
Hi all,
I'm working with your PPL library by using the Prolog interface, although my question is generic. Say we have a 3D polyhedron P defined by (for example): [A > 0, A + B < 3, C=4]
Now we project *ppl_Polyhedron_remove_space_dimensions* over {A,C} getting P': [ 3>A>0, C=4 ].
Later on we want to intersect the projected eq.system P'(that might be altered during the program) with P. The problem is that our 'C' is VAR(2) in P but VAR(1) in P' so a call to *ppl_Polyhedron_add_constraints_and_minimize* adding to P the Constraint System of P' produces an undesired output-> ... A = '$VAR'(0),B = '$VAR'(1), C = '$VAR'(2), ppl_new_Polyhedron_from_constraints(nnc,[A>0,A+B<3,C=2],Poly1), ppl_new_Polyhedron_from_Polyhedron(nnc,Poly1,nnc,Poly2), ppl_Polyhedron_remove_space_dimensions(Poly2,[B]), ppl_Polyhedron_get_minimized_constraints(Poly2,Poly2_Cons), ppl_Polyhedron_add_constraints_and_minimize(Poly1,Poly2_Cons),!, ... no
, although the original system is solvable
How can we add dimensions to P' so the intersection with P saves the original information about B? Note that B is not the first dimension (so we cannot use *concatenate_assign* nor the last one (so we cannot use *add_space_dimension*. An "ad-hoc" solution might store the deleted indexes and then make a version of the Constraint System of P' with all the VAR(X) renumbered according to that information, so a call to *add_constraints* makes sense, but I wonder if there's an alternative just using the API.
Thanks!
PS: Assume that the projection is mandatory
Mario

Mario Mendez wrote:
I'm working with your PPL library by using the Prolog interface, although my question is generic. Say we have a 3D polyhedron P defined by (for example): [A > 0, A + B < 3, C=4]
Now we project *ppl_Polyhedron_remove_space_dimensions* over {A,C} getting P': [ 3>A>0, C=4 ].
Dear Mario,
if you project over {A, C} it means that you remove the B dimension, hence the C dimension becomes the B dimension. You should thus end up with [ A>0, B=4 ].
If, instead, you project over {B, C} it means that you remove the A dimension, hence the B dimension becomes A and C becomes B. You should thus and up with [ A<3, B=4 ].
I do not see how you could obtain [ 3>A>0, C=4 ].
Later on we want to intersect the projected eq.system P'(that might be altered during the program) with P. The problem is that our 'C' is VAR(2) in P but VAR(1) in P' so a call to *ppl_Polyhedron_add_constraints_and_minimize* adding to P the Constraint System of P' produces an undesired output-> ... A = '$VAR'(0),B = '$VAR'(1), C = '$VAR'(2), ppl_new_Polyhedron_from_constraints(nnc,[A>0,A+B<3,C=2],Poly1), ppl_new_Polyhedron_from_Polyhedron(nnc,Poly1,nnc,Poly2), ppl_Polyhedron_remove_space_dimensions(Poly2,[B]), ppl_Polyhedron_get_minimized_constraints(Poly2,Poly2_Cons), ppl_Polyhedron_add_constraints_and_minimize(Poly1,Poly2_Cons),!, ... no
, although the original system is solvable
How can we add dimensions to P' so the intersection with P saves the original information about B? Note that B is not the first dimension (so we cannot use *concatenate_assign* nor the last one (so we cannot use *add_space_dimension*. An "ad-hoc" solution might store the deleted indexes and then make a version of the Constraint System of P' with all the VAR(X) renumbered according to that information, so a call to *add_constraints* makes sense, but I wonder if there's an alternative just using the API.
Thanks!
PS: Assume that the projection is mandatory
Even though I do not understand the example you give, if you want to add dimensions "in the middle" then the `map_space_dimensions' operation is may be what you want. Or maybe you did not want to remove dimensions, but simply to "forget" the information about some dimensions. Then the `add_generators' operation may be what you want. Please, come back to us with a clearer explanation of what you have, what you do and what you want, so that we can see what is your best option. All the best,
Roberto
participants (2)
-
Mario Mendez
-
Roberto Bagnara