moldyn
-
Upload
muhammad-shahid-bilal -
Category
Documents
-
view
216 -
download
0
Transcript of moldyn
-
7/29/2019 moldyn
1/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 1 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1) Molecular Dynamics Simulation
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 2of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.1. Introduction
Remember what Scientific Computing deals with:From phenomena to predictions!
phenomenon, process etc.
mathematical modelc
modelling
numerical algorithmc
numerical treatment
simulation codec
implementation
results to interpretc
visualization
%
rrrrj embedding
statement tool
E
E
E
va
li
da
t
ion
-
7/29/2019 moldyn
2/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 3of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Overview
modelling aspects of molecular dynamics simulations:
why to leave the classical continuum mechanics point of view?
where appropriate?
which models, i.e. which equations?
numerical aspects of molecular dynamics simulations?
how to discretize the resulting modelling equations?
efficient algorithms?
implementation aspects of molecular dynamics simulations?
suitable data structures?
parallelisation?
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 4of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Hierarchy of Models
Different points of view for simulating human beings:
issue level of resolution model basis (e.g.!)global increasein population
countries, regions population dynamics
local increase in
population
villages, individuals population dynamics
man circulations, organs system simulatorblood circulation pump/channels/valves network simulatorheart blood cells continuum mechanicscell macro molecules continuum mechanicsmacromolecules
atoms molecular dynamics
atoms electrons or finer quantum mechanics
-
7/29/2019 moldyn
3/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 5of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Scales an Important Issue
length scales in simulations:
from 109m (atoms)
to 1023m (galaxy clusters)
time scales in simulations: from 1015s
to 1017s
mass scales in simulations:
from 1024g (atoms)
to 1043g (galaxies)
obviously impossible to take all scales into acount in an explicit
and simultaneous way first molecular dynamics simulations reported in 1957
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 6of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Applications for Micro and Nano Simulations
Lab-on-a-chip, used in brewing technology (Siemens)
-
7/29/2019 moldyn
4/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 7of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Applications for Micro and Nano Simulations
Flow through a nanotube (where the assumptions of continuummechanics are no longer valid)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 8of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Applications for Micro and Nano Simulations
Protein simulation: actin, important component of muscles (overlayof macromolecular model with electron density obtained by X-ray
crystallography (brown) and simulation (blue))
-
7/29/2019 moldyn
5/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 9of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Applications for Micro and Nano Simulations
Protein simulation: human haemoglobin (light blue and purple:alpha chains; red and green: beta chains; yellow, black, and darkblue: docked stabilizers or potential docking positions for oxygen)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 10of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Applications for Micro and Nano Simulations
Material science: hexagonal crystal grid of Bornitrid
-
7/29/2019 moldyn
6/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 11 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
A Prominent Recent Example:
Gordon-Bell-Prize 2005 (most important annual supercomput-ing award)
phenomenon studied: solidification processes in Tantalum andUranium
method: 3D molecular dynamics, up to 524,000,000 atoms sim-ulated
machine: IBM Blue Gene/L, 131,072 processors (worlds #1 inNovember 2005)
performance: more than 101 TeraFlops (almost 30% of thepeak performance)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 12of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.2. Essentials from Continuum Mechanics
Fluids
fluid: notion covering liquids and gases
liquids: hardly compressible
gases: volume depends on pressure
small resistance to changes of form continuum:
space, continuously filled with mass
homogeneous
subdivision into small fluid voxels with constant physicalproperties is possible
idea valid on micro scale upward (where we consider con-tinuous masses and not discrete particles)
-
7/29/2019 moldyn
7/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 13of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Lagrange and Euler Formulation
Lagrange formulation:
considers the change of position of material particles
example: velocities of material particles
typical for structural dynamics
Euler formulation:
considers a fixed point in space
example: velocity field, describing the velocities of virtualparticles at fixed positions
typical for fluid mechanics
Arbitrary Lagrangean Eulerian (ALE) formulation:
refers to an arbitrary configuration
example: a particle, moving with its own system of refer-ence through a fluid, having a fixed system of reference
widely used in fluid-structure interactions
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 14of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Description of State
consideration of a control volume V0 (Eulerian perspective)
description of the fluids state via
the velocity field v(x, t) and two thermodynamical quanti-ties, typically
the pressure p(x, t) and
the density (x, t)
for incompressible fluids, the density is constant (if there areno chemical reactions)
-
7/29/2019 moldyn
8/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 15of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Hard Sphere Model
first, atoms are described in a simplified way
as suspended round disks in 2D space, or
as suspended round balls in 3D space
this is called the hard sphere model the fixed (and rather dense) arrangement in a crystal leads to a
solid
for gases, the average distances between particles are verylarge
atoms and molecules of liquids do not have a fixed arrange-ment, but the density compared with gases is higher
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 16of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Solids in 2D
D
3D
D
the maximum density in 2D comes with a triangu-lar packing
an ideal crystal without thermal motion provides aperfect 2D grid
the elementary cell for disks of diameter D forms
a rhombus of areaAE = 2 12 D
123D = 1
23D2 0.866D2
in an elementary cell, there are parts that can becombined to a complete disk:AP =
4 D
2 0.785D2
density Cm = APAE =4D2
12
3D2
= 6
3 0.9069
-
7/29/2019 moldyn
9/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 17of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Solids in 3D
D2D
2D
fill a box with balls of diameter D
periodic arrangement is neither formed automati-cally nor unique
face-centered cubic (fcc) lattice with basis vectorscycl2
2D 2
2D 0
elementary cell: brick,
VE =
2D3
= 2
2D3 2.828D3 VP =
6 1
2+ 8 1
8
6
D3 = 23
D3 2.094D3
Cm =VPVE
=23D3
2
2D3=
6
2 0.7405
VE = volume of elementary cell
VP = volume of atoms
Cm = maximum density ratio
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 18of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Fluids
intermediate states for the transition solid-liquid:
plasticity: irreversible deformation of the crystal lattice alonggliding planes
thixotropy: liquid state after destruction of frame-like ag-glomerations of molecules
for C0 0.8 the disks can escape from their positions 2Dliquid without a periodic crystal lattice; average distance of twoparticles: O(D)
gases: smaller densities (C C0) with an average distance ofO( D
C)
C = density ratio of material to reference cell volume
-
7/29/2019 moldyn
10/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 19of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Scope of Application
number of molecules to be taken into account:
Loschmidt number 2, 687 1019cm3: number of moleculesin 1 cm3 of an ideal gas
Avogadro constant 6.02214151023mol1: number of carbon-
12 atoms in 12g of carbon-12, or number of molecules in 1mol (1 mol, under normal conditions, taking a volume of22.4 litres)
Avogadro number: notion being used in different ways forboth of the above constants, which depend on each other(2, 687 1019cm3 22.413996 103cm3mol1 = 6.0221415 1023mol1)
time steps for numerical simulations are typically in the or-der of femtoseconds (1f s := 1015s)
hence: scope of application is limited to nanoscale simulations(at least for the near future)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 20of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.3. Molecular Dynamics the Physical Model
Quantum Mechanics a Tour de Force
particle dynamics described by the Schrdinger equation
its solution (state or wave function ) only provides probabilitydistributions for the particles (i.e. nuclei and electrons) positionand momentum
Heisenbergs uncertainty principle: position and momentum cannot be measured with arbitrary accuracy simultaneously
there are discrete values/units (for the energy of bonded elec-trons, e.g.)
in general, no analytical solution available
high dimensional problems: dimensionality corresponds to num-ber of nuclei and electrons
= (R1, . . . , RN, r1, . . . , rK, t) - wave functionR - position of nucleusr - position of electront - time
-
7/29/2019 moldyn
11/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 21 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
hence, numerical solution is possible for rather small systemsonly
therefore, various (simplifying and approximating) approachessuch as density functional method or Hartree-Fock approach(ab-initio Molecular Dynamics, see next slide)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 22of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Classical Molecular Dynamics
Quantum mechanicsapproximation classical Molecular Dynamics
classical Molecular Dynamics is based on Newtons equationsof motion
molecules are modelled as particles; simplest case: point masses
there are interactions between molecules multibody potential functions describe the potential energy of
the system; the velocities of the molecules (kinetic energy) area composition of
the Brownian motion (high velocities, no macroscopic movement),
flow velocity (for fluids)
ab-initio Molecular Dynamics uses quantum mechanical cal-culations to determine the potential hypersurface, apart from
semi-empirical potential functions (cf. Car Parrinello MolecularDynamics (CPMD) methods)
total energy is constant energy conservation
-
7/29/2019 moldyn
12/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 23of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Fundamental Interactions
Classification of the fundamental in-teractions:
strong nuclear force
electromagnetic force weak nuclear force
gravity
O
rk
ri
rj
interaction potential energy the total potential of N particles is the sum of multibody poten-
tials:
U := 0
-
7/29/2019 moldyn
13/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 25of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Well-Known Potentials
i j
rij
some potentials from mechanics:
harmonic potential (Hookes law): Uharm (rij) = 12 k (rij r0)2;potential energy of a spring with length r0, stretched/clinchedto a length rij
gravitational potential: Ugrav (rij) = g mimjrij ;potential energy caused by a mass attraction of two bodies(planets, e.g.)
the resulting force isFij = gradU(rij) =
U
rijintegration of the force over the displacement results in the energy or a potential
difference
Newtons 3rd law (actio=reactio):Fij = Fji
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 26of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Intermolecular Two-Body Potentials
2
1.5
1
0.5
0
0.5
1
1.5
2
0 0.5 1 1.5 2 2.5 3
potentialU
distance r
hard sphere potentials
hard sphereSquarewell
Sutherland
2
1.5
1
0.5
0
0.5
1
1.5
2
0 0.5 1 1.5 2 2.5 3
potentialU
distance r
soft sphere potentials
soft sphereLennardJonesvan der Waals
-
7/29/2019 moldyn
14/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 27of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Intermolecular Two-Body Potentials
hard sphere potential: UHS(rij) =
rij d0 rij > d
Force: Dirac Funktion
soft sphere potential: USS (rij) = rijn
Square-well potential: USW (rij) =
rij d1 d1 < rij < d20 rij d2
Sutherland potential: USu (rij) =
rij dr6ij
rij > d Lennard Jones potential
van der Waals potential UW (rij) = 46
1rij6
Coulomb potential: UC (rij) = 140qiqjrij
= energy parameter
= length parameter (corresponds to atom diameter, cmp. van der Waals
radius)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 28of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Lennard Jones Potential
e s
e,s
O
ri
rj
rij
Lennard Jones potential: ULJ (rij) =
rij
n
rij
mwith n > m and = 1
n
m n
n
mm
1
nm
continuous and differentiable (C), since rij > 0
LJ 12-6 potential
ULJ (rij) = 4
rij
12
rij
6 m = 6: van der Waals attraction (van der Waals potential) n = 12: Pauli repulsion (softsphere potential): heuristic application: simulation of inert gases (e.g. Argon)
force between 2 molecules:Fij = U(rij)rij = 24rij
2
rij
12 rij
6 very fast fade away short range (m = 6 > 3 = d dimen-
sion)
-
7/29/2019 moldyn
15/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 29of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
LJ Atom-Interaction Parameters
e s
atom [1.38066 1023J]1 [101nm]2
H 8.6 2.81He 10.2 2.28C 51.2 3.35N 37.3 3.31O 61.6 2.95F 52.8 2.83
Ne 47.0 2.72S 183.0 3.52Cl 173.5 3.35Ar 119.8 3.41Br 257.2 3.54Kr 164.0 3.83
= energy parameter = length parameter (cmp. van der Waals radius)
parameter fitting to real world experiments
1 Boltzmann-constant: kB := 1.38066 1023 JK2 101nm = 1010m = 1 (ngstrm)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 30of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Dimensionsless Formulation
using reference values such as , , reduced forms of the equationscan be derived and implemented transformation of the problem
position, distancer :=
1
r (1a)
timet
:=
1
mt
(1b)
velocityv :=
t
v (1c)
potential (atom-interaction parameters are eliminated!): U := U
U
LJ (rij) :=ULJ (rij)
= 4
r
ij26
r
ij23
(1d)
U
kin :=Ukin
=1
mv2
2
=v
2
2t2
(1e)
forceF
ij :=Fij
= 24
2r
ij26
r
ij23
rij
rij2
(1f)
-
7/29/2019 moldyn
16/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 31 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Multi-Centered Molecules
CA1 CA2
CA
CB1
CB2
CB
FA1B1FA1B2
FA2B1FA2B2
FB1A1FB1A2
FB2A1
FB2A2
FAB
FBA
molecules can be composed with multipleLJ-centers rigid bodies without internal degrees offreedom
additionally: orientation (quarternions), an-gular velocity
additionally: moment of inertia (principalaxes transformation)
calculation of the interactions betweeneach center of one molecule to each centerof the other
resulting force (sum) acts at the center ofgravity, additional calculation of torque
MBS (Multi Body System) point of view: instead of moving multi-centered molecules, there is a holonomically constrained mo-tion of atoms (for a constraint to be holonomic it can be expressible as a function f(r,v,t) = 0)
advantage: better approximation of unsymmetric molecules
there is not necessarily one LJ center for each atom
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 32of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Mixtures of Fluids
simulation of various components (molecule types)
modified Lorentz-Berthelot rules for interaction of molecules ofdifferent types
AB :=A + B
2(2a)
AB := aB (2b)with 1e.g. N2 + O2 = 1.007, O2 + CO2 = 0.979 . . .
A A
B B
e ,sA A
e ,sA A
e ,sAB AB
-
7/29/2019 moldyn
17/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 33of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
NVT-Ensemble, Thermostat
statistical (thermodynamics) ensemble: set of possible states a sys-tem might be in
for the simulation of a (canonical) NVT-ensemble, the followingvalues have to be kept constant:
N: number of molecules V: volume
T: temperature
a thermostat regulates and controls the temperature (the kineticenergy), which is fluctuating in a simulation
the kinetic energy is specified by the velocity of the molecules:Ekin =
12
i miv
2i
the temperatur is defined by T = 23NkB
Ekin(N: number of molecules, kB : Boltzmann-constant)
simple method: the isokinetic (velocity) scaling:
vcorr := vact mit =
TrefTact
further methods e.g. Berendsen-, NosHoover-thermostat
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 34of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Domain
aa
b
b
Periodic Boundary Conditions (PBC):
modelling an infinite space, built from identical cells domain with torus topology
-
7/29/2019 moldyn
18/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 35of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Domain
Minimum Image Convention (MIC):
with PBC, each molecule and the associated interactionsexist several times
with MIC, only interactions between the closest represen-tants of a molecule are taken into consideration
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 36of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.4. Molecular Dynamics the Mathematical Model
System of ODE
resulting force acting on a molecule: Fi =
j=i Fij
acceleration of a molecule (Newtons 2nd law):
ir =Fimi
=
j=iFij
mi= j=i
U(ri,rj)
|rij |mi
(3)
system of dN coupled ordinary differential equations of 2nd or-der transferable (as compared to Hamilton formalism) to 2dNcoupled ordinary differential equations of 1st order (N: number ofmolecules, d: dimension), e.g. independent variables q := r and pwith
pi := miri (4a)
pi = Fi (4b)
-
7/29/2019 moldyn
19/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 37of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Boundary Conditions
Initial Value Problem:position of the molecules and velocities have to be given;initial configuration e.g.:
molecules in crystal lattice (body-/face-centered cell) initial velocity
* random direction
* absolute value dependent of the temperature
(normal distribution or uniform), e.g.32
N kBT =12
Ni=1 mv
2i with vi := v0
v0 :=
3kBTm
resp. v0 :=
3Tt
time discretisation: t := t0 + i t time integration procedure
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 38of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.5. MD Approximations and Discretization
Euler Method
Taylor series expansion of the positions in time:
r(t + t) = r(t) + tr(t) +1
2t2r(t) +
ti
i!r(i)(t) + . . . (5)
(r, r, r(i)
: derivatives)
approximation of (5), neglecting terms of higher order of t, aswell as an analogous formulation of v(t) := r(t) with a(t) :=v(t) = r(t) =
F(t)m
leads to the explicit Euler method:
v(t + t).
= v(t) + t a(t) (6a)
r(t + t).
= r(t) + t v(t) (6b)
implicit Euler method derivatives at the time step end:v(t + t) .= v(t) + t a(t + t) (7a)
r(t + t).
= r(t) + t v(t + t) (7b)
(6a) in (7b) r(t + t) .= r(t) + tv(t) + t2a(t)
-
7/29/2019 moldyn
20/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 39of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Strmer Verlet Method
the Taylor series expansion in (5) can also be performed fort: (Richardson extrapolation for = 1)
r(t t) = r(t) tr(t) + 12
t2r(t) +(t)i
i!r(i)(t) + . . . (8)
from (5) and (8) the classical Verlet algorithm can be derived:
r(t + t) = 2r(t) r(t t) + t2r(t) + O(t4) 2r(t) r(t t) + t2a(t) (9)
direct calculation of r(t + t) from r(t) and F(t)
the velocity can be estimated with
v(t) = r(t).
=r(t + t)
r(t
t)
2t (10)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 40of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Crank Nicolson Method
explicit approximation (11a) for half step [t, t + t2
] inserted intoimplicit approximation (11b) for half step [t + t2 , t + t] gives forv (11c):
v(t +t
2) = v(t) +
t
2a(t) (11a)
v(t + t) = v(t +t
2 ) +t
2 a(t + t) (11b)
v(t + t) = v(t) +t
2(a(t) + a(t + t)) (11c)
alternative conversion to integral equation
v(t + t) v(t) =t+tt
a() d
numerical integration with trapezoidal rule (11c) generalization with further subdivisions in subintervals leads to
Runge Kutta methods. More force-evaluations necessary!
-
7/29/2019 moldyn
21/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 41 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Velocity Strmer Verlet Method
The velocity Strmer Verlet method is a composition of a
Taylor series expansion of 2nd order for the positions (5), and a
Crank Nicolson method for the velocities (11c)
r(t + t) = r(t) + tv(t) +t2
2 a(t) (12a)
v(t + t) = v(t) +t
2(a(t) + a(t + t)) (12b)
tt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt
r
v
F
tt-Dt t+Dt
Forward Euler r Force calculation Crank-Nicolson v
memory requirements: (3 + 1) 3N(calculation of v(t + t) requires not only v(t), r(t + t) and F(t + t), but also F(t) and
therefore 3 + 1 vector fields)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 42of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Dimensionless Velocity Strmer Verlet Method
insertion of the equations (1) leads to:(r := r, v :=
v, t2 := 2 m
t2, r = 1
mF := 1
m
F)
r(t + t) = r(t) + v(t) +t2
2F(t) (13a)
v(t + t) = v(t) +t2
2
F(t) +t2
2
F(t + t) (13b)
procedure:
tt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt
r
v
F
tt-Dt t+Dttt-Dt t+Dt
Forward Euler r Forward Euler v Force calculation Backward Euler v
1. calculate new positions (13a),partial velocity update: +t
2
2F(t) in (13b)
2. calculate new forces, accelerations (computationally intensive!)3. calculate new velocities: +t22 F(t + t) in (13b)
memory requirements: 3 3N
-
7/29/2019 moldyn
22/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 43of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Leapfrog Method
for the Leapfrog method, the velocity calculations are shifted fora half time step with respect to the position calculations:
v(t +t
2
) = v(t
t
2
) + t a(t) (14a)
r(t + t) = r(t) + t v(t +t
2) (14b)
tt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt
r
v
F
t-Dt/2 t+Dt/2t-Dt/2 t+Dt/2 t-Dt/2 t+Dt/2 t-Dt/2 t+Dt/2
tt-Dt t+Dt
exact arithmetic: Strmer Verlet, Velocity Strmer Verlet and
Leapfrog schemes are equivalent the latter two are more robust with respect to roundoff errors
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 44of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Leapfrog Method Thermostat
for the Leapfrog method, the velocity calculations are shifted fora half time step with respect to the position calculations:
v(t +t
2) = v(t t
2) + t a(t)
r(t + t) = r(t) + t v(t +t
2
)
t-Dt t+Dt t+Dt t-Dt t+Dt
r
v
F
tt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt
Thermostat
an intermediate step may be introduced for the thermostat v(t) :=v(t+t2 )+v(tt2 )
2to synchronize the velocity:
vact(t) = v(t t
2 ) +t
2 a(t) (16a)
v(t +t
2) = (2 1)vact(t) + t
2a(t) (16b)
-
7/29/2019 moldyn
23/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 45of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Multistep, Predictor Corrector Methods
Multistep methods:
results are stored for several time steps, which define a(polynomial) interpolant
use the interpolant (extrapolation) for the integration
initialization with single-step-methods
increased memory requirements caused by storage of dataof previous steps data!
Predictor Corrector methods:
1. use explicit method to determine predictor values for t + t
2. implicit method uses predictor values instead of the un-known ones for t + t
3. increased computational effort!4. quality of the predictor step caused by the complex chaotic
behaviour is often not very good
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 46of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Multi-Centered Molecules
for multi-centered molecules, besides position r and velocitiesv, orientations qand angular velocities w have to be also calcu-lated
candidate: explicit or implicit version of the Fincham Leapfrogrotational algorithm
r,v,F using classical Leapfrog method additional orientation q, angular velocity w as well as angu-
lar momentum j
t-Dt/2 t+Dt/2t+Dt/2 t-Dt/2 t+Dt/2
r
v
F,a
tt-Dt t+Dttt-Dt t+Dt tt-Dt t+Dt tt-Dt t+Dt
q
t-Dt/2 t+Dt/2t+Dt/2 t-Dt/2 t+Dt/2j,w
t
t+Dt/2
-
7/29/2019 moldyn
24/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 47of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Evaluation of Time Integration Methods
accuracy (not of great importance)
stability
conservation
of phase space density (symplectic) of energy
of momentum (especially with PBC (Periodic Boundary Conditions))
reversibility of time
use of resources:
computational effort (number of force evaluations)
maximum time step size
memory usage
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 48of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Reversibility of Time
time reversal for a closed system means
a turnaround of the velocities and also momentums; posi-tions at the inversion point stay constant
traverse of a trajectory back in the direction of the origin
demand for symmetry for time integration methods
+ e.g. Verlet method
- e.g. Euler method, Predictor Corrector methods
contradiction with
the H-theorem (increase of entropy, irreversible processes)?(Loschmidt objection)
the second theorem of thermodynamics?
reversibility in theory only for a very short time
Lyapunov instability Kolmogorov entropy
-
7/29/2019 moldyn
25/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 49of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Lyapunov Instability
Example of a simple system:
stable case:jumping ball on a plane with slightly disturbed initial hori-zontal velocity
linear increase of the disturbance
instable case:jumping ball on a sphere with slightly disturbed initial hor-izontal velocity exponential increase of the disturbance(Lyapunov exponent)
for the instable case, small disturbances result in large changes:chaotic behaviour (butterfly hurricane?)
non-linear differential equations are often dynamically instable
calculation of the trajectories: badly conditioned problem;
a small change of the initial position of a molecule may result ina distance to the comparable original position, after some time,in the magnitude of the whole domain!
there are also conserved quantities for whichnumerical simulations make sense!
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 50of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Lyapunov Instability: A Numerical Experiment
setup of 4000 fcc atoms
for a second setup, the position of a singleatom was changed with a displacement of0.001
tracing the movement of the atom for bothsetups, the distance increases for eachstep
colours indicate velocity
2.53
3.54
4.55
5.57.1
7.2
7.3
7.4
7.5
7.6
7.7
3.5
3.6
3.7
3.8
3.9
4
tracing a Molecule (with initial displacement)
Molecule 25, run1
Molecule 25, run2
4.1
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
2.5 3 3.5 4 4.5 5 5.5
Molecule deviation (with initial displacement)
-
7/29/2019 moldyn
26/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 51 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Short-Range Potential
choosing m = 6 (negative exponent in the LJ-potential) fast decay of po-tential and force
for each molecule, an influence volume (closed sphere) withcut-off radius rc (Euclidian metrics) can be assumed where ev-
ery molecule outside this influence volume is neglected
ULJ,rc
rij
=
4
rij26 rij23 for rij rc
0 for rij > rc(17a)
Fij,rc
rij
=
24
2
rij26 rij23 rijrij2 for rij rc
0 for rij > rc(17b)
consider only a subgraph of the interaction-graph the anti-symmetric force matrix, related to this graph, is sparse
the complexity of the calculation can be reducedfrom O (N2) to O(N).
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 52of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Short-Range Interactions
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0 1 2 3 4 5
U*
r*
dim. red. LennardJones 126 Potential
2.5
2
1.5
1
0.5
0
0.5
0 1 2 3 4 5
F*
r*
dim. red. LennardJones 126 Force
1
2
3
4
5
Fij Force matrix/Interaction-graph- F12 F13 F14 F15
F12 - F23 F24 F25F13 F23 - F34 F35F14 F24 F34 - F45F15 F25 F35 F45 -
fast decay of force contributions with increasing distance
dense force matrix with O(n2
), mostly very small, entries
-
7/29/2019 moldyn
27/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 53of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Short-Range Interactions
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0 1 2 3 4 5
U*
r*
dim. red. finites LennardJones 126 Potential (rc=2)
2.5
2
1.5
1
0.5
0
0.5
0 1 2 3 4 5
F*
r*
dim. red. finite LennardJones 126 Force (rc=2)
1
2
3
4
5
Fij Force matrix/Interaction-graph- F12 F13 F14 0
F12 - 0 F24 F25F13 0 - F34 0F14 F24 F34 - F45
0 F25 0 F45 -
cut-off radius leads to a reduction of the computational effortsparse force matrix with O(n) entries
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 54of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Shifted Potentials
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0 1 2 3 4 5
U*
r*
shifted dim. red. finites LennardJones 126 Potential (rc=2)
2.5
2
1.5
1
0.5
0
0.5
0 1 2 3 4 5
F*
r*
dim. red. finite LennardJones 126 Force (rc=2)
ULJ,rc,shifted
rij
=
ULJ
rij ULJ (rc ) for rij rc
0 for rij > rc
Fij,rc
rij
=
Fij
rij
for rij rc0 for rij > r
c
additionally, constant additive term for the potential continuous potentialreduced error for the overall potential
-
7/29/2019 moldyn
28/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 55of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Shifted Potentials
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0 1 2 3 4 5
U*
r*
shifted dim. red. finites LennardJones 126 Potential (rc=2)
2.5
2
1.5
1
0.5
0
0.5
0 1 2 3 4 5
F*
r*
shifted dim. red. finite LennardJones 126 Force (rc=2)
ULJ,rc,shifted
rij
=
ULJ
rij ULJ (rc) FLJ (rc) rij rc for rij rc
0 for rij > rc
Fij,rc,shifted
rij
=
Fij
rij FLJ (rc) for rij rc
0 for rij > rc
additionally, constant additive term for the potential
continuous potential additionally, linear additive term for the potential
continuous force
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 56of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Cut-Off Corrections
due to the cut-off radius, the calculation of
the potential energy
the pressure
neglects some addends with small absolute values
(small) errors
cut-off correction tries to correct this error
constant density and a homogeneus distribution are a prerequi-site
physical values in the calculated volume can be approximatelyextrapolated
-
7/29/2019 moldyn
29/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 57of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.6. MD Implementational Aspects
Verlet Neighbour Lists
rc
rmax
every molecule stores its neigh-bours for a distance rmax > rc
every nupd time steps (dep. on rmax),the lists are updated
the "buffer" has to be larger than thecovered distance of a molecule forthat time:
rmax rc > nupd t vm
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 58of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Classical Linked-Cell Algorithm
molecules are ranged in a lattice of cu-bic cells of side length rc
hash table with"geometrically motivated" hashfunction
"Binning" resp. "Bucketing"-techniques from "ComputationalGeometry"
direct volume representation(voxel) of the influence region
runtime: O(n) only half (point symetry) of the neigh-
bour cells are explicitly traversed (New-tons 3rd law)
erase and generate the data structurein each time step
-
7/29/2019 moldyn
30/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 59of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Variable Linked-Cell Algorithm
lattice might be built up from cellsof side length rc
twith t R+
preferable integer numbers are usedfor the divisor t
N
for t , the examinatedinfluence volume will converge tothe (optimal) sphere
1
2
3
4
5
6
7
8
9
10
1 2 3 4 5 6 7 8 9 1 0
rc/cellwidth
searchvolume/hemispherevolume
t = 1 t = 2 t = 4 t = 3
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 60of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Linked-Cell Algorithm Data Structure I
0 1 2 3 4 5 6 7 8 9 10 11
12 23
24 35
108 119
inner zoneboundary zoneHalo
rc
molecule 1
data
nextincell
molecule 2
data
nextincell
molecule 3
data
nextincell
molecule 4
data
nextincell
molecule 5
data
nextincell
molecule x
data
nextincell
molecule N
data
nextincell
cellseq. 1 cellseq. 2cellseq. x cellseq. i
cells are stored as a one-dimensional array (vector)
intrusive list for the cell molecules
list to determine the processing sequence
-
7/29/2019 moldyn
31/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 61 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Linked-Cell Algorithm Data Structure II
-50 -49 -48 -47 -46
-39 -38 -37 -36 -35 -34 -33
-28 -27 -26 -25 -24 -23 -22 -21 -20
-16 -15 -14 -13 -12 -11 -10 -9 -8
-4 -3 -2 -1
rc
-25 -24 -23
-19 -18 -17 -16
-14 -13 -12 -11 -10
-8 -7 -6 -5 -4
-2 -1
rc
-20
offset mask to determine the neighbours
cache efficiency is influenced by the processing order(temporal locality)
1
2
3
4
5
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 62of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.7. MD Parallelisation
Profiling
!"#$%
&
'&%
()&!*
+,
,-%
,-%
,--%
,--%
+,
'&%
/0
force calculation is dominating
-
7/29/2019 moldyn
32/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 63of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Shared Memory Parallelisation
each process calculates one part (Np
) of the molecules (cells)
availability of all relevant data (position) because of commonmemory
Shared Memory algorithm: Velocity Strmer Verlet method1. parallel explicit Euler method r,v (half step) for N
pmolecules
2. parallel force calculations for Np
molecules or the respectivecells(force summation critical, respecting Newtons 3rd law: reduction; same with
linked-cell algorithm)
3. parallel implicit Euler method v (half step) for Np
molecules
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 64of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Replicated Data I
"Atom Decomposition" Shared Memory parallelisation every node has to store all position data
collective communication for thesynchronization of redundant data
"Atom Decomposition" algorithm: Velocity Strmer Verlet method
1. explicit Euler method r,v (half step) for Np
molecules
2. distribute (gather-to-all) the Np
position data for each PEto all other PEs
3. force calculation for Np
molecules
4. possible distribution of partial forces to the appropriate PEs
5. implicit Euler method for v (half step) for Np
molecules
-
7/29/2019 moldyn
33/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 65of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Replicated Data II
F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14
F15
F16
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
S
Force matrix is only virtual and not allocated/set up
costs
calculation: Np
communication partners per PE: p 1 memory requirements: N positions and N
p
forces
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 66of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Replicated Data II
F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14
F15
F16
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
S
Force matrix is only virtual and not allocated/set up
F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14
F15
F16
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
S
Force matrix is only virtual and not allocated/set up
costs
calculation: N2p communication partners per PE: p 1 memory requirements: taking advantage of Newtons 3rd
law needs a vector for N (partial) forces and additional com-
munication
-
7/29/2019 moldyn
34/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 67of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Force Decomposition I
each process calculates a part of the molecules and the forcematrix
on each node: position data of 2 Np
molecules
communication: distribution of positions and calculated forces "Force Decomposition" algorithm: Velocity Strmer Verlet method
1. explicit Euler method for r,v (half step) for Np
molecules
2. distribution of Np
position data per PE to 2
Np
1
PEs
3. force calculation of a
p Np
sub-matrix
4. distribution of partial forces to
p 1 PEs5. implicit Euler method for v (half step) for N
p
molecules
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 68of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Force Decomposition II
1 2 3 4
5 6 8
9 10 11 1213141516
7
F1
F2
F3
F4
F5
F6
F7
F8
F9
F10
F11
F12
F13
F14
F15
F16
S
1 5 9 13 2 6 10 14 3 7 11 15 4 8 12 16
Force matrix is only virtual and not allocated/set up
costs
calculation: Np
communication partners per PE: 2
p 1
-
7/29/2019 moldyn
35/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 69of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Spatial Decomposition
domain is decomposed into subdomains
each processor handles one subdomain
amount of molecules per processor is variable(molecules are moving!)
overlapping buffer regions (halo, rc) have to be synchronized
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 70of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Spatial Decomposition
domain is decomposed into subdomains
each processor handles one subdomain
amount of molecules per processor is variable(molecules are moving!)
overlapping buffer regions (halo, rc) have to be synchronized
point-to-point communication, dependent of decomposition molecule movement (flow velocity) communication method: "x-y-z" vs. "direct"
-
7/29/2019 moldyn
36/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 71 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Domain Decomposition: Cubes Slices (1)
assumption:
homogeneous molecule distribution
subdomains with Np
molecules and volume Ld
p: N Ld
communication size proportional to halo volume full halo
slices:
2 neighbour PEs
halo volume: Ld12 rc = 2Ld rcL bad scaling properties
relatively easy to implement
special case of a cartesian topol-ogy ( cube)
communication
amount: 2p comm./PE: 2 rc
LN
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 72of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Domain Decomposition: Cubes Slices (2) assumption:
homogeneous molecule distribution
subdomains with Np
molecules and volume Ld
p: N Ld
communication size proportional to halo volume "complete" halo
cubes: 3d 1 neighbour PEs side length: l = d
Ld
p= Ldp
halo volume (l + 2rc)d ld =di=1
d
i
ldi(2rc)i
d1ld12rc = 2 d ld rcl =2 d Ldp
1d1 rc
L
communication
amount:
3d 1p (direct) or 2d p (x-y-z) comm./PE: 2d rc
LN p
1d1
-
7/29/2019 moldyn
37/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 73of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.8. Molecular Dynamics Examples of Nanofluidic
Simulations
1.8.1. Simulating Diffusion
Fluid
a fluid is a continuum (a space continuously filled with mass)without a rigid crystal structure:
liquids: hardly compressible
gases: volume depends on pressure
(looking at isothermal processes)
small resistance to changes of form
length scales of a system have to be large compared to themean free path of the molecules Knudsen number
Kn < 0.01: ideal fluid
0.01 < Kn < 0.1: viscous fluid (Navier Stokes equation)
0.005 < Kn: kinetic (Boltzmann theory)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 74of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
the transport of properties in a fluid is caused by
diffusion
advection
-
7/29/2019 moldyn
38/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 75of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Diffusion
diffusion and thermal conduction are triggered due to the Brow-nian motion
stochastic models of a microscopic model lead to macroscopicmodels
the process is driven for instance through a concentration or atemperature gradient, (i.e. a spatial difference)
00
x
Concentration profiles
Dt=0.1Dt=0.5
Dt=1Dt=2
Dt=10
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 76of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
MD Diffusion Simulation
t = 1 t = 5 t = 10 t = 15 t = 20
0
0.05
0.1
0.15
0.2
0.25
0 5 10 15 20 25 30 35 40 45 50
Diffusion
t=1t=1t=5
t=10t=15t=20
-
7/29/2019 moldyn
39/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 77of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Gradient Diffusion
the Ficks equation provides the power density of the materialdiffusion (D: molecular diffusion coefficient [m2s1]):
M = Dgrad (18)
the thermal flow of a thermal conduction equation (k: thermalconductivity [k g m s3 K1]):
W = k gradT (19)
Gradient: multiplication of the Nabla operator with a scalar function(vector):
grad f = f =
x
y
z
T f =
fx
fy
fz
T
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 78of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Diffusion Equation
change in the state variable in the control volume denotes thesum or the integral of all flows over the surface
m
t=
t
d =
D grad dn (20)
applying Gaussian integration leads to
t d =
div (D grad)dand to the diffusion equation (parabolic differential equation):
t= D (21)
divergence: multiplication of the Nabla operator with a vector (scalarvalue):
div v = v =
x
y
z
T v = vxx
+vyy
+vzz
Laplace operator: = 2, z.B. f = div gradf = 2fx2
+ 2f
y2+
2fz2
Gaussian integration:
div v d =
v dn
-
7/29/2019 moldyn
40/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 79of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.8.2. Nucleation
Nucleation process of supersaturated Argon
t = 10000 t = 125000 t = 230000 t = 360000
nucleation process for an oversaturated Argon vapour at 0.97 Mol/land 80k
the simulation program automatically detects clusters (droplets)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 80of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
0
2
4
6
8
10
12
35030025020015010050
numberofclusters
timeste s 103
Clusters in a supersaturated Argon vapor, 80 K, 0.97 Mol/l
f1(x) = 0.00731 x -2.36f2(x) = 0.00774 x -4.09f3(x) = 0.00749 x -4.698
cluster size > 20cluster size > 30cluster size > 40
f1(x)f2(x)f3(x)
counting and grouping clusters of certain size ranges, a statisticcan be generated
the growth of the clusters (slope) is known as nucleation rateand important for macroscopic simulations
-
7/29/2019 moldyn
41/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 81 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.9. Numerical Methods for Long-Range Potentials
1.9.1. Introduction
so far: focus on short-range potentials such as Lennard-Jones,e.g.
resulting mutual interactions are restricted to particles insome local neighbourhood
facilitates numerical treatment and algorithmic organization:no quadratic complexity induced by an "each-with-each"behaviour
now: tackle long-range potentials, too
examples: Coulomb or gravitation potential
interactions between remote particles must not be neglected
simple cut-off not possible nevertheless need for approaches that avoid quadratic com-
plexity
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 82of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
what is long-range?
intuitively: potential function U(r) does not decrease rapidlywith increasing r
formally (one possibility): for d > 2, potentials not decreas-ing faster than rd for increasing r (criterion: integrabilityover IRd)
typical potentials in applications have both a short-range part
(to be dealt with according to the previous sections) and a long-range part, represented as two additive components:
U(r) := Ushort + Ulong
-
7/29/2019 moldyn
42/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 83of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Two Main Classes
grid-based methods:
decompose Ulong itself into a smooth but long-range and asingular but short-range part
for the latter, use the linked-cell approach again
for the first, there exist special so-called grid-based meth-ods:
* P3M (Particle-ParticleParticle-Mesh) method
* PME (Particle-Mesh-Ewald) method
* SPME (Smooth-Particle-Mesh-Ewald) method)
starting point: PDE-representation of the potential
(x) = 10
(x)
* potential as solution of a potential (Poisson) equation
* efficient solution with standard discretisation techniquessuch as Finite Differences or Finite Elements in case ofsmooth solutions
* hence, feasible for the smooth long-range part and forhomogeneous particle distributions
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 84of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
hierarchical or tree-based methods:
starting point: integral representation of the potential
(x) =1
40
(y)
1
y xdy
advantageous especially for heterogeneous particle distri-butions (frequent in astrophysics, relevant also for molecu-lar dynamics)
examples:
* panel clustering
* Barnes-Hut method
* (fast) multipole methods
-
7/29/2019 moldyn
43/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 85of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
1.9.2. Tree-Based Methods
based on integral representation of the potential
hierarchical decompositions of the domain of simulation
adaptive approximation of the particle distribution
widespread scheme: octrees
allow for separation of near-field and far-field influences
log-linear or even linear complexity can be obtained
high flexibility with respect to more general potentials (as neededfor special applications, such as biomolecular problems)
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 86of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Series Expansion of the Potential
general (integral) representation of the potential:
(x) =
G(x, y)(y)dy
(general kernel G, particle density (y), and domain )
Taylor expansion of the kernel G (if sufficiently smooth apart
from the singularity in x = y) in y around y0:G(x, y) =
j1p
1
j!G0,j(x, y0)(y y0)j + Rp(x, y)
(multi-index j = (j1, j2, j3), j! = j1!j2!j3!, Gk,j(x, y) mixed (k, j)-th derivative (k-th w.r.t. x, j-th w.r.t. y), remainder Rp(x, y))
leads to expansion (and approximation) of the potential:
(x) j1p1
j!Mj(, y0)G0,j(x, y0)
with the so-called moments
Mj(, y0) :=
(y)(y y0)jdy
-
7/29/2019 moldyn
44/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 87of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Subdivision of the Domain
separation of near-field and far-field for given x:
= near far , near far =
decomposition of the far-field into disjoint, convex subdomains:
far =i
fari
note: this decomposition depends on x, i.e. it is done for eachparticle position x (efficient derivation possible from one hierar-chical tree structure)
each fari has an associated point yi0
how to choose the subdivision?
diam
x yi0:=
supyfari
y yi0x yi0
for some suitable constant 0 < < 1
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 88of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
resulting approximation for (x):
(x) =
(y)G(x, y)dy
=
near
(y)G(x, y)dy +
far
(y)G(x, y)dy
= near
(y)G(x, y)dy + i
far
i
(y)G(x, y)dy
near(y)G(x, y)dy +
i
j1p
1
j!Mj(
fari , y
i0)G0,j(x, y
i0)
-
7/29/2019 moldyn
45/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 89of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Error Estimates
error characteristics for one fix particle position x : local relative approximation error for one fari can be shown
to be of order O(p+1)
global relative approximation error (summation over wholefar-field) can be shown to be of order O(p+1)
this clarifies the role of :
allows to control the global approximation error in x
geometric requirement to the far-field subdivision: the closerfari is located to x, the smaller it has to be to fulfil the -condition
hence: a typical "level of detail"
the closer, the higher resolved cf. terrain representation in flight simulators
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 90of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Tree Structures
central question: how can we construct all these necessaryseparations of near-fields and far-fields and subdivisions of far-fields in an efficient way?
idea: recursive decomposition of (a square in 2D, a cube in3D without loss of generality) in cells of different size, termi-nating the subdivision process if a cell is either empty or con-tains just one particle
concepts:
kd-tree: alternate subdivision in coordinate direction (x, y,and z), such that the separation produces two subdomainsthat roughly contain the same number of particles each
quadtree (2D) or octree (3D): subdivision into four congruentsubsquares or eight congruent subcubes, respectively
the following algorithms (Barnes-Hut etc.) use the octree ap-proach
-
7/29/2019 moldyn
46/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 91 of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
kd-Trees Example
3
4
5
7
6
11
9
8
10
12
13
1 2 3 4 5
6
7
9
8 10 1312
11
2
1
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 92of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Quadtrees and Octrees Examples
-
7/29/2019 moldyn
47/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 93of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Recursive Computation of the Far-Field Subdivision
starting point: create the octree corresponding to the set of par-ticles
each node of the octree represents a subdomain of orone cell
for each cell i, define some yi0 (the centre point or the centre
of gravity of all particles contained, e.g.) for doing the Taylorexpansion for each cell i, let the parameter diam just denote the diam-
eter of the smallest surrounding sphere, e.g.
objective: for each particle position x, use as few cells as pos-sible (i.e. as big cells as possible) for fulfilling the diam- rule
hence: start from root node, checkdiam
x
yi0
,
stop if fulfilled (no need for further subdivision) and proceed ifnot yet fulfilled
note that for each x, we typically get a different subdivision
but note also that all these subdivisions are just subtrees of ourconstructed octree
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 94of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Recursive Computation of the Moments
now: use this subdivision for the efficient calculation of the localmoments Mj(fari , y
i0)
direct (numerical) integration or direct summation are not effi-cient
therefore: use hierarchical tree structure to calculate all mo-
ments for all cells in one run and store them crucial property for that:
Mj(1 2, y0) = Mj(1, y0) + Mj(2, y0) ,if the point of expansion y0 is the same
in the (standard) case of different y10 and y20, there are simple
conversion formulas:
Mj(1, y0) =kjj
k
(y0 y0)j
k
Mk(1, y0)
(k j component-wise, multiplicative binomial coefficients)
-
7/29/2019 moldyn
48/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 95of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
this allows for a bottom-up calculation of the moments from theleaves to the root
in the leaves:
if no particle present: zero
if one particle of mass m there in x: m(x yi0)j
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 96of124
Algorithmen des WissenschaftlichenRechnens II
1. Molecular DynamicsSimulation
Hans-Joachim Bungartz
Using these Building Blocks
still to be done for a numerical routine:
how to construct the tree, starting from a given set of particles?
how to store the tree?
how to choose cells and expansion points?
how to determine far-field and near-field?
several algorithmic variants to be discussed in the following
-
7/29/2019 moldyn
49/62
Introduction
Examples
Essentials from. . .
Molecular Dynamics . . .
Molecular Dynamics . . .
MD Approximations. . .
MD Implement...
MD Parallelisation
Molecular Dynamics . . .
Numerical Methods for. . .
Page 97o