def get_stuff_for_the_local_version(cpa_space): if cpa_space.tess not in ['I','II']: raise ValueError(cpa_space.tess) # compute_maps = cpa_space.dim_domain > 1 or cpa_space.tess == 'I' compute_maps = cpa_space.dim_domain==1 or cpa_space.tess == 'I' if not compute_maps: return None nC = cpa_space.nC nHomoCoo = cpa_space.nHomoCoo lengthAvee = cpa_space.lengthAvee dim_domain = cpa_space.dim_domain dim_range = cpa_space.dim_range b = Bunch() cells_verts_homo_coo = cpa_space.tessellation.cells_verts_homo_coo if compute_maps: X = np.zeros((nC,lengthAvee,lengthAvee)) Xinv = np.zeros_like(X) if dim_domain == 1: for (x,xinv,(vrt0,vrt1)) in zip(X,Xinv,cells_verts_homo_coo): x[0,:2]=vrt0 x[1,:2]=vrt1 xinv[:]=inv(x) elif dim_domain == 2: for (x,xinv,(vrt0,vrt1,vrt2)) in zip(X,Xinv,cells_verts_homo_coo): x[0,:3]=x[1,3:]=vrt0 x[2,:3]=x[3,3:]=vrt1 x[4,:3]=x[5,3:]=vrt2 xinv[:]=inv(x) elif dim_domain == 3: for (x,xinv,(vrt0,vrt1,vrt2,vrt3)) in zip(X,Xinv,cells_verts_homo_coo): x[0,:4]=x[1,4:8]=x[2,8:]=vrt0 x[3,:4]=x[4,4:8]=x[5,8:]=vrt1 x[6,:4]=x[7,4:8]=x[8,8:]=vrt2 x[9,:4]=x[10,4:8]=x[11,8:]=vrt3 xinv[:]=inv(x) else: raise NotImplementedError(dim_domain) vert_tess = [] vert_tess_one_cell = [] ind_into_vert_tess = np.zeros((nC,nHomoCoo),np.int) for c,cell_verts in enumerate(cells_verts_homo_coo): for j,v in enumerate(cell_verts): t = tuple(v.tolist()) if t not in vert_tess: vert_tess.append(t) # c is the cell index # j is the index of this vertex within that cell vert_tess_one_cell.append((c,j)) ind_into_vert_tess[c,j]=vert_tess.index(t) vert_tess = np.asarray(vert_tess) vert_tess_one_cell = np.asarray(vert_tess_one_cell) b.vert_tess = vert_tess b.ind_into_vert_tess = ind_into_vert_tess b.Xinv = Xinv b.X = X """ Build a sparse matrix H such that Avees = H times velTess The values of H, which is sparse, are dictated by vertTess. H.shape = (lengthAvee*nC,len(vert_tess)*dim_range) """ H = np.zeros((lengthAvee*nC,len(vert_tess)*dim_range)) for c in range(nC): ind = ind_into_vert_tess[c] ind_all_coo = np.zeros((len(ind),dim_range),np.int) for coo in range(dim_range): ind_all_coo[:,coo]=ind*dim_range+coo H[c*lengthAvee:(c+1)*lengthAvee,ind_all_coo.ravel()]=Xinv[c] # """ Build a sparse matrix H such that velTess = G times Avees G.shape = (len(vert_tess)*dim_range,lengthAvee*nC) """ G = np.zeros((len(vert_tess)*dim_range,lengthAvee*nC)) for i in range(vert_tess.shape[0]): # c is the cell index # j is the index of this vertex within this cell c,j = vert_tess_one_cell[i] for coo in range(dim_range): G[i*dim_range+coo,lengthAvee*c:lengthAvee*(c+1)]=X[c][j*dim_range+coo] # ipshell('hi') H = lil_matrix(H) G = lil_matrix(G) b._mat_velTess2Avees = H b._mat_Avees2velTess = G # if 1: def mv1(v): return H.dot(v) def mv2(v): return G.dot(v) def rmv1(v): return H.T.dot(v) def rmv2(v): return G.T.dot(v) def mm1(V): return H.dot(V) def mm2(V): return G.dot(V) _H = ssl.LinearOperator(H.shape,matvec=mv1, rmatvec=rmv1, matmat=mm1) _G = ssl.LinearOperator(lil_matrix(G).shape,matvec=mv2, rmatvec=rmv2, matmat=mm2) b.linop_velTess2Avees = _H b.linop_Avees2velTess = _G return b