mdp options

Main Table of Contents

Sun 18 Jan 2009

Table of Contents


Default values are given in parentheses. The first option in the list is always the default option. Units are given in square brackets The difference between a dash and an underscore is ignored.

A sample .mdp file is available. This should be appropriate to start a normal simulation. Edit it to suit your specific needs and desires.


directories to include in your topology. Format:
-I/home/john/mylib -I../otherlib
defines to pass to the preprocessor, default is no defines. You can use any defines to control options in your customized topology files. Options that are already available by default are:
Will tell grompp to include flexible water in stead of rigid water into your topology, this can be useful for normal mode analysis.
Will tell grompp to include posre.itp into your topology, used for position restraints.

Run control

integrator: (Despite the name, this list includes algorithms that are not actually integrators. steep and all entries following it are in this category)
A leap-frog algorithm for integrating Newton's equations of motion.
A velocity Verlet algorithm for integrating Newton's equations of motion. For constant NVE simulations started from corresponding points in the same trajectory, the trajectories are analytically, but not binary, identical to the md leap-frog integrator. The the kinetic energy, which is determined from the whole step velocities and is therefore slightly too high. The advantage of this integrator is more accurate, reversible Nose-Hoover and Parrinello-Rahman coupling integration based on Trotter expansion, as well as (slightly too small) full step velocity output. This all comes at the cost off extra computation, especially with constraints and extra communication in parallel. Note that for nearly all production simulations the md integrator is accurate enough.
A velocity Verlet algorithm identical to md-vv, except that the kinetic energy is determined as the average of the two half step kinetic energies as in the md integrator, and this thus more accurate. With Nose-Hoover and/or Parrinello-Rahman coupling this comes with a slight increase in computational cost.
An accurate leap-frog stochastic dynamics integrator. Four Gaussian random number are required per integration step per degree of freedom. With constraints, coordinates needs to be constrained twice per integration step. Depending on the computational cost of the force calculation, this can take a significant part of the simulation time. The temperature for one or more groups of atoms (tc_grps) is set with ref_t [K], the inverse friction constant for each group is set with tau_t [ps]. The parameter tcoupl is ignored. The random generator is initialized with ld_seed. When used as a thermostat, an appropriate value for tau_t is 2 ps, since this results in a friction that is lower than the internal friction of water, while it is high enough to remove excess heat (unless cut-off or reaction-field electrostatics is used). NOTE: temperature deviations decay twice as fast as with a Berendsen thermostat with the same tau_t.
An efficient leap-frog stochastic dynamics integrator. This integrator is equivalent to sd, except that it requires only one Gaussian random number and one constraint step. This integrator is less accurate. For water the relative error in the temperature with this integrator is 0.5 delta_t/tau_t, but for other systems it can be higher. Use with care and check the temperature.
An Euler integrator for Brownian or position Langevin dynamics, the velocity is the force divided by a friction coefficient (bd_fric [amu ps-1]) plus random thermal noise (ref_t). When bd_fric=0, the friction coefficient for each particle is calculated as mass/tau_t, as for the integrator sd. The random generator is initialized with ld_seed.
A steepest descent algorithm for energy minimization. The maximum step size is emstep [nm], the tolerance is emtol [kJ mol-1 nm-1].
A conjugate gradient algorithm for energy minimization, the tolerance is emtol [kJ mol-1 nm-1]. CG is more efficient when a steepest descent step is done every once in a while, this is determined by nstcgsteep. For a minimization prior to a normal mode analysis, which requires a very high accuracy, GROMACS should be compiled in double precision.
A quasi-Newtonian algorithm for energy minimization according to the low-memory Broyden-Fletcher-Goldfarb-Shanno approach. In practice this seems to converge faster than Conjugate Gradients, but due to the correction steps necessary it is not (yet) parallelized.
Normal mode analysis is performed on the structure in the tpr file. GROMACS should be compiled in double precision.
Test particle insertion. The last molecule in the topology is the test particle. A trajectory should be provided with the -rerun option of mdrun. This trajectory should not contain the molecule to be inserted. Insertions are performed nsteps times in each frame at random locations and with random orientiations of the molecule. When nstlist is larger than one, nstlist insertions are performed in a sphere with radius rtpi around a the same random location using the same neighborlist (and the same long-range energy when rvdw or rcoulomb>rlist, which is only allowed for single-atom molecules). Since neighborlist construction is expensive, one can perform several extra insertions with the same list almost for free. The random seed is set with ld_seed. The temperature for the Boltzmann weighting is set with ref_t, this should match the temperature of the simulation of the original trajectory. Dispersion correction is implemented correctly for tpi. All relevant quantities are written to the file specified with the -tpi option of mdrun. The distribution of insertion energies is written to the file specified with the -tpid option of mdrun. No trajectory or energy file is written. Parallel tpi gives identical results to single node tpi. For charged molecules, using PME with a fine grid is most accurate and also efficient, since the potential in the system only needs to be calculated once per frame.
Test particle insertion into a predefined cavity location. The procedure is the same as for tpi, except that one coordinate extra is read from the trajectory, which is used as the insertion location. The molecule to be inserted should be centered at 0,0,0. Gromacs does not do this for you, since for different situations a different way of centering might be optimal. Also rtpi sets the radius for the sphere around this location. Neighbor searching is done only once per frame, nstlist is not used. Parallel tpic gives identical results to single node tpic.
tinit: (0) [ps]
starting time for your run (only makes sense for integrators md, sd and bd)
dt: (0.001) [ps]
time step for integration (only makes sense for integrators md, sd and bd)
nsteps: (0)
maximum number of steps to integrate or minimize, -1 is no maximum
init_step: (0)
The starting step. The time at an step i in a run is calculated as: t = tinit + dt*(init_step + i). The free-energy lambda is calculated as: lambda = init_lambda + delta_lambda*(init_step + i). Also non-equilibrium MD parameters can depend on the step number. Thus for exact restarts or redoing part of a run it might be necessary to set init_step to the step number of the restart frame. tpbconv does this automatically.
Remove center of mass translation
Remove center of mass translation and rotation around the center of mass
No restriction on the center of mass motion
nstcomm: (10) [steps]
frequency for center of mass motion removal
group(s) for center of mass motion removal, default is the whole system

Langevin dynamics

bd_fric: (0) [amu ps-1]
Brownian dynamics friction coefficient. When bd_fric=0, the friction coefficient for each particle is calculated as mass/tau_t.
ld_seed: (1993) [integer]
used to initialize random generator for thermal noise for stochastic and Brownian dynamics. When ld_seed is set to -1, the seed is calculated from the process ID. When running BD or SD on multiple processors, each processor uses a seed equal to ld_seed plus the processor number.

Energy minimization

emtol: (10.0) [kJ mol-1 nm-1]
the minimization is converged when the maximum force is smaller than this value
emstep: (0.01) [nm]
initial step-size
nstcgsteep: (1000) [steps]
frequency of performing 1 steepest descent step while doing conjugate gradient energy minimization.
nbfgscorr: (10)
Number of correction steps to use for L-BFGS minimization. A higher number is (at least theoretically) more accurate, but slower.

Shell Molecular Dynamics

When shells or flexible constraints are present in the system the positions of the shells and the lengths of the flexible constraints are optimized at every time step until either the RMS force on the shells and constraints is less than emtol, or a maximum number of iterations (niter) has been reached
emtol: (10.0) [kJ mol-1 nm-1]
the minimization is converged when the maximum force is smaller than this value. For shell MD this value should be 1.0 at most, but since the variable is used for energy minimization as well the default is 10.0.
niter: (20)
maximum number of iterations for optimizing the shell positions and the flexible constraints.
fcstep: (0) [ps2]
the step size for optimizing the flexible constraints. Should be chosen as mu/(d2V/d q2) where mu is the reduced mass of two particles in a flexible constraint and d2V/d q2 is the second derivative of the potential in the constraint direction. Hopefully this number does not differ too much between the flexible constraints, as the number of iterations and thus the runtime is very sensitive to fcstep. Try several values!

Test particle insertion

rtpi: (0.05) [nm]
the test particle insertion radius see integrators tpi and tpic

Output control

nstxout: (100) [steps]
frequency to write coordinates to output trajectory file, the last coordinates are always written
nstvout: (100) [steps]
frequency to write velocities to output trajectory, the last velocities are always written
nstfout: (0) [steps]
frequency to write forces to output trajectory.
nstlog: (100) [steps]
frequency to write energies to log file, the last energies are always written
nstcalcenergy: (-1)
frequency for calculating the energies, 0 is never. This option is only relevant with dynamics. With a twin-range cut-off setup nstcalcenergy should be equal to or a multiple of nstlist. This option affects the performance in parallel simulations, because calculating energies requires global communication between all processes which can become a bottleneck at high parallelization. The default value of -1 sets nstcalcenergy equal to nstlist, unless nstlist ≤0, then a value of 10 is used. If nstenergy is smaller than the automatically generated value, the lowest common denominator of nstenergy and nstlist is used.
nstenergy: (100) [steps]
frequency to write energies to energy file, the last energies are always written, should be a multiple of nstcalcenergy. Note that the exact sums and fluctuations over all MD steps modulo nstcalcenergy are stored in the energy file, so g_energy can report exact energy averages and fluctuations also when nstenergy>1
nstxtcout: (0) [steps]
frequency to write coordinates to xtc trajectory
xtc_precision: (1000) [real]
precision to write to xtc trajectory
group(s) to write to xtc trajectory, default the whole system is written (if nstxtcout is larger than zero)
group(s) to write to energy file

Neighbor searching

nstlist: (10) [steps]
Frequency to update the neighbor list (and the long-range forces, when using twin-range cut-off's). When this is 0, the neighbor list is made only once. With energy minimization the neighborlist will be updated for every energy evaluation when nstlist>0.
The neighbor list is only constructed once and never updated. This is mainly useful for vacuum simulations in which all particles see each other.
Automated update frequency. This can only be used with switched, shifted or user potentials where the cut-off can be smaller than rlist. One then has a buffer of size rlist minus the longest cut-off. The neighbor list is only updated when one or more particles have moved further than half the buffer size from the center of geometry of their charge group as determined at the previous neighbor search. Coordinate scaling due to pressure coupling or the deform option is taken into account. This option guarantees that their are no cut-off artifacts. But for larger systems this can come at a high computational cost, since the neighbor list update frequency will be determined by just one or two particles moving slightly beyond the half buffer length (which not even necessarily implies that the neighbor list is invalid), while 99.99% of the particles are fine.
Make a grid in the box and only check atoms in neighboring grid cells when constructing a new neighbor list every nstlist steps. In large systems grid search is much faster than simple search.
Check every atom in the box when constructing a new neighbor list every nstlist steps.
Use periodic boundary conditions in all directions.
Use no periodic boundary conditions, ignore the box. To simulate without cut-offs, set all cut-offs to 0 and nstlist=0. For best performance without cut-offs, use nstlist=0, ns_type=simple and particle decomposition instead of domain decomposition.
Use periodic boundary conditions in x and y directions only. This works only with ns_type=grid and can be used in combination with walls. Without walls or with only one wall the system size is infinite in the z direction. Therefore pressure coupling or Ewald summation methods can not be used. These disadvantages do not apply when two walls are used.
molecules are finite, fast molecular pbc can be used
for systems with molecules that couple to themselves through the periodic boundary conditions, this requires a slower pbc algorithm and molecules are not made whole in the output
rlist: (1) [nm]
cut-off distance for the short-range neighbor list
rlistlong: (-1) [nm]
Cut-off distance for the long-range neighbor list. This parameter is only relevant for a twin-range cut-off setup with switched potentials. In that case a buffer region is required to account for the size of charge groups. In all other cases this parameter is automatically set to the longest cut-off distance.


Twin range cut-off's with neighborlist cut-off rlist and Coulomb cut-off rcoulomb, where rcoulombrlist.
Classical Ewald sum electrostatics. The real-space cut-off rcoulomb should be equal to rlist. Use e.g. rlist=0.9, rcoulomb=0.9. The highest magnitude of wave vectors used in reciprocal space is controlled by fourierspacing. The relative accuracy of direct/reciprocal space is controlled by ewald_rtol.
NOTE: Ewald scales as O(N3/2) and is thus extremely slow for large systems. It is included mainly for reference - in most cases PME will perform much better.
Fast Particle-Mesh Ewald electrostatics. Direct space is similar to the Ewald sum, while the reciprocal part is performed with FFTs. Grid dimensions are controlled with fourierspacing and the interpolation order with pme_order. With a grid spacing of 0.1 nm and cubic interpolation the electrostatic forces have an accuracy of 2-3*10-4. Since the error from the vdw-cutoff is larger than this you might try 0.15 nm. When running in parallel the interpolation parallelizes better than the FFT, so try decreasing grid dimensions while increasing interpolation.
Particle-Particle Particle-Mesh algorithm for long range electrostatic interactions. Use for example rlist=0.9, rcoulomb=0.9. The grid dimensions are controlled by fourierspacing. Reasonable grid spacing for PPPM is 0.05-0.1 nm. See Shift for the details of the particle-particle potential.
NOTE: PPPM is not functional in the current version, we plan to implement PPPM through a small modification of the PME code.
Reaction field with Coulomb cut-off rcoulomb, where rcoulomb &ge rlist. The dielectric constant beyond the cut-off is epsilon_rf. The dielectric constant can be set to infinity by setting epsilon_rf=0.
Generalized reaction field with Coulomb cut-off rcoulomb, where rcoulomb &ge rlist. The dielectric constant beyond the cut-off is epsilon_rf. The ionic strength is computed from the number of charged (i.e. with non zero charge) charge groups. The temperature for the GRF potential is set with ref_t [K].
In GROMACS normal reaction-field electrostatics leads to bad energy conservation. Reaction-Field-zero solves this by making the potential zero beyond the cut-off. It can only be used with an infinite dielectric constant (epsilon_rf=0), because only for that value the force vanishes at the cut-off. rlist should be 0.1 to 0.3 nm larger than rcoulomb to accommodate for the size of charge groups and diffusion between neighbor list updates. This, and the fact that table lookups are used instead of analytical functions make Reaction-Field-zero computationally more expensive than normal reaction-field.
The same as Reaction-Field, but implemented as in GROMACS versions before 3.3. No reaction-field correction is applied to excluded atom pairs and self pairs. The 1-4 interactions are calculated using a reaction-field. The missing correction due to the excluded pairs that do not have a 1-4 interaction is up to a few percent of the total electrostatic energy and causes a minor difference in the forces and the pressure.
Analogous to Shift for vdwtype. You might want to use Reaction-Field-zero instead, which has a similar potential shape, but has a physical interpretation and has better energies due to the exclusion correction terms.
The Coulomb potential is decreased over the whole range, using the definition from the Encad simulation package.
Analogous to Switch for vdwtype. Switching the Coulomb potential can lead to serious artifacts, advice: use Reaction-Field-zero instead.
mdrun will now expect to find a file table.xvg with user-defined potential functions for repulsion, dispersion and Coulomb. When pair interactions are present, mdrun also expects to find a file tablep.xvg for the pair interactions. When the same interactions should be used for non-bonded and pair interactions the user can specify the same file name for both table files. These files should contain 7 columns: the x value, f(x), -f'(x), g(x), -g'(x), h(x), -h'(x), where f(x) is the Coulomb function, g(x) the dispersion function and h(x) the repulsion function. When vdwtype is not set to User the values for g, -g', h and -h' are ignored. For the non-bonded interactions x values should run from 0 to the largest cut-off distance + table-extension and should be uniformly spaced. For the pair interactions the table length in the file will be used. The optimal spacing, which is used for non-user tables, is 0.002 [nm] when you run in single precision or 0.0005 [nm] when you run in double precision. The function value at x=0 is not important. More information is in the printed manual.
A combination of PME and a switch function for the direct-space part (see above). rcoulomb is allowed to be smaller than rlist. This is mainly useful constant energy simulations. For constant temperature simulations the advantage of improved energy conservation is usually outweighed by the small loss in accuracy of the electrostatics.
A combination of PME and user tables (see above). rcoulomb is allowed to be smaller than rlist. The PME mesh contribution is subtracted from the user table by mdrun. Because of this subtraction the user tables should contain about 10 decimal places.
A combination of PME-User and a switching function (see above). The switching function is applied to final particle-particle interaction, i.e. both to the user supplied function and the PME Mesh correction part.
rcoulomb_switch: (0) [nm]
where to start switching the Coulomb potential
rcoulomb: (1) [nm]
distance for the Coulomb cut-off
epsilon_r: (1)
The relative dielectric constant. A value of 0 means infinity.
epsilon_rf: (1)
The relative dielectric constant of the reaction field. This is only used with reaction-field electrostatics. A value of 0 means infinity.


Twin range cut-off's with neighbor list cut-off rlist and VdW cut-off rvdw, where rvdw &ge rlist.
The LJ (not Buckingham) potential is decreased over the whole range and the forces decay smoothly to zero between rvdw_switch and rvdw. The neighbor search cut-off rlist should be 0.1 to 0.3 nm larger than rvdw to accommodate for the size of charge groups and diffusion between neighbor list updates.
The LJ (not Buckingham) potential is normal out to rvdw_switch, after which it is switched off to reach zero at rvdw. Both the potential and force functions are continuously smooth, but be aware that all switch functions will give rise to a bulge (increase) in the force (since we are switching the potential). The neighbor search cut-off rlist should be 0.1 to 0.3 nm larger than rvdw to accommodate for the size of charge groups and diffusion between neighbor list updates.
The LJ (not Buckingham) potential is decreased over the whole range, using the definition from the Encad simulation package.
See user for coulombtype. The function value at x=0 is not important. When you want to use LJ correction, make sure that rvdw corresponds to the cut-off in the user-defined function. When coulombtype is not set to User the values for f and -f' are ignored.
rvdw_switch: (0) [nm]
where to start switching the LJ potential
rvdw: (1) [nm]
distance for the LJ or Buckingham cut-off
don't apply any correction
apply long range dispersion corrections for Energy and Pressure
apply long range dispersion corrections for Energy only


table-extension: (1) [nm]
Extension of the non-bonded potential lookup tables beyond the largest cut-off distance. The value should be large enough to account for charge group sizes and the diffusion between neighbor-list updates. Without user defined potential the same table length is used for the lookup tables for the 1-4 interactions, which are always tabulated irrespective of the use of tables for the non-bonded interactions.
When user tables are used for electrostatics and/or VdW, here one can give pairs of energy groups for which seperate user tables should be used. The two energy groups will be appended to the table file name, in order of their definition in energygrps, seperated by underscores. For example, if energygrps = Na Cl Sol and energygrp_table = Na Na Na Cl, mdrun will read table_Na_Na.xvg and table_Na_Cl.xvg in addition to the normal table.xvg which will be used for all other energy group pairs.


fourierspacing: (0.12) [nm]
The maximum grid spacing for the FFT grid when using PPPM or PME. For ordinary Ewald the spacing times the box dimensions determines the highest magnitude to use in each direction. In all cases each direction can be overridden by entering a non-zero value for fourier_n*. For optimizing the relative load of the particle-particle interactions and the mesh part of PME it is useful to know that the accuracy of the electrostatics remains nearly constant when the Coulomb cut-off and the PME grid spacing are scaled by the same factor.
fourier_nx (0) ; fourier_ny (0) ; fourier_nz: (0)
Highest magnitude of wave vectors in reciprocal space when using Ewald.
Grid size when using PPPM or PME. These values override fourierspacing per direction. The best choice is powers of 2, 3, 5 and 7. Avoid large primes.
pme_order (4)
Interpolation order for PME. 4 equals cubic interpolation. You might try 6/8/10 when running in parallel and simultaneously decrease grid dimension.
ewald_rtol (1e-5)
The relative strength of the Ewald-shifted direct potential at rcoulomb is given by ewald_rtol. Decreasing this will give a more accurate direct sum, but then you need more wave vectors for the reciprocal sum.
ewald_geometry: (3d)
The Ewald sum is performed in all three dimensions.
The reciprocal sum is still performed in 3d, but a force and potential correction applied in the z dimension to produce a pseudo-2d summation. If your system has a slab geometry in the x-y plane you can try to increase the z-dimension of the box (a box height of 3 times the slab height is usually ok) and use this option.
epsilon_surface: (0)
This controls the dipole correction to the Ewald summation in 3d. The default value of zero means it is turned off. Turn it on by setting it to the value of the relative permittivity of the imaginary surface around your infinite system. Be careful - you shouldn't use this if you have free mobile charges in your system. This value does not affect the slab 3DC variant of the long range corrections.
Don't calculate the optimal FFT plan for the grid at startup.
Calculate the optimal FFT plan for the grid at startup. This saves a few percent for long simulations, but takes a couple of minutes at start.

Temperature coupling

No temperature coupling.
Temperature coupling with a Berendsen-thermostat to a bath with temperature ref_t [K], with time constant tau_t [ps]. Several groups can be coupled separately, these are specified in the tc_grps field separated by spaces.
Temperature coupling using a Nose-Hoover extended ensemble. The reference temperature and coupling groups are selected as above, but in this case tau_t [ps] controls the period of the temperature fluctuations at equilibrium, which is slightly different from a relaxation time. For NVT simulations the conserved energy quantity is written to energy and log file.
Temperature coupling using velocity rescaling with a stochastic term (JCP 126, 014101). This thermostat is similar to Berendsen coupling, with the same scaling using tau_t, but the stochastic term ensures that a proper canonical ensemble is generated. The random seed is set with ld_seed. This thermostat works correctly even for tau_t=0. For NVT simulations the conserved energy quantity is written to the energy and log file.
nsttcouple: (-1)
The frequency for coupling the temperature. The default value of -1 sets nsttcouple equal to nstlist, unless nstlist≤0, then a value of 10 is used. For velocity Verlet integrators nsttcouple is set to 1.
nh-chain-length (10)
the number of chained Nose-Hoover thermostats for velocity Verlet integrators, the leap-frog md integrator only supports 1. Data for the NH chain variables is not printed to the .edr, but can be using the GMX_NOSEHOOVER_CHAINS environment variable
groups to couple separately to temperature bath
tau_t: [ps]
time constant for coupling (one for each group in tc_grps), -1 means no temperature coupling
ref_t: [K]
reference temperature for coupling (one for each group in tc_grps)

Pressure coupling

No pressure coupling. This means a fixed box size.
Exponential relaxation pressure coupling with time constant tau_p [ps]. The box is scaled every timestep. It has been argued that this does not yield a correct thermodynamic ensemble, but it is the most efficient way to scale a box at the beginning of a run.
Extended-ensemble pressure coupling where the box vectors are subject to an equation of motion. The equation of motion for the atoms is coupled to this. No instantaneous scaling takes place. As for Nose-Hoover temperature coupling the time constant tau_p [ps] is the period of pressure fluctuations at equilibrium. This is probably a better method when you want to apply pressure scaling during data collection, but beware that you can get very large oscillations if you are starting from a different pressure. For simulations where the exact fluctation of the NPT ensemble are important, or if the pressure coupling time is very short it may not be appropriate, as the previous time step pressure is used in some steps of the GROMACS implementation for the current time step pressure.
Martyna-Tuckerman-Tobias-Klein implementation, only useable with md-vv or md-vv-avek, very similar to Parrinello-Rahman. As for Nose-Hoover temperature coupling the time constant tau_p [ps] is the period of pressure fluctuations at equilibrium. This is probably a better method when you want to apply pressure scaling during data collection, but beware that you can get very large oscillations if you are starting from a different pressure. Currently only supports isotropic scaling.
Isotropic pressure coupling with time constant tau_p [ps]. The compressibility and reference pressure are set with compressibility [bar-1] and ref_p [bar], one value is needed.
Pressure coupling which is isotropic in the x and y direction, but different in the z direction. This can be useful for membrane simulations. 2 values are needed for x/y and z directions respectively.
Idem, but 6 values are needed for xx, yy, zz, xy/yx, xz/zx and yz/zy components, respectively. When the off-diagonal compressibilities are set to zero, a rectangular box will stay rectangular. Beware that anisotropic scaling can lead to extreme deformation of the simulation box.
Surface tension coupling for surfaces parallel to the xy-plane. Uses normal pressure coupling for the z-direction, while the surface tension is coupled to the x/y dimensions of the box. The first ref_p value is the reference surface tension times the number of surfaces [bar nm], the second value is the reference z-pressure [bar]. The two compressibility [bar-1] values are the compressibility in the x/y and z direction respectively. The value for the z-compressibility should be reasonably accurate since it influences the convergence of the surface-tension, it can also be set to zero to have a box with constant height.
nstpcouple: (-1)
The frequency for coupling the pressure. The default value of -1 sets nstpcouple equal to nstlist, unless nstlist ≤0, then a value of 10 is used. For velocity Verlet integrators nstpcouple is set to 1.
tau_p: (1) [ps]
time constant for coupling
compressibility: [bar-1]
compressibility (NOTE: this is now really in bar-1) For water at 1 atm and 300 K the compressibility is 4.5e-5 [bar-1].
ref_p: [bar]
reference pressure for coupling
The reference coordinates for position restraints are not modified. Note that with this option the virial and pressure will depend on the absolute positions of the reference coordinates.
The reference coordinates are scaled with the scaling matrix of the pressure coupling.
Scale the center of mass of the reference coordinates with the scaling matrix of the pressure coupling. The vectors of each reference coordinate to the center of mass are not scaled. Only one COM is used, even when there are multiple molecules with position restraints. For calculating the COM of the reference coordinates in the starting configuration, periodic boundary conditions are not taken into account.

Simulated annealing

Simulated annealing is controlled separately for each temperature group in GROMACS. The reference temperature is a piecewise linear function, but you can use an arbitrary number of points for each group, and choose either a single sequence or a periodic behaviour for each group. The actual annealing is performed by dynamically changing the reference temperature used in the thermostat algorithm selected, so remember that the system will usually not instantaneously reach the reference temperature!
Type of annealing for each temperature group
No simulated annealing - just couple to reference temperature value.
A single sequence of annealing points. If your simulation is longer than the time of the last point, the temperature will be coupled to this constant value after the annealing sequence has reached the last time point.
The annealing will start over at the first reference point once the last reference time is reached. This is repeated until the simulation ends.
A list with the number of annealing reference/control points used for each temperature group. Use 0 for groups that are not annealed. The number of entries should equal the number of temperature groups.
List of times at the annealing reference/control points for each group. If you are using periodic annealing, the times will be used modulo the last value, i.e. if the values are 0, 5, 10, and 15, the coupling will restart at the 0ps value after 15ps, 30ps, 45ps, etc. The number of entries should equal the sum of the numbers given in annealing_npoints.
List of temperatures at the annealing reference/control points for each group. The number of entries should equal the sum of the numbers given in annealing_npoints.

Confused? OK, let's use an example. Assume you have two temperature groups, set the group selections to annealing = single periodic, the number of points of each group to annealing_npoints = 3 4, the times to annealing_time = 0 3 6 0 2 4 6 and finally temperatures to annealing_temp = 298 280 270 298 320 320 298. The first group will be coupled to 298K at 0ps, but the reference temperature will drop linearly to reach 280K at 3ps, and then linearly between 280K and 270K from 3ps to 6ps. After this is stays constant, at 270K. The second group is coupled to 298K at 0ps, it increases linearly to 320K at 2ps, where it stays constant until 4ps. Between 4ps and 6ps it decreases to 298K, and then it starts over with the same pattern again, i.e. rising linearly from 298K to 320K between 6ps and 8ps. Check the summary printed by grompp if you are unsure!

Velocity generation

Do not generate velocities. The velocities are set to zero when there are no velocities in the input structure file.
Generate velocities in grompp according to a Maxwell distribution at temperature gen_temp [K], with random seed gen_seed. This is only meaningful with integrator md.
gen_temp: (300) [K]
temperature for Maxwell distribution
gen_seed: (173529) [integer]
used to initialize random generator for random velocities, when gen_seed is set to -1, the seed is calculated from the process ID number.


No constraints except for those defined explicitly in the topology, i.e. bonds are represented by a harmonic (or other) potential or a Morse potential (depending on the setting of morse) and angles by a harmonic (or other) potential.
Convert the bonds with H-atoms to constraints.
Convert all bonds to constraints.
Convert all bonds and additionally the angles that involve H-atoms to bond-constraints.
Convert all bonds and angles to bond-constraints.
LINear Constraint Solver. With domain decomposition the parallel version P-LINCS is used. The accuracy in set with lincs_order, which sets the number of matrices in the expansion for the matrix inversion. After the matrix inversion correction the algorithm does an iterative correction to compensate for lengthening due to rotation. The number of such iterations can be controlled with lincs_iter. The root mean square relative constraint deviation is printed to the log file every nstlog steps. If a bond rotates more than lincs_warnangle [degrees] in one step, a warning will be printed both to the log file and to stderr. LINCS should not be used with coupled angle constraints.
SHAKE is slightly slower and less stable than LINCS, but does work with angle constraints. The relative tolerance is set with shake_tol, 0.0001 is a good value for ``normal'' MD. SHAKE does not support constraints between atoms on different nodes, thus it can not be used with domain decompositon when inter charge-group constraints are present. SHAKE can not be used with energy minimization.
This option was formerly known as unconstrained_start.
apply constraints to the start configuration and reset shells
do not apply constraints to the start configuration and do not reset shells, useful for exact coninuation and reruns
shake_tol: (0.0001)
relative tolerance for SHAKE
lincs_order: (4)
Highest order in the expansion of the constraint coupling matrix. When constraints form triangles, an additional expansion of the same order is applied on top of the normal expansion only for the couplings within such triangles. For ``normal'' MD simulations an order of 4 usually suffices, 6 is needed for large time-steps with virtual sites or BD. For accurate energy minimization an order of 8 or more might be required. With domain decomposition, the cell size is limited by the distance spanned by lincs_order+1 constraints. When one wants to scale further than this limit, one can decrease lincs_order and increase lincs_iter, since the accuracy does not deteriorate when (1+lincs_iter)*lincs_order remains constant.
lincs_iter: (1)
Number of iterations to correct for rotational lengthening in LINCS. For normal runs a single step is sufficient, but for NVE runs where you want to conserve energy accurately or for accurate energy minimization you might want to increase it to 2.
lincs_warnangle: (30) [degrees]
maximum angle that a bond can rotate before LINCS will complain
bonds are represented by a harmonic potential
bonds are represented by a Morse potential

Energy group exclusions

Pairs of energy groups for which all non-bonded interactions are excluded. An example: if you have two energy groups Protein and SOL, specifying
energygrp_excl = Protein Protein  SOL SOL
would give only the non-bonded interactions between the protein and the solvent. This is especially useful for speeding up energy calculations with mdrun -rerun and for excluding interactions within frozen groups.


nwall: 0
When set to 1 there is a wall at z=0, when set to 2 there is also a wall at z=z_box. Walls can only be used with pbc=xy. When set to 2 pressure coupling and Ewald summation can be used (it is usually best to use semiisotropic pressure coupling with the x/y compressibility set to 0, as otherwise the surface area will change). Walls interact wit the rest of the system through an optional wall_atomtype. Energy groups wall0 and wall1 (for nwall=2) are added automatically to monitor the interaction of energy groups with each wall. The center of mass motion removal will be turned off in the z-direction.
the atom type name in the force field for each wall. By (for example) defining a special wall atom type in the topology with its own combination rules, this allows for independent tuning of the interaction of each atomtype with the walls.
LJ integrated over the volume behind the wall: 9-3 potential
LJ integrated over the wall surface: 10-4 potential
direct LJ potential with the z distance from the wall
user defined potentials indexed with the z distance from the wall, the tables are read analogously to the energygrp_table option, where the first name is for a ``normal'' energy group and the second name is wall0 or wall1, only the dispersion and repulsion columns are used
wall_r_linpot: -1 (nm)
Below this distance from the wall the potential is continued linearly and thus the force is constant. Setting this option to a postive value is especially useful for equilibration when some atoms are beyond a wall. When the value is ≤0 (<0 for wall_type=table), a fatal error is generated when atoms are beyond a wall.
wall_density: [nm-3/nm-2]
the number density of the atoms for each wall for wall types 9-3 and 10-4
wall_ewald_zfac: 3
The scaling factor for the third box vector for Ewald summation only, the minimum is 2. Ewald summation can only be used with nwall=2, where one should use ewald_geometry=3dc. The empty layer in the box serves to decrease the unphysical Coulomb interaction between periodic images.

COM pulling

No center of mass pulling. All the following pull options will be ignored (and if present in the .mdp file, they unfortunately generate warnings)
Center of mass pulling using an umbrella potential between the reference group and one or more groups.
Center of mass pulling using a constraint between the reference group and one or more groups. The setup is identical to the option umbrella, except for the fact that a rigid constraint is applied instead of a harmonic potential.
Center of mass pulling using a linear potential and therefore a constant force. For this option there is no reference position and therefore the parameters pull_init and pull_rate are not used.
Pull along the vector connecting the two groups. Components can be selected with pull_dim.
Pull in the direction of pull_vec.
As direction, but allows the distance to be larger than half the box size. With this geometry the box should not be dynamic (e.g. no pressure scaling) in the pull dimensions and the pull force is not added to virial.
Designed for pulling with respect to a layer where the reference COM is given by a local cylindrical part of the reference group. The pulling is in the direction of pull_vec. From the reference group a cylinder is selected around the axis going through the pull group with direction pull_vec using two radii. The radius pull_r1 gives the radius within which all the relative weights are one, between pull_r1 and pull_r0 the weights are switched to zero. Mass weighting is also used. Note that the radii should be smaller than half the box size. For tilted cylinders they should be even smaller than half the box size since the distance of an atom in the reference group from the COM of the pull group has both a radial and an axial component.
Pull to the position of the reference group plus pull_init + time*pull_rate*pull_vec.
pull_dim: (Y Y Y)
the distance components to be used with geometry distance and position, and also sets which components are printed to the output files
pull_r1: (1) [nm]
the inner radius of the cylinder for geometry cylinder
pull_r0: (1) [nm]
the outer radius of the cylinder for geometry cylinder
pull_constr_tol: (1e-6)
the relative constraint tolerance for constraint pulling
do not modify pull_init
add the COM distance of the starting conformation to pull_init
pull_nstxout: (10)
frequency for writing out the COMs of all the pull group
pull_nstfout: (1)
frequency for writing out the force of all the pulled group
pull_ngroups: (1)
The number of pull groups, not including the reference group. If there is only one group, there is no difference in treatment of the reference and pulled group (except with the cylinder geometry). Below only the pull options for the reference group (ending on 0) and the first group (ending on 1) are given, further groups work analogously, but with the number 1 replaced by the group number.
The name of the reference group. When this is empty an absolute reference of (0,0,0) is used. With an absolute reference the system is no longer translation invariant and one should think about what to do with the center of mass motion.
see pull_weights1
pull_pbcatom0: (0)
see pull_pbcatom1
The name of the pull group.
Optional relative weights which are multiplied with the masses of the atoms to give the total weight for the COM. The number should be 0, meaning all 1, or the number of atoms in the pull group.
pull_pbcatom1: (0)
The reference atom for the treatment of periodic boundary conditions inside the group (this has no effect on the treatment of the pbc between groups). This option is only important when the diameter of the pull group is larger than half the shortest box vector. For determining the COM, all atoms in the group are put at their periodic image which is closest to pull_pbcatom1. A value of 0 means that the middle atom (number wise) is used. This parameter is not used with geometry cylinder. A value of -1 turns on cosine weighting, which is useful for a group of molecules in a periodic system, e.g. a water slab (see Engin et al. J. Chem. Phys. B 2010).
pull_vec1: (0.0 0.0 0.0)
The pull direction. grompp normalizes the vector.
pull_init1: (0.0) / (0.0 0.0 0.0) [nm]
The reference distance at t=0. This is a single value, except for geometry position which uses a vector.
pull_rate1: (0) [nm/ps]
The rate of change of the reference position.
pull_k1: (0) [kJ mol-1 nm-2] / [kJ mol-1 nm-1]
The force constant. For umbrella pulling this is the harmonic force constant in [kJ mol-1 nm-2]. For constant force pulling this is the force constant of the linear potential, and thus minus (!) the constant force in [kJ mol-1 nm-1].
pull_kB1: (pull_k1) [kJ mol-1 nm-2] / [kJ mol-1 nm-1]
As pull_k1, but for state B. This is only used when free_energy is turned on. The force constant is then (1 - lambda)*pull_k1 + lambda*pull_kB1.

NMR refinement

no distance restraints (ignore distance restraint information in topology file)
simple (per-molecule) distance restraints, ensemble averaging can be performed with mdrun -multi where the environment variable GMX_DISRE_ENSEMBLE_SIZE sets the number of systems within each ensemble (usually equal to the mdrun -multi value)
distance restraints over an ensemble of molecules in one simulation box, should only be used for special cases, such as dimers (this option is not fuctional in the current version of GROMACS)
the forces are the derivative of the restraint potential, this results in an r-7 weighting of the atom pairs
divide the restraint force equally over all atom pairs in the restraint
the violation used in the calculation of the restraint force is the time averaged violation
the violation used in the calculation of the restraint force is the square root of the time averaged violation times the instantaneous violation
disre_fc: (1000) [kJ mol-1 nm-2]
force constant for distance restraints, which is multiplied by a (possibly) different factor for each restraint
disre_tau: (0) [ps]
time constant for distance restraints running average
nstdisreout: (100) [steps]
frequency to write the running time averaged and instantaneous distances of all atom pairs involved in restraints to the energy file (can make the energy file very large)
no orientation restraints (ignore orientation restraint information in topology file)
use orientation restraints, ensemble averaging can be performed with mdrun -multi
orire_fc: (0) [kJ mol]
force constant for orientation restraints, which is multiplied by a (possibly) different factor for each restraint, can be set to zero to obtain the orientations from a free simulation
orire_tau: (0) [ps]
time constant for orientation restraints running average
fit group for orientation restraining, for a protein backbone is a good choice
nstorireout: (100) [steps]
frequency to write the running time averaged and instantaneous orientations for all restraints and the molecular order tensor to the energy file (can make the energy file very large)

Free energy calculations

Only use topology A.
Interpolate between topology A (lambda=0) to topology B (lambda=1) and write the derivative of the Hamiltonian with respect to lambda (as specified with dhdl_derivatives), or the Hamiltonian differences with respect to other lambda values (as specified with foreign_lambda) to the energy file and/or to dhdl.xvg, where they can be processed by, for example g_bar. The potentials, bond-lengths and angles are interpolated linearly as described in the manual. When sc_alpha is larger than zero, soft-core potentials are used for the LJ and Coulomb interactions.
init_lambda: (0)
starting value for lambda
delta_lambda: (0)
increment per time step for lambda
foreign_lambda: ()
Zero, one or more lambda values for which Delta H values will be determined and written to dhdl.xvg every nstdhdl steps. Free energy differences between different lambda values can then be determined with g_bar.
dhdl_derivatives: (yes)
If yes (the default), the derivatives of the Hamiltonian with respect to lambda at each nstdhdl step are written out. These values are needed for interpolation of linear energy differences with g_bar (although the same can also be achieved with the right foreign lambda setting, that may not be as flexible), or with thermodynamic integration
sc_alpha: (0)
the soft-core parameter, a value of 0 results in linear interpolation of the LJ and Coulomb interactions
sc_power: (0)
the power for lambda in the soft-core function, only the values 1 and 2 are supported
sc_sigma: (0.3) [nm]
the soft-core sigma for particles which have a C6 or C12 parameter equal to zero or a sigma smaller than sc_sigma
Here one can supply a molecule type (as defined in the topology) for calculating solvation or coupling free energies. There is a special option system that couples all molecule types in the system. This can be useful for equilibrating a system starting from (nearly) random coordinates. free_energy has to be turned on. The Van der Waals interactions and/or charges in this molecule type can be turned on or off between lambda=0 and lambda=1, depending on the settings of couple-lambda0 and couple-lambda1. If you want to decouple one of several copies of a molecule, you need to copy and rename the molecule definition in the topology.
all interactions are on at lambda=0
the charges are zero (no Coulomb interactions) at lambda=0
the Van der Waals interactions are turned at lambda=0; soft-core interactions will be required to avoid singularities
the Van der Waals interactions are turned off and the charges are zero at lambda=0; soft-core interactions will be required to avoid singularities
analogous to couple-lambda1, but for lambda=1
All intra-molecular non-bonded interactions for moleculetype couple-moltype are replaced by exclusions and explicit pair interactions. In this manner the decoupled state of the molecule corresponds to the proper vacuum state without periodicity effects.
The intra-molecular Van der Waals and Coulomb interactions are also turned on/off. This can be useful for partitioning free-energies of relatively large molecules, where the intra-molecular non-bonded interactions might lead to kinetically trapped vacuum conformations. The 1-4 pair interactions are not turned off.
nstdhdl: (10)
the frequency for writing dH/dlambda and possibly Delta H to dhdl.xvg, 0 means no ouput, should be a multiple of nstcalcenergy
separate_dhdl_file: (yes)
the free energy values that are calculated (as specified with the foreign-lambda and dhdl_derivatives settings) are written out to a separate file, with the default name dhdl.xvg. This file can be used directly with g_bar.
The free energy values are written out to the energy output file (ener.edr, in accumulated blocks at every nstenergy steps), where they can be extracted with g_energy or used directly with g_bar.
dh_hist_size: (0)
If nonzero, specifies the size of the histogram into which the Delta H values (specified with foreign_lambda) and the derivative dH/dl values are binned, and written to ener.edr. This can be used to save disk space while calculating free energy differences. One histogram gets written for each foreign lambda and two for the dH/dl, at every nstenergy step. Be aware that incorrect histogram settings (too small size or too wide bins) can introduce errors. Do not use histograms unless you're certain you need it.
dh_hist_spacing (0.1)
Specifies the bin width of the histograms, in energy units. Used in conjunction with dh_hist_size. This size limits the accuracy with which free energies can be calculated. Do not use histograms unless you're certain you need it.

Non-equilibrium MD

groups for constant acceleration (e.g.: Protein Sol) all atoms in groups Protein and Sol will experience constant acceleration as specified in the accelerate line
accelerate: (0) [nm ps-2]
acceleration for acc_grps; x, y and z for each group (e.g. 0.1 0.0 0.0 -0.1 0.0 0.0 means that first group has constant acceleration of 0.1 nm ps-2 in X direction, second group the opposite).
Groups that are to be frozen (i.e. their X, Y, and/or Z position will not be updated; e.g. Lipid SOL). freezedim specifies for which dimension the freezing applies. To avoid spurious contibrutions to the virial and pressure due to large forces between completely frozen atoms you need to use energy group exclusions, this also saves computing time. Note that frozen coordinates are not subject to pressure scaling.
dimensions for which groups in freezegrps should be frozen, specify Y or N for X, Y and Z and for each group (e.g. Y Y N N N N means that particles in the first group can move only in Z direction. The particles in the second group can move in any direction).
cos_acceleration: (0) [nm ps-2]
the amplitude of the acceleration profile for calculating the viscosity. The acceleration is in the X-direction and the magnitude is cos_acceleration cos(2 pi z/boxheight). Two terms are added to the energy file: the amplitude of the velocity profile and 1/viscosity.
deform: (0 0 0 0 0 0) [nm ps-1]
The velocities of deformation for the box elements: a(x) b(y) c(z) b(x) c(x) c(y). Each step the box elements for which deform is non-zero are calculated as: box(ts)+(t-ts)*deform, off-diagonal elements are corrected for periodicity. The coordinates are transformed accordingly. Frozen degrees of freedom are (purposely) also transformed. The time ts is set to t at the first step and at steps at which x and v are written to trajectory to ensure exact restarts. Deformation can be used together with semiisotropic or anisotropic pressure coupling when the appropriate compressibilities are set to zero. The diagonal elements can be used to strain a solid. The off-diagonal elements can be used to shear a solid or a liquid.

Electric fields

E_x ; E_y ; E_z:
If you want to use an electric field in a direction, enter 3 numbers after the appropriate E_*, the first number: the number of cosines, only 1 is implemented (with frequency 0) so enter 1, the second number: the strength of the electric field in V nm-1, the third number: the phase of the cosine, you can enter any number here since a cosine of frequency zero has no phase.
E_xt; E_yt; E_zt:
not implemented yet

Mixed quantum/classical molecular dynamics

Do a QM/MM simulation. Several groups can be described at different QM levels separately. These are specified in the QMMM-grps field separated by spaces. The level of ab initio theory at which the groups are described is speficied by QMmethod and QMbasis Fields. Describing the groups at different levels of theory is only possible with the ONIOM QM/MM scheme, specified by QMMMscheme.
groups to be descibed at the QM level
normal QM/MM. There can only be one QMMM-grps that is modelled at the QMmethod and QMbasis level of ab initio theory. The rest of the system is described at the MM level. The QM and MM subsystems interact as follows: MM point charges are included in the QM one-electron hamiltonian and all Lennard-Jones interactions are described at the MM level.
The interaction between the subsystem is described using the ONIOM method by Morokuma and co-workers. There can be more than one QMMM-grps each modeled at a different level of QM theory (QMmethod and QMbasis).
QMmethod: (RHF)
Method used to compute the energy and gradients on the QM atoms. Available methods are AM1, PM3, RHF, UHF, DFT, B3LYP, MP2, CASSCF, and MMVB. For CASSCF, the number of electrons and orbitals included in the active space is specified by CASelectrons and CASorbitals.
QMbasis: (STO-3G)
Basisset used to expand the electronic wavefuntion. Only gaussian bassisets are currently available, i.e. STO-3G, 3-21G, 3-21G*, 3-21+G*, 6-21G, 6-31G, 6-31G*, 6-31+G*, and 6-311G.
QMcharge: (0) [integer]
The total charge in e of the QMMM-grps. In case there are more than one QMMM-grps, the total charge of each ONIOM layer needs to be specified separately.
QMmult: (1) [integer]
The multiplicity of the QMMM-grps. In case there are more than one QMMM-grps, the multiplicity of each ONIOM layer needs to be specified separately.
CASorbitals: (0) [integer]
The number of orbitals to be included in the active space when doing a CASSCF computation.
CASelectrons: (0) [integer]
The number of electrons to be included in the active space when doing a CASSCF computation.
No surface hopping. The system is always in the electronic ground-state.
Do a QM/MM MD simulation on the excited state-potential energy surface and enforce a diabatic hop to the ground-state when the system hits the conical intersection hyperline in the course the simulation. This option only works in combination with the CASSCF method.

Implicit solvent

No implicit solvent
Do a simulation with implicit solvent using the Generalized Born formalism. Three different methods for calculating the Born radii are available, Still, HCT and OBC. These are specified with the gb_algorithm field. The non-polar solvation is specified with the sa_algorithm field.
Use the Still method to calculate the Born radii
Use the Hawkins-Cramer-Truhlar method to calculate the Born radii
Use the Onufriev-Bashford-Case method to calculate the Born radii
nstgbradii: (1) [steps]
Frequency to (re)-calculate the Born radii. For most practial purposes, setting a value larger than 1 violates energy conservation and leads to unstable trajectories.
rgbradii: (1.0) [nm]
Cut-off for the calculation of the Born radii. Currently must be equal to rlist
gb_epsilon_solvent: (80)
Dielectric constant for the implicit solvent
gb_saltconc: (0) [M]
Salt concentration for implicit solvent models, currently not used
gb_obc_alpha (1); gb_obc_beta (0.8); gb_obc_gamma (4.85);
Scale factors for the OBC model. Default values are OBC(II). Values for OBC(I) are 0.8, 0 and 2.91 respectively
gb_dielectric_offset: (0.009) [nm]
Distance for the di-electric offset when calculating the Born radii. This is the offset between the center of each atom the center of the polarization energy for the corresponding atom
Use an Ace-type approximation (default)
No non-polar solvation calculation done. For GBSA only the polar part gets calculated
sa_surface_tension: [kJ mol-1 nm-2]
Default value for surface tension with SA algorithms. The default value is -1; Note that if this default value is not changed it will be overridden by grompp using values that are specific for the choice of radii algorithm (0.0049 kcal/mol/Angstrom2 for Still, 0.0054 kcal/mol/Angstrom2 for HCT/OBC) Setting it to 0 will while using an sa_algorithm other than None means no non-polar calculations are done.

User defined thingies

user1_grps; user2_grps:
userint1 (0); userint2 (0); userint3 (0); userint4 (0)
userreal1 (0); userreal2 (0); userreal3 (0); userreal4 (0)
These you can use if you modify code. You can pass integers and reals to your subroutine. Check the inputrec definition in src/include/types/inputrec.h