}=999$
Dynamic memory allocation failed.
See
Section 3.7 in How to Use the NAG Library and its Documentation for further information.
7
Accuracy
The accuracy of integration is determined by the arguments
tol and
thresh in a prior call to
d02pqf (see the routine document for
d02pqf for further details and advice). Note that only the local error at each step is controlled by these arguments. The error estimates obtained are not strict bounds but are usually reliable over one step. Over a number of steps the overall error may accumulate in various ways, depending on the properties of the differential system.
8
Parallelism and Performance
d02pgf 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 implementationspecific information.
If
d02pgf returns with
${\mathbf{ifail}}>{\mathbf{1}}$ then the values returned in
t and
y are for the last successful step, or the initial conditions supplied if no successful step has been taken.
If
d02pgf returns with
${\mathbf{ifail}}={\mathbf{5}}$ and the accuracy specified by
tol and
thresh is really required then you should consider whether there is a more fundamental difficulty. For example, the solution may contain a singularity. In such a region the solution components will usually be large in magnitude. Successive output values of
y should be monitored with the aim of trapping the solution before the singularity. In any case, numerical integration cannot be continued through a singularity and analytical treatment may be necessary.
Performance statistics are available after any return from
d02pgf (except when
${\mathbf{ifail}}={\mathbf{1}}$) by a call to
d02ptf. If
${\mathbf{method}}>0$ in the call to
d02pqf, global error assessment is available after any return from
d02pgf (except when
${\mathbf{ifail}}={\mathbf{1}}$) by a call to
d02puf.
After a failure with
${\mathbf{ifail}}={\mathbf{5}}$ or
${\mathbf{6}}$ each of the diagnostic routines
d02ptf and
d02puf
may be called only once.
If d02pgf returns with ${\mathbf{ifail}}={\mathbf{4}}$ then it is advisable to change to another code more suited to the solution of stiff problems. d02pgf will not return with ${\mathbf{ifail}}={\mathbf{4}}$ if the problem is actually stiff but it is estimated that integration can be completed using less function evaluations than already computed.
10
Example
This example solves the equation
reposed as
over the range
$\left[0,2\pi \right]$ with initial conditions
${y}_{1}=0.0$ and
${y}_{2}=1.0$. We use relative error control with threshold values of
$\text{1.0E\u22128}$ for each solution component and print the solution at regular intervals using the interpolation routines
d02phf and
d02pjf within integration steps across the range; points on the range at which
${y}_{1}$ or
${y}_{2}$ change sign are also evaluated using a combination of the root finding routine
c05azf and the interpolation routines. We use a medium order Runge–Kutta method
(
${\mathbf{method}}=2$) with tolerance
${\mathbf{tol}}=\text{1.0E\u22125}$.
10.1
Program Text
Program Text (d02pgfe.f90)
10.2
Program Data
Program Data (d02pgfe.d)
10.3
Program Results
Program Results (d02pgfe.r)