From abca9e39eacb7ef2c3aadf6d4163aacec9159f9f Mon Sep 17 00:00:00 2001 From: Nicolas Boumal Date: Fri, 5 Jul 2024 16:13:09 +0200 Subject: [PATCH] refined comments --- tests/test_penrose.m | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tests/test_penrose.m b/tests/test_penrose.m index 56db16f8..067f3adf 100644 --- a/tests/test_penrose.m +++ b/tests/test_penrose.m @@ -11,12 +11,14 @@ problem.egrad = @(X) A'*(A*X*B - C)*B'; % grad f(X) in R^(nxn) % Wrong. -% This code treats Xdot as if it were in the embedding space. +% This code treats the second input of ehess as if it were in +% the embedding space. Optimization algorithms won't be amused. problem.ehess = @(X, Xdot) A'*(A*Xdot*B)*B'; % Correct. -% This code transforms Omg to X*Omg, which is the direction that is actually -% encoded by Omg, and which we usually refer to as Xdot. +% This code notes that the second input to ehess is a representation of +% a tangent vector. It transforms Omg to X*Omg, which is the direction +% that is actually encoded by Omg, and which we often call Xdot. % Equivalently, we can call SOn.tangent2ambient(X, Omg) instead of X*Omg. problem.ehess = @(X, Omg) A'*(A*X*Omg*B)*B';