login  home  contents  what's new  discussion  bug reports     help  links  subscribe  changes  refresh  edit

Edit detail for #191 exquo and therefore gcd cannot handle UP(x, EXPR INT) revision 5 of 6

1 2 3 4 5 6
Editor: test1
Time: 2017/02/23 19:30:27 GMT+0
Note:

added:

In fact, to ensure that algebraic relations are respected inside expressions one has to normalize
all involved expressions:
\begin{axiom}
uP := UP(x, EXPR INT)
p1 := (x-2^a)::uP
p2 := simplify((x-2^a)*(x+2^a))::uP
cl1 := coefficients(p1)
cl2 := coefficients(p2)
ncl := normalize(concat(cl1, cl2))
ncl1 := first(ncl, #cl1)
ncl2 := rest(ncl, #cl1)
np1 := reduce(_+, [monomial(c,degree(m))$uP for m in monomials(p1) for c in ncl1])
np2 := reduce(_+, [monomial(c,degree(m))$uP for m in monomials(p2) for c in ncl2])
gcd(np1, np2)
\end{axiom}
Not nice, but in general this is what is needed (below are simpler attempts which sometimes
work, but may fail).


changed:
-in terms of independent kernels.
in terms of independent kernels.  In more general contexts (in
particular involving 'abs') this is undecidable, but for
elementary functions 'normalize' works.  More precisely,
currently 'normalize' ignores possible dependencies
between algebraic quantities like roots.  And to simplification
of transcendental constants depends on Schanuel conjecture
(if Schanuel conjecture is false we may miss some dependencies).

changed:
-I'm afraid that this cannot be fixed easily, since there is no
-general mechanism to determine whether an expression is zero or not,
-which is needed in exquo. The instance of exquo involved is the
-one in SUP.
-
-From billpage Wed Jul 13 00:47:27 -0500 2005
-From: billpage
-Date: Wed, 13 Jul 2005 00:47:27 -0500
-Subject: $2^{{a2}}$ vs $4^a$
-Message-ID: <20050713004727-0500@page.axiom-developer.org>
-
-The problem seems to be that Axiom does not always treat $2^{a2}$
-the same as $4^a$.
This shows that using normalize _separately_ on polynomials does not help, we need
to normalize list of coefficients as a whole.

Few examples showing that without normalization FriCAS does not always treat $2^{a2}$
the same as $4^a$:

removed:
-Comments from wyscc:
-
-Martin wrote:
-> I'm afraid that this cannot be fixed easily, since there is no
-> general mechanism to determine whether an expression is zero or not,
-> which is needed in exquo.
-
-Your analysis seems to be the correct diagnosis. The problem has nothing to do with 'gcd' or 'exquo', but with the fact that simplification is an art as there is no canonical form for expressions and hence no way to test equality (which is in general *different* from testing zero, a special case needed for 'exquo'). More frequent use of simplification will help but will not eliminate the problem. Here, it is because the expressions $4^a$ and $2^{2a}$ are not handled by an automatic simplification in 'UP(x, R)' (that is, not pushed down to the level of 'R').
-
-From billpage Tue Jul 12 22:06:08 -0500 2005
-From: billpage
-Date: Tue, 12 Jul 2005 22:06:08 -0500
-Subject: property change
-Message-ID: <20050712220608-0500@page.axiom-developer.org>
-
-Category:  => Axiom Compiler 
-Severity:  => serious 
-Status:  => open 
-
-
-
-
-
-From kratt6 Thu Dec 20 01:07:34 -0800 2007
-From: kratt6
-Date: Thu, 20 Dec 2007 01:07:34 -0800
-Subject: 
-Message-ID: <20071220010734-0800@axiom-wiki.newsynthesis.org>
-
-Category: Axiom Compiler => Axiom Library 
-
-
-From test1 Tue Apr 21 17:38:07 +0000 2015
-From: test1
-Date: Tue, 21 Apr 2015 17:38:07 +0000
-Subject: 
-Message-ID: <20150421173807+0000@axiom-wiki.newsynthesis.org>
-
-Category: Axiom Library => Axiom Documentation 
-Severity: serious => normal 
-

Submitted by : (unknown) at: 2007-11-17T22:05:44-08:00 (16 years ago)
Name :
Axiom Version :
Category : Severity : Status :
Optional subject :  
Optional comment :

(new) exquo and therefore gcd cannot handle UP(x, EXPR INT) --Bill Page, Mon, 11 Jul 2005 15:56:25 -0500 reply
Update of bug #10530 (project axiom):

Status: None => transferred

fricas
gcd((x-2^a)::UP(x, EXPR INT), simplify((x-2^a)*(x+2^a))::UP(x, EXPR INT))

\label{eq1}1(1)
Type: UnivariatePolynomial?(x,Expression(Integer))

Gives 1, while the correct answer should be x-2^a, as given by

fricas
gcd((x-2^a)::UP(x, EXPR INT),((x-2^a)*(x+2^a))::UP(x, EXPR INT))

\label{eq2}x -{{2}^{a}}(2)
Type: UnivariatePolynomial?(x,Expression(Integer))

In fact, to ensure that algebraic relations are respected inside expressions one has to normalize all involved expressions:

fricas
uP := UP(x, EXPR INT)

\label{eq3}\hbox{\axiomType{UnivariatePolynomial}\ } (x , \hbox{\axiomType{Expression}\ } (\hbox{\axiomType{Integer}\ }))(3)
Type: Type
fricas
p1 := (x-2^a)::uP

\label{eq4}x -{{2}^{a}}(4)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
p2 := simplify((x-2^a)*(x+2^a))::uP

\label{eq5}{{x}^{2}}-{{4}^{a}}(5)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
cl1 := coefficients(p1)

\label{eq6}\left[ 1, \: -{{2}^{a}}\right](6)
Type: List(Expression(Integer))
fricas
cl2 := coefficients(p2)

\label{eq7}\left[ 1, \: -{{4}^{a}}\right](7)
Type: List(Expression(Integer))
fricas
ncl := normalize(concat(cl1, cl2))

\label{eq8}\left[ 1, \: -{{e}^{a \ {\log \left({2}\right)}}}, \: 1, \: -{{{e}^{a \ {\log \left({2}\right)}}}^{2}}\right](8)
Type: List(Expression(Integer))
fricas
ncl1 := first(ncl, #cl1)

\label{eq9}\left[ 1, \: -{{e}^{a \ {\log \left({2}\right)}}}\right](9)
Type: List(Expression(Integer))
fricas
ncl2 := rest(ncl, #cl1)

\label{eq10}\left[ 1, \: -{{{e}^{a \ {\log \left({2}\right)}}}^{2}}\right](10)
Type: List(Expression(Integer))
fricas
np1 := reduce(_+, [monomial(c,degree(m))$uP for m in monomials(p1) for c in ncl1])

\label{eq11}x -{{e}^{a \ {\log \left({2}\right)}}}(11)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
np2 := reduce(_+, [monomial(c,degree(m))$uP for m in monomials(p2) for c in ncl2])

\label{eq12}{{x}^{2}}-{{{e}^{a \ {\log \left({2}\right)}}}^{2}}(12)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
gcd(np1, np2)

\label{eq13}x -{{e}^{a \ {\log \left({2}\right)}}}(13)
Type: UnivariatePolynomial?(x,Expression(Integer))

Not nice, but in general this is what is needed (below are simpler attempts which sometimes work, but may fail).

Internal Cause

In EXPR INT, 4^a and 2^{(2a)} are treated as two variables without relations in EXPR INT. But

fricas
simplify((x-2^a)*(x+2^a))::UP(x, EXPR INT)

\label{eq14}{{x}^{2}}-{{4}^{a}}(14)
Type: UnivariatePolynomial?(x,Expression(Integer))

Therefore exquo in:

  exquo(simplify((x-2^a)*(x+2^a))::UP(x,EXPR INT),(x-2^a)::UP(x,EXPR INT))

fails. In details, it calls exquo$UP(x,EXPR INT). This implements exact division of polynomials p1 by p2 as usual. After each subtraction - done via fmecg$UP - the result is again stored in p1. exquo terminates when p1 is the empty list - note that UPs? are stored as lists of pairs (degree, coefficient) - or the degree of p2 is larger than p1. In the latter case, exquo fails.

Thus, in our case, at one point p1 is 4^a-2^{(2a)}, which is zero mathematically, but FriCAS? does not know it. In particular, p1 is not the empty list, but rather a constant polynomial...

To get correct result we need to express coefficients of p1 and p2 in terms of independent kernels. In more general contexts (in particular involving abs) this is undecidable, but for elementary functions normalize works. More precisely, currently normalize ignores possible dependencies between algebraic quantities like roots. And to simplification of transcendental constants depends on Schanuel conjecture (if Schanuel conjecture is false we may miss some dependencies).

Wed 09/29/2004 at 16:02, comment #1:

I should have added:

fricas
exquo(normalize(simplify(((A-2^a)*(A+2^a)))::EXPR INT),normalize((A-2^a)::EXPR INT))

\label{eq15}{{{e}^{a \ {\log \left({4}\right)}}}-{{A}^{2}}}\over{{{e}^{a \ {\log \left({2}\right)}}}- A}(15)
Type: Union(Expression(Integer),...)
fricas
exquo(simplify((A-2^a)*(A+2^a))::UP(A,EXPR INT),(A-2^a)::UP(A,EXPR INT))

\label{eq16}\verb#"failed"#(16)
Type: Union("failed",...)

This shows that using normalize separately on polynomials does not help, we need to normalize list of coefficients as a whole.

Few examples showing that without normalization FriCAS? does not always treat 2^{a2} the same as 4^a:

fricas
dom:=UP('x,EXPR INT)

\label{eq17}\hbox{\axiomType{UnivariatePolynomial}\ } (x , \hbox{\axiomType{Expression}\ } (\hbox{\axiomType{Integer}\ }))(17)
Type: Type
fricas
p:dom:=x-2^a

\label{eq18}x -{{2}^{a}}(18)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
q:=(x-2^a)*(x+2^a)

\label{eq19}-{{{2}^{a}}^{2}}+{{x}^{2}}(19)
Type: Expression(Integer)
fricas
qq:= simplify(q)

\label{eq20}-{{4}^{a}}+{{x}^{2}}(20)
Type: Expression(Integer)
fricas
r:= q::dom

\label{eq21}{{x}^{2}}-{{{2}^{a}}^{2}}(21)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
rr:= qq::dom

\label{eq22}{{x}^{2}}-{{4}^{a}}(22)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
gcd(p,q)

\label{eq23}x -{{2}^{a}}(23)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
exquo(q,p)

\label{eq24}{{2}^{a}}+ x(24)
Type: Union(Expression(Integer),...)
fricas
gcd(p,r)

\label{eq25}x -{{2}^{a}}(25)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
gcd(p,qq)

\label{eq26}1(26)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
gcd(p,rr)

\label{eq27}1(27)
Type: UnivariatePolynomial?(x,Expression(Integer))
fricas
q - qq

\label{eq28}{{4}^{a}}-{{{2}^{a}}^{2}}(28)
Type: Expression(Integer)
fricas
simplify q - qq

\label{eq29}0(29)
Type: Expression(Integer)
fricas
simplify ((r - rr)::EXPR INT)

\label{eq30}0(30)
Type: Expression(Integer)
fricas
t:Boolean:=(r = rr)

\label{eq31} \mbox{\rm false} (31)
Type: Boolean