cartesian_to_irc produces 180 degree dihedrals with random sign #52
Description
Hello, I hope I can explain this right, but I observed the issue as described in the title. Take ethane as an example. it produces the following coordinates:
New internal
(redundant) coordinates: 28
*************************************
7 bonds / Å
*************************************
( 0, 1) 1.4613
( 0, 2) 1.0941
( 1, 3) 1.0941
( 0, 4) 1.0941
( 1, 5) 1.0941
( 0, 6) 1.0941
( 1, 7) 1.0941
*************************************
12 angles / °:
*************************************
( 1, 0, 2) 111.84
( 0, 1, 3) 111.84
( 1, 0, 4) 111.84
( 2, 0, 4) 107.00
( 0, 1, 5) 111.84
( 3, 1, 5) 107.00
( 1, 0, 6) 111.84
( 2, 0, 6) 107.00
( 4, 0, 6) 107.00
( 0, 1, 7) 111.84
( 3, 1, 7) 107.00
( 5, 1, 7) 107.00
*************************************
9 dihedrals / °
*************************************
( 2, 0, 1, 3) -60.00
( 3, 1, 0, 4) 60.00
( 2, 0, 1, 5) 180.00
( 4, 0, 1, 5) -60.00
( 3, 1, 0, 6) -180.00
( 5, 1, 0, 6) 60.00
( 2, 0, 1, 7) 60.00
( 4, 0, 1, 7) -180.00
( 6, 0, 1, 7) -60.00
*************************************
If you run cartesian_to_iirc successive times the outcome of the 180 dihedral angles flip sign. it may give
( 2, 0, 1, 5) 180.00
( 3, 1, 0, 6) -180.00
( 4, 0, 1, 7) -180.00
or sometimes
( 2, 0, 1, 5) -180.00
( 3, 1, 0, 6) 180.00
( 4, 0, 1, 7) 180.00
but only for the 180 degree case.
As you can imagine this can pay havoc with geometry optimization because dihedrals that have flipped 360 degrees during a geometry optimization step can give rogue values for dq,
one can get dq = 2 pi when it should have been dq = 0, or something very small.
Yesterday, I put a plaster in my own code to detect this flip, so I could work around the issue. Of course, it would be nice to see what the issue in libirc is.
Thus far, I've made the following change; It works in all my tests so far, but not knowing your code base very well I don't if there could be unwanted consequences
In constsnts.h I added
// dihedrals near 2 pi
constexpr double dihedral_near_180_tol{1e-6};
and in connectivity.h I changed the following lines, around 440 thereabouts
// Compute dihedral angle in radians (in the interval [-pi,pi])
double angle{std::atan2(y, x)};
if (std::fabs(std::fabs(angle) - tools::constants::pi) < tools::constants::dihedral_near_180_tol)
angle = std::fabs(angle);
return angle;
What do you think ?
Best regards.
oh ad thanks for the library, great stuff.