Short Course on Molecular
Dynamics Simulation
Lecture 4:
Temperature Control
Professor A. Martini
Purdue University
High Level Course Outline
1. MD Basics
2. Potential Energy Functions
3. Integration Algorithms
4. Temperature Control
5. Boundary Conditions
6. Neighbor Lists
7. Initialization and Equilibrium
8. Extracting Static Properties
9. Extracting Dynamic Properties
10. Non-Equilibrium MD
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Temperature Control
Macroscopic systems
– Conduction
– Convection
– Radiation
Model atomic system
– Numerical temperature control (thermostats)
• Velocity scaling
• Berendsen
• Andersen
• Langevin dynamics
• Nose-Hoover
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Velocity Scaling
Energy and temperature in an N-body system
H ( p, r ) = Ek ( p N ) + U (rN ) = const
1 N v2 3
Ek = ∑
2m i
pi = NkbT
2
N
v 2
∑ (mi vi ) = 3MNkbT
i
N
M v v
∑ vi ⋅ vi
N
v2
M ∑ vi = 3 NkbT T=
3 Nkb i
i
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Velocity Scaling
Controlling temperature during an
NVE simulation
N
M v new v new
T new 3 Nkb
∑ vi ⋅ vi
= i
T old M N
v old v old
3 Nkb
∑ i ⋅ vi
v
i
N
v new v new
T new
∑ vi ⋅ vi current
= i new
v
i =v
current
i T new
T
old N
T v old v old
∑ vi ⋅ vi
i
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Berendsen
Berendsen; adjustable parameter determines
how tightly bath and system are coupled
– Simple velocity scaling as the coupling
parameter approaches the integration
timestep
– Suggested value Δt / τ ~ 0.0025
dT (t ) 1 Δt
= (Tbath − T (t ) ) ΔT = (Tbath − T (t ) )
dt τ τ
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Andersen
Andersen heat bath Æ stochastic impulsive
forces acting on randomly selected particles
– Coupling strength Æ frequency of stochastic
collisions, ν
1. Start with a set of positions and momenta
2. Integrate equations of motion
3. Select particles to undergo collision with heat
bath; probability of selection νΔt
4. The velocities of particles selected for collision
are taken from a Boltzmann distribution at the
desired temperature
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Andersen
Coupling strength Æ frequency of stochastic
collisions, ν
– The value of ν is related to the intermolecular
collision frequency νc (a function of the density and
thermal conductivity)
νc
ν=
N2 3
– For each particle, a random number is selected
between 0 and 1
– If that number is less than νΔt, its momenta are
reset
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Langevin Dynamics
Langevin method simulates interactions with
a solvent
Two terms added to the equation of motion
– Friction term removes energy Æ frictional drag on
the system moving through a solvent
– Random term adds energy Æ
collisions/interactions with solvent molecules
v v v v
ma = F − mγv + R
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Nosé-Hoover
Includes the heat bath explicitly as an
additional degree of freedom
Widely used algorithm
Introduce an artificial variable s that plays the
role of a time scaling parameter
– “mass” Q
.
– “velocity” s
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
Nosé-Hoover
Nosé-Hoover Hamiltonian
v
N
pi2 vN ps2
H Nose − Hoover =∑ 2
+ U (r ) + + ( f + 1)kbT ln s
i =1 2mi s 2Q
Magnitude of Q controls coupling strength
– Too large Æ loose coupling Æ poor temperature control
– Too small Æ tight coupling Æ high frequency temperature
oscillation
– Can be related to effective relaxation time
Q = CfkbT
found by testing the system oscillation about a desired T
Molecular Dynamics Simulation A. Martini Last Updated 8/2009
