OFF-EP-0010 #61
OFF-EP-0010 #61
Conversation
lilyminium
requested review from
jchodera,
davidlmobley,
mattwthompson and
j-wags
|
|
||
| ## Motivation and Scope | ||
|
|
||
| A ProperTorsion is defined between a connected quartet of atoms i-j-k-l. The dihedral angle can be calculated using input vectors in two directions: ($r_{ij}$, $r_{jk}$, $r_{kl}$) or ($r_{ij}$, $r_{kl}$, $r_{kl}$), where $r_{ij} = x_{j} - x{i}$. It is currently unclear which standard to follow. While both directions yield the same result for symmetric torsions where the phase is 0 or pi, the choice of direction affects the sign for asymmetric torsions. |
There was a problem hiding this comment.
I'm not sure what you mean by "The dihedral angle can be calculated using input vectors in two directions:" where you give triples of vectors (a, b, c). Can you give an explicit mathematical expression for the dihedral angle phi in terms of these vectors (a,b,c)?
Also, the second ($r_{ij}$, $r_{kl}$, $r_{kl}$) contains two repeated vectors.
There was a problem hiding this comment.
I've changed this section to a more qualitative description in words, and left the mathematical definition for the detailed description / specification itself.
|
|
||
| A ProperTorsion is defined between a connected quartet of atoms i-j-k-l. The dihedral angle can be calculated using input vectors in two directions: ($r_{ij}$, $r_{jk}$, $r_{kl}$) or ($r_{ij}$, $r_{kl}$, $r_{kl}$), where $r_{ij} = x_{j} - x{i}$. It is currently unclear which standard to follow. While both directions yield the same result for symmetric torsions where the phase is 0 or pi, the choice of direction affects the sign for asymmetric torsions. | ||
|
|
||
| In addition, ProperTorsions are matched symmetrically, even if the SMIRKS pattern is written asymmetrically; no order is guaranteed in applying the torsion parameter. While atom order does not affect the angle calculation in many implementations, it is important to highlight this potentially undefined behaviour in case some software does take atom order into account. |
There was a problem hiding this comment.
In addition, ProperTorsions are matched symmetrically, even if the SMIRKS pattern is written asymmetrically;
Since this is an implementation detail, it may be more useful to say something like "A SMIRKS patterns that can match a particular bonded quartet in either i-j-k-l or l-k-j-i order is ambiguous, and the specification cannot guarantee the match will be performed in any predetermined or deterministic order, potentially leading to undesired results."
There was a problem hiding this comment.
Thanks John, that's much clearer -- I've changed the wording accordingly!
|
I'm removing myself as a reviewer as I'm not a member of the committee. I glanced through the linked notes from the September meeting and don't think there's an explicit call-out for what to do when a member of the committee submitting their own proposal. (Still, I figure that it's reasonable to assume an author approves of their own work and 3/3 counts as 4/4.) In passing my, feedback on the "auto" section is positive and I'm confused about the directionality section for reasons other than this change (namely that I haven't thought through it). |
There was a problem hiding this comment.
Awesome, thanks @lilyminium! I read through this and rendered locally and everything looks good. Great writeup.
Co-authored-by: Jeff Wagner <[email protected]>
Co-authored-by: Jeff Wagner <[email protected]>
Co-authored-by: Jeff Wagner <[email protected]>
Co-authored-by: Jeff Wagner <[email protected]>
| > defined by the four atoms of the torsion `i-j-k-l`. | ||
| > Where the vector ``r_ij`` is defined as the vector from atom `j` to atom `i`: | ||
| > ``` | ||
| > r_ij = x_i - x_j |
There was a problem hiding this comment.
This definition is backwards, isn't it? This vector points from j to i
There was a problem hiding this comment.
Yes -- I chose to call it r_ij instead of r_ji for easier comparison to OpenMM's implementation, which also names vectors like that, but happy to switch to r_ji instead if that reads better.
There was a problem hiding this comment.
Which OpenMM docs are you referring to? This is not in the OpenMM User Guide.
There was a problem hiding this comment.
Let's use standard vector notation r_ij = x_j - x_i
| > r_ij = x_i - x_j | ||
| > ``` | ||
| > the angle ``theta`` should be calculated using the input vectors ``r_ij``, ``r_kj``, and ``r_kl``. | ||
| > The directionality or sign of the angle is determined by comparing the `r_ij` vector to the `u_jkl` plane. If the angle is acute, the sign is positive; if obtuse, the sign is negative. |
There was a problem hiding this comment.
This statement is confusing to me. Can we drop it and let the math speak for itself? Or add a figure?
| > defined by the four atoms of the torsion `i-j-k-l`. | ||
| > Where the vector ``r_ij`` is defined as the vector from atom `j` to atom `i`: | ||
| > ``` | ||
| > r_ij = x_i - x_j |
There was a problem hiding this comment.
Which OpenMM docs are you referring to? This is not in the OpenMM User Guide.
| > defined by the four atoms of the torsion `i-j-k-l`. | ||
| > Where the vector ``r_ij`` is defined as the vector from atom `j` to atom `i`: | ||
| > ``` | ||
| > r_ij = x_i - x_j |
There was a problem hiding this comment.
Let's use standard vector notation r_ij = x_j - x_i
|
Changed the notation -- thank you for the reviews, everyone! |
Fixes #60
Resolves #59
This PR:
- clarifies and defines
idivf- defines how the dihedral should be calculated
- adds a warning flagging that ProperTorsion SMIRKS are symmetric across the central bond
As John pointed out, a dihedral angle from i-j-k-l should evaluate the same as from l-k-j-i, at least using our definition. I'm 99% sure that this should always be the case, and also briefly checked was true at least for OpenMM -- but thought it may be worth including the warning anyway just in case there's engines where for any reason this is not the case.