Changeset 66872e

Timestamp:

May 18, 2017, 7:14:56 PM (8 years ago)

Author:

Frederik Heber <frederik.heber@…>

Branches:

ForceAnnealing_goodresults, ForceAnnealing_tocheck

Children:

396aab

Parents:

b94e74

git-author:

Frederik Heber <heber@…> (03/01/17 20:34:07)

git-committer:

Frederik Heber <frederik.heber@…> (05/18/17 19:14:56)

Message:

DOCU: Enhanced documentation of fit-potential to explain new implementation.

File:

• doc/userguide/userguide.xml (modified) (2 diffs)

doc/userguide/userguide.xml

@@ -2090 +2090 @@
         <section xml:id="potentials.fit-potential">
           <title xml:id="potentials.fit-potential.title">Fitting empirical potentials</title>
+          <para>Empirical potentials are function that represent a certain aspect
+          of binding forces in molecular dynamics, e.g. a bond potential represents
+          the force between two atoms arising because of a binding orbital in
+          between that may be (in some approximartion) represented by a Hooke's
+          spring law. 
+          In a more abstract view, an empirical potential consists of the 
+          following: a binding model representable as a graph consisting of nodes
+          and edges, a function that takes the distance between nodes and the
+          set of edges (representing bonds in the binding model) and last but
+          not least a set of parameters that represent the respective strength
+          of the bonds. For example, a torsion potential has a binding model
+          containing four nodes and three edges connecting node 1 and 2, 2 and
+          3, and 3 and 4. The function requires the six interatomic distances. And
+          the parameters are coefficients in the functions that need to be
+          evaluated to obtain the resulting force vector.
+          </para>
+          <para>In this manner, empirical potentials are implemented in 
+          MoleCuilder. They are not just a function but always an additional
+          binding model that allows to associate atoms with a specific node
+          in the model and thereby to assicate it with a particular interatomic
+          distance. And said model determines in what order the elements have 
+          to be given.
+          </para>
           <para>Let&apos;s take a look at an exemplary call to the fit potential
           action.</para>
@@ -2103 +2126 @@
           we want to look at. Here, obviously we are interested in water
           molecules, consisting of a single oxygen (8) and two hydrogen atoms (1).
-          Next, we specify the chemical element type of the potential, here a potential between oxygen (8) and hydrogen (1). We give
-          the type of the potential as morse, which requires a single distance
-          or two nuclear coordinates and the distance taken between the two. Finally, we state that the non-linear regression should be
-          done with five random starting positions, i.e. five individual minimizations, and the set of parameters
-          with the smallest L2 norm wins.</para>
+          Next, we specify the chemical element type of the potential, here a
+          potential between oxygen (8) and hydrogen (1). We give the type of 
+          the potential as morse, which requires a single distance or two 
+          nuclear coordinates and the distance taken between the two. Finally, 
+          we state that the non-linear regression should be done with five 
+          random starting positions, i.e. five individual minimizations, and 
+          the set of parameters with the smallest L2 norm wins.</para>
           <note>
             <para>Due to translational and rotational degrees of freedom for
             fragments smaller than 7 atoms, it is appropriate to look at the
-            pair-wise distances and not at the absolute coordinates. Hence,
-            the two atomic positions, here for oxygen and hydrogen, are
-            converted to a single distance. If we had given an harmonic
-            angular potential and the then required three charges/elements, &quot;8 1 1&quot;, i.e. oxygen
-            and two hydrogens, we would have obtained three distances.</para>
+            pair-wise distances and not at the absolute coordinates. In the
+            case of the water molecule as a the fragment whose energy we
+            want to represent by a empirical potential, there are 3 atoms
+            and therefore 3 unique distances between any pair of atoms.
+            From this set of distances MoleCuilder needs to pick any subset
+            that matches with the ones required by the binding model.
+            In our case of the Morse potential, we need two atoms, oxygen 
+            and hydrogen, i.e. a single distance. If we had given a harmonic
+            angular potential and the then required three charges/elements, 
+            &quot;1 8 1&quot;, i.e. oxygen and two hydrogens, we would have 
+            obtained three distances. The order of the elements, i.e. 
+            &quot;8 1 1&quot; would match a different angular interaction
+            in the same fragment, depends on the binding model of the
+            potential. In the case of the harmonic angle, the second element
+            in the list is the central atom in the angle, while the first and
+            third atom define either arm of the angle.Naturally, for the Morse
+            potential the order does not matter as each distance is symmetric.
+            </para>
             <para>MoleCuilder always adds a so-called constant potential to
             the fit containing only a single parameter, the energy offset.
             This offset compensates for the interaction energy associated with
-            a fragment of order 1, e.g. a single hydrogen atom. Essentially, this captures the atomic energy that is not associated to any bonding interactions.</para>
+            a fragment of order 1, e.g. a single hydrogen atom. Essentially, 
+            this captures the atomic energy that is not associated to any 
+            binding interactions.</para>
             <para>Note that by choosing "set-max-iterations" and "take-best-of"
             one can force the optimization to try either a single set of random