Improper expressions from SR(string)

[rwst]

Strings are not translated into proper Sage expressions for some functions:

sage: SR("sin(x)")
sin(x)
sage: type(_.operator())
<class 'sage.functions.trig.Function_sin'>
sage: SR("arcsin(x)")
arcsin(x)
sage: type(_.operator())
<class 'sage.functions.trig.Function_arcsin'>
sage: SR("asin(x)")
asin(x)
sage: type(_.operator())
<class 'sage.symbolic.function_factory.NewSymbolicFunction'>

Previous ticket description:

sage: integrate(((-1-3*x)^(-1/2)*(-1+3*x)^(-1/2)),x,algorithm='fricas')
-2/3*atan(1/3*(sqrt(3*x - 1)*sqrt(-3*x - 1) - 1)/x)
sage: _.subs(x==.1)
-2/3*atan(-6.51313067138982 + 3.89412863198815e-16*I)
sage: _.n()
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-29-875d0b089ddd> in <module>()
----> 1 _.n()

/home/ralf/sage/src/sage/structure/element.pyx in sage.structure.element.Element.n (build/cythonized/sage/structure/element.c:7987)()
    836             0.666666666666667
    837         """
--> 838         return self.numerical_approx(prec, digits, algorithm)
    839 
    840     N = deprecated_function_alias(13055, n)

/home/ralf/sage/src/sage/symbolic/expression.pyx in sage.symbolic.expression.Expression.numerical_approx (build/cythonized/sage/symbolic/expression.cpp:33844)()
   5566             res = x.pyobject()
   5567         else:
-> 5568             raise TypeError("cannot evaluate symbolic expression numerically")
   5569 
   5570         # Important -- the  we get might not be a valid output for numerical_approx in

TypeError: cannot evaluate symbolic expression numerically

Already the second command should have evaluated to a floating point value.

CC: @mantepse

Component: symbolics

Keywords: FriCAS

Author: Frédéric Chapoton

Branch/Commit: f9b0635

Reviewer: Martin Rubey

Issue created by migration from https://trac.sagemath.org/ticket/22525

[mantepse]

Changed keywords from none to FriCAS

[mantepse]

comment:1

A simpler way to see the bug:

sage: fricas(atan(x)).sage() == atan(x)
atan(x) == arctan(x)

[mantepse]

comment:2

I don't quite get what's happening. Is there something wrong with my "a"?

sage: [f(x)._fricas_().sage().subs(x=1.0) for f in [sin, cos, sec, csc, cot, tan, asin, acos, 
....: atan, acot, acsc, asec, arcsin, arccos, arctan, arccot, arccsc, arcsec]]
[0.841470984807897,
 0.540302305868140,
 1.85081571768093,
 1.18839510577812,
 0.642092615934331,
 1.55740772465490,
 asin(1.00000000000000),
 acos(1.00000000000000),
 atan(1.00000000000000),
 acot(1.00000000000000),
 acsc(1.00000000000000),
 asec(1.00000000000000),
 asin(1.00000000000000),
 acos(1.00000000000000),
 atan(1.00000000000000),
 acot(1.00000000000000),
 acsc(1.00000000000000),
 asec(1.00000000000000)]
sage: [f(x)._fricas_().sage().subs(x=1.0) for f in [tanh, sinh, cosh, coth, sech, csch, asinh,
....:  acosh, atanh, acoth, asech, acsch, arcsinh, arccosh, arctanh, arccoth, arcsech, arccsch
....: ]]

[0.761594155955765,
 1.17520119364380,
 1.54308063481524,
 1.31303528549933,
 0.648054273663885,
 0.850918128239322,
 asinh(1.00000000000000),
 acosh(1.00000000000000),
 atanh(1.00000000000000),
 acoth(1.00000000000000),
 asech(1.00000000000000),
 acsch(1.00000000000000),
 asinh(1.00000000000000),
 acosh(1.00000000000000),
 atanh(1.00000000000000),
 acoth(1.00000000000000),
 asech(1.00000000000000),
 acsch(1.00000000000000)]

[mantepse]

comment:3

I just learned:

sage: type(fricas(atan(x)).sage().operator())
<class 'sage.symbolic.function_factory.NewSymbolicFunction'>

sage: type(atan(x).operator())
<class 'sage.functions.trig.Function_arctan'>

sage: type(fricas(sin(x)).sage().operator())
<class 'sage.functions.trig.Function_sin'>

but I do not know where this NewSymbolicFunction thing comes from, and why it only appears for atan.

[rwst]

Changed keywords from FriCAS to none

[rwst]

comment:4

Interesting. It's the conversion via SR(string).

[rwst]

comment:5

In calculus/calculus.py:symbolic_expression_from_string() the arc versions are recognized but not the a versions.

[rwst]

comment:6

However if I put these into the global _augmented dict (as string/function pairs) the dict will be empty when doctesting via sage -tp. Can someone explain this to a Python amateur?

[mantepse]

comment:7

not sure whether the following is helpful:

sage: from sage.libs.pynac.pynac import symbol_table
sage: type(symbol_table["functions"]["sin"])
<class 'sage.functions.trig.Function_sin'>

sage: type(symbol_table["functions"]["asin"])
<class 'sage.symbolic.function_factory.NewSymbolicFunction'>

sage: type(symbol_table["functions"]["arcsin"])
<class 'sage.functions.trig.Function_arcsin'>

[mantepse]

comment:8

Sorry, mistake: in a fresh session we have

sage: from sage.libs.pynac.pynac import symbol_table
sage: type(symbol_table["functions"]["asin"])
---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-3-95eddacf035c> in <module>()
----> 1 type(symbol_table["functions"]["asin"])

KeyError: 'asin'

[mantepse]

comment:9

Is it possible that SR("asin"), SR("atan"), etc. actually are supposed to fail?

I don't know how the "conversion" machinery is supposed to work, but symbol_table["fricas"] does contain the correct translations. For example:

sage: from sage.libs.pynac.pynac import symbol_table
sage: symbol_table["fricas"]["asin"]
arcsin

sage: type(_)
<class 'sage.functions.trig.Function_arcsin'>

Also, it does seem to be a problem with the fricas interface:

sage: type(maxima("asin(x)").sage().operator())
<class 'sage.functions.trig.Function_arcsin'>
sage: type(fricas("asin(x)").sage().operator())
<class 'sage.symbolic.function_factory.NewSymbolicFunction'>

In fact, maxima_abstract.py:_sage_ contains (skipping the docstring)

    def _sage_(self):
        """
        ...
        """
        import sage.calculus.calculus as calculus
        return calculus.symbolic_expression_from_maxima_string(self.name(),
                maxima=self.parent())

which looks more plausible than the fricas.py counterpart (which I wrote...):

        from sage.symbolic.ring import SR
        s = unparsed_InputForm
        replacements = [('pi()', 'pi '),
                        ('::Symbol', ' ')]
        for old, new in replacements:
            s = s.replace(old, new)
        try:
            return SR(s)
        except TypeError:
            raise NotImplementedError("The translation of the FriCAS Expression %s to sage is not yet implemented." %s)

[mantepse]

comment:10

I'm now pretty much convinced that it is wrongdoing of the FriCAS interface. An essential step in symbolic_expression_from_maxima_string is to replace all known functions with their sage equivalent, using symbol_table.

So, we should do the following: rewrite fricas._sage_expression, which currently takes the unparsed InputForm of a FriCAS Expression Integer or OrderedCompletion Expression Integer object as follows:

• input should be the InputForm, instead of the unparsed InputForm, which is essentially a nested list of the form (function arg1 arg2 ...) (lisp syntax)
• replace function by its sage equivalent, and recurse.

I am unlikely to do that.

[mantepse]

Changed keywords from none to FriCAS

[rwst]

comment:12

I think your analysis didn't consider the fact that return SR(s) in fricas.py calls symbolic_expression_from_string and therefore symbolic_expression_from_string would be the code to fix.

[mantepse]

comment:13

Yes, but it shouldn't. symbolic_expression_from_string hopes for expressions in terms of the dictionary symbol_table["functions"], not symbol_table["fricas"]. I didn't know about symbol_table when I coded the interface.

[fchapoton]

comment:14

Here is a tentative of enhancement.

---

New commits:

7775304

trac 22525 better conversion from fricas expressions

[fchapoton]

Branch: public/22525

[fchapoton]

Commit: 7775304

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

f9b0635

add a test

[sagetrac-git]

Changed commit from 7775304 to f9b0635

[mantepse]

comment:17

This looks great! I added a test checking that we can use it to compute numerical values, which is probably the most interesting part of the fix.

Unfortunately, I have to recompile almost from scratch. I'll set it to positive review afterwards.

[mantepse]

Author: Frédéric Chapoton

[mantepse]

Reviewer: Martin Rubey

[mantepse]

comment:19

(most of the #optional - fricas tests will fail here, because I made a mistake in #22501, which is fixed in #23782)

[mantepse]

comment:20

Looks good!

[vbraun]

Changed branch from public/22525 to f9b0635