27 |
|
\end{equation} |
28 |
|
A point mass interacting with other bodies moves with the |
29 |
|
acceleration along the direction of the force acting on it. Let |
30 |
< |
$F_ij$ be the force that particle $i$ exerts on particle $j$, and |
31 |
< |
$F_ji$ be the force that particle $j$ exerts on particle $i$. |
30 |
> |
$F_{ij}$ be the force that particle $i$ exerts on particle $j$, and |
31 |
> |
$F_{ji}$ be the force that particle $j$ exerts on particle $i$. |
32 |
|
Newton¡¯s third law states that |
33 |
|
\begin{equation} |
34 |
< |
F_ij = -F_ji |
34 |
> |
F_{ij} = -F_{ji} |
35 |
|
\label{introEquation:newtonThirdLaw} |
36 |
|
\end{equation} |
37 |
|
|
315 |
|
isolated and conserve energy, Microcanonical ensemble(NVE) has a |
316 |
|
partition function like, |
317 |
|
\begin{equation} |
318 |
< |
\Omega (N,V,E) = e^{\beta TS} |
319 |
< |
\label{introEqaution:NVEPartition}. |
318 |
> |
\Omega (N,V,E) = e^{\beta TS} \label{introEquation:NVEPartition}. |
319 |
|
\end{equation} |
320 |
|
A canonical ensemble(NVT)is an ensemble of systems, each of which |
321 |
|
can share its energy with a large heat reservoir. The distribution |
393 |
|
\begin{equation} |
394 |
|
\rho \propto e^{ - \beta H} |
395 |
|
\label{introEquation:densityAndHamiltonian} |
396 |
+ |
\end{equation} |
397 |
+ |
|
398 |
+ |
\subsubsection{\label{introSection:phaseSpaceConservation}Conservation of Phase Space} |
399 |
+ |
Lets consider a region in the phase space, |
400 |
+ |
\begin{equation} |
401 |
+ |
\delta v = \int { \ldots \int {dq_1 } ...dq_f dp_1 } ..dp_f . |
402 |
+ |
\end{equation} |
403 |
+ |
If this region is small enough, the density $\rho$ can be regarded |
404 |
+ |
as uniform over the whole phase space. Thus, the number of phase |
405 |
+ |
points inside this region is given by, |
406 |
+ |
\begin{equation} |
407 |
+ |
\delta N = \rho \delta v = \rho \int { \ldots \int {dq_1 } ...dq_f |
408 |
+ |
dp_1 } ..dp_f. |
409 |
|
\end{equation} |
410 |
+ |
|
411 |
+ |
\begin{equation} |
412 |
+ |
\frac{{d(\delta N)}}{{dt}} = \frac{{d\rho }}{{dt}}\delta v + \rho |
413 |
+ |
\frac{d}{{dt}}(\delta v) = 0. |
414 |
+ |
\end{equation} |
415 |
+ |
With the help of stationary assumption |
416 |
+ |
(\ref{introEquation:stationary}), we obtain the principle of the |
417 |
+ |
\emph{conservation of extension in phase space}, |
418 |
+ |
\begin{equation} |
419 |
+ |
\frac{d}{{dt}}(\delta v) = \frac{d}{{dt}}\int { \ldots \int {dq_1 } |
420 |
+ |
...dq_f dp_1 } ..dp_f = 0. |
421 |
+ |
\label{introEquation:volumePreserving} |
422 |
+ |
\end{equation} |
423 |
+ |
|
424 |
+ |
\subsubsection{\label{introSection:liouvilleInOtherForms}Liouville's Theorem in Other Forms} |
425 |
|
|
426 |
|
Liouville's theorem can be expresses in a variety of different forms |
427 |
|
which are convenient within different contexts. For any two function |
458 |
|
\label{introEquation:liouvilleTheoremInOperator} |
459 |
|
\end{equation} |
460 |
|
|
434 |
– |
|
461 |
|
\subsection{\label{introSection:ergodic}The Ergodic Hypothesis} |
462 |
|
|
463 |
|
Various thermodynamic properties can be calculated from Molecular |
586 |
|
}}{{I_2 }} + \frac{{\pi _3^2 }}{{I_3 }}} \right) |
587 |
|
\end{equation} |
588 |
|
|
589 |
< |
\subsection{\label{introSection:geometricProperties}Geometric Properties} |
589 |
> |
\subsection{\label{introSection:exactFlow}Exact Flow} |
590 |
> |
|
591 |
|
Let $x(t)$ be the exact solution of the ODE system, |
592 |
|
\begin{equation} |
593 |
|
\frac{{dx}}{{dt}} = f(x) \label{introEquation:ODE} |
597 |
|
x(t+\tau) =\varphi_\tau(x(t)) |
598 |
|
\] |
599 |
|
where $\tau$ is a fixed time step and $\varphi$ is a map from phase |
600 |
< |
space to itself. In most cases, it is not easy to find the exact |
574 |
< |
flow $\varphi_\tau$. Instead, we use a approximate map, $\psi_\tau$, |
575 |
< |
which is usually called integrator. The order of an integrator |
576 |
< |
$\psi_\tau$ is $p$, if the Taylor series of $\psi_\tau$ agree to |
577 |
< |
order $p$, |
600 |
> |
space to itself. The flow has the continuous group property, |
601 |
|
\begin{equation} |
602 |
+ |
\varphi _{\tau _1 } \circ \varphi _{\tau _2 } = \varphi _{\tau _1 |
603 |
+ |
+ \tau _2 } . |
604 |
+ |
\end{equation} |
605 |
+ |
In particular, |
606 |
+ |
\begin{equation} |
607 |
+ |
\varphi _\tau \circ \varphi _{ - \tau } = I |
608 |
+ |
\end{equation} |
609 |
+ |
Therefore, the exact flow is self-adjoint, |
610 |
+ |
\begin{equation} |
611 |
+ |
\varphi _\tau = \varphi _{ - \tau }^{ - 1}. |
612 |
+ |
\end{equation} |
613 |
+ |
The exact flow can also be written in terms of the of an operator, |
614 |
+ |
\begin{equation} |
615 |
+ |
\varphi _\tau (x) = e^{\tau \sum\limits_i {f_i (x)\frac{\partial |
616 |
+ |
}{{\partial x_i }}} } (x) \equiv \exp (\tau f)(x). |
617 |
+ |
\label{introEquation:exponentialOperator} |
618 |
+ |
\end{equation} |
619 |
+ |
|
620 |
+ |
In most cases, it is not easy to find the exact flow $\varphi_\tau$. |
621 |
+ |
Instead, we use a approximate map, $\psi_\tau$, which is usually |
622 |
+ |
called integrator. The order of an integrator $\psi_\tau$ is $p$, if |
623 |
+ |
the Taylor series of $\psi_\tau$ agree to order $p$, |
624 |
+ |
\begin{equation} |
625 |
|
\psi_tau(x) = x + \tau f(x) + O(\tau^{p+1}) |
626 |
|
\end{equation} |
627 |
|
|
628 |
+ |
\subsection{\label{introSection:geometricProperties}Geometric Properties} |
629 |
+ |
|
630 |
|
The hidden geometric properties of ODE and its flow play important |
631 |
< |
roles in numerical studies. Let $\varphi$ be the flow of Hamiltonian |
632 |
< |
vector field, $\varphi$ is a \emph{symplectic} flow if it satisfies, |
631 |
> |
roles in numerical studies. Many of them can be found in systems |
632 |
> |
which occur naturally in applications. |
633 |
> |
|
634 |
> |
Let $\varphi$ be the flow of Hamiltonian vector field, $\varphi$ is |
635 |
> |
a \emph{symplectic} flow if it satisfies, |
636 |
|
\begin{equation} |
637 |
< |
'\varphi^T J '\varphi = J. |
637 |
> |
{\varphi '}^T J \varphi ' = J. |
638 |
|
\end{equation} |
639 |
|
According to Liouville's theorem, the symplectic volume is invariant |
640 |
|
under a Hamiltonian flow, which is the basis for classical |
642 |
|
field on a symplectic manifold can be shown to be a |
643 |
|
symplectomorphism. As to the Poisson system, |
644 |
|
\begin{equation} |
645 |
< |
'\varphi ^T J '\varphi = J \circ \varphi |
645 |
> |
{\varphi '}^T J \varphi ' = J \circ \varphi |
646 |
|
\end{equation} |
647 |
< |
is the property must be preserved by the integrator. It is possible |
648 |
< |
to construct a \emph{volume-preserving} flow for a source free($ |
649 |
< |
\nabla \cdot f = 0 $) ODE, if the flow satisfies $ \det d\varphi = |
650 |
< |
1$. Changing the variables $y = h(x)$ in a |
651 |
< |
ODE\ref{introEquation:ODE} will result in a new system, |
647 |
> |
is the property must be preserved by the integrator. |
648 |
> |
|
649 |
> |
It is possible to construct a \emph{volume-preserving} flow for a |
650 |
> |
source free($ \nabla \cdot f = 0 $) ODE, if the flow satisfies $ |
651 |
> |
\det d\varphi = 1$. One can show easily that a symplectic flow will |
652 |
> |
be volume-preserving. |
653 |
> |
|
654 |
> |
Changing the variables $y = h(x)$ in a ODE\ref{introEquation:ODE} |
655 |
> |
will result in a new system, |
656 |
|
\[ |
657 |
|
\dot y = \tilde f(y) = ((dh \cdot f)h^{ - 1} )(y). |
658 |
|
\] |
659 |
|
The vector filed $f$ has reversing symmetry $h$ if $f = - \tilde f$. |
660 |
|
In other words, the flow of this vector field is reversible if and |
661 |
< |
only if $ h \circ \varphi ^{ - 1} = \varphi \circ h $. When |
607 |
< |
designing any numerical methods, one should always try to preserve |
608 |
< |
the structural properties of the original ODE and its flow. |
661 |
> |
only if $ h \circ \varphi ^{ - 1} = \varphi \circ h $. |
662 |
|
|
663 |
+ |
A \emph{first integral}, or conserved quantity of a general |
664 |
+ |
differential function is a function $ G:R^{2d} \to R^d $ which is |
665 |
+ |
constant for all solutions of the ODE $\frac{{dx}}{{dt}} = f(x)$ , |
666 |
+ |
\[ |
667 |
+ |
\frac{{dG(x(t))}}{{dt}} = 0. |
668 |
+ |
\] |
669 |
+ |
Using chain rule, one may obtain, |
670 |
+ |
\[ |
671 |
+ |
\sum\limits_i {\frac{{dG}}{{dx_i }}} f_i (x) = f \bullet \nabla G, |
672 |
+ |
\] |
673 |
+ |
which is the condition for conserving \emph{first integral}. For a |
674 |
+ |
canonical Hamiltonian system, the time evolution of an arbitrary |
675 |
+ |
smooth function $G$ is given by, |
676 |
+ |
\begin{equation} |
677 |
+ |
\begin{array}{c} |
678 |
+ |
\frac{{dG(x(t))}}{{dt}} = [\nabla _x G(x(t))]^T \dot x(t) \\ |
679 |
+ |
= [\nabla _x G(x(t))]^T J\nabla _x H(x(t)). \\ |
680 |
+ |
\end{array} |
681 |
+ |
\label{introEquation:firstIntegral1} |
682 |
+ |
\end{equation} |
683 |
+ |
Using poisson bracket notion, Equation |
684 |
+ |
\ref{introEquation:firstIntegral1} can be rewritten as |
685 |
+ |
\[ |
686 |
+ |
\frac{d}{{dt}}G(x(t)) = \left\{ {G,H} \right\}(x(t)). |
687 |
+ |
\] |
688 |
+ |
Therefore, the sufficient condition for $G$ to be the \emph{first |
689 |
+ |
integral} of a Hamiltonian system is |
690 |
+ |
\[ |
691 |
+ |
\left\{ {G,H} \right\} = 0. |
692 |
+ |
\] |
693 |
+ |
As well known, the Hamiltonian (or energy) H of a Hamiltonian system |
694 |
+ |
is a \emph{first integral}, which is due to the fact $\{ H,H\} = |
695 |
+ |
0$. |
696 |
+ |
|
697 |
+ |
|
698 |
+ |
When designing any numerical methods, one should always try to |
699 |
+ |
preserve the structural properties of the original ODE and its flow. |
700 |
+ |
|
701 |
|
\subsection{\label{introSection:constructionSymplectic}Construction of Symplectic Methods} |
702 |
|
A lot of well established and very effective numerical methods have |
703 |
|
been successful precisely because of their symplecticities even |
713 |
|
\end{enumerate} |
714 |
|
|
715 |
|
Generating function tends to lead to methods which are cumbersome |
716 |
< |
and difficult to use\cite{}. In dissipative systems, variational |
717 |
< |
methods can capture the decay of energy accurately\cite{}. Since |
718 |
< |
their geometrically unstable nature against non-Hamiltonian |
719 |
< |
perturbations, ordinary implicit Runge-Kutta methods are not |
720 |
< |
suitable for Hamiltonian system. Recently, various high-order |
721 |
< |
explicit Runge--Kutta methods have been developed to overcome this |
722 |
< |
instability \cite{}. However, due to computational penalty involved |
723 |
< |
in implementing the Runge-Kutta methods, they do not attract too |
724 |
< |
much attention from Molecular Dynamics community. Instead, splitting |
725 |
< |
have been widely accepted since they exploit natural decompositions |
726 |
< |
of the system\cite{Tuckerman92}. The main idea behind splitting |
727 |
< |
methods is to decompose the discrete $\varphi_h$ as a composition of |
728 |
< |
simpler flows, |
716 |
> |
and difficult to use. In dissipative systems, variational methods |
717 |
> |
can capture the decay of energy accurately. Since their |
718 |
> |
geometrically unstable nature against non-Hamiltonian perturbations, |
719 |
> |
ordinary implicit Runge-Kutta methods are not suitable for |
720 |
> |
Hamiltonian system. Recently, various high-order explicit |
721 |
> |
Runge--Kutta methods have been developed to overcome this |
722 |
> |
instability. However, due to computational penalty involved in |
723 |
> |
implementing the Runge-Kutta methods, they do not attract too much |
724 |
> |
attention from Molecular Dynamics community. Instead, splitting have |
725 |
> |
been widely accepted since they exploit natural decompositions of |
726 |
> |
the system\cite{Tuckerman92}. |
727 |
> |
|
728 |
> |
\subsubsection{\label{introSection:splittingMethod}Splitting Method} |
729 |
> |
|
730 |
> |
The main idea behind splitting methods is to decompose the discrete |
731 |
> |
$\varphi_h$ as a composition of simpler flows, |
732 |
|
\begin{equation} |
733 |
|
\varphi _h = \varphi _{h_1 } \circ \varphi _{h_2 } \ldots \circ |
734 |
|
\varphi _{h_n } |
735 |
|
\label{introEquation:FlowDecomposition} |
736 |
|
\end{equation} |
737 |
|
where each of the sub-flow is chosen such that each represent a |
738 |
< |
simpler integration of the system. Let $\phi$ and $\psi$ both be |
739 |
< |
symplectic maps, it is easy to show that any composition of |
740 |
< |
symplectic flows yields a symplectic map, |
738 |
> |
simpler integration of the system. |
739 |
> |
|
740 |
> |
Suppose that a Hamiltonian system takes the form, |
741 |
> |
\[ |
742 |
> |
H = H_1 + H_2. |
743 |
> |
\] |
744 |
> |
Here, $H_1$ and $H_2$ may represent different physical processes of |
745 |
> |
the system. For instance, they may relate to kinetic and potential |
746 |
> |
energy respectively, which is a natural decomposition of the |
747 |
> |
problem. If $H_1$ and $H_2$ can be integrated using exact flows |
748 |
> |
$\varphi_1(t)$ and $\varphi_2(t)$, respectively, a simple first |
749 |
> |
order is then given by the Lie-Trotter formula |
750 |
|
\begin{equation} |
751 |
+ |
\varphi _h = \varphi _{1,h} \circ \varphi _{2,h}, |
752 |
+ |
\label{introEquation:firstOrderSplitting} |
753 |
+ |
\end{equation} |
754 |
+ |
where $\varphi _h$ is the result of applying the corresponding |
755 |
+ |
continuous $\varphi _i$ over a time $h$. By definition, as |
756 |
+ |
$\varphi_i(t)$ is the exact solution of a Hamiltonian system, it |
757 |
+ |
must follow that each operator $\varphi_i(t)$ is a symplectic map. |
758 |
+ |
It is easy to show that any composition of symplectic flows yields a |
759 |
+ |
symplectic map, |
760 |
+ |
\begin{equation} |
761 |
|
(\varphi '\phi ')^T J\varphi '\phi ' = \phi '^T \varphi '^T J\varphi |
762 |
< |
'\phi ' = \phi '^T J\phi ' = J. |
762 |
> |
'\phi ' = \phi '^T J\phi ' = J, |
763 |
|
\label{introEquation:SymplecticFlowComposition} |
764 |
|
\end{equation} |
765 |
< |
Suppose that a Hamiltonian system has a form with $H = T + V$ |
765 |
> |
where $\phi$ and $\psi$ both are symplectic maps. Thus operator |
766 |
> |
splitting in this context automatically generates a symplectic map. |
767 |
> |
|
768 |
> |
The Lie-Trotter splitting(\ref{introEquation:firstOrderSplitting}) |
769 |
> |
introduces local errors proportional to $h^2$, while Strang |
770 |
> |
splitting gives a second-order decomposition, |
771 |
> |
\begin{equation} |
772 |
> |
\varphi _h = \varphi _{1,h/2} \circ \varphi _{2,h} \circ \varphi |
773 |
> |
_{1,h/2} , \label{introEquation:secondOrderSplitting} |
774 |
> |
\end{equation} |
775 |
> |
which has a local error proportional to $h^3$. Sprang splitting's |
776 |
> |
popularity in molecular simulation community attribute to its |
777 |
> |
symmetric property, |
778 |
> |
\begin{equation} |
779 |
> |
\varphi _h^{ - 1} = \varphi _{ - h}. |
780 |
> |
\label{introEquation:timeReversible} |
781 |
> |
\end{equation} |
782 |
> |
|
783 |
> |
\subsubsection{\label{introSection:exampleSplittingMethod}Example of Splitting Method} |
784 |
> |
The classical equation for a system consisting of interacting |
785 |
> |
particles can be written in Hamiltonian form, |
786 |
> |
\[ |
787 |
> |
H = T + V |
788 |
> |
\] |
789 |
> |
where $T$ is the kinetic energy and $V$ is the potential energy. |
790 |
> |
Setting $H_1 = T, H_2 = V$ and applying Strang splitting, one |
791 |
> |
obtains the following: |
792 |
> |
\begin{align} |
793 |
> |
q(\Delta t) &= q(0) + \dot{q}(0)\Delta t + |
794 |
> |
\frac{F[q(0)]}{m}\frac{\Delta t^2}{2}, % |
795 |
> |
\label{introEquation:Lp10a} \\% |
796 |
> |
% |
797 |
> |
\dot{q}(\Delta t) &= \dot{q}(0) + \frac{\Delta t}{2m} |
798 |
> |
\biggl [F[q(0)] + F[q(\Delta t)] \biggr]. % |
799 |
> |
\label{introEquation:Lp10b} |
800 |
> |
\end{align} |
801 |
> |
where $F(t)$ is the force at time $t$. This integration scheme is |
802 |
> |
known as \emph{velocity verlet} which is |
803 |
> |
symplectic(\ref{introEquation:SymplecticFlowComposition}), |
804 |
> |
time-reversible(\ref{introEquation:timeReversible}) and |
805 |
> |
volume-preserving (\ref{introEquation:volumePreserving}). These |
806 |
> |
geometric properties attribute to its long-time stability and its |
807 |
> |
popularity in the community. However, the most commonly used |
808 |
> |
velocity verlet integration scheme is written as below, |
809 |
> |
\begin{align} |
810 |
> |
\dot{q}\biggl (\frac{\Delta t}{2}\biggr ) &= |
811 |
> |
\dot{q}(0) + \frac{\Delta t}{2m}\, F[q(0)], \label{introEquation:Lp9a}\\% |
812 |
> |
% |
813 |
> |
q(\Delta t) &= q(0) + \Delta t\, \dot{q}\biggl (\frac{\Delta t}{2}\biggr ),% |
814 |
> |
\label{introEquation:Lp9b}\\% |
815 |
> |
% |
816 |
> |
\dot{q}(\Delta t) &= \dot{q}\biggl (\frac{\Delta t}{2}\biggr ) + |
817 |
> |
\frac{\Delta t}{2m}\, F[q(0)]. \label{introEquation:Lp9c} |
818 |
> |
\end{align} |
819 |
> |
From the preceding splitting, one can see that the integration of |
820 |
> |
the equations of motion would follow: |
821 |
> |
\begin{enumerate} |
822 |
> |
\item calculate the velocities at the half step, $\frac{\Delta t}{2}$, from the forces calculated at the initial position. |
823 |
> |
|
824 |
> |
\item Use the half step velocities to move positions one whole step, $\Delta t$. |
825 |
> |
|
826 |
> |
\item Evaluate the forces at the new positions, $\mathbf{r}(\Delta t)$, and use the new forces to complete the velocity move. |
827 |
> |
|
828 |
> |
\item Repeat from step 1 with the new position, velocities, and forces assuming the roles of the initial values. |
829 |
> |
\end{enumerate} |
830 |
> |
|
831 |
> |
Simply switching the order of splitting and composing, a new |
832 |
> |
integrator, the \emph{position verlet} integrator, can be generated, |
833 |
> |
\begin{align} |
834 |
> |
\dot q(\Delta t) &= \dot q(0) + \Delta tF(q(0))\left[ {q(0) + |
835 |
> |
\frac{{\Delta t}}{{2m}}\dot q(0)} \right], % |
836 |
> |
\label{introEquation:positionVerlet1} \\% |
837 |
> |
% |
838 |
> |
q(\Delta t) &= q(0) + \frac{{\Delta t}}{2}\left[ {\dot q(0) + \dot |
839 |
> |
q(\Delta t)} \right]. % |
840 |
> |
\label{introEquation:positionVerlet1} |
841 |
> |
\end{align} |
842 |
> |
|
843 |
> |
\subsubsection{\label{introSection:errorAnalysis}Error Analysis and Higher Order Methods} |
844 |
> |
|
845 |
> |
Baker-Campbell-Hausdorff formula can be used to determine the local |
846 |
> |
error of splitting method in terms of commutator of the |
847 |
> |
operators(\ref{introEquation:exponentialOperator}) associated with |
848 |
> |
the sub-flow. For operators $hX$ and $hY$ which are associate to |
849 |
> |
$\varphi_1(t)$ and $\varphi_2(t$ respectively , we have |
850 |
> |
\begin{equation} |
851 |
> |
\exp (hX + hY) = \exp (hZ) |
852 |
> |
\end{equation} |
853 |
> |
where |
854 |
> |
\begin{equation} |
855 |
> |
hZ = hX + hY + \frac{{h^2 }}{2}[X,Y] + \frac{{h^3 }}{2}\left( |
856 |
> |
{[X,[X,Y]] + [Y,[Y,X]]} \right) + \ldots . |
857 |
> |
\end{equation} |
858 |
> |
Here, $[X,Y]$ is the commutators of operator $X$ and $Y$ given by |
859 |
> |
\[ |
860 |
> |
[X,Y] = XY - YX . |
861 |
> |
\] |
862 |
> |
Applying Baker-Campbell-Hausdorff formula to Sprang splitting, we |
863 |
> |
can obtain |
864 |
> |
\begin{eqnarray*} |
865 |
> |
\exp (h X/2)\exp (h Y)\exp (h X/2) & = & \exp (h X + h Y + h^2 |
866 |
> |
[X,Y]/4 + h^2 [Y,X]/4 \\ & & \mbox{} + h^2 [X,X]/8 + h^2 [Y,Y]/8 \\ |
867 |
> |
& & \mbox{} + h^3 [Y,[Y,X]]/12 - h^3 [X,[X,Y]]/24 & & \mbox{} + |
868 |
> |
\ldots ) |
869 |
> |
\end{eqnarray*} |
870 |
> |
Since \[ [X,Y] + [Y,X] = 0\] and \[ [X,X] = 0\], the dominant local |
871 |
> |
error of Spring splitting is proportional to $h^3$. The same |
872 |
> |
procedure can be applied to general splitting, of the form |
873 |
> |
\begin{equation} |
874 |
> |
\varphi _{b_m h}^2 \circ \varphi _{a_m h}^1 \circ \varphi _{b_{m - |
875 |
> |
1} h}^2 \circ \ldots \circ \varphi _{a_1 h}^1 . |
876 |
> |
\end{equation} |
877 |
> |
Careful choice of coefficient $a_1 ,\ldot , b_m$ will lead to higher |
878 |
> |
order method. Yoshida proposed an elegant way to compose higher |
879 |
> |
order methods based on symmetric splitting. Given a symmetric second |
880 |
> |
order base method $ \varphi _h^{(2)} $, a fourth-order symmetric |
881 |
> |
method can be constructed by composing, |
882 |
> |
\[ |
883 |
> |
\varphi _h^{(4)} = \varphi _{\alpha h}^{(2)} \circ \varphi _{\beta |
884 |
> |
h}^{(2)} \circ \varphi _{\alpha h}^{(2)} |
885 |
> |
\] |
886 |
> |
where $ \alpha = - \frac{{2^{1/3} }}{{2 - 2^{1/3} }}$ and $ \beta |
887 |
> |
= \frac{{2^{1/3} }}{{2 - 2^{1/3} }}$. Moreover, a symmetric |
888 |
> |
integrator $ \varphi _h^{(2n + 2)}$ can be composed by |
889 |
> |
\begin{equation} |
890 |
> |
\varphi _h^{(2n + 2)} = \varphi _{\alpha h}^{(2n)} \circ \varphi |
891 |
> |
_{\beta h}^{(2n)} \circ \varphi _{\alpha h}^{(2n)} |
892 |
> |
\end{equation} |
893 |
> |
, if the weights are chosen as |
894 |
> |
\[ |
895 |
> |
\alpha = - \frac{{2^{1/(2n + 1)} }}{{2 - 2^{1/(2n + 1)} }},\beta = |
896 |
> |
\frac{{2^{1/(2n + 1)} }}{{2 - 2^{1/(2n + 1)} }} . |
897 |
> |
\] |
898 |
|
|
899 |
|
\section{\label{introSection:molecularDynamics}Molecular Dynamics} |
900 |
|
|
905 |
|
|
906 |
|
\subsection{\label{introSec:mdInit}Initialization} |
907 |
|
|
908 |
+ |
\subsection{\label{introSec:forceEvaluation}Force Evaluation} |
909 |
+ |
|
910 |
|
\subsection{\label{introSection:mdIntegration} Integration of the Equations of Motion} |
911 |
|
|
912 |
|
\section{\label{introSection:rigidBody}Dynamics of Rigid Bodies} |
913 |
|
|
914 |
< |
A rigid body is a body in which the distance between any two given |
915 |
< |
points of a rigid body remains constant regardless of external |
916 |
< |
forces exerted on it. A rigid body therefore conserves its shape |
917 |
< |
during its motion. |
914 |
> |
Rigid bodies are frequently involved in the modeling of different |
915 |
> |
areas, from engineering, physics, to chemistry. For example, |
916 |
> |
missiles and vehicle are usually modeled by rigid bodies. The |
917 |
> |
movement of the objects in 3D gaming engine or other physics |
918 |
> |
simulator is governed by the rigid body dynamics. In molecular |
919 |
> |
simulation, rigid body is used to simplify the model in |
920 |
> |
protein-protein docking study{\cite{Gray03}}. |
921 |
|
|
922 |
< |
Applications of dynamics of rigid bodies. |
922 |
> |
It is very important to develop stable and efficient methods to |
923 |
> |
integrate the equations of motion of orientational degrees of |
924 |
> |
freedom. Euler angles are the nature choice to describe the |
925 |
> |
rotational degrees of freedom. However, due to its singularity, the |
926 |
> |
numerical integration of corresponding equations of motion is very |
927 |
> |
inefficient and inaccurate. Although an alternative integrator using |
928 |
> |
different sets of Euler angles can overcome this difficulty\cite{}, |
929 |
> |
the computational penalty and the lost of angular momentum |
930 |
> |
conservation still remain. A singularity free representation |
931 |
> |
utilizing quaternions was developed by Evans in 1977. Unfortunately, |
932 |
> |
this approach suffer from the nonseparable Hamiltonian resulted from |
933 |
> |
quaternion representation, which prevents the symplectic algorithm |
934 |
> |
to be utilized. Another different approach is to apply holonomic |
935 |
> |
constraints to the atoms belonging to the rigid body. Each atom |
936 |
> |
moves independently under the normal forces deriving from potential |
937 |
> |
energy and constraint forces which are used to guarantee the |
938 |
> |
rigidness. However, due to their iterative nature, SHAKE and Rattle |
939 |
> |
algorithm converge very slowly when the number of constraint |
940 |
> |
increases. |
941 |
|
|
942 |
+ |
The break through in geometric literature suggests that, in order to |
943 |
+ |
develop a long-term integration scheme, one should preserve the |
944 |
+ |
symplectic structure of the flow. Introducing conjugate momentum to |
945 |
+ |
rotation matrix $A$ and re-formulating Hamiltonian's equation, a |
946 |
+ |
symplectic integrator, RSHAKE, was proposed to evolve the |
947 |
+ |
Hamiltonian system in a constraint manifold by iteratively |
948 |
+ |
satisfying the orthogonality constraint $A_t A = 1$. An alternative |
949 |
+ |
method using quaternion representation was developed by Omelyan. |
950 |
+ |
However, both of these methods are iterative and inefficient. In |
951 |
+ |
this section, we will present a symplectic Lie-Poisson integrator |
952 |
+ |
for rigid body developed by Dullweber and his |
953 |
+ |
coworkers\cite{Dullweber1997}. |
954 |
+ |
|
955 |
|
\subsection{\label{introSection:lieAlgebra}Lie Algebra} |
956 |
|
|
957 |
< |
\subsection{\label{introSection:DLMMotionEquation}The Euler Equations of Rigid Body Motion} |
957 |
> |
\subsection{\label{introSection:constrainedHamiltonianRB}Constrained Hamiltonian for Rigid Body} |
958 |
|
|
959 |
< |
\subsection{\label{introSection:otherRBMotionEquation}Other Formulations for Rigid Body Motion} |
959 |
> |
\begin{equation} |
960 |
> |
H = \frac{1}{2}(p^T m^{ - 1} p) + \frac{1}{2}tr(PJ^{ - 1} P) + |
961 |
> |
V(q,Q) + \frac{1}{2}tr[(QQ^T - 1)\Lambda ]. |
962 |
> |
\label{introEquation:RBHamiltonian} |
963 |
> |
\end{equation} |
964 |
> |
Here, $q$ and $Q$ are the position and rotation matrix for the |
965 |
> |
rigid-body, $p$ and $P$ are conjugate momenta to $q$ and $Q$ , and |
966 |
> |
$J$, a diagonal matrix, is defined by |
967 |
> |
\[ |
968 |
> |
I_{ii}^{ - 1} = \frac{1}{2}\sum\limits_{i \ne j} {J_{jj}^{ - 1} } |
969 |
> |
\] |
970 |
> |
where $I_{ii}$ is the diagonal element of the inertia tensor. This |
971 |
> |
constrained Hamiltonian equation subjects to a holonomic constraint, |
972 |
> |
\begin{equation} |
973 |
> |
Q^T Q = 1$, \label{introEquation:orthogonalConstraint} |
974 |
> |
\end{equation} |
975 |
> |
which is used to ensure rotation matrix's orthogonality. |
976 |
> |
Differentiating \ref{introEquation:orthogonalConstraint} and using |
977 |
> |
Equation \ref{introEquation:RBMotionMomentum}, one may obtain, |
978 |
> |
\begin{equation} |
979 |
> |
Q^T PJ^{ - 1} + J^{ - 1} P^T Q = 0 . \\ |
980 |
> |
\label{introEquation:RBFirstOrderConstraint} |
981 |
> |
\end{equation} |
982 |
|
|
983 |
< |
%\subsection{\label{introSection:poissonBrackets}Poisson Brackets} |
983 |
> |
Using Equation (\ref{introEquation:motionHamiltonianCoordinate}, |
984 |
> |
\ref{introEquation:motionHamiltonianMomentum}), one can write down |
985 |
> |
the equations of motion, |
986 |
> |
\[ |
987 |
> |
\begin{array}{c} |
988 |
> |
\frac{{dq}}{{dt}} = \frac{p}{m} \label{introEquation:RBMotionPosition}\\ |
989 |
> |
\frac{{dp}}{{dt}} = - \nabla _q V(q,Q) \label{introEquation:RBMotionMomentum}\\ |
990 |
> |
\frac{{dQ}}{{dt}} = PJ^{ - 1} \label{introEquation:RBMotionRotation}\\ |
991 |
> |
\frac{{dP}}{{dt}} = - \nabla _Q V(q,Q) - 2Q\Lambda . \label{introEquation:RBMotionP}\\ |
992 |
> |
\end{array} |
993 |
> |
\] |
994 |
> |
|
995 |
> |
In general, there are two ways to satisfy the holonomic constraints. |
996 |
> |
We can use constraint force provided by lagrange multiplier on the |
997 |
> |
normal manifold to keep the motion on constraint space. Or we can |
998 |
> |
simply evolve the system in constraint manifold. The two method are |
999 |
> |
proved to be equivalent. The holonomic constraint and equations of |
1000 |
> |
motions define a constraint manifold for rigid body |
1001 |
> |
\[ |
1002 |
> |
M = \left\{ {(Q,P):Q^T Q = 1,Q^T PJ^{ - 1} + J^{ - 1} P^T Q = 0} |
1003 |
> |
\right\}. |
1004 |
> |
\] |
1005 |
|
|
1006 |
< |
\section{\label{introSection:correlationFunctions}Correlation Functions} |
1006 |
> |
Unfortunately, this constraint manifold is not the cotangent bundle |
1007 |
> |
$T_{\star}SO(3)$. However, it turns out that under symplectic |
1008 |
> |
transformation, the cotangent space and the phase space are |
1009 |
> |
diffeomorphic. Introducing |
1010 |
> |
\[ |
1011 |
> |
\tilde Q = Q,\tilde P = \frac{1}{2}\left( {P - QP^T Q} \right), |
1012 |
> |
\] |
1013 |
> |
the mechanical system subject to a holonomic constraint manifold $M$ |
1014 |
> |
can be re-formulated as a Hamiltonian system on the cotangent space |
1015 |
> |
\[ |
1016 |
> |
T^* SO(3) = \left\{ {(\tilde Q,\tilde P):\tilde Q^T \tilde Q = |
1017 |
> |
1,\tilde Q^T \tilde PJ^{ - 1} + J^{ - 1} P^T \tilde Q = 0} \right\} |
1018 |
> |
\] |
1019 |
> |
|
1020 |
> |
For a body fixed vector $X_i$ with respect to the center of mass of |
1021 |
> |
the rigid body, its corresponding lab fixed vector $X_0^{lab}$ is |
1022 |
> |
given as |
1023 |
> |
\begin{equation} |
1024 |
> |
X_i^{lab} = Q X_i + q. |
1025 |
> |
\end{equation} |
1026 |
> |
Therefore, potential energy $V(q,Q)$ is defined by |
1027 |
> |
\[ |
1028 |
> |
V(q,Q) = V(Q X_0 + q). |
1029 |
> |
\] |
1030 |
> |
Hence, |
1031 |
> |
\[ |
1032 |
> |
\nabla _q V(q,Q) = F(q,Q) = \sum\limits_i {F_i (q,Q)} |
1033 |
> |
\] |
1034 |
> |
|
1035 |
> |
\[ |
1036 |
> |
\nabla _Q V(q,Q) = F(q,Q)X_i^t |
1037 |
> |
\] |
1038 |
> |
|
1039 |
> |
As a common choice to describe the rotation dynamics of the rigid |
1040 |
> |
body, angular momentum on body frame $\Pi = Q^t P$ is introduced to |
1041 |
> |
rewrite the equations of motion, |
1042 |
> |
\begin{equation} |
1043 |
> |
\begin{array}{l} |
1044 |
> |
\mathop \Pi \limits^ \bullet = J^{ - 1} \Pi ^T \Pi + Q^T \sum\limits_i {F_i (q,Q)X_i^T } - \Lambda \\ |
1045 |
> |
\mathop Q\limits^{{\rm{ }} \bullet } = Q\Pi {\rm{ }}J^{ - 1} \\ |
1046 |
> |
\end{array} |
1047 |
> |
\label{introEqaution:RBMotionPI} |
1048 |
> |
\end{equation} |
1049 |
> |
, as well as holonomic constraints, |
1050 |
> |
\[ |
1051 |
> |
\begin{array}{l} |
1052 |
> |
\Pi J^{ - 1} + J^{ - 1} \Pi ^t = 0 \\ |
1053 |
> |
Q^T Q = 1 \\ |
1054 |
> |
\end{array} |
1055 |
> |
\] |
1056 |
> |
|
1057 |
> |
For a vector $v(v_1 ,v_2 ,v_3 ) \in R^3$ and a matrix $\hat v \in |
1058 |
> |
so(3)^ \star$, the hat-map isomorphism, |
1059 |
> |
\begin{equation} |
1060 |
> |
v(v_1 ,v_2 ,v_3 ) \Leftrightarrow \hat v = \left( |
1061 |
> |
{\begin{array}{*{20}c} |
1062 |
> |
0 & { - v_3 } & {v_2 } \\ |
1063 |
> |
{v_3 } & 0 & { - v_1 } \\ |
1064 |
> |
{ - v_2 } & {v_1 } & 0 \\ |
1065 |
> |
\end{array}} \right), |
1066 |
> |
\label{introEquation:hatmapIsomorphism} |
1067 |
> |
\end{equation} |
1068 |
> |
will let us associate the matrix products with traditional vector |
1069 |
> |
operations |
1070 |
> |
\[ |
1071 |
> |
\hat vu = v \times u |
1072 |
> |
\] |
1073 |
|
|
1074 |
+ |
Using \ref{introEqaution:RBMotionPI}, one can construct a skew |
1075 |
+ |
matrix, |
1076 |
+ |
\begin{equation} |
1077 |
+ |
(\mathop \Pi \limits^ \bullet - \mathop \Pi \limits^ \bullet ^T |
1078 |
+ |
){\rm{ }} = {\rm{ }}(\Pi - \Pi ^T ){\rm{ }}(J^{ - 1} \Pi + \Pi J^{ |
1079 |
+ |
- 1} ) + \sum\limits_i {[Q^T F_i (r,Q)X_i^T - X_i F_i (r,Q)^T Q]} - |
1080 |
+ |
(\Lambda - \Lambda ^T ) . \label{introEquation:skewMatrixPI} |
1081 |
+ |
\end{equation} |
1082 |
+ |
Since $\Lambda$ is symmetric, the last term of Equation |
1083 |
+ |
\ref{introEquation:skewMatrixPI}, which implies the Lagrange |
1084 |
+ |
multiplier $\Lambda$ is ignored in the integration. |
1085 |
+ |
|
1086 |
+ |
Hence, applying hat-map isomorphism, we obtain the equation of |
1087 |
+ |
motion for angular momentum on body frame |
1088 |
+ |
\[ |
1089 |
+ |
\dot \pi = \pi \times I^{ - 1} \pi + Q^T \sum\limits_i {F_i (r,Q) |
1090 |
+ |
\times X_i } |
1091 |
+ |
\] |
1092 |
+ |
In the same manner, the equation of motion for rotation matrix is |
1093 |
+ |
given by |
1094 |
+ |
\[ |
1095 |
+ |
\dot Q = Qskew(M^{ - 1} \pi ) |
1096 |
+ |
\] |
1097 |
+ |
|
1098 |
+ |
The free rigid body equation is an example of a non-canonical |
1099 |
+ |
Hamiltonian system. |
1100 |
+ |
|
1101 |
+ |
\subsection{\label{introSection:symplecticDiscretizationRB}Symplectic Integration of Euler Equations} |
1102 |
+ |
|
1103 |
+ |
\[ |
1104 |
+ |
\varphi _{\Delta t} = \varphi _{\Delta t/2,V} \circ \varphi |
1105 |
+ |
_{\Delta t,T} \circ \varphi _{\Delta t/2,V} |
1106 |
+ |
\] |
1107 |
+ |
|
1108 |
+ |
\[ |
1109 |
+ |
\varphi _{\Delta t,T} = \varphi _{\Delta t,R} \circ \varphi |
1110 |
+ |
_{\Delta t,\pi } |
1111 |
+ |
\] |
1112 |
+ |
|
1113 |
+ |
\[ |
1114 |
+ |
\varphi _{\Delta t,\pi } = \varphi _{\Delta t/2,\pi _1 } \circ |
1115 |
+ |
\varphi _{\Delta t/2,\pi _2 } \circ \varphi _{\Delta t,\pi _3 } |
1116 |
+ |
\circ \varphi _{\Delta t/2,\pi _2 } \circ \varphi _{\Delta t/2,\pi |
1117 |
+ |
_1 } |
1118 |
+ |
\] |
1119 |
+ |
|
1120 |
+ |
\[ |
1121 |
+ |
\varphi _{\Delta t/2,V} = \varphi _{\Delta t/2,F} \circ \varphi |
1122 |
+ |
_{\Delta t/2,\tau } |
1123 |
+ |
\] |
1124 |
+ |
|
1125 |
+ |
|
1126 |
|
\section{\label{introSection:langevinDynamics}Langevin Dynamics} |
1127 |
|
|
1128 |
|
\subsection{\label{introSection:LDIntroduction}Introduction and application of Langevin Dynamics} |
1171 |
|
\dot p &= - \frac{{\partial H}}{{\partial x}} |
1172 |
|
&= m\ddot x |
1173 |
|
&= - \frac{{\partial W(x)}}{{\partial x}} - \sum\limits_{\alpha = 1}^N {g_\alpha \left( {x_\alpha - \frac{{g_\alpha }}{{m_\alpha w_\alpha ^2 }}x} \right)} |
1174 |
< |
\label{introEq:Lp5} |
1174 |
> |
\label{introEquation:Lp5} |
1175 |
|
\end{align} |
1176 |
|
, and |
1177 |
|
\begin{align} |
1330 |
|
|
1331 |
|
\subsection{\label{introSection:centersRigidBody}Centers of Rigid |
1332 |
|
Body} |
1333 |
+ |
|
1334 |
+ |
\section{\label{introSection:correlationFunctions}Correlation Functions} |