(Note: These tutorials are meant to provide illustrative examples of how to use the AMBER software suite to carry out simulations that can be run on a simple workstation in a reasonable period of time. They do not necessarily provide the optimal choice of parameters or methods for the particular application area.)
Copyright Ross Walker 2004

TUTORIAL 7 - SECTION 6

The Nudged Elastic Band Approach to Finding the Lowest Energy
Pathway Between two States

By Ross Walker

Return to Index Page

6. Slow Cooling

The final stage is to slowly cool the system down so that the images freeze out along the pathway. We will do the cooling in 2 stages. In the first stage we shall gradually cool down in 50 K steps over 120 ps using the following profile:

0 to 10ps : 300 K -> 250 K
10 to 20ps : Equil at 250 K
20 to 30ps : 250 K -> 200 K
30 to 40ps : Equil at 200 K
40 to 50ps : 200 -> 150 K
50 to 60ps : Equil at 150 K
60 to 70ps : 150 K -> 100 K
70 to 80ps : Equil at 100 K
80 to 90ps : 100 K -> 50 K
90 to 100ps : Equil at 50 K
100 to 110 ps : 50 K -> 0 K
110 to 120 ps : Equil at 0 K

Note since we are now back at 300 K and below we can safely switch back to using a 1 fs time step. Here is the input file:

md-ala-neb_cool1.in

120ps NEB ALA-ALA cool stage 1
 &cntrl
  imin = 0, irest = 1, ntx=5,
  ntc=1, ntf=1,
  ntpr=500, ntwx=10000,
  ntb = 0, cut = 999.0, rgbmax=999.0,
  igb = 1, saltcon=0.2,
  nstlim = 120000, nscm= 0,
  dt = 0.001,
  ntt = 3, gamma_ln=1000.0,
  temp0=300,
  ineb = 1,skmin = 50,skmax = 50,
  nmropt=1
 /
 &wt type='TEMP0', istep1=0,istep2=10000,
   value1=300.0, value2=250.0
 /
 &wt type='TEMP0', istep1=10001,istep2=20000,
   value1=250.0, value2=250.0
 /
 &wt type='TEMP0', istep1=20001,istep2=30000,
   value1=250.0, value2=200.0
 /
 &wt type='TEMP0', istep1=30001,istep2=40000,
   value1=200.0, value2=200.0
 /
 &wt type='TEMP0', istep1=40001,istep2=50000,
   value1=200.0, value2=150.0
 /
 &wt type='TEMP0', istep1=50001,istep2=60000,
   value1=150.0, value2=150.0
 /
 &wt type='TEMP0', istep1=60001,istep2=70000,
   value1=150.0, value2=100.0
 /
 &wt type='TEMP0', istep1=70001,istep2=80000,
   value1=100.0, value2=100.0
 /
 &wt type='TEMP0', istep1=80001,istep2=90000,
   value1=100.0, value2=50.0
 /
 &wt type='TEMP0', istep1=90001,istep2=100000,
   value1=50.0, value2=50.0
 /
 &wt type='TEMP0', istep1=100001,istep2=110000,
   value1=50.0, value2=0.0
 /
 &wt type='TEMP0', istep1=110001,istep2=120000,
   value1=0.0, value2=0.0
 /
 &wt type='END'
 /

Lets run this now, it will take about 13 minutes to run on 1 cpu of a Pentium-D 3.2GHz machine:

$AMBERHOME/exe/sander.PIMD -O -i md-ala-neb_cool1.in -o md-ala-neb_cool1.out -p neb.prmtop -c md-ala-neb_anneal.rst -r md-ala-neb_cool1.rst -x md-ala-neb_cool1.mdcrd

Input Files: md-ala-neb_cool1.in, neb.prmtop, md-ala-neb_anneal.rst
Output Files: md-ala-neb_cool1.out, md-ala-neb_cool1.rst, md-ala-neb_cool1.mdcrd

Normally we would be done at this point, however, the NEB method is extremely sensitive to there being any kinetic energy remaining. Even a small amount of kinetic energy can greatly effect the predicted pathway. E.g. take a look at the final results from the cooling run above:

 NSTEP =   120000   TIME(PS) =     540.000  TEMP(K) =     0.00  PRESS =     0.0
 Etot   =      -888.6417  EKtot   =         0.0046  EPtot      =      -888.6463
 BOND   =        16.1334  ANGLE   =        72.4586  DIHED      =       316.1304
 1-4 NB =        71.5424  1-4 EEL =      1312.9108  VDWAALS    =       -37.4381
 EELEC  =     -2093.7117  EGB     =      -546.6721  RESTRAINT  =         0.0000

As you can see there is still 0.0046 KCal/mol of kinetic energy present. This might seem like a tiny amount but it is essential that we try to remove as much of this as possible. To do this we will do one final 200ps long stage of cooling where we will set temp0=0.0 K and we will turn on the quenched velocity verlet method that was discussed in section 1 (vv=1). This should remove the last of the kinetic energy. Here is the input file:

md-ala-neb_cool2.in

50ps NEB ALA-ALA cool stage 2
 &cntrl
  imin = 0, irest = 1, ntx=5,
  ntc=1, ntf=1,
  ntpr=100, ntwx=10000,
  ntb = 0, cut = 999.0, rgbmax=999.0,
  igb = 1, saltcon=0.2,
  nstlim = 200000, nscm= 0,
  dt = 0.001,
  ntt = 3, gamma_ln=1000.0,
  temp0=0.0,
  ineb = 1,skmin = 50,skmax = 50,
  vv=1,vfac=0.1
 /

This will take about 22 minutes to run on 1 cpu of a Pentium-D 3.2GHz machine:

$AMBERHOME/exe/sander.PIMD -O -i md-ala-neb_cool2.in -o md-ala-neb_cool2.out -p neb.prmtop -c md-ala-neb_cool1.rst -r md-ala-neb_cool2.rst -x md-ala-neb_cool2.mdcrd

Input Files: md-ala-neb_cool2.in, neb.prmtop, md-ala-neb_cool1.rst
Output Files: md-ala-neb_cool2.out, md-ala-neb_cool2.rst, md-ala-neb_cool2.mdcrd

Lets take a look at the final step of this final cooling:

 NSTEP =   200000   TIME(PS) =     740.000  TEMP(K) =     0.00  PRESS =     0.0
 Etot   =      -896.9015  EKtot   =         0.0021  EPtot      =      -896.9036
 BOND   =        16.1942  ANGLE   =        71.7284  DIHED      =       309.1552
 1-4 NB =        71.5022  1-4 EEL =      1311.8589  VDWAALS    =       -37.1966
 EELEC  =     -2104.9263  EGB     =      -535.2198  RESTRAINT  =         0.0000
NEB replicate breakdown:
Energy for replicate   1 =      -33.7654
Energy for replicate   2 =      -33.0651
Energy for replicate   3 =      -32.6502
Energy for replicate   4 =      -32.1257
Energy for replicate   5 =      -31.4913
Energy for replicate   6 =      -30.7533
Energy for replicate   7 =      -29.9331
Energy for replicate   8 =      -29.0721
Energy for replicate   9 =      -28.2335
Energy for replicate  10 =      -27.4946
Energy for replicate  11 =      -26.9349
Energy for replicate  12 =      -26.6137
Energy for replicate  13 =      -26.5543
Energy for replicate  14 =      -26.7422
Energy for replicate  15 =      -27.1331
Energy for replicate  16 =      -27.6610
Energy for replicate  17 =      -28.2468
Energy for replicate  18 =      -28.8051
Energy for replicate  19 =      -29.2529
Energy for replicate  20 =      -29.5457
Energy for replicate  21 =      -29.5970
Energy for replicate  22 =      -29.5503
Energy for replicate  23 =      -29.7765
Energy for replicate  24 =      -30.4228
Energy for replicate  25 =      -31.1069
Energy for replicate  26 =      -31.6143
Energy for replicate  27 =      -31.9435
Energy for replicate  28 =      -32.1101
Energy for replicate  29 =      -32.3055
Energy for replicate  30 =      -32.4029
Total Energy of replicates =     -896.9036
NEB RMS =      0.138640
 ------------------------------------------------------------------------------

Notice how the energies fluctuate along the pathway. It is probably easier to visualise this if we plot the energies. I have manually extracted these here: final_pathway_energies.dat.

As you can see there is 1 major maxima along this pathway and the overall barrier height is approximately 7 KCal/mol in one direction and 5 KCal/mol in the reverse direction.

Next take a look at the structures in the final restart file in VMD. Load the neb.prmtop file (select parm7 in vmd 1.8.3 or AMBER7 Parm in vmd 1.8.4) and then into this molecule load the md-ala-neb_cool2.rst file (select rst7 in vmd 1.8.3 or AMBER7 Restart in vmd 1.8.4):

Notice how a number of the structures have the N-H groups on top of each other. This corresponds to moving along the valley on the right hand side of the energy surface. Here the C-O group is moving but the N-H group is remaining fairly static. The final step is to extract the pathway data from this restart file, create an animation of the pathway and then plot the position of each of the replicas on our energy surface.

Return to section 5: Simulated Annealing

Proceed to section 7: Extract Final Pathway