Add MD plot and make small updates

docs/CONQUEST-manual/basissets.rst

@@ -178,8 +178,8 @@ Some form of self-consistency between the MSSF and the charge density
 is required (as the MSSF will determine the Hamiltonian and hence the
 output charge density).  At present, this is performed as a complete
 SCF cycle for each set of MSSF coefficients (though this is likely to
-be updated soon for improved efficiency).  This is selected by setting
-the parameter ``Multisite.LFD.Minimise T``.
+be updated soon for improved efficiency).  This is selected by default
+(but can be turned off by setting the parameter ``Multisite.LFD.NonSCF T``).
 
 This iterative process is not variational, but is terminated when the
 absolute energy change between iterations is less than
@@ -192,7 +192,6 @@ An example input block for this process would be as follows:
 ::
 
    Multisite.LFD T
-   Multisite.LFD.Minimise T
    Multisite.LFD.Min.ThreshE 1.0e-6
    Multisite.LFD.Min.ThreshD 1.0e-6
 
@@ -230,7 +229,6 @@ and the coefficients will then be optimised.
 
    Basis.MultisiteSF T
    Multisite.LFD T
-   Multisite.LFD.Minimise T
    minE.VaryBasis T
 
 If good initial coefficient values have been found in a previous

docs/CONQUEST-manual/examples.rst

@@ -262,7 +262,9 @@ this simulation, the conserved quantity is the total energy (the sum
 of ionic kinetic energy and potential energy of the system) which is
 maintained to better than 0.1mHa in this instance.  More importantly,
 the variation in this quantity is much smaller than the variation in
-the potential energy.
+the potential energy.  This can be seen in the plot below.
+
+.. image:: MDPlot.png
 
 Go to :ref:`top <examples>`.
 

docs/CONQUEST-manual/moldyn.rst

@@ -20,20 +20,22 @@ Go to :ref:`top <moldyn>`.
 Self-consistency tolerance and XL-BOMD
 --------------------------------------
 
-Self-consistency tolerance is a critical parameter, because it can be
-responsible for "drift" in the conserved quantity of the dynamics if it is not
-tight enough. Although the molecular dynamics integrators used in CONQUEST are
-time reversible, *the SCF procedure is not*. Therefore a tight SCF tolerance
-(``minE.SCTolerance`` for diagonalisation, ``minE.LTolerance`` for linear
-scaling) is necessary. In the case of diagonalisation, a value of ``1E-7`` is
+The convergence of the electronic structure is important in MD, as
+insufficient convergence can be responsible for "drift" in the
+conserved quantity of the dynamics. Although the molecular dynamics
+integrators used in CONQUEST are time reversible, *the SCF procedure
+is not*. Therefore tight convergence (``minE.SCTolerance`` for
+diagonalisation, ``minE.LTolerance`` for linear scaling) is
+necessary. In the case of diagonalisation, SCF tolerance of ``1E-6`` is
 typically enough to negate the drift. However, extended-Lagrangian
-Born-Oppenheimer MD (XL-BOMD) :cite:`md-Niklasson2008`, currently only implemented for O(N),
-essentially makes the SCF component of the MD time-reversible by adding the
-electronic degrees of freedom to the Lagrangian, relaxing the constraint on
-``minE.LTolerance`` --- although it is still somewhat dependent on the ensemble.
-In the NVE and NVT ensembles, a L-tolerance of ``1E-5`` has been found to be
-sufficient to give good energy conservations, decreasing to ``1E-6`` in the NPT
-ensemble. The following flags are required for XL-BOMD:
+Born-Oppenheimer MD (XL-BOMD) :cite:`md-Niklasson2008`, currently only
+implemented for O(N), essentially makes the SCF component of the MD
+time-reversible by adding the electronic degrees of freedom to the
+Lagrangian, relaxing the constraint on ``minE.LTolerance`` ---
+although it is still somewhat dependent on the ensemble.  In the NVE
+and NVT ensembles, a L-tolerance of ``1E-5`` has been found to be
+sufficient to give good energy conservations, decreasing to ``1E-6``
+in the NPT ensemble. The following flags are required for XL-BOMD:
 
 ::
 
@@ -55,7 +57,7 @@ setting,
    AtomMove.RestartRun T
 
 This will do several things: it will read the atomic coordinates from
-``md.positions`` and read the ``md.checkpoint`` file, which contains the
+``md.position`` and read the ``md.checkpoint`` file, which contains the
 velocities and extended system (Nose-Hoover chain and cell) variables. Depending
 on the value of ``DM.SolutionMethod``, it will read the K-matrix files
 (``diagon``) or the L-matrix files (``ordern``), and if XL-BOMD is being used,

docs/CONQUEST-manual/strucrelax.rst

@@ -16,7 +16,7 @@ conjugate gradients at present, though L-BFGS will be implemented.
 Setting ``AtomMove.WriteXSF T`` for all flavours of optimisation will dump the
 trajectory to the file ``trajectory.xsf``, which can be visualised using `VMD
 <https://www.ks.uiuc.edu/Research/vmd/>`_. Setting ``AtomMove.AppendCoords T``
-will the structure at each step to ``UpdatedAtoms.dat`` in the format of a
+will append the structure at each step to ``UpdatedAtoms.dat`` in the format of a
 CONQUEST structure input.
 
 For the L-BFGS and conjugate gradients relaxations, the progress of the calculation can be
@@ -157,8 +157,7 @@ may cause the calculation to crash, particlularly in the case of combined
 optimisation. In such cases, it may help to try ``AtomMove.OptCellMethod 2``,
 which uses a simple but robust double-loop minimisation: a full ionic conjugate
 gradients relaxation for the inner loop and a single cell steepest descent
-relaxation for the outer loop. This is considerable less efficient, and is not
-guaranteed to converge to the same minimum as ``AtomMove.OptCellMethod 3``, but
+relaxation for the outer loop. This is considerably less efficient, but
 may help in particularly problematic cases.
 
 Go to :ref:`top <strucrelax>`.