pimplem (c46b2)


Implementation of the Thermodynamic Simulation Method



* Description | How Chemical Perturbation works.
* File Formats | Output File Formats for Chemical Perturbation.
* IC Implementation | Implementation and File Formats for Internal
Coordinate Perturbation


Top


How the Chemical Perturbation Energy Calculation Works


For thermodynamic perturbation calculations the atoms making up
the system described by the hybrid Hamiltonian, H(lambda), can be divided
into four groups. 1) The environment part - all atoms that do not change
during the perturbation. E.g., for ethanol -> propane the solvent and
the terminal methyl group. 2) The reactant atoms - the atoms that are
present at lambda = 0 and absent at lambda = 1. 3) The product atoms -
the atoms that are absent at lambda = 0 and present at lambda = 1. 4)
The COLO atoms - atoms that are present in both the reactant and product
but change charge in going from one to the other.

Certain basic premises underly our approach. Energy values are
factored by lambda (or functions thereof), never the energy functions
themselves. The standard energy routines are called unchanged and can be
modified without requiring changes to the perturbation routines as long
as the calling sequence remains the same. Potential energy terms are
written to output during a trajectory and in the case of the window
method trajectories can be combined. Futhermore any lambda -> lambda'
can be calculated post priori and additional lambda points can be added
as desired. Most other implementations do not appear to allow this.
There is, however, a price entailed namely a certain amount of redundant
calculation. Furthermore , purely as a matter of conceptual preference,
the entire perturbation part of the Hamiltonian is facter by lambda in
the same way. There has been some advocacy of factoring the attractive
and repulsive part of the Lennard-Jones potential with different powers
of lambda (see Cross).

We want to calculate the potential energy U(lambda) = Uenv +
(1-lambda)**N Ureac + lambda**N Uprod, where N is positive integer
exponent and Uenv is the energy of the common environment part of the
system. The residue topology file for the system undergoing the
perturbation has all the internal coordinate terms for both the product
and reactant parts and the regular CHARMM energy routine calculates an
energy term that in it's sum contains part of Ureac and Uprod along with
Uenv and in certain cases, as will be discussed shortly, an additional
term that needs to be removed. The residue description must contain
non-bonded exclusions between the product and reactant atoms. Of course,
none of this is factored correctly, or at all, by lambda.

The approach to obtaining a the correct U(lambda) is an indirect
one. Instead of making it so that the normal energy routine calculates
Uenv only and having the perturbation energy routine calcuated determine
(1-lambda)**N Ureac + lambda**N Uprod, we have it instead calculate the
amount that must be subtracted from the normal energy routine value (here
after referred to as Unorm) to get U(lambda). The previous statement
must be amended for the case where there are COLO atoms. Then, Unorm
contains a term that must be totally removed and is missing some terms
completely, which must be added.

For the internal coordinate energy terms and the non-bonded van
der Waals interactions, the amount that must be subtracted from Unorm to
obtain U(lambda) is given by:

U(lambda) = Unorm + Ucorr

since,

U(lambda) = U(env) + (1 - lambda)**N Ureac + lambda**N Uprod
and

Unorm = U(env) + Ureac + Uprod
then
-Ucorr = [1-(1-lambda)**N]Ureac + [1-lambda**N]Uprod .

We have currently ignored the electrostatic terms. If there are no COLO
atoms the above expressions hold true for those terms as well.

If there are COLO atoms , the situation becomes a bit more
complicated. To discuss this the following nomenclature is introduced:

[reac| reac,colo-r,env]

The expression above indicates the calculation of the electrostatic
interaction between reactant atoms and 1) other reactant atoms 2) COLO
atoms with the reactant energy charges and 3) with environment atoms.

Unorm contains the following electrostatic terms:

[reac| reac, colo-r, env] + [color | prod, colo-r, env] +
[prod | prod, env]

The term [ colo-r | prod ] must be removed in it's entirety (product
atoms do not interact with reactant charges (colo-r). And the missing
interactions involving colo-p (product) charges must be added (suitably
factored by lambda). To do this Ucorr must contain:

(1 - (1 - lambda)**N) { [reac | reac, colo-r, env] +
[colo-r | colo-r, env] }
+ (1 - lambda**N)[prod | prod, env] + 1[color|prod]
- lambda**N [colo-p | colo-p, prod, env]

Note that -Ucorr is passed from the perturbation energy routine, thus the
negative term (last one) actually adds what is totally missing from
Unorm. The electrostatic contribution to Ucorr is actually calculated in
an even more round-about fashion than that which is given above.

First both the van der Waal's and electrostatic interactions
involving reactant and product atoms with everything (except interactions
between reactant and product atoms) are calculated. The reactant colo-r
charges are used for this. This provides the term:

(1 -(1-lambda)**N)[reac | env, colo-r, reac ]
and
(1 - lambda**N)[prod | env, colo-r, prod ].

If there are no COLO atoms, this is all we need (absent the colo-r term
in the expressions). Otherwise, three more calculations, involving only
the electrostatic energy, are required. The first involves interactions
between colo-r charges with environment and other colo-r charges:

(1-(1-lambda)**N)[colo-r | env, colo-r]

Next the colo-r product atom electrostatic interaction is calculated and
factored by a function of lambda that compensates for the amount in the
second ([prod | colo-r ...] ) calculation. In that term,
1-lambda**N[prod | colo-r] is included so we must determine,

(lambda**N)[colo-r| prod]

(Since the quantity (1-lambda**N) is calculated once we actually use,

(1 - (1 - lambda**N))[colo-r| prod]

Following this the colo-r charges are exchanged, temporarily, for colo-p
and the last calculations is done. The final expression is:

-lambda**N [colo-p | prod, env, colo-p]

Which actually adds (see above) the missing interaction into the total
potential energy.

The colo-r charges are restored after this. The same procedure
is done for the image atom calculation.

It is obvious that some optimization of this method is
achievable. One possibility is that by sorting the atom list so that
COLO, reactant and product atoms appear at the top of the list in that
order, most of the non-bonded list checking can be avoided and the
copying of data structures on the heap eliminated. A more radical change
would be to edit the non-bonded lists so that the normal energy routine
calculates only Uenv and the perturbation routines calculated Ureac and
Uprod directly. The presence of the COLO atoms makes both procedures
more complicated. However, there does not appear to be a viable
alternative to the COLO atoms that is consistant with our approach.


Top

File Formats

This node provides information on the FES output file format.
The data file created during dynamics can only be written as an ASCII
formatted file. It starts with a title that is written using the
subroutine WRITITL and thus has the standard CHARMM title format. After
terminating the title with a line containing an asterisk in the first
column and nothing else, an information line follows, containing:

NSTEP, PERFRQ, NDEGF, NPUMB, LPOWER - 5(I6,1X)

The first two numbers are not currently used by the post-processor.
NDEGF, the number of degrees of freedom is used if the CTEMp flag is set.
Npumb is the number of umbrella dihedral angles. If the UMBR flag is set
in the PROCess command and NPUMB is non-zero, the umbrella sampling
correction will be effected. LPOWER is the exponent for lambda scaling.

Every PFREQ steps the FES information is written out the unit
specified in the SAVE statement. If umbrella sampling is invoked the
format is as follows:


NPRIV,AKMATI,LAMBDA,E,VPRTTR,VPRTTP,VPRTNR,VPRTNP,
VPRTVR,VPRTVP,VPRTER,VPRTEP,TOTE,TOTKE,PUMEP

FORMAT(I12,2(1X,1PG24.16E2),/,3(2(1PG24.16E2,1X),
1 1PG24.16E2,/),2(1PG24.16E2,1X),1PG24.16E2)

If umbrella sampling is not invoked it is as follows:

NPRIV,AKMATI,LAMBDA,E,VPRTTR,VPRTTP,VPRTNR,VPRTNP,
VPRTVR,VPRTVP,VPRTER,VPRTEP,TOTE,TOTKE

FORMAT(I12,2(1X,1PG24.16E2),/,3(2(1PG24.16E2,1X),
1 1PG24.16E2,/),1PG24.16E2,1X,1PG24.16E2)

Where:
NPRIV step number
AKMATI timestep in wierd CHARMM units
LAMBDA value of lambda at timestep
E total potential energy
VPRTTR V(reactant) potential energy
VPRTTP V(product) ""
VPRTNR V(reactant) potential energy vdw + electrostatic
VPRTNP V(product) ""
VPRTVR V(reactant) potential energy vdw
VPRTVP V(product) ""
VPRTER V(reactant) potential energy electrostatic
VPRTEP V(product) ""
TOTE Total energy (potential + kinetic)
TOTKE Total kinetic energy.
and with umbrella sampling:
PUMEP The exp[-beta(Vsur - Vact)] term.


Top
Internal Coordinate Implementation and File Formats

We describe how we have incorporated the double-wide, multiple-point,
window method for computing conformational free energy surfaces into
internal coordinate constraint and perturbation code with other CHARMM
routines, and it also shows the order in which the tasks are carried out,
as well as the format of the perturbation data file.

The primary internal coordinate (i.c.) constraint, perturbation, and
post-processing commands, as well as other TSM commands, are parsed in the
subroutine TSMS. When an i.c. constraint command is read, TSMS calls
ICFSET to parse the remainder of the command and to set-up the data needed
for the constraint resetting algorithm. When an i.c. perturbation command
is read, TSMS calls ICPSET to parse the remainder of the command and to
set-up the data needed to do the i.c. perturbations. Post-processing
command parsing and set-up is handled by the subroutine TSMP.

Some time after the constraints and perturbations are specified, a
dynamics command is issued and the dynamics is set up. During the dynamics
set-up, a "header" is written to the i.c. perturbation data file (opened on
unit iunicp) using the following fortran write statement in the subroutine
DCNTRL:

write(iunicp,100) nicp,icpinc,ndegf,delta
100 format(3i6,f12.6)

The variable nicp is the number of internal coordinates that will be
perturbed, icpinc is the number of subintervals, ndegf is the number of
degrees of freedom, and delta is the timestep in AKMA units. After the
dynamics is set-up, DCNTRL calls DYNAMC to integrate the equations of
motion. The main dynamics loop in DYNAMC is summarized in the following
pseudo-fortran code:

do istep = istart,istop * loop over number of steps
...
call ENERGY * get U(z) and forces
take unconstrained dynamics step
call SHAKEA * satisfy shake and i.c. cons.
...
call DYNICP * do pert. and get int. EÕs;
...
end do

The subroutine ENERGY calculates the total potential energy and the forces
needed to propagate the dynamics. After an unconstrained dynamics step is
taken, SHAKEA is called to satisfy the SHAKE and i.c. constraints in an
iterative fashion. We have to iterate the SHAKE and i.c. constraints
together, because the SHAKE constraint resetting may cause an ic constraint
to be violated, and vice versa. This constraint resetting procedure is
illustrated with the following pseudo-code:

do while (.not.done) * e.g. until shake has converged
perform iteration of shake cons. resetting
call icfcns * satisfy i.c. constraints
done = done.and..not.anyadj * if any i.c. constraints
* were adjusted, anyadj =
.true.
end do while

After the constraints are satisfied, DYNICP is called to do the double-
wide, multiple-point perturbations and calculate the interaction energies.
In DYNICP, the subroutine EIICP is called first to compute the interaction
energy of the unperturbed system (esbnp). Then the internal coordinate
values are obtained and some data is written to the perturbation data file:

call EIICP * get E for unperturbed system (esbnp)
do i = 1,nicp
get icval(1,i) * get unperturbed i.c. values
end do
c write data for unperturbed system:
write(iunicp,100) npriv,akmati,tote,totke,esbnp
100 format(i7,f10.4,3d16.8)

The two-dimensional array icval holds the internal coordinate values. The
unperturbed values are held in the first row, the values from the forward
perturbation in the second row, and the values from the reverse
perturbation in the third row. The data written to the data file includes
the number of the current dynamics step (npriv), the current simulation
time in AKMA units (akmati), the total energy (tote), total kinetic energy
(totke), and the interaction energy of the unperturbed system (esbnp).

Next, the unperturbed coordinates are copied into temporary arrays so
they can be restored after the perturbations have been carried out. Then
the double-wide, multiple point perturbations are carried out in a loop
over subintervals. The forward perturbation in each subinterval is done
first, followed by the reverse perturbation. The subroutine MVICP moves
the atoms involved in the perturbations using the algorithms described
above, and EIICP computed the interaction energies. The following pseudo-
code shows how these tasks are dispatched:

copy coords. into temp. arrays
scale = 0.0
dscale = 1.0/icpinc
do inc = 1,icpinc * loop over subintervals
scale = scale + dscale
call mvicp * move atoms by scale*dz
call eiicp * get int. E for forward pert. (esbfp)
do i = 1,nicp
get icval(2,i) * get perturbed i.c. values
end do
restore coords. from temp. arrays
call mvicp * move atoms by Ðscale*dz
call eiicp * get E for reverse pert. (esbrp)
do i = 1,nicp
get icval(3,i) * get perturbed i.c. values
end do

After all of the atoms have been moved and the interaction energies have
been computed for the forward and reverse perturbations in a subinterval,
the interaction energies and internal coordinate values are written to the
data file, and the unperturbed coordinates are restored in preparation for
the next subinterval (or the next dynamics step):

c write interaction energies of perturbed systems
write(iunicp,101) scale,esbfp,esbrp
101 format(7x,f10.4,2d16.8)
c write internal coordinate values
do i = 1,nicp
write(iunicp,102)
ic,icptyp(i),icval(1,i),icval(2,i),icval(3,i)
102 format(9x,2i4,3d16.8)
end do
restore coordinates from temp. arrays
end do