Thank you, this is an impressive result. :) 2009/8/20 Aaron S. Meurer <asmeu...@gmail.com>
> > I have finished up my Google Summer of Code 2009 project, which was to > implement various ODE solvers in SymPy. Please pull fromhttp:// > github.com/asmeurer/sympy/tree/odes-review > . Note that I have elected to do only moderate rebasing from my > original tree (branch odes), so you should really only be reviewing > the HEAD. > > This includes the following features: > > - All ODE solving now happens in ode.py, instead of solvers.py where > it was crowding stuff. > - dsolve() remains the function to use to solve an ODE. The syntax is > the same for default solving: dsolve(ode, func). > For example: > >>> dsolve(diff(f(x), x, 2) - 1, f(x)) > f(x) == C1 + C2*x + x**2/2 > > dsolve() always returns an Equality instance, because it very often > cannot solve for f(x) and must return an implicit solution. > > - The function classify_ode() was added. This function classifies > ODEs into the various methods (or "hints", as I also call them) that > dsolve() can use to solve them. > For example: > >>> classify_ode(f(x).diff(x) - 1, f(x)) > ('separable', '1st_exact', '1st_linear', 'Bernoulli', > '1st_homogeneous_coeff_best', > '1st_homogeneous_coeff_subs_indep_div_dep', > '1st_homogeneous_coeff_subs_dep_div_indep', > 'nth_linear_constant_coeff_undetermined_coefficients', > 'nth_linear_constant_coeff_variation_of_parameters', > 'separable_Integral', '1st_exact_Integral', '1st_linear_Integral', > 'Bernoulli_Integral', > '1st_homogeneous_coeff_subs_indep_div_dep_Integral', > '1st_homogeneous_coeff_subs_dep_div_indep_Integral', > 'nth_linear_constant_coeff_variation_of_parameters_Integral') > > - The function deriv_degree() remains the function to use for > determining the degree, or order, of an ODE. > For example: > >>> deriv_degree(f(x).diff(x, 5) + sin(x) + f(x).diff(x, 2), f(x)) > 5 > > - The new function checkodesol() checks if a solution to an ODE is > valid by substituting it in and either returning True if it finds 0 > equivalence or the expression that it could not reduce to 0. > For example: > >>> checkodesol(f(x).diff(x, 2) - 1, f(x), Eq(f(x), C1 + C2*x + > x**2/2)) > True > >>> checkodesol(f(x).diff(x, 2) - 1, f(x), Eq(f(x), C1 + C2*x + x**2)) > 1 This is problematic, because this way if not checkodesol(...): print 'not a solution' does not work, because most expressions evaluate to True. I'd prefer returning just 0 instead of True, this is more consistent and useful. > > > - The function separatevars(), which I wrote to help with the 1st > order separable method, attempts to intelligently separate variables > an expression multiplicatively. > For example: > >>> separatevars(x**2 + x**2*sin(y)) > x**2*(1 + sin(y)) > >>> separatevars(exp(x + y)) > exp(x)*exp(y) > > This function resides in simplify.py. > > - The new function homogeneous_order(), which I wrote to help with the > 1st order ODE with homogeneous coefficients method, determines the > homogeneous order of an expression. An expression F(x, y, ...) is > homogeneous of order n if F(x*t, y*t, t*...) == t**n*F(x, y, ...). > For example: > >>> homogeneous_order(x**2*sqrt(x**2 + y**2) + x*y**2*sin(x/y), x, y) > 3 > > - dsolve() will simplify arbitrary constants, and renumber them in > order in the expression. You should never see something like x + 2*C1 > or 1 + C2*x + C1*x**2 returned from dsolve(). You can disable this by > using simplify=False in dsolve(). > > - Internally, the solving engine has been completely refactored. The > old way was to match the ODE to various solvers in a predetermined > order, and as soon as it matched one, it would solve it using that > method and return a result. This had several problems with it. > > First, several ODEs can be solved with more than one method (c.f. the > simple ODE in the classify_ode() example above), and sometimes you > want to see the result from a different method. > > Second, sometimes the default solver would return a result that was > much more complex than another one would have returned if the ODE had > gotten to it. For example, it may return an implicit solution that is > difficult for solve() to solve, whereas the other method would return > the equation already solver. typo > Another common case is that one method > would set up an integral that integrate() could not do, whereas > another method would set up a simpler integral. That leads to the > third point. > > Third, whenever a solver would set up a hard integral, integrate() > would either hand for a while then return an Integral, or it would > hang forever. Either way, you would get an unevaluatable (by SymPy) > integral, when sometimes another method might not give you that. > > Fourth, there was a lot of code duplicated this way. For example, > whenever a method had the potential to return an implicit solution, it > had to duplicate the same code chunk that tried to solve it and return > a solved solution if it could be solved or else an implicit solution. > > The new way is for dsolve to have an optional hint argument that you > can use to specify whatever method you want to use, if it fits. > dsolve() calls classify_ode() on the ODE and gets all of the possible > hints. By default, hint="default" is used, which uses a predetermined > order, much like the old way. But you can easily use something like > hint="1st_exact" to change the hint from the default method to the 1st > order exact ODE solver. Aside from "default", there are also some > other 'meta-hints' that you can use. "all" will return a dictionary > of hint: solution terms for all matching hints. "best" will use a > heuristic to return the simplest of all possible solutions. This > heuristic is based on whether the solution has unevaluated Integrals > in it, whether it can be solved explicitly, and finally, just which > one is the shortest. All hints that set up integrals also have a > coresponding "hint_Integral" hint, which will not attempt to evaluate > the integrals but instead return unevaluated Integrals in the > solution. "all_Integral" acts like "all", but it skips hints that > attempt to evaluate integrals. This is so you can quickly get an > unevaluated integral of something you know SymPy cannot do, instead of > letting integrate() hang. It also lets you discover how the solvers > work by showing you what integrals they do. > > - Along with extensive documentation on all of the functions, I have > included a guide in the module docstring on how to add more methods to > the module using the hints system. It can give you an idea of how the > system works internally. > > > The following ODE solving methods have been implemented. See the > internal docstrings for more information on each method: > - 1st order separable differential equations > - 1st order differential equations whose coefficients or dx and dy > are functions homogeneous of the same order. > - 1st order exact differential equations. > - 1st order linear differential equations. (this is already > implemented in the current SymPy) > - 1st order Bernoulli differential equations. (this is already > implemented in the current SymPy by a patch I sent in in the spring) > - 2nd order Liouville differential equations. > - nth order linear homogeneous differential equation with constant > coefficients. > - nth order linear inhomogeneous differential equation with constant > coefficients using the method of undetermined coefficients. > - nth order linear inhomogeneous differential equation with constant > coefficients using the method of variation of parameters. > > In addition to the ode module, this branch includes fixes to the > following issues, which were necessary for me throughout my work: > - Issues 1582, 1566, 1530, and 1429 (the first part only). > - Integral.__getnewargs__ now properly works. If r is an Integral > instance, then r == r.new(r.__getnewargs__(r.args)) should always be > True. > - Changed two assertions and one raise TypeError in solve() to raise > NotImplementedError. This way, you do not have to check for > AssertionError or TypeError to see if solve() has failed, only > NotImplementedError. See issues 1425 and 1338. I'll have to play around with this. :) Vinzent --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "sympy-patches" group. To post to this group, send email to sympy-patches@googlegroups.com To unsubscribe from this group, send email to sympy-patches+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sympy-patches?hl=en -~----------~----~----~----~------~----~------~--~---