Dear Lorenzo and Kostas, Adding a context_check() in convex_index() method of the mim object (line 74 of getfm_mesh_im.h) solves the problem. I committed this fix. Lorenzo, thank you very much for the report of this problem.
Best regards, Yves De: "Konstantinos Poulios" <logar...@googlemail.com> À: "Lorenzo Ferro" <lorenzo.i...@gmail.com> Cc: "getfem-users" <getfem-users@nongnu.org> Envoyé: Jeudi 21 Juillet 2022 15:36:26 Objet: Re: gf.asm_generic mistake immediately after a mesh change interesting, if you just add a print(mim) before calling the second assembly, then it works correctly. This is a bug, somewhere we have forgotten to update mesh_im when mesh is changed. On Thu, Jul 21, 2022 at 11:02 AM Lorenzo Ferro < [ mailto:lorenzo.i...@gmail.com | lorenzo.i...@gmail.com ] > wrote: Hello Konstantinos, I've realized for this troubleshooting a simplified example, where I generate a 1D mesh corresponding to the radius of a circular section. Then I've added the mim and I've calculated the corresponding area both analytically and numerically with asm_generic considering the axisymmetric configuration. I removed convexes and points from the mesh with the del_convex and del_point methods and then I generated other elements on the same mesh object. The numerical area calculation turns out to be wrong. ####### CODE START ####### import getfem as gf import numpy as np RoundRadius=0.01 NumberOfElements = 5 nodes = np.arange(0, RoundRadius+RoundRadius/NumberOfElements, RoundRadius/NumberOfElements) mesh = gf.Mesh("cartesian", nodes) im_degree = 3 mim = gf.MeshIm(mesh, im_degree) RoundSection = np.pi*R**2 RoundSection_asm = gf.asm_generic(mim, 0, '2*pi*X' , -1) print ("STARTING AREA CALCULATION") print ("analitical section =", RoundSection, "VS numerical section = ", RoundSection_asm) # Change completely the mesh mesh.del_convex(mesh.cvid()) mesh.del_point(mesh.pid()) NumberOfElements = 10 nodes = np.arange(0, RoundRadius+RoundRadius/NumberOfElements, RoundRadius/NumberOfElements) mesh.add_point([nodes]) for x in range(NumberOfElements): mesh.add_convex(gf.GeoTrans('GT_QK(1,1)'), [[nodes[x], nodes[x+1]]]) RoundSection_asm = gf.asm_generic(mim, 0, '2*pi*X' , -1) print ("AREA CALCULATION AFTER MESH CHANGE") print ("analytical section =", RoundSection, "VS numerical section = ", RoundSection_asm) # Let's try to apply another mesh method mesh.set_pts([nodes]) RoundSection_asm = gf.asm_generic(mim, 0, '2*pi*X' , -1) print ("AREA CALCULATION AFTER MESH.SET_PTS METHOD") print ("analytical section =", RoundSection, "VS numerical section = ", RoundSection_asm) ####### CODE FINISH####### Thank you, Lorenzo Il giorno mer 20 lug 2022 alle ore 11:43 Konstantinos Poulios < [ mailto:logar...@googlemail.com | logar...@googlemail.com ] > ha scritto: BQ_BEGIN Dear Lorenzo, It would be very useful if you could provide a minimal test case that demonstrates this issue. Strange that md.set_time() helps updating the model correctly. Best regards Kostas On Mon, Jul 18, 2022 at 4:36 PM Lorenzo Ferro < [ mailto:lorenzo.i...@gmail.com | lorenzo.i...@gmail.com ] > wrote: BQ_BEGIN Dear All, I'm experimenting with mesh changes (a sequence like del_convex -> del_point -> add_point -> add_convex -> set_region) and relevant MeshFEM automatic adaptation with the Python interface. I found out that as soon as I update mesh and regions, if after that I try to run a test assembly method for example just to calculate the total area of the mesh elements (via gf.asm_generic) I achieve a completely wrong result, but, after just running a Model method like "md.set_time(0.)" and then running immediately "gf.asm_generic" then I achieve the correct result! Is this a bug? or is there some logic behind this behaviour? Thank you, Lorenzo BQ_END BQ_END
