diff --git a/Surface_mesh_approximation/doc/Surface_mesh_approximation/tofix.txt b/Surface_mesh_approximation/doc/Surface_mesh_approximation/tofix.txt
index 9f698d17c8e..c6c6655457f 100644
--- a/Surface_mesh_approximation/doc/Surface_mesh_approximation/tofix.txt
+++ b/Surface_mesh_approximation/doc/Surface_mesh_approximation/tofix.txt
@@ -1,5 +1,9 @@
 - verify that L2,1 is square root of integral of L2 error over normals.
 - very first partition must use one proxy per connected component.
 - verify that teleport can be forced even when heuristic test fails. about the test: why not simply comparing increase of error and decrease of error ?
+- use Intel TBB to parallelize where possible? eg, when it is trivially parallelizable such as measure of total error, or re-fitting. partitioning seems harder.
+
+harder:
+
 - when meshing, use discrete analog of a *constrained* 2D Delaunay triangulation ? (involving visibility)
 - add option to optimize the vertex locations of the output mesh