conversion of Symbolic Ring to FriCAS Expression Integer

[mantepse]

This ticket modifies the FriCAS interface so that symbolic ring elements are always converted to FriCAS Expression Integer or Expression Complex Integer, depending on whether the complex unit appears. Doing so, we fix the failure below.

see https://ask.sagemath.org/question/48431/why-this-integral-fail-using-fricas-algorithm/

for the original bug report:

integrate(sqrt(2)*x^2 + 2*x,x, algorithm="fricas")

CC: @fchapoton

Component: interfaces: optional

Keywords: FriCAS

Author: Martin Rubey

Branch/Commit: c92aec8

Reviewer: Travis Scrimshaw

Issue created by migration from https://trac.sagemath.org/ticket/28647

[mantepse]

Branch: u/mantepse/conversion_of_symbolic_ring_to_fricas_expression_integer

[mantepse]

New commits:

4c672a0

make sagemath SR convert to FriCAS EXPR INT

[mantepse]

Commit: 4c672a0

[mantepse]

Author: Martin Rubey

[sagetrac-tmonteil]

comment:3

Possible collision with #28641 which refers to the same ask.sagemath.org question.

[mantepse]

comment:4

Replying to @sagetrac-tmonteil:

Possible collision with #28641 which refers to the same ask.sagemath.org question.

Yes, but the solutions are orthogonal to each other. In other words: after applying this branch, #28641 solves a different problem.

[fchapoton]

comment:5

some failing doctests in src/sage/functions/generalized.py and src/sage/functions/exp_integral.py

[mantepse]

comment:6

There are unfortunately more problems. In particular, fricas(I) calls FriCASConverter.pyobject(I, I) (as it should). But this now returns I instead of %i.

[mantepse]

comment:7

I'm afraid my approach won't work, because FriCAS distinguishes between Expression Integer and Expression Complex Integer.

Put differently, the complex unit I is not an element of Expression Integer, but would have to be expressed as sqrt(-1). I see the following options:

1. rewrite I as sqrt(-1),
2. always use Expression Complex Integer,
3. detect whether I appears and distinguish accordingly,
4. somehow let FriCAS decide

I tend towards option 1., but I am not sure at all.

[mantepse]

comment:8

help appreciated

[sagetrac-git]

Changed commit from 4c672a0 to 8948d38

[sagetrac-git]

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

8948d38

try to fix conversion of complex expressions to fricas

[mantepse]

comment:11

I think I have to make an exception for the conversion of symbols, because otherwise fricas.integrate(ex, x) won't work anymore.

[sagetrac-git]

Changed commit from 8948d38 to d1b4490

[sagetrac-git]

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

8601700	fix doctests
4205759	Merge branch 'u/mantepse/conversion_of_symbolic_ring_to_fricas_expression_integer' of git://trac.sagemath.org/sage into t/28647/conversion_of_symbolic_ring_to_fricas_expression_integer
d1b4490	Merge branch 'develop' into t/28647/conversion_of_symbolic_ring_to_fricas_expression_integer

[mantepse]

Changed keywords from none to FriCAS

[sagetrac-git]

Changed commit from d1b4490 to b95f2a4

[sagetrac-git]

Changed commit from 09bee26 to da22da2

[sagetrac-git]

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

da22da2

Merge branch 'develop' into t/28647/conversion_of_symbolic_ring_to_fricas_expression_integer

[mantepse]

comment:23

Unfortunately, #18036 breaks this.

[mantepse]

comment:24

(and I have no idea currently how to fix this)

[mkoeppe]

comment:25

Setting new milestone based on a cursory review of ticket status, priority, and last modification date.

[sagetrac-git]

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

f9c4dfd	Merge branch 'develop' into t/28647/conversion_of_symbolic_ring_to_fricas_expression_integer
1624c45	adapt to new parent of I

[sagetrac-git]

Changed commit from da22da2 to 1624c45

[tscrim]

comment:28

I believe you can simplify

-            if (isinstance(obj, NumberFieldElement_quadratic) and
-                obj.parent() is GaussianField()):
+            if isinstance(obj, NumberFieldElement_gaussian):

I probably would also add a test that includes SR(I) (or sqrt(-1)) to make sure that alternative version yields what you want. Other than that, LGTM.

[sagetrac-git]

Changed commit from 1624c45 to 0ead458

[sagetrac-git]

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

0ead458

simplify test for complex i

[mantepse]

comment:30

Yes, you are right! I copied these lines from the generic class InterfaceInit, so I applied your suggestion there, too.

[sagetrac-git]

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

c92aec8

remove superfluous import of I

[sagetrac-git]

Changed commit from 0ead458 to c92aec8

[tscrim]

comment:32

Thank you. LGTM.

[tscrim]

Reviewer: Travis Scrimshaw

[vbraun]

Changed branch from u/mantepse/conversion_of_symbolic_ring_to_fricas_expression_integer to c92aec8