Computer simulations and the Laplace demon
Alessandro Laio, SISSA (Trieste)
Capability to predict the future Lots of demon-like features Super-human capabilities not to get bored with numbers
Requires continuous attentions and sacrifices. Otherwise it gets angry.
Computer simulations and the Laplace demon
Alessandro Laio, SISSA (Trieste)
Capability to predict the future Lots of demon-like features Super-human capabilities not to get bored with numbers
Requires continuous attentions and sacrifices. Otherwise it gets angry.
Computer simulations: ... deriving from simple equations complex and realistic predictions ... Simple equations, althogh beautiful, contain the description of our world only virtually
Given a potential energy surface: The dynamics is determined from Newton’s equation: Molecular dynamics Extremely efficient
Parallel and highly scalable implementation
Etc…. Modern MD code More than 3 decades of work by hundreds of people!!!
Accuracy:
the more accurate the description, the more computationally expensive. Size:
interesting systems are large and inhomogeneus Time-scale:
chemical reactions, phase transitions, conformational changes are “rare events“ Three compeeting demands
Which level of description should one choose? Accuracy:
the more accurate the description, the more computationally expensive.
The cheap option:
Classical Potentials Many popular force fields (Amber, Charmm, Gromos, OPLS, etc.) differ only for the value of the parameters (charges, torsions,…) . Bonded Electrostatic Van der Waals
Schrödinger equation The accurate option:
dealing with the electrons Newton equation + = Car-Parrinello molecular dynamics
500000 “moves”= 1/1,000,000,000 OF A SECOND IN ONE DAY!!!!! Simulation of "realistic" systems: what we can afford. Example:simulation of HIV protease
(classical potential)
50000 atoms (protein+water) Each atom “interacts” with ~ 100 atoms (its neighbors) In order to calculate the forces, 50000*100 operations A computer can perform 5000000 operations in 0.2 seconds In one day I can “move” the system 3600*24/0.2~500000 times
Quantum potentials (electrons are explicitly treated: chemical reactions):
1/100,000,0000,000 of a second for a 100 atoms system Classical potentials (no chemical reactions):
1/1,000,000,000 of a second for a 50000 atoms system
Simulation of "realistic" systems: what we can afford
(one day of simulation)
what we will be able to afford in the future Blue Gene (IBM):
65,536 "Compute Nodes" and 1024 "IO nodes“.
360 TFLOPS=360000 desktop PCs One millisecond of molecular dynamics of a protein in one day!!!!
what they will be able to afford in the future Blue Gene (IBM):
65,536 "Compute Nodes" and 1024 "IO nodes“.
360 TFLOPS=360000 desktop PCs One millisecond of molecular dynamics of a protein in one day!!!!
what they will be able to afford in the future In Italy:
MD simulation of the satellite tobacco mosaic virus
P.L. Freddolino, A.S. Arkhipov, S.B. Larson, A. McPherson & K. Schulten 1 million atoms!!!
Simulation time: 50 ns, program: NAMD
The simulation would take a single 2008 desktop computer around 15 years to complete!!!
CAPSIDE (60 copies)
1194 atoms,
10 GUA-CYT pairs
200 water molecules
3960 electrons!!! A single configuration of the system occupies ~20 Gbytes of memory!! Car-Parrinello simulation of Z-DNA
(F.L. Gervasio, P. Carloni & M. Parrinello)
Accuracy:
the more accurate the description, the more computationally expensive. Size:
interesting systems are large and inhomogeneus Time-scale:
chemical reactions, phase transitions, conformational changes are “rare events“
Time-scale:
chemical reactions, phase transitions, conformational changes are “rare events“
Direct simulation is hopeless, even if you have access to a Blue Gene supercomputer. Azulene Naftalene ? Time-scale:
chemical reactions, phase transitions, conformational changes are “rare events“ Car-Parrinello molecular dynamics
Simulating rare events requires some „computational wizardry“ Local elevation, Wang-Landau sampling, metadynamics: in order to observe a transition, fill the wells with “computational sand”
Azulene Naftalene Molecular dynamics with “computational sand” Normal molecular dynamics
Solid Liquid Freezing water on a computer
(D. Donadio, P. Raiteri & M. Parrinello)
Comments