Dion Vlachos

1
The emerging field of multiscale simulation in
the chemical and biological sciences

• Dion Vlachos
• Department of Chemical Engineering University of
  Delaware
• Newark, DE 19716
• www.che.udel.edu/vlachos, vlachos_at_udel.edu

Based on the review paper D. G. Vlachos, A
Review of multiscale analysis Examples from
systems biology, materials engineering, and other
fluid-surface interacting systems , Adv. Chem.
Eng., Vol. 30, 1 (2005).
2
What is multiscale simulation?

• Most complex science and engineering systems
  exhibit a wide spectrum of time and length scales
• Multiscaling is the science dealing with modeling
  and simulation of phenomena and models across
  multiple time and/or length scales
• Traditionally, modeling has focused on a single
  model this is a relatively mature approach
• Constitutive eqs. contain coarse-grained physics
• Multiple time scales in IVPs Stiff ODEs
  implicit solvers
• Multiple length scales in BVPs Boundary layer
  theory, asymptotics, homogenization

3
Breakthrough work on materials mechanics
(dislocation, crack initiation)
Broughton et al., PRB 1999
4
DOE vision of multiscale modeling of biological
systems
David A. Dixon, Dept. of Chemistry, Univ. of
Alabama, Tuscaloosa, AL
5
DOE sets goals for multiscale modeling of
biological systems
PNNL report
6
All roadmaps identify multiscaling as a critical,
enabling technology

• A National Research Council (NRC) report
  entitled, Beyond the Molecular Frontier
  Challenges for Chemistry and Chemical Engineering
  (2003), advances thirteen Grand Challenges for
  the field.
• Advancing Chemical Theory and Modeling is
  viewed as one of the critical, enabling
  technologies. Quoting from the report Chemistry
  covers an enormous span of time and space from
  atoms and molecules to industrial-scale
  processing. Advances in computing and modeling
  could help us connect phenomena at the electronic
  and molecular scale to the commercial
  processing.

7
The Multiscale Simulation Paradigm

• Multiscale simulation Predict macroscopic
  behavior from first principles (bottom-up)
• Model may involve two or more models and scales
• As one goes up the ladder, degrees of freedom are
  left out (coarse-grained)

Reviews Chem. Eng. J. 90, 3 (2002) Adv. Chem.
Eng. (Invited, 2005)
8
Reversing the ladder

• The emerging fields of nanotechnology,
  biotechnology, and miniaturization demand passing
  of information from coarser to finer scales
• Multiscale simulation Enables design and control
  of nanoscopic scale behavior via macroscopic
  variable manipulation (top-down)

Reviews Chem. Eng. J. 90, 3 (2002) Adv. Chem.
Eng. (Invited, 2005)
9
Several important achievements catalyze
multiscaling

• Increased computational power
• The observation and prediction made in 1965 by
  Gordon Moore the number of transistors per
  square in had doubled every year since the
  integrated circuit was invented (current
  definition data density doubles every 18
  months)
• Tremendous improvement in mathematics and
  algorithms
• Simulations of more complex systems with higher
  accuracy and efficiency than ever before

10
Computing has changed dramatically

• Early 1990s Alpha workstation

Late 1970s
Late 1980s
11
Older generation of Beowulf clusters
Computing in 1998
12
Beowulf clusters keep involving
2003
2001
13
Newest addition, 2006
14
Cluster Specifications in our Group Hardware

• Grendel, 60 Processors, 268.8 GFLOPS
• 16 1.8 Ghz Intel Xeon, 1 GB RAM, 100-BaseT
• 44 2.4 Ghz Intel Xeon, 2 GB RAM, 1000-BaseT
• 734GB SATA RAID 5
• Spongebob, 36 Processors, 201.6 GFLOPS
• 36 2.8 Ghz Intel Xeon, 2 GB RAM, 1000-BaseT
• 500GB ATA RAID 5
• Patrick, 42 Processors, 362.34 GFLOPS
• 42 AMD Opteron Model 248, 2 GB RAM, RDMA
• 440GB Ultra 320 RAID 5
• Totals
• 842 GFLOPS, ½ of entry 500 in Top500!
• 1.7 Terrabytes of Storage
• 260 Gigabytes of RAM

15
Cluster Specifications in our Group Software

• All systems running Centos 4.2 deployed with
  Warewulf 2.6
• Complete binary compatibility for non-parallel
  jobs
• Commercial Packages
• Fluent, Femlab, Gaussian, ADF, Chemkin, MATLAB
• Compilers
• GNU Compiler Collection, Portland Group
  Workstation, Intel Fortran Compiler with Math
  Kernel Library
• Other codes
• DACAPO, DL-POLY

16
Experimental capabilities have changed
dramatically

• It is now possible to obtain high resolution
  information across multiple scales that enables
  comparison to corresponding simulations

Quantum dots follow Stranski-Krastanov growth
mode
InAs/GaAs, Nottingham, UK
MC simulations (Chatterjee and Vlachos)
17
New multiscale methods are being developed

• Algorithms
• Hybrid models
• Coarse-graining, accelerated method
• Model reduction
• Mathematics
• Theory
• Error estimates

18
An example from materials deposition and control
of thin films
19
Full coupling between fluid (macro) and surface
(meso) scales
Expt parameters Composition, flow rate, P, T
Gas-phase continuum model/Finite difference
Continuum heat mass transfer, fluid mechanics
equations for a stagnation flow geometry
Out
In
Dynamic coupling through fluxes
Substrate
Surface atomistic model, KMC simulation
Solid-on-solid lattice First-nearest neighbors
interactions Adsorption-Desorption-Migration
Vlachos, Appl. Phys. Lett. 74, 2797 (1999)
20
Macroscopic parameters control microstructure
Low temperature rough interface 2D Nucleation
growth mode
High temperature smooth interface Step flow
growth mode
Lam and Vlachos, Phys. Rev. B 64(3), 35401 (2001)
21
Role of macroscopic variables in growth mode
transition
Kisker et al., J. Cryst. Growth 146, 104 (1995)
Lam and Vlachos, Phys. Rev. B 64(3), 35401 (2001)

• Simulations of prototype model are in
  qualitative agreement with experiments

22
An example from cellular engineering
23
Early events of EGF signaling

• Spatial organization of EGF receptors can
  influence characteristics (dynamics) of
• EGF receptor dimerization
• EGF binding
• Intracellular activation and signaling

Cell
EGF (ligand)
Ligand binding
Dimerization
EGF receptors (EGFR)
Activation
24
Proposed Approach Bio-imaging-informaticsComputa
tional Integration of Experimental Data
Time
Images taken from Sako et al., 2000 Van Belzen
et al., 1988
25
Experimental data Single Particle Tracking

• Experiment by Sako et al., Nature Cell Biology
  (2000) High temporal resolution
• Attached fluorescence to ligand and tracked low
  intensity and high intensity spots.
• Concluded Sequence 1 is the predominant ligand
  binding sequence.
• Experimental Limitations
• Low spatial resolution
• Low ligand concentration
• Ability to see only bound receptors

6µm
Sequence 1
invisible
Sequence 3
Low intensity
High intensity
Invisible
Microscopy images taken from Sako et al., 2000
26
Comparison of MC with Single Particle Tracking
Experiment Data
Points Experimental data Curves and error bars
Simulation results
Sequence 2 0-4.9
Sequence 3 0-0.1
Sequence 1 95-100
27
An example from micropower generation
28
Drivers for renewable energy and improved energy
efficiency

• Exhaustion of natural resources
• Dependence on overseas crude oil
• Energy security for the US
• Emissions and global worming
• Climate change
• Distributed energy
• On-board H2 production
• Electric reliability
• Local solutions, e.g., farmers based on biomass
• Portable energy (electronics)

29
The need for distributed energy increases

• On-board production of H2
• In vehicles?
• At gas-stations? Retains existing fuel
  infrastructure
• Local solutions

NH3
Fertilizers
Electricity
Biomass
Biofuel
PEM fuel cell
H2
30
Compact micropower generation

• Increasing amount of portable electronics
• Current batteries are
• Too heavy
• Dont last long enough
• Need to develop alternative portable power
  generation devices!

31
Converting chemical energy into electricity
Fuel processing ?
Electrical Power
Fuel

• High energy density of fuels compared to
  batteries
• More environmentally benign
• Ease of refuel/recharge
• Ammonia leads to COx -free H2 2NH3 N23H2

32
Many problems exhibit multiple geometric length
scales

• Many systems contain hundreds to millions of
  small features
• Resolving each and every feature requires
  tremendous computational resources
• Homogenization theory (1970s) is an excellent
  tool to deal with this geometric disparity in
  scales

Example of mreactor for portable power (Masel)
Example of high surface area insert for portable
power (Vlachos)
33
Things have evolved since the early days

• Established modeling paradigms are already in
  place
• The Bird, Stewart and Lightfoot (1960) paradigm
  came about by 3 drivers
• Establishment of theory (e.g., transport
  equations)
• Development of powerful computational algorithms
  (linear algebra, stiff ODEs, adaptive meshes for
  PDEs,)
• Growth of computer power

• Nowadays CFD is considered as a mature subject!

34
H2 Dissociating Trajectory on a Pt surface

• MD trajectories on potential energy surfaces
  (PES) obtained by DFT (Ludwig and Vlachos, 2004
  2005)
• Coupling Training of PES once

• Challenges
• Phenomena and models are strongly coupled
• Develop bridges between models of various scales
  to enable accurate, robust, efficient, seamless
  coupling

DFT/MD Coupling
Ludwig and Vlachos, Mol. Simul. (2004)
35
Microreactor-thermoelectric integration produces
electricity
Heat Sink
Alumina Electrical Insulators
Thermoelectric Device
Microcombustor
Copper Thermal Spreader
Alumina Insulation
Power Generated W
I
Outlet
Inlet
V
Peripheral Electronic Device
Fuel
Air

• As much as 1 W of power is generated
• Approximately 1 efficient, as 117 W of fuel
  burned

• Hot side of thermoelectric heated by
  microcombustor
• Cold side cooled by heat sink
• Thin alumina electrical insulators used to
  eliminate shorting

Norton, D. G., et al. Proc. of The 24th Army
Science Conference, Orlando, FL.
36
Outlook

• Multiscale modeling can revolutionize how we
  design, model, interpret and control systems
• Enablers
• Computational power
• Computational methods
• Experimental capabilities
• Max potential of multiscaling in emerging
  technologies
• Nano, micro, bio

37
Acknowledgements

• Funding
• NSF, DOE, ARO, ARL/CCM, chemical companies
• Collaborators
• Markos Katsoulakis (UMass)
• Eric Wetzel (ARL)
• Students and postdocs
• Raimondeau, Lam, Snyder, Chatterjee, Ludwig,
  Deshmukh, Lebedeva, Kaisare, Mhadeshwar