+ Computational Materials Design w/ 1st Principles Quantum Mechanics
atomistic architecture of a material is key to material sciences.
first principle quantum mechanical calculations allows capability to
predict atomistic structures.
investigation of site-selective adsorption of carbon atoms on A1 surfaces.
Carbon is a common contaminant of aluminum surfaces, carbon is adsorbed on
Al surface in 2 geometries
(1) a fourfold-coordinated site
(2) a sixfold-coordinated site. more stable by .1ev / atom
Based on first-principles density functional theory with a local density
approximation and a pseudopotential plane wave method.
"Density functional" relates the total electron density and the
total energy. Thus derive effective 1-particle Schrödinger equations, which
solve iteratively to give the electronic states, the charge density and the
total energy.
+ Off-the Shelf chips conquer the Heights of computing
Supercomputer industry is taking a turn.
Through history, supercomputers have been built w/ specialized processors.
pricey custom-made computer chips for unmatched performance .
But massproduced microprocessors used in computer workshtations are
fast enough now, that silicon graphics , IBM, et al. are
building state-of-the-art supercomputers
by stringing together less powerful but far less expensive
components. Commodity microprocessors continue to improve
far faster than the custom-built couterparts.
Applications reflect what problems are amenable to the particular
type of supercomputer. Since the new generation can deal w/
larger databases, "datamining" looking for relationships in
large data sets will grow in popularity
In the 80's battle of the vector supercomputers.
Cray was the most powerful because
it used custom processors to perform calculations simultaneously on
long strings of numbers - vectors - instead of +,-,*,/ two at a time.
Instead of relying on a few high power, high cost processors, why not
amass hundreds of thousands of cheaper, lowpower processors and provide the
same computing power at a fraction of the cost? MPP Born!
(Massively parallel processing).
powerful, Commodity microprocessors changed everything.
Mass-produced chips at the heart of computer workstations.
1985 : Workstation microprocessors 1megaflop
Cray's Processors 200megaflops. (1 flop = 15digit#*15digit#)
1995 : Cray Processor = 2gigaflobs
Microprocessors DEC Alpha 600megaflop
Microprocessors HP's 800megaflop (12/95)
Microprocessor cost is a fraction of traditional vector processor.
Vector processors, consist of complex collections of chips, the
microprocessor are designed for mass production. Each is a self-contained
miniature computer made on a single bit of silicon only a centimeter or
so across, in a process honed by 2 decades of experience making integrated
circuits.
The inexpensive power of the microprocessor has made possible a
new type of parallel supercomputer. Known as the symmetric multiprocesser
(SMP) , it connects up to 220 commodity microprocessors to a single large
memory instead of distributing the memory among its processors as
MPPs do.
350 |____
n | \____ _______massively parallel
u | _\_/_
m | ___/ \____parallel vector
b 150 |___/
e | symmetric multiprocessor (microprocessor based)
r | ___/
| ___/
0 +---------------------
Silicon graphics, 18microprocessor, each 300+megaflops
peak speeds : >5gigaflops.
cost : $1 million
Cray 16procC90(flagship), peak speed : 16gigflops
cost : $30 million
The lower price tag has made SMPs *very* popular.
It's possible to make an SMP cluster. Several machines
connected with fiber optic cable coordinated through S/W.
MPPs are coming to dominate the top end of the market thanks to
commodity microprocessors - which has more memory than
traditional supercomputers since each processor
has its own memory. The extra memory suits the machines to
data-intensive applications. Imaging or comparing
observational data w/ predictions of models.
100 | \ MIPS R2000
| \ microprocessors
n | \ HP 7000
a | ___ \ MIPS R4000
n | \___ \
s 10| \___ \
e | \___\ DEC Alpha
c | vector
1| supercomputers
| Cray 1S, XMP,YMP,C90
+---------------------
75 1995
Chips are up. Microprocessors now rival vector
processors in speed and number of top supercomputers.
+Computer Scientists rethink their discipline's foundations
Quantum Dot Computers :
Possible Speed Increase : 100-10000 depending on size of the smallest dots
that can be produced in regular arrays.
Current Status : Experimental tests are just beginning (1995)
Negatives : Early systems will operate at low temperatures
How : replace transistors with spots of semiconductor so small they
accomodate 1 electron each.
At small scales, electrons behave not like point particles, but as
waves. So, hard to handle when circuit elements @ those scales.
As e- move wave function spread out, making them apt to tunnel through
thin walls between circuit elements and cause circuit to malfunction.
Trap individual e- wi/ quantum dots, blobs of semiconductor
so tiny an e- around an atomic nucleus the e- wave in a dot is
forced to occupy a specific energy state. Narrowing the wave and
holding it tight. If laid out at corners of a square, electron's
mutual repulsion will force them to burrow to opposite corners of
pattern.
o---X X---o
\ / \ /
o "1" or o "0"
/ \ / \
X---o o---X
Quantum Computers :
Possible Speed Increase : A trillion or more
Current Status : Experiments to create single logic gates are under way.
Negatives : Limited range of problems that such computers can
solve; exponential accumulation of errors.
How : Build components of quantum computers take advantage of
single electron wave's ability to exist in many different state
at once. Eachstate representing a piece of information. All
processed in parallel.
Holographic Association
Possible Speed Increase : <=100,000
Current Status : Experiments in a variety of holographic media
Negatives : Limited to specific applications e.g. pattern recognition, AI
How : use a single large laser, bright enough to force a strong
response from existing nonlinear materials to drive many computations
at once. encode light beams with images or digital information and
mix them in nonlinear media as diverse as gases , crystals, and
bacterial proteins. The nonlinear medium allows information in one
light beam to affect how the material "processes" a second beam. In
effect, the medium performs many computations in parallel. Enabling
"holographic association" to compare two data sets tens of thousands
of times faster than existing supercomputers.
Optical Computers :
Possible Speed Increase : 1000-100000 wo/ accounting for extra parallelism
since photons can pass through one another
Current Status : Individual components exist
Negatives : Better switches need to be developed
How : Light pulses fastest messenger in nature, pass through one another
without effect. Allows any number of activities to take place
simultaneously in an optical circuit.
Some optics researchers aare learning how to make light guides to
carry photons around on a chip. Tiny, superefficient lasers
generate pulses of light at play in an optical circuit.
Induced lasing in a ring of semiconductor 4.5microns across. Ring=
resonating cavity intensifying laser light. Since circumference
is only several wavelengths long, other wavelengths can't
resonate within it and siphon off energy.
The laser photons escape from the ring to a surrounding
waveguide, where they drive an optical circuit. problen:Such a circuit
switches 1 light signal to alter another (qv transistor)
DNA Computers :
Possible Speed Increase : <=1,000,000,000,000
Current Status : Proof or principle and more sophisticated tests begin
Negatives : Limited range of problems, errors, potential practical barriers
How : simultaneous parallelism via soup of DNA molecules.
2^70 DNA molecules (a few liters) act as individual processors.
Nucleotide sequence of each molecule encodes a possible solution
to the problem. By applying the techniques of molecular biology
to clone,combine and select subsets of the molecules, the operator
of this biochemical computer can force the system to sort through
the entire astronomical range of possible solutions, leaving the
correct sequence to be extracted and read out. Each step
takes minutes to hours, vs billionth of a second / operation for a
supercomputer - but each step also acts on the whole panoply of
molecules at once. The result in the most optimisic scenarios, is
a theoretical advantage of 10^8 - 10^12 in computing speed.
The advantage holds only for problems solved efficienty by following
many parallel computational paths (e.g. cryptography).
Applications :
model global climate,
embryonic development
elementary particle interactions,
simultaneously optimizing jet's structural mechanics acoustics, manuf & $
+ Keplerian complexity:Numerical simulations of accretion disk transport
Supercomputer simulations have been used in conjunction with analytic studies
to investigate the central issue of astrophysical accretion-disk dynamics:
the nature of the angular momentum transport. Simulations provide the means
to investigate and experiment with candidate mechanisms , including global
hydrodynamic instabilities, spiral shock waves, and local magnetohydrodynamic
instabilities. Simulations have demonstrated that accretion disks are generally
MHD turbulent. These results suggest that the fundamental physical mechanism
for angular momentum transport in accretion disks has now been identified.
--fin
