`solve` returns cell array of structs, which differs from SMT

[MaxGaspa]

Using latest version of Octave and symbolic package on Windows 10

>> pkg load symbolic
>> syms x y z a
Symbolic pkg v3.2.2: Python communication link active, SymPy v1.10.1.
>> eqna = x + y + z == 0
eqna = (sym) x + y + z = 0
>> eqnb = 2*x*y-4==0
eqnb = (sym) 2*x*y - 4 = 0
>> S = solve(eqna,eqnb)
S =
{
  [1,1] =

    scalar structure containing the fields:

      x =

        <class sym>

      y =

        <class sym>


  [1,2] =

    scalar structure containing the fields:

      x =

        <class sym>

      y =

        <class sym>


}


>> S.x
error: cell cannot be indexed with .
>> S.y
error: cell cannot be indexed with .
>>

But with Matlab 2024b I can get the corret behavior

>> syms x y z a
>> eqna = x + y + z == 0
 
eqna =
 
x + y + z == 0
 
>> eqnb = 2*x*y-4==0
 
eqnb =
 
2*x*y - 4 == 0
 
>> S = solve(eqna,eqnb)

S = 

  struct with fields:

    x: [2×1 sym]
    y: [2×1 sym]

>> S.x
 
ans =
 
  (z^2 - 8)^(1/2)/2 - z/2
- z/2 - (z^2 - 8)^(1/2)/2
 
>> S.y
 
ans =
 
- z/2 - (z^2 - 8)^(1/2)/2
  (z^2 - 8)^(1/2)/2 - z/2
 
>> 

Thank you for your attention

Regards

Max

[MaxGaspa]

For cbm755

using

S = solve(eqna,eqnb,[x y])

does not fix th issue. It's still exactly the same.

The funny thing is that even the example listed in the Symbolic Package manual is not working

In the manual

syms x y
d = solve(x^2 == 4, x + y == 1);

% the first solution
d{1}.x
  ⇒ (sym) -2
d{1}.y
  ⇒ (sym) 3

% the second solution
d{2}.x
  ⇒ (sym) 2
d{2}.y
  ⇒ (sym) -1

in real octave

>> syms x y
>> d = solve(x^2 == 4, x + y == 1);
>> d.x
error: cell cannot be indexed with .
>> d.y
error: cell cannot be indexed with .
>> d
d =
{
  [1,1] =

    scalar structure containing the fields:

      x =

        <class sym>

      y =

        <class sym>


  [1,2] =

    scalar structure containing the fields:

      x =

        <class sym>

      y =

        <class sym>


}

>>

[cbm755]

Its possible this is a difference between Octave and Matlab, and nothing to do with Symbolic.

For example,

>>> C = {struct('x', 1, 'y', 2), struct('x', 3, 'y', 4)}
C =
{
  [1,1] =
    scalar structure containing the fields:
      x =     1
      y =     2
  [1,2] =
    scalar structure containing the fields:
      x =     3
      y =     4
}

(note no symbolic stuff here). Now try:

C.x
error: cell cannot be indexed with .

Perhaps you can try that example on Matlab and include the output here?

BTW, please try to format your posts: the "triple backtic" will make the computer output more readable.

[cbm755]

The funny thing is that even the example listed in the Symbolic Package manual is not working

Hmmm, as far as I can tell, its working fine. You need to type d{1}.x. Or S{1}.x in your original example.

[cbm755]

Perhaps Matlab's SMT is making this sort of struct:

octave:58> C = struct('x', [1; 2], 'y', [3; 4])
C =
  scalar structure containing the fields:
    x =
                1
                2
    y =
                3
                4

From that, its possible to index as you tried:

octave:59> C.x
ans =
            1
            2

octave:60> C.y
ans =
            3
            4

I think I was aware of this difference before... but I can't remember if there was a conscious decision to do something different from SMT.

[cbm755]

The test suite has this:

%!test
%! % Multiple solutions, multiple variables
%! d = solve(x^2 == 4, x + y == 1);
%! assert (length(d) == 2);
%! % FIXME: SMT has d.x gives vector and d.y giving vector, what is
%! % more intuitive?
%! for i = 1:2
%!   assert (isequal (d{i}.x + d{i}.y, 1))
%!   assert (isequal ((d{i}.x)^2, 4))
%! end

[MaxGaspa]

The funny thing is that even the example listed in the Symbolic Package manual is not working

Hmmm, as far as I can tell, its working fine. You need to type d{1}.x. Or S{1}.x in your original example.

d = solve(x^2 == 4, x + y == 1);
d{1}.x
ans = (sym) -2

And for the first example

S{1}.x
ans = (sym)

       ________
      /  2
z   \/  z  - 8

---

2        2

And in Matlab

S{1}.x
Brace indexing is not supported for variables of this type.

Ah! So curly brackets are needed! For which reason Octave is using a syntax completely different from Matlab? Is it expected that Octave is Matlab compatible? Anyway if it is not a Symbolic issue but an Octave incompatibility with Matlab I think the issue should be closed here.

[MaxGaspa]

Perhaps you can try that example on Matlab and include the output here?

The Matlab output is included in the fisrt message

BTW, please try to format your posts: the "triple backtic" will make the computer output more readable.

I apologiza but i don't know what's "triple backtic"...

Max

[cbm755]

I've been editting your messages, you can edit one to see what a triple backtic is. For my keyboard, its the key left of 1, pressed three times.

There are various things being conflated in this issue report:

1. The documentation (of Symbolic) clearly shows the use of d{1}.x (with curly braces). Thus in some sense everything is functioning as designed.
2. I confused things by thinking Matlab and Octave behaved differently about cell arrays with structs inside; this does not appear to be so.
3. The real issue is that Symbolic and SMT return two different things. It would be theoretically possible to fix this in Symbolic. See my comment
   and my other comment