2011-05-02 Brian Gough * gsl_odeiv2.h: fix header guard to __GSL_ODEIV2_H__ 2011-05-01 Tuomo Keskitalo * Replaced ChangeLog with contents of ode-initval2 development log file (ChangeLog-odeiv2_development) and reformatted the log file. 2010-10-16 Tuomo Keskitalo * Modified evolve and driver so that when user function returns GSL_EBADFUNC, then ode-initval2 functions return immediately with that same return value. * Corrected bug in msadams for saving ordm1coeff to state. Thanks to Andrew Benson for pointing out this bug! 2010-10-10 Tuomo Keskitalo * driver.c: Corrected error in driver_evolve_apply, integration to negative direction should work now. * Added evolve_apply_fixed_step and driver_apply_fixed_step for users who want to use a constant step size to evolve the system. * test.c: added test cases for driver (negative evolution, apply_fixed_step). * gsl_odeiv2.h: cleaning and add apply_fixed_step functions. * doc/ode-initval.texi: Replaced derivative apostrophe with \prime (because my new system fails to make ps otherwise). Added new documentation about apply_fixed_step functions. 2010-10-09 Tuomo Keskitalo * ode-initval2 bzr development branch now at savannah: http://bzr.savannah.gnu.org/r/gsl/ode-initval2/ 2010-01-02 Tuomo Keskitalo * Changes from ode-initval2-0.9 to ode-initval2-1.0: * Removed set_control member from step object and added a set_driver member to evolve, control and step objects, so that the driver object and consequently evolve, control and step objects are all accessible from another. Currently only some steppers utilize this. They access the control_errlevel function via state->driver->c instead of state->control. This modification to the framework enables implementation of more efficient ODE solvers, which utilize communication among evolve, control and stepper objects. This also means that some steppers now require a driver object, and it is best to use those stepper methods through driver object functions. * cstd.c and cscal.c: Added a sanity check for errlev <= 0 * run indent -gnu -nut for *.c *.h * driver.c: added function odeiv2_driver_alloc_standard_new and odeiv2_driver_alloc_yp_new * driver.c: changed n and nmax in driver struct to unsigned long int * modified documentation accordingly with small nomenclature changes and made year updates to source files 2009-11-01 Tuomo Keskitalo * Merged with master and modified source to be published as a package for gsl-1.13. 2009-05-09 Tuomo Keskitalo * Agreed with Brian Gough that the modification I've made cannot be incorporated to GSL until a major version number change, due to changes in API. * Replated rk2imp with imprk2 and rk4imp with imprk4. Also renamed impeuler to rk1imp for consistency. * Deleted steppers that are deprecated considering new major release: gear1, gear2, and rk2simp. Modified test.c accordingly. 2009-05-01 Tuomo Keskitalo * Merged with master (evolve.c and test.c needed some manual work). * driver.c: Modified alloc-functions to minimal interface, added set_hmax, set_nmax and set_hmin functions. * test.c: Added test for minimum allowed step size. Applied indent -gnu -nut. * doc/ode-initval.texi: Updated documentation, added Driver section, updated examples. 2009-04-05 Tuomo Keskitalo * Several uninitialized variable corrections from Brian Gough added * modnewton1, impeuler, imprk2, imprk4: removed unneeded arguments from solve and init * msadams and msbdf modifications: * added ordprevbackup, ordwaitbackup (backups) and failcount (to detect repeated steps rejections). * added checks to make sure order is not changed by more than one order. Added sanity check to catch it. * corrected failure and rejection handling for first step * Added a heuristic stability enhancement for msbdf. This is yet to be tested in practice. * test.c: Added more benchmark test cases. The results seem sensible. * test.c: Note for benchmark_precision: msbdf performs badly in rhs_func_exp. Newton iteration starts to fail to converge at some point, and this forces very small step sizes to be used. I currently think that this is not a bug, but that msbdf is just not suitable for this problem. I think that msbdf's prediction step does not give a good initial guess for Newton iteration in this case. * I tried to increase max_iter in msbdf from 3 to 7 (value used in modnewton1), and as a result, the performance was better for this case (Newton iteration convergence was better). However, I would not increase max_iter, because the default value of 3 is the value used by the authors in the references, and because it would decrease the efficiency of msbdf (method would use more function evaluations per step, and convergence might still not be achieved). It is best to useanother stepper in this kind of a case. 2009-02-28 Tuomo Keskitalo * new stepper: msbdf. This is a variable-coefficient linear multistep backward differentiation formula (BDF) method in Nordsieck form. It appears to handle stiff cases well and it is efficient! * test.c: Ordered functions in file. Added extreme test case. * modnewton1: Convergence test now takes the desired error level via control object into account. Modified steppers that use modnewton1 accordingly. * modnewton1: Removed reuse_decomp. The variable was useless since iteration matrix is reconstructed at each stepper call. * modnewton1: Dropped max_iter from 10 to 7. Hairer and Wanner suggests a "relatively high number", but many iterations per step is costly. * steppers that use step doubling for error estimates: Introduced constant GSL_ODEIV_ERR_SAFETY in gsl_odeiv.h that is used to multiply the error estimates (yerr). This way user can modify this coefficient in one place if a change is wanted. * driver.c: This is a high level wrapper for ode-initval, which was wished for in TODO. Added test_driver to test.c to test it. Is this functionality enough? * ode-initval.texi: Added documentation for new steppers. Moved note about user error codes to system definition. * Some finetuning / code cleanup. 2009-01-21 Tuomo Keskitalo * New explicit stepper: msadams (An explicit variable-coefficient linear multistep Adams method in Nordsieck form). Functional iteration is used to solve the non-linear equation system. The algorithms described in the references (see msadams.c) have been adapted to fit GSL odeiv framework. Currently this is beta, but the stepper now passes the odeiv test suite. * control.c, cstd.c, cscal.c: Added errlev function which calculates desired error level. This is used in masadams (and can be used in imp* steppers, too) to control accuracy of functional iteration. User must call gsl_odeiv_step_set_control after allocating step and control objects to pass the information about control object to stepper. This change (or something similar) is needed to pass tolerance for the stepper routines that need it. Modified stepper routines accordingly. * evolve.c: Changed the step length scaling coefficient for stepper failure from 0.1 to 0.5, because msadams seems to cope much better with this change in stiff[0,1] test case. * gsl_odeiv.h: gsl_odeiv_control and control_type introduced earlier because gsl_odeiv_step_type now refers to it. 2008-11-01 Tuomo Keskitalo * New implicit solvers: imprk4 (replaces imprk42 in previous test package), imprk2 (replaces imprk21 in previous test package) and impeuler, which use modified Newton iteration to solve the implicit equations. I decided to dump imprk42 and imprk21 because I was unsure how to calculate the error estimates for them. I think that these new implicit methods could be used instead of rk4imp, rk2imp, gear2 and gear1, which use functional iteration (which is inefficient for stiff problems.) However, even these new solvers can not be used efficiently for truly nasty stiff problems, and better solvers are still needed. * modnewton1.c: a modified Newton solver for solution of non-linear equations of implicit solvers. Jacobian evaluation has been moved from modnewton1 to imprk2, imprk4 and impeuler. * evolve.c: modification to decrease step size if stepper fails, instead of giving up immediately. * evolve.c: A bug fix: Exit with GSL_FAILURE if step size reaches machine precision instead of continuing with old step size. * test.c: added Robertson stiff test * test.c: embedded test_oregonator to test_compare_stiff_probelms to test several stiff problems. Note: the tested integration interval is rather short due to inefficiency of solvers other than bsimp in these problems. * test.c: modified test_compare_vanderpol and test_compare_stiff_problems to check against results from first ode-solver. Changed rk4 or bsimp to first place. * test.c: modified stepfn and stepfn2 and expanded these to be tested with all explicit solvers * test.c: removed stepfn3 and added test_broken and test_stepsize_fail to check that evolve decreases step size below machine precision and exits with a failure code in the end if user functions fail continuously. * Changes in ode-initval.texi: * removed untrue sentence concerning stepping functions: "The step-size @var{h} will be set to the step-size which caused the error." * Changed gsl_odeiv_system variable name from dydt to sys for clarity * added "explicit" or "implicit" to stepper introductions for clarity * reformulated description of evolve_apply and step_apply and made changes to reflect the modifications to the code. * clarified the point on integrating over discontinuities. It is best to integrate in sequences. * removed the text suggesting the user to force a step size to integrate over the edge of a discontinuity. The applicability of this kind of a numerical tweak depends on the case, and should in my opinion not to be included in a reference book.