# tps (c49b1)

Transition Path Sampling

The Transition Path Sampling (TPS) methods introduced by Chandler and

co-workers to sample rare events (see References) are implemented as

extensions of the RXNCoor, DYNAmics, and USER commands in CHARMM. The

TPS keyword must be included in pref.dat for the code to be compiled.

* Syntax | Syntax required to invoke TPS

* Description | Description of TPS specific keywords

* References | Some references of use

The Transition Path Sampling (TPS) methods introduced by Chandler and

co-workers to sample rare events (see References) are implemented as

extensions of the RXNCoor, DYNAmics, and USER commands in CHARMM. The

TPS keyword must be included in pref.dat for the code to be compiled.

* Syntax | Syntax required to invoke TPS

* Description | Description of TPS specific keywords

* References | Some references of use

Top

Syntax required to invoke TPS

[Syntax RXNCoor]

RXNCoor [ standard RXNCoor keywords ] [ set-spec ] [ bas-spec ] [wri-spec]

set-spec ::= SET [ NRXN 1 ] NRXN{ name }

where NRXN{ name } is an NRXN-long list of the names of

the order parameters to calculate.

bas-spec ::= BASIn NRXN{ name alo ahi blo bhi }

wri-spec ::= TPUNit NRXN{ name unit }

[Syntax DYNAmics]

DYNAmics [ RTRJ ] [ standard DYNAmics keywords ] -

PATH [ mod-spec ] [ tps-spec ] [ sht-spec ] [ trj-spec ] [ hsa-spec ]

mod-spec ::= [ ISVFrequency 0 ]

tps-spec ::= [ NTPAth 0 ] [ NSAVP 0 ] [ NPRAccept 0 ] -

[ ITPRint 0 ] [ ITPUnit STDOUT ] [ ACCU STDOUT ] -

[ USER 0 ] [ PSHOot 1.0 ] [ IMXShift 1 ]

[ SDUNit 0 ] [ SDINit 0 ]

sht-spec ::= [ VFRAction 0.0 ] [ TFRAction 1.0 ] [ NTFRaction NTPAth ] -

[ IFSHoot -1 ] [ IRST 0 ] [ PHALf 0 ] -

[ ISLO 0 ] [ ISLN 0 ]

trj-spec ::= [ NUNIt 1 ] [ FIRStu -1 ] [ VFIRst FIRStu+NUNIt ] -

[ BEGIn 0 ] [ SKIP -1 ] [ STOP 0 ] -

[ PNSAve NSTEP / NSAVC + 1]

hsa-spec ::= [ HSAM 0 ] [ IHUN STDOUT ] [ IHFR 0 ] -

[ NHSV -1 ] [ IHPR 1 ] [ NHST 0 ]

Syntax required to invoke TPS

[Syntax RXNCoor]

RXNCoor [ standard RXNCoor keywords ] [ set-spec ] [ bas-spec ] [wri-spec]

set-spec ::= SET [ NRXN 1 ] NRXN{ name }

where NRXN{ name } is an NRXN-long list of the names of

the order parameters to calculate.

bas-spec ::= BASIn NRXN{ name alo ahi blo bhi }

wri-spec ::= TPUNit NRXN{ name unit }

[Syntax DYNAmics]

DYNAmics [ RTRJ ] [ standard DYNAmics keywords ] -

PATH [ mod-spec ] [ tps-spec ] [ sht-spec ] [ trj-spec ] [ hsa-spec ]

mod-spec ::= [ ISVFrequency 0 ]

tps-spec ::= [ NTPAth 0 ] [ NSAVP 0 ] [ NPRAccept 0 ] -

[ ITPRint 0 ] [ ITPUnit STDOUT ] [ ACCU STDOUT ] -

[ USER 0 ] [ PSHOot 1.0 ] [ IMXShift 1 ]

[ SDUNit 0 ] [ SDINit 0 ]

sht-spec ::= [ VFRAction 0.0 ] [ TFRAction 1.0 ] [ NTFRaction NTPAth ] -

[ IFSHoot -1 ] [ IRST 0 ] [ PHALf 0 ] -

[ ISLO 0 ] [ ISLN 0 ]

trj-spec ::= [ NUNIt 1 ] [ FIRStu -1 ] [ VFIRst FIRStu+NUNIt ] -

[ BEGIn 0 ] [ SKIP -1 ] [ STOP 0 ] -

[ PNSAve NSTEP / NSAVC + 1]

hsa-spec ::= [ HSAM 0 ] [ IHUN STDOUT ] [ IHFR 0 ] -

[ NHSV -1 ] [ IHPR 1 ] [ NHST 0 ]

Top

RXNCoor

The RXNCoor command has been modified to facilitate the definition of the

basins that constrain the endpoints of the path.

Multiple order parameters can be specified using the NRXN keyword

following the SET keyword (» umbrel for a description of the

latter). However, note that the tree structure associated with the

RXNCoor command is not dynamically allocated, so that MAXNOD in

rxncom.fcm must be increased with for larger numbers of order parameters.

The BASIn keyword is used to specify the boundaries of the basins (A

and B). For each order parameter, one must give the name followed by

four real numbers: alo, ahi, blo, and bhi (the order is important).

Here, alo (blo) is the lower bound for basin A (B) and ahi (bhi) is

the upper bound for basin A (B).

The TPUNit keyword designates units to which values of the order

parameters are written. Values are written for each saved trajectory.

A unit number less than 1 suppresses writeout.

In addition, a steered molecular dynamics (SMD) has been implemented

through RXNCOR and integrated with transition path sampling (TPS) to allow

generation of reactive trajectories de novo. In this bias-annealing method,

SMD allows transitions between basins (defined in the RXNCOR module) of the

reaction coordinates where symmetric (harmonic) biasing potentials are

advanced in a ratchet-like manner. TPS shooting moves can then be made with

progressively milder biasing potentials until unbiased reactive trajectories

could be obtained. This facility is activated by including the SMDDel keyword

(along with a non-zero KUMB value) in the UMBRella subcommand. This facility

can be used with DYNAmics without TPS so long as the basins are defined.

DYNAmics

The DYNAmics command has been modified to provide a looping structure to

calculate multiple trajectories (paths). TPS is invoked by including the PATH

keyword.

The additional keywords that are specific to TPS are:

NTPAth The number of paths to calculate.

NSAVP The frequency of saving paths to the trajectory and velocity

files.

USER Whether to user USERSB to calculated the order parameters

used to determine the stable states.

ITPRint The frequency with which to write order parameter values

at basin evaluations.

ITPUnit The unit number on which to write order parameter values

at basin evaluations.

SDUNit The unit number to which to write random number generator

seeds for TPS with Langevin dynamics.

SDINit The unit number from which to read random number generator

seeds for TPS with Langevin dynamics.

NPRAccept The frequency of printing acceptance statistics.

ACCUnit The unit number to which to write acceptance statistics

as a function of saved structure. The columns are the same

as those printed by TPSACC to the CHARMM output file.

PSHOot The fraction of moves that are shooting moves.

IMXShift The maximum number of saved phase space points by which to

reptate the path in a shifting move. In other words, a

shift can be up to IMXS*NSAVC molecular dynamics steps long.

VFRAction The amount to perturb the velocities in shooting moves. A

random vector is chosen from a Gaussian (Maxwell-Boltzmann)

distribution and then scaled by VFRAction. The scaled vector

is added to the current velocity vector and the result is

scaled to conserve kinetic energy after correcting for SHAKE

if necessary. This procedure has the effect of rotating the

3N-dimensional velocity vector without changing its magnitude.

TFRAction The amount by which to scale the kinetic energy in each

shooting move if annealing is desired.

NTFRaction The number of ACCEPTED shooting moves in which to scale the

kinetic energy by TFRAction.

IFSHoot The first point from which to shoot in units of NSAVC. A

value of -1 indicates that IFSHot is chosen randomly if RTRJ

is specified and it is set to the middle of the path

if shooting from a structure.

IRST The saved phase space point to save to the restart file. It

is best if IRST is chosen to be close to the transition state.

Otherwise, numerical errors can prevent one from regenerating

a valid path using the saved phase space point.

PHALf The probability of shooting half a trajectory. A stochastic

element should be included if PHALf is greater than

zero. For example, see discussion about Langevin dynamics.

ISLO The lowest saved structure from which to shoot.

ISLN The number of saved structures from which to shoot (i.e.,

the last saved structure shot from is ISLO + ISLN - 1).

RTRJ If this keyword is present (in place of STARt or RESTart),

an entire trajectory is read at the beginning of a restart.

The keywords BEGIn, SKIP, and STOP have their usual meaning.

NUNIt The number of trajectory files to read if RTRJ is specified.

IFIRst The first trajectory file to read if RTRJ is specified.

VFIRst The first velocity file to read if RTRJ is specified.

PNSAve Allow TPS to read shorter trajectories into longer ones in

order to lengthen the allowed transition time. PNSAve specifies

the number of phase points in the short trajectories (default

is the number of phase points to be saved in the new trajectories).

HSAMple If this keyword is present, paths are accepted if they

start in basin A and ever go through basin B. Also, the

probability that the system is in basin B as a function of

time is calculated [<h_B(t)>].

NHSTart The path at which to start calculating <h_B(t)>.

IHUNit The unit number to which to write <h_B(t)>.

IHFRequency The frequency with which to write <h_B(t)>.

NHSV The frequency of evaluating whether the path is in

basin B (h_B[x(t)]). A value of -1 sets NHSV to NSAVC.

IHPRint h_B[x(t)] is printed every IHPRint*NHSV steps.

In addition, note that the meaning of the ISVFrequency keyword is changed

during TPS. It refers to the number of PATHS, not the the number of

molecular dynamics steps, between writes to the restart file.

Note that if PHALf is greater than 0, shooting moves are carried out in

which the path is only updated in one direction. In this case a stochastic

element should be included in the integration, such as Langevin dynamics.

It is possible to apply Langevin integration to only the periphery of the

simulation using the RBUF keyword. When using Langevin dynamics with TPS,

the random number seed used to generate the random forces is recorded for

every saved structure. This is necessary to regenerate the appropriate

displacement vectors from the coordinates and velocities in 2-step dynamics

during a shooting move. When writing and reading trajectories, these seeds

can be written/read as designated by the keywords SDUNit and SDINit. If

seeds are not read in with a trajectory, seeds are generated randomly for

the initial shooting move, which results in a different displacement vector

than in the original structure. If SHAKe is used in conjunction with Langevin

dynamics, the new displacement vectors have some velocity components along

the constrained bond. These components are zeroed, and the overall kinetic

energy will be reduced for that step. Also note that, if SHAKe is used,

there is a small error in the regeneration of displacement vectors for

atoms that have different values of FBETA and are connected by a shaken bond.

RXNCoor

The RXNCoor command has been modified to facilitate the definition of the

basins that constrain the endpoints of the path.

Multiple order parameters can be specified using the NRXN keyword

following the SET keyword (» umbrel for a description of the

latter). However, note that the tree structure associated with the

RXNCoor command is not dynamically allocated, so that MAXNOD in

rxncom.fcm must be increased with for larger numbers of order parameters.

The BASIn keyword is used to specify the boundaries of the basins (A

and B). For each order parameter, one must give the name followed by

four real numbers: alo, ahi, blo, and bhi (the order is important).

Here, alo (blo) is the lower bound for basin A (B) and ahi (bhi) is

the upper bound for basin A (B).

The TPUNit keyword designates units to which values of the order

parameters are written. Values are written for each saved trajectory.

A unit number less than 1 suppresses writeout.

In addition, a steered molecular dynamics (SMD) has been implemented

through RXNCOR and integrated with transition path sampling (TPS) to allow

generation of reactive trajectories de novo. In this bias-annealing method,

SMD allows transitions between basins (defined in the RXNCOR module) of the

reaction coordinates where symmetric (harmonic) biasing potentials are

advanced in a ratchet-like manner. TPS shooting moves can then be made with

progressively milder biasing potentials until unbiased reactive trajectories

could be obtained. This facility is activated by including the SMDDel keyword

(along with a non-zero KUMB value) in the UMBRella subcommand. This facility

can be used with DYNAmics without TPS so long as the basins are defined.

DYNAmics

The DYNAmics command has been modified to provide a looping structure to

calculate multiple trajectories (paths). TPS is invoked by including the PATH

keyword.

The additional keywords that are specific to TPS are:

NTPAth The number of paths to calculate.

NSAVP The frequency of saving paths to the trajectory and velocity

files.

USER Whether to user USERSB to calculated the order parameters

used to determine the stable states.

ITPRint The frequency with which to write order parameter values

at basin evaluations.

ITPUnit The unit number on which to write order parameter values

at basin evaluations.

SDUNit The unit number to which to write random number generator

seeds for TPS with Langevin dynamics.

SDINit The unit number from which to read random number generator

seeds for TPS with Langevin dynamics.

NPRAccept The frequency of printing acceptance statistics.

ACCUnit The unit number to which to write acceptance statistics

as a function of saved structure. The columns are the same

as those printed by TPSACC to the CHARMM output file.

PSHOot The fraction of moves that are shooting moves.

IMXShift The maximum number of saved phase space points by which to

reptate the path in a shifting move. In other words, a

shift can be up to IMXS*NSAVC molecular dynamics steps long.

VFRAction The amount to perturb the velocities in shooting moves. A

random vector is chosen from a Gaussian (Maxwell-Boltzmann)

distribution and then scaled by VFRAction. The scaled vector

is added to the current velocity vector and the result is

scaled to conserve kinetic energy after correcting for SHAKE

if necessary. This procedure has the effect of rotating the

3N-dimensional velocity vector without changing its magnitude.

TFRAction The amount by which to scale the kinetic energy in each

shooting move if annealing is desired.

NTFRaction The number of ACCEPTED shooting moves in which to scale the

kinetic energy by TFRAction.

IFSHoot The first point from which to shoot in units of NSAVC. A

value of -1 indicates that IFSHot is chosen randomly if RTRJ

is specified and it is set to the middle of the path

if shooting from a structure.

IRST The saved phase space point to save to the restart file. It

is best if IRST is chosen to be close to the transition state.

Otherwise, numerical errors can prevent one from regenerating

a valid path using the saved phase space point.

PHALf The probability of shooting half a trajectory. A stochastic

element should be included if PHALf is greater than

zero. For example, see discussion about Langevin dynamics.

ISLO The lowest saved structure from which to shoot.

ISLN The number of saved structures from which to shoot (i.e.,

the last saved structure shot from is ISLO + ISLN - 1).

RTRJ If this keyword is present (in place of STARt or RESTart),

an entire trajectory is read at the beginning of a restart.

The keywords BEGIn, SKIP, and STOP have their usual meaning.

NUNIt The number of trajectory files to read if RTRJ is specified.

IFIRst The first trajectory file to read if RTRJ is specified.

VFIRst The first velocity file to read if RTRJ is specified.

PNSAve Allow TPS to read shorter trajectories into longer ones in

order to lengthen the allowed transition time. PNSAve specifies

the number of phase points in the short trajectories (default

is the number of phase points to be saved in the new trajectories).

HSAMple If this keyword is present, paths are accepted if they

start in basin A and ever go through basin B. Also, the

probability that the system is in basin B as a function of

time is calculated [<h_B(t)>].

NHSTart The path at which to start calculating <h_B(t)>.

IHUNit The unit number to which to write <h_B(t)>.

IHFRequency The frequency with which to write <h_B(t)>.

NHSV The frequency of evaluating whether the path is in

basin B (h_B[x(t)]). A value of -1 sets NHSV to NSAVC.

IHPRint h_B[x(t)] is printed every IHPRint*NHSV steps.

In addition, note that the meaning of the ISVFrequency keyword is changed

during TPS. It refers to the number of PATHS, not the the number of

molecular dynamics steps, between writes to the restart file.

Note that if PHALf is greater than 0, shooting moves are carried out in

which the path is only updated in one direction. In this case a stochastic

element should be included in the integration, such as Langevin dynamics.

It is possible to apply Langevin integration to only the periphery of the

simulation using the RBUF keyword. When using Langevin dynamics with TPS,

the random number seed used to generate the random forces is recorded for

every saved structure. This is necessary to regenerate the appropriate

displacement vectors from the coordinates and velocities in 2-step dynamics

during a shooting move. When writing and reading trajectories, these seeds

can be written/read as designated by the keywords SDUNit and SDINit. If

seeds are not read in with a trajectory, seeds are generated randomly for

the initial shooting move, which results in a different displacement vector

than in the original structure. If SHAKe is used in conjunction with Langevin

dynamics, the new displacement vectors have some velocity components along

the constrained bond. These components are zeroed, and the overall kinetic

energy will be reduced for that step. Also note that, if SHAKe is used,

there is a small error in the regeneration of displacement vectors for

atoms that have different values of FBETA and are connected by a shaken bond.

Top

REFERENCES

All studies that employ TPS in CHARMM should reference:

Hagan, M. F., Dinner, A. R., Chandler, D. and Chakraborty, A. K. (2003)

Atomistic understanding of kinetic pathways for single base-pair binding

and unbinding in DNA. Proc. Natl. Acad. Sci. USA 100, 13922-13927.

In addition, studies that employ SMD based on RXNCOR should reference:

Hu, J., Ma, A. and Dinner, A. R. (2006) Bias annealing: A method for obtaining

transition paths de novo. J. Chem. Phys., submittted.

Additional references on TPS:

Dellago, C., Bolhuis, P., Csajka, F. and Chandler, D. (1998)

Transition Path Sampling and the Calculation of Rate Constants.

J. Chem. Phys . 108, 1964.

Dellago, C., Bolhuis, P. and Chandler, D. (1998) Efficient

Transition Path Sampling: Application to Lennard-Jones Cluster

Rearrangements. J. Chem. Phys. 108, 9236.

Dellago C., Bolhuis, P. G., Geissler, P. L. (2002) Transition path

sampling. Adv. Chem. Phys. 123, 1.

Bolhuis, P. G., Chandler, D., Dellago, C. and Geissler, P. (2002)

Transition Path Sampling: Throwing ropes over mountain passes, in

the dark. Ann. Rev. Phys. Chem. 59, 291.

REFERENCES

All studies that employ TPS in CHARMM should reference:

Hagan, M. F., Dinner, A. R., Chandler, D. and Chakraborty, A. K. (2003)

Atomistic understanding of kinetic pathways for single base-pair binding

and unbinding in DNA. Proc. Natl. Acad. Sci. USA 100, 13922-13927.

In addition, studies that employ SMD based on RXNCOR should reference:

Hu, J., Ma, A. and Dinner, A. R. (2006) Bias annealing: A method for obtaining

transition paths de novo. J. Chem. Phys., submittted.

Additional references on TPS:

Dellago, C., Bolhuis, P., Csajka, F. and Chandler, D. (1998)

Transition Path Sampling and the Calculation of Rate Constants.

J. Chem. Phys . 108, 1964.

Dellago, C., Bolhuis, P. and Chandler, D. (1998) Efficient

Transition Path Sampling: Application to Lennard-Jones Cluster

Rearrangements. J. Chem. Phys. 108, 9236.

Dellago C., Bolhuis, P. G., Geissler, P. L. (2002) Transition path

sampling. Adv. Chem. Phys. 123, 1.

Bolhuis, P. G., Chandler, D., Dellago, C. and Geissler, P. (2002)

Transition Path Sampling: Throwing ropes over mountain passes, in

the dark. Ann. Rev. Phys. Chem. 59, 291.