November 2002 - PURRS-devel

purrs/src rr_solver.cc
by Tatiana Zolo 20 Nov '02

20 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: zolo(a)cs.unipr.it 2002-11-20 15:05:51 Modified files: src : rr_solver.cc Log message: Modified `solve_constant_coeff_order_2()' so that it returns the symbolic solution with the object "sum" if the inhomogeneous term is not in the form of exponentials or polynomials times exponentials. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/rr_solver.cc.diff?cvsro…

1 0

purrs/tests heap
by Tatiana Zolo 20 Nov '02

20 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: zolo(a)cs.unipr.it 2002-11-20 12:04:10 Modified files: tests : heap Log message: Marked with `y' the recurrences of the first order that have the sums not Gosper-summable. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/tests/heap.diff?cvsroot=pur…

1 0

purrs/src rr_solver.cc Recurrence.defs.hh
by Tatiana Zolo 20 Nov '02

20 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: zolo(a)cs.unipr.it 2002-11-20 12:02:11 Modified files: src : rr_solver.cc Recurrence.defs.hh Log message: Moved the decomposition of the inhomogeneous term inside the functions `solve_constant_coeff_order_1()', `solve_constant_coeff_order_2()' and `solve_constant_coeff_order_k()'. Furthermore, now returns the symbolic solution with the "sum" object when the summation is not Gosper-summable, instead of to return TOO_COMPLEX. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/rr_solver.cc.diff?cvsro… http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/Recurrence.defs.hh.diff…

1 0

[Fwd: Re: Automatic evaluation/simplification]
by Roberto Bagnara 20 Nov '02

20 Nov '02

-------- Original Message -------- Subject: Re: Automatic evaluation/simplification Date: Wed, 20 Nov 2002 10:07:32 +0100 From: KRECKEL Richard NGLT <Richard.Kreckel(a)framatome-anp.com> To: 'bagnara(a)cs.unipr.it' <bagnara(a)cs.unipr.it> Dear Roberto, > Yes, the FAQ says it is completely automatic. However, is > there a way to switch it off? > > Forgive me for the very imprecise question. The situation > is as follows: we have some monster expressions where we need > to substitute a few thousands of symbol's occurrences with > something else. We did that by implementing our own recursive > tree traversal. However, we estimate that more than 80% of the > time is spent in attempted simplifications by GiNaC (which we > know are deemed to fail). Is there any way to implement this > substitution process so as to avoid this big overhead? If you really do think that the "anonymous evaluation"(*) process in GiNaC is your bottleneck you could try arranging your "monster expressions" manually in an exvector (i.e. std::vector<GiNaC::ex>) or an epvector (i.e. std::vector<GiNaC::expair>) and then use the GiNaC::add ctor from those structures. Regards -richy. (*) Definition: We call "evaluation" a top-down tree-traversal and -rearrangement algorithm. "Anonymous" evaluation is the one you don't call explicitly, as opposed to a "named" evaluation which you call by hand, e.g. evalf(). "Simplification" is more general, it does not have to go top-down (or bottom-up, this doesn't really matter, it's just preorder vs. postorder traversal) and can thus be regarded as non-local on the expression-tree's nodes. This is my own definition of the two concepts I came up with about two years ago because the literature seems to be utterly confused and I was unable to see a consensus. This definition has turned out to be useful and well-defined. GiNaC provides evaluation, mostly. The function objects transforming trigonometric functions mentioned in the FAQ are simplifications, according to this definition. I will soon add some clarification about the two concepts in the documentation. PS: Sorry for the late answer, I am stuck in Erlangen for a while and can only view the GiNaC lists via the web-interface. Email sent to my old address will not be read until x-mas, so please reply to this new address. -- Dr. Richard B. Kreckel Framatome ANP GmbH, NGLTS <Richard.Kreckel(a)Framatome-ANP.COM> Tel: (+49)9131/189-5275, Fax: (+49)9131/189-9908 -- Prof. Roberto Bagnara Computer Science Group Department of Mathematics, University of Parma, Italy http://www.cs.unipr.it/~bagnara/ mailto:bagnara@cs.unipr.it

1 0

purrs/tests heap
by Roberto Bagnara 19 Nov '02

19 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: roberto(a)cs.unipr.it 2002-11-19 21:27:39 Modified files: tests : heap Log message: Some new recurrences. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/tests/heap.diff?cvsroot=pur…

1 0

purrs/src approximate_impl.hh
by Roberto Bagnara 19 Nov '02

19 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: roberto(a)cs.unipr.it 2002-11-19 20:57:07 Modified files: src : approximate_impl.hh Log message: Provide a better workaround for the limitations of CoStLy's pow(). Can now approximate the solutions of all the solvable recurrences in tests/recurrences and tests/heap. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/approximate_impl.hh.dif…

1 0

purrs/src approximate_impl.hh
by Roberto Bagnara 19 Nov '02

19 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: roberto(a)cs.unipr.it 2002-11-19 16:05:25 Modified files: src : approximate_impl.hh Log message: Workaround for CoStLy fixed. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/approximate_impl.hh.dif…

1 0

Re: Problem with CoStLy 0.2
by Roberto Bagnara 19 Nov '02

19 Nov '02

Markus Neher wrote: > thank you for your email. We'll check the problem and will hopefully > find a solution to it. Thanks a lot. We are indeed stuck at this and we do not know how we should proceed. We have written a temporary workaround for the problem but we are unsure whether this is the way to go. In fact, new cases are arising: for instance, squaring a negative real interval causes the function_undefined() exception to be thrown: (gdb) frame 4 #4 0x080a20e9 in ln(cinterval const&) (z=@0xbfffeba0) at /home/roberto/purrs/purrs/CoStLy/cimath.cpp:1368 1368 throw function_not_defined(); (gdb) frame 5 #5 0x080ac01b in pow(cinterval const&, cinterval const&) (bas=@0xbfffeba0, n=@0xbfffea10) at /home/roberto/purrs/purrs/CoStLy/cimath.cpp:2719 2719 return exp(n*ln(bas)); (gdb) print bas $5 = (const cinterval &) @0xbfffeba0: { real_part = { INF = -0.41421356237309914, SUP = -0.4142135623730902 }, imag_part = { INF = 0, SUP = 0 } } (gdb) print n $6 = (const cinterval &) @0xbfffea10: { real_part = { INF = 2, SUP = 2 }, imag_part = { INF = -0, SUP = -0 } } (gdb) All the best Roberto -- Prof. Roberto Bagnara Computer Science Group Department of Mathematics, University of Parma, Italy http://www.cs.unipr.it/~bagnara/ mailto:bagnara@cs.unipr.it

1 0

purrs/src approximate_impl.hh
by Roberto Bagnara 19 Nov '02

19 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: roberto(a)cs.unipr.it 2002-11-19 15:38:06 Modified files: src : approximate_impl.hh Log message: Workaround for CoStLy problem corrected and extended. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/approximate_impl.hh.dif…

1 0

purrs/src Number.inlines.hh approximate_impl.hh
by Roberto Bagnara 19 Nov '02

19 Nov '02

CVSROOT: /cvs/purrs Module name: purrs Changes by: roberto(a)cs.unipr.it 2002-11-19 15:21:57 Modified files: src : Number.inlines.hh approximate_impl.hh Log message: Work around the "pow(-1, -1) problem" of CoStLy 0.2. Temporary kludge in Interval approximate_integer(const Number&) using Number::to_double() when Number::to_int() fails. Patches: http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/Number.inlines.hh.diff?… http://www.cs.unipr.it/cgi-bin/cvsweb.cgi/purrs/src/approximate_impl.hh.dif…

1 0