Page Menu
Home
c4science
Search
Configure Global Search
Log In
Files
F93115861
fix_rhoKUmbrella.cpp
No One
Temporary
Actions
Download File
Edit File
Delete File
View Transforms
Subscribe
Mute Notifications
Award Token
Subscribers
None
File Metadata
Details
File Info
Storage
Attached
Created
Tue, Nov 26, 08:04
Size
6 KB
Mime Type
text/x-c
Expires
Thu, Nov 28, 08:04 (1 d, 21 h)
Engine
blob
Format
Raw Data
Handle
22578203
Attached To
rLAMMPS lammps
fix_rhoKUmbrella.cpp
View Options
/*
fix_rhoK_umbrella.cpp
A fix to do umbrella sampling on rho(k).
The usage is as follows:
fix [name] [groupID] rhoKUmbrella [nx] [ny] [nz] [kappa = spring constant] [rhoK0]
where k_i = (2 pi / L_i) * n_i
Written by Ulf Pedersen and Patrick Varilly, 4 Feb 2010
Tweaked for LAMMPS 15 Jan 2010 version by Ulf Pedersen, 19 Aug 2010
Tweaked again March 4th 2012 by Ulf Pedersen.
*/
#include "fix_rhoKUmbrella.h"
#include "error.h"
#include "update.h"
#include "respa.h"
#include "atom.h"
#include "domain.h"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <math.h>
using
namespace
LAMMPS_NS
;
using
namespace
FixConst
;
// Constructor: all the parameters to this fix specified in
// the LAMMPS input get passed in
FixRhoKUmbrella
::
FixRhoKUmbrella
(
LAMMPS
*
inLMP
,
int
inArgc
,
char
**
inArgv
)
:
Fix
(
inLMP
,
inArgc
,
inArgv
)
{
// Check arguments
if
(
inArgc
!=
8
)
error
->
all
(
FLERR
,
"Illegal fix rhoKUmbrella command"
);
// Set up fix flags
scalar_flag
=
1
;
// have compute_scalar
vector_flag
=
1
;
// have compute_vector...
size_vector
=
3
;
// ...with this many components
global_freq
=
1
;
// whose value can be computed at every timestep
//scalar_vector_freq = 1; // OLD lammps: whose value can be computed at every timestep
thermo_energy
=
1
;
// this fix changes system's potential energy
extscalar
=
0
;
// but the deltaPE might not scale with # of atoms
extvector
=
0
;
// neither do the components of the vector
// Parse fix options
int
n
[
3
];
n
[
0
]
=
atoi
(
inArgv
[
3
]
);
n
[
1
]
=
atoi
(
inArgv
[
4
]
);
n
[
2
]
=
atoi
(
inArgv
[
5
]
);
mK
[
0
]
=
n
[
0
]
*
(
2
*
M_PI
/
(
domain
->
boxhi
[
0
]
-
domain
->
boxlo
[
0
]));
mK
[
1
]
=
n
[
1
]
*
(
2
*
M_PI
/
(
domain
->
boxhi
[
1
]
-
domain
->
boxlo
[
1
]));
mK
[
2
]
=
n
[
2
]
*
(
2
*
M_PI
/
(
domain
->
boxhi
[
2
]
-
domain
->
boxlo
[
2
]));
mKappa
=
atof
(
inArgv
[
6
]
);
mRhoK0
=
atof
(
inArgv
[
7
]
);
}
FixRhoKUmbrella
::~
FixRhoKUmbrella
()
{
}
// Methods that this fix implements
// --------------------------------
// Tells LAMMPS where this fix should act
int
FixRhoKUmbrella
::
setmask
()
{
int
mask
=
0
;
// This fix modifies forces...
mask
|=
POST_FORCE
;
mask
|=
POST_FORCE_RESPA
;
mask
|=
MIN_POST_FORCE
;
// ...and potential energies
mask
|=
THERMO_ENERGY
;
return
mask
;
}
/*int FixRhoKUmbrella::setmask()
{
int mask = 0;
mask |= POST_FORCE;
mask |= POST_FORCE_RESPA;
mask |= MIN_POST_FORCE;
return mask;
}*/
// Initializes the fix at the beginning of a run
void
FixRhoKUmbrella
::
init
()
{
// RESPA boilerplate
if
(
strcmp
(
update
->
integrate_style
,
"respa"
)
==
0
)
mNLevelsRESPA
=
((
Respa
*
)
update
->
integrate
)
->
nlevels
;
// Count the number of affected particles
int
nThisLocal
=
0
;
int
*
mask
=
atom
->
mask
;
int
nlocal
=
atom
->
nlocal
;
for
(
int
i
=
0
;
i
<
nlocal
;
i
++
)
{
// Iterate through all atoms on this CPU
if
(
mask
[
i
]
&
groupbit
)
{
// ...only those affected by this fix
nThisLocal
++
;
}
}
MPI_Allreduce
(
&
nThisLocal
,
&
mNThis
,
1
,
MPI_INT
,
MPI_SUM
,
world
);
mSqrtNThis
=
sqrt
(
mNThis
);
}
// Initial application of the fix to a system (when doing MD)
void
FixRhoKUmbrella
::
setup
(
int
inVFlag
)
{
if
(
strcmp
(
update
->
integrate_style
,
"verlet"
)
==
0
)
post_force
(
inVFlag
);
else
{
((
Respa
*
)
update
->
integrate
)
->
copy_flevel_f
(
mNLevelsRESPA
-
1
);
post_force_respa
(
inVFlag
,
mNLevelsRESPA
-
1
,
0
);
((
Respa
*
)
update
->
integrate
)
->
copy_f_flevel
(
mNLevelsRESPA
-
1
);
}
}
// Initial application of the fix to a system (when doing minimization)
void
FixRhoKUmbrella
::
min_setup
(
int
inVFlag
)
{
post_force
(
inVFlag
);
}
// Modify the forces calculated in the main force loop of ordinary MD
void
FixRhoKUmbrella
::
post_force
(
int
inVFlag
)
{
double
**
x
=
atom
->
x
;
double
**
f
=
atom
->
f
;
int
*
mask
=
atom
->
mask
;
int
nlocal
=
atom
->
nlocal
;
// Loop over locally-owned atoms affected by this fix and calculate the
// partial rhoK's
mRhoKLocal
[
0
]
=
0.0
;
mRhoKLocal
[
1
]
=
0.0
;
for
(
int
i
=
0
;
i
<
nlocal
;
i
++
)
{
// Iterate through all atoms on this CPU
if
(
mask
[
i
]
&
groupbit
)
{
// ...only those affected by this fix
// rho_k = sum_i exp( - i k.r_i )
mRhoKLocal
[
0
]
+=
cos
(
mK
[
0
]
*
x
[
i
][
0
]
+
mK
[
1
]
*
x
[
i
][
1
]
+
mK
[
2
]
*
x
[
i
][
2
]
);
mRhoKLocal
[
1
]
-=
sin
(
mK
[
0
]
*
x
[
i
][
0
]
+
mK
[
1
]
*
x
[
i
][
1
]
+
mK
[
2
]
*
x
[
i
][
2
]
);
}
}
// Now calculate mRhoKGlobal
MPI_Allreduce
(
mRhoKLocal
,
mRhoKGlobal
,
2
,
MPI_DOUBLE
,
MPI_SUM
,
world
);
// WARNING!!!!! < \sum_{i,j} e^{-ik.(r_i - r_j)} > ~ N, so
// we define rho_k as (1 / sqrt(N)) \sum_i e^{-i k.r_i}, so that
// <rho_k^2> is intensive.
//
// Don't forget this two years from now when you change the system size!!!
mRhoKGlobal
[
0
]
/=
mSqrtNThis
;
mRhoKGlobal
[
1
]
/=
mSqrtNThis
;
// We'll need magnitude of rho_k
double
rhoK
=
sqrt
(
mRhoKGlobal
[
0
]
*
mRhoKGlobal
[
0
]
+
mRhoKGlobal
[
1
]
*
mRhoKGlobal
[
1
]
);
for
(
int
i
=
0
;
i
<
nlocal
;
i
++
)
{
// Iterate through all atoms on this CPU
if
(
mask
[
i
]
&
groupbit
)
{
// ...only those affected by this fix
// Calculate forces
// U = kappa/2 ( |rho_k| - rho_k^0 )^2
// f_i = -grad_i U = -kappa ( |rho_k| - rho_k^0 ) grad_i |rho_k|
// grad_i |rho_k| = Re( rho_k* (-i k e^{-i k . r_i} / sqrt(N)) ) / |rho_k|
//
// In terms of real and imag parts of rho_k,
//
// Re( rho_k* (-i k e^{-i k . r_i}) ) =
// (- Re[rho_k] * sin( k . r_i ) - Im[rho_k] * cos( k . r_i )) * k
double
sinKRi
=
sin
(
mK
[
0
]
*
x
[
i
][
0
]
+
mK
[
1
]
*
x
[
i
][
1
]
+
mK
[
2
]
*
x
[
i
][
2
]
);
double
cosKRi
=
cos
(
mK
[
0
]
*
x
[
i
][
0
]
+
mK
[
1
]
*
x
[
i
][
1
]
+
mK
[
2
]
*
x
[
i
][
2
]
);
double
prefactor
=
mKappa
*
(
rhoK
-
mRhoK0
)
/
rhoK
*
(
-
mRhoKGlobal
[
0
]
*
sinKRi
-
mRhoKGlobal
[
1
]
*
cosKRi
)
/
mSqrtNThis
;
f
[
i
][
0
]
-=
prefactor
*
mK
[
0
];
f
[
i
][
1
]
-=
prefactor
*
mK
[
1
];
f
[
i
][
2
]
-=
prefactor
*
mK
[
2
];
}
}
}
// Forces in RESPA loop
void
FixRhoKUmbrella
::
post_force_respa
(
int
inVFlag
,
int
inILevel
,
int
inILoop
)
{
if
(
inILevel
==
mNLevelsRESPA
-
1
)
post_force
(
inVFlag
);
}
// Forces in minimization loop
void
FixRhoKUmbrella
::
min_post_force
(
int
inVFlag
)
{
post_force
(
inVFlag
);
}
// Compute the change in the potential energy induced by this fix
double
FixRhoKUmbrella
::
compute_scalar
()
{
double
rhoK
=
sqrt
(
mRhoKGlobal
[
0
]
*
mRhoKGlobal
[
0
]
+
mRhoKGlobal
[
1
]
*
mRhoKGlobal
[
1
]
);
return
0.5
*
mKappa
*
(
rhoK
-
mRhoK0
)
*
(
rhoK
-
mRhoK0
);
}
// Compute the ith component of the vector
double
FixRhoKUmbrella
::
compute_vector
(
int
inI
)
{
if
(
inI
==
0
)
return
mRhoKGlobal
[
0
];
// Real part
else
if
(
inI
==
1
)
return
mRhoKGlobal
[
1
];
// Imagniary part
else
if
(
inI
==
2
)
return
sqrt
(
mRhoKGlobal
[
0
]
*
mRhoKGlobal
[
0
]
+
mRhoKGlobal
[
1
]
*
mRhoKGlobal
[
1
]
);
else
return
12345.0
;
}
Event Timeline
Log In to Comment