2.2.1.1 Limitations in Calculating Potential Energy Function
The empirical potential function has several limitations, which result in inaccuracies in the
calculated potential energy.
One limitation is due to the fixed set of atom types employed when determining the
parameters for the force field. Atom types are used to define an atom in a particular bonding
situation, for example an aliphatic carbon atom in a sp3 bonding situation has different properties
than a carbon atom found in the benzene ring. Instead of presenting each atom in the molecule as
a unique one described by unique set of parameters, there is certain amounts of grouping in order
minimize the number of atom types. This can lead to type-specific errors. The properties of
certain atoms, like aliphatic carbon or hydrogen atoms, are less sensitive to their surroundings
and a single set of parameters may work quite well, while other atoms like oxygen and nitrogen
are much more influenced by their neighboring atoms. These atoms require more types and
parameters to account for the different bonding environments.
An approximation introduced to decrease the computational demand is the pair-wise
additive approximation, i.e., interaction energy between one atom and the rest of the system is
calculated as a sum of pair-wise (one atom to one atom) interactions, or as if the pair of atoms do
not see the other atoms in the system. The simultaneous interaction between three or more atoms
is not calculated, so certain polarization effects are not explicitly included in the force field. This
can lead to subtle differences between calculated and experimental results.
Another important point to take into consideration is that the potential energy function
does not include entropic effects. Thus, a minimum value of V calculated as a sum of potential
functions does not necessarily correspond to the equilibrium, or the most probable state; this
corresponds to the minimum of free energy. Because of the fact that experiments are generally
carried out under isothermal-isobaric conditions (constant pressure, constant system size and
constant temperature) the equilibrium state corresponds to the minimum of Gibb's Free Energy,
G. While just an energy calculation ignores entropic effects, these are included in molecular
dynamics simulations.
2.2.2 36BIntegration Algorithms
The potential energy is a function of the atomic positions (3N) of all the atoms in the
system. Due to the complicated nature of this function, there is no analytical solution to the
equations of motion; they must be solved numerically. In order to get a solution as accurate as
possible, a high accuracy time integration algorithm to integrate Eq. 2.1 with small steps would
be very much necessary [57, 58]. It does not matter how often the forces had to be calculated.
The situation in macromolecular systems usually studied with molecular dynamics is however
very different. In this case it is unnecessary to determine a very detailed solution for individual
atoms since in the dynamics; small numerical errors will grow exponentially and affect the
trajectories. This might strike bad at first since it affects the whole concept of simulations, but it
only reflects real systems-equilibrium properties are not sensitive to details of individual
trajectories. It is thus fruitless to reproduce motions exactly. Instead, one should make sure that
any reasonably long part extracted from a trajectory would be a fair description of a particle with
the same initial conditions Lindahl, 2001 [58].
Numerous numerical algorithms have been developed for integrating the equations of
motion. These include:
• Verlet algorithm
• Leap-frog algorithm
• Velocity Verlet algorithm
• Beeman’s algorithm
All the integration algorithms assume that the positions, velocities and accelerations can be
approximated by a Taylor series expression:
(2.11)
(2.12)
(2.13)
Where is the position, is the velocity (the first derivative of position with respect to time),
is the acceleration (the second derivative of position with respect to time), and b is the third
derivative of position with time.
2.2.2.1 Verlet Algorithm
Verlet algorithm is one of the most commonly used algorithms developed by Verlet [59].
The Verlet algorithm uses positions and accelerations at time t and the positions from time
to calculate new positions at time .
(2.14)
(2.15)
Summing Eq. 2.14 and Eq. 2.15 we get
(2.16)
This offers an advantage that the first and third-order term from the Taylor expression cancels
out, thus making the Verlet algorithm more accurate than integration Taylor expansion alone.
30
Note that if using this equation at t = 0, one needs the position at time , . At first sight
this could cause problems, because the initial conditions are known only at the initial time. This
can be solved by doing the first time step using the equation
(2.17)
The advantages of the Verlet algorithm are, i) it is straightforward, and ii) the storage
requirements are modest. The disadvantage is that the algorithm is of moderate precision.
2.2.2.2 Leap-Frog Algorithm
A slightly modified, but theoretically equivalent, algorithm is the Leap-Frog algorithm
[60] which handles velocities somewhat better than Verlet Algorithm. In this algorithm, the
velocities are first calculated at time ; these are used to calculate the positions, r, at time
. In this way, the velocities leap over the positions, then the positions leap over the
velocities. The Leap-Frog algorithm is not self-starting. The current state and a prior state must
both be known to advance the solution. Since the prior state is not known for the initial
conditions, a prior state is estimated when the initialize method is invoked.
(2.18)
(2.19)
The advantage of this algorithm is that the velocities are explicitly calculated and it is much
faster than Verlet algorithm, however the disadvantage is that they are not calculated at the same
time as the positions. The velocities at time t can be approximated by the relationship:
(2.20)
31
2.2.2.3 Velocity Verlet Algorithm
Another commonly used algorithm is velocity Verlet algorithm; this uses a similar approach but
explicitly incorporates velocity, solving the first-time step problem in the Basic Verlet algorithm:
(2.21)
(2.22)
The advantage of velocity algorithm is that it consumes less memory compared to Verlet
algorithm.
2.2.2.4 48BBeeman’s Algorithm
Beeman’s algorithm is closely related to Verlet algorithm. It produces identical positions to
Verlet, but is more accurate in velocities and gives better energy conservation.
(2.23)
(2.24)
The disadvantage is that the more complex expressions make the calculation more expensive.
2.2.3 37BConstraint Algorithms
Constraint algorithms are often applied to molecular dynamics simulations. Although
such simulations are sometimes carried out in internal coordinates that automatically satisfy the
bond-length and bond-angle constraints. To extend the length of the simulation we have to use a
longer time step, but due to the increase in the time step there will be successively larger errors
in the motions, and after a few steps the fluctuations will diverge, causing the whole simulation
to crash. To solve these problems ‘Constraint Dynamics’ is often employed in the simulations. It
completely removes the bond and/or angle degrees of freedom from the system. Explicit
32
constraint forces typically shorten the time-step significantly, making the simulation less
efficient computationally; in other words, more computer power is required to compute a
trajectory of a given length but avoids the errors when integrating bond oscillations. The constant
bond lengths are also fairly good approximations of the ground states of quantum mechanical
oscillators. Explicit constraint forces typically shorten the time-step significantly, making the
simulation less efficient computationally; in other words, more computer power is required to
compute a trajectory of a given length. Most commonly used constraint algorithms are:
• SHAKES Algorithm
• LINCS Algorithm
2.2.3.1 49BSHAKES Algorithm
The SHAKE algorithm was the first and most widespread algorithm developed to satisfy bond
geometry constraints during molecular dynamics simulations. It solves the system of non-linear
constraint equations using the Gauss-Seidel method to approximate the solution of the linear
system of equations. In this algorithm for each pair of atoms involved in a bond (or triplet in an
angle), force necessary to restore then to the equilibrium value, is calculated. In a
macromolecular system since a lot of bonds are connected, the algorithm has to be iterated
continuously until convergence is achieved. This limits the applicability somewhat; for time
steps greater than 2-3 fs it does not always converge, and the iteration makes it unsuitable for
parallel computers since it incurs a lot of extra communication between processors.
2.2.3.2 50BLINCS Algorithm
An alternative constraint method, LINCS (Linear Constraint Solver) was developed in
1997, but was based on earlier method, EEM [61], and a modification thereof [62]. This
algorithm resets bonds to their correct lengths after an unconstrained update [63]. This is noniterative
approach, as it always uses two steps. This advantage makes it possible to extend time
33
steps at least to 3-4 fs. Although LINCS is based on matrices, no matrix-matrix multiplication is
involved. LINCS has been reported to be 3-4 times faster than SHAKE [63], but it can only be
used with bond constraints and isolated angle constraints.
2.2.4 38BLimitations of Molecular Dynamic Simulations
Although Molecular dynamic simulation is a powerful technique, it has number of
limitations. Using Newton’s equation implies the use of Classical mechanics to describe the
motion of atoms, but it is known that systems at atomistic level obey quantum laws rather than
classical laws. Therefore one cannot hope to describe chemical reactions in which bonds form or
break