First French-Japanese WorkshopPetascale Applications, Algorithms and Programming(PAAP)A grand challenge application for the next-generation supercomputer in the soft nano-scienceFumio HirataInstitute for Molecular Science

Grand challenge applications in nano-science(The next-generation Integrated Nanoscience Simulations)Material science for the information technologypost-silico electronic devise) molecular switch, nano-wire, etc.

Material science for the biotechnology (medicine, pharmaceutical) virus, cancer drug, drug delivery, etc.

3Material science related to effective use of the solar energy (environment and energy shortage) solar cells, enzyme (cellulase), super capacitor, etc.These problems are grand challenges in dual senses. 1. important for the society (economy, medicine, environment) 2. unsolved scientific problems

What is nanoscience?Why makes nanoscience so challenging? micro nano macro (10-1110-8 M) (10-910-6 M) ( 10-6 )electrons, atoms,moleculesmacromoleculesmolecular assemblyQuantum mechanicsMechanicsThermodynamicsHydrodynamicsElectromagnetismStatistical Mechanicsvisible materialsmicro chip (IC, LSI)car designetc.molecular deviceanticancer drugEnzymatic reactionsetc.???Multi-scale & multi-physics

Molecular (microscopic) theories to be applied to nano-phenomenaHard nano-phenomena: (ex. electron conduction in molecular wire)

Band theory, DFT, Hubbard model, ab-initio MD (Car-Parrinello), QEDSoft nano-phenomena: (ex. Enzymatic reactions) Molecular simulation (MC, MD), Car-Parrinello method Generalized ensemble method (Replica exchange, etc.) Statistical mechanics of liquids (RISM, .) Molecular orbital(MO) theory (FMO, DFT, ONIOM) None of a single theory can explain an entire nano-phenomenon

celluloseethanolenzymatic reaction(The most efficeint way of decomposing cellulose)

Energy cyclecellulosesugarethanol+Carbon dioxideenzymeFermentationwith enzymePhoto synthesisAs foodCellulose as Energy resorcecellulaseEnzyme to decompose cellulosehuman being does not haveexist in bacteria (yeast)Celulase CELCWe use the peta machine to design an enzyme to decomposeCellulose.

catalystsenhances reaction rate dramatically.(1000 to 1 million timesexample: binap by Prof. Noyorienzymebiocatalystcatalyses almost all chemical reactions occurring in our body

featureit works in watertheory of water is essentialit should accommodate substrate molecules in the active-sitemolecular recognition

The reaction to alcohol from cellulose stepscellulosesugar (glucosesugar alcohol

Enzymes concern both reactions, but the mechanismof the first step has not been well understood.What is the enzimatic reaction?Why is it difficult to treat by theory?reactantproductwith enzymeor catalystwithout enzymeaccelerate a reactionrecognitionreactionreleasingcellulose-glucoseenzymatic reaction

Important factors in enzymatic reactions:

Intake and release of substrate molecules by enzyme (molecular recognition) affinity (free energy) between protein and substrate structural fluctuations of protein

methodology: RISM/3D-RISM(statistical mechanics), Gneralized Langevin Dynamics (statistical mechanics)(2)Chemical reactions (hydrolysis, redox, etc.) in ProteinChange in the electronic structure

methodology: 3D-RISM-FMO

In any case, water (solvent) plays essential role

SN2()SN2SN2()!

RISM-SCF theory predicts chemical reactions in solutionsMenshutkin reactionH3N + CH3Cl H3N+CH3 + ClSolvent effect

3D-RISM theoryInput:o protein structure protein data bank(PDB) o the solute-solvent interactions, uuv(r)o the correlation functions of solvent, wvv(r), hvv(r)o temperature, b = 1/kBT; the density of solvent, rOutput: The distribution function of solvent atoms normalized by solvent density gg(r) =hg(r) + 1 gg(r) = rg(r) / rgO(r) > 2Isosurface representation of the 3D-distribution function of water oxygen

Results (1-1) Cavity 1: 3D distribution gO(r) > 8gH(r) > 8(a) 3D distribution functions of water(b) Hydration model reproduced from (a)(c) X-ray structure=o Hydration structure in Cavity 1 determined by 3D-RISM is in excellent agreement with the X-ray structure(Imai, Hiraoka, FH JACS communcation, 2005)

A92DIn NaCl solutionWild typeQ86DThe wild type andQ86D do not bind Na+, while A92DDoes.A92D/Q86DIn CaCl2 solutionA92D/Q86D bindsCa2+Selective ion binding by different mutants of human lysozimeCorrespondign experiments done by Kuroki-Yutani

Questions asked for aquaprins.Why aquaporin does not permeate proton? Does aquaporin permeate Ions? How and what extent?What is the gating mechanism of the channels?What is the conduction mechanism of aquaporin?What is the role of c-GMP in aquaporin as an ionChannel? Aquaporin (water channel)Works in our body to control water concentration (kidney, eye, etc.)

ExtracellularIntracellularWater channelIon channel ?

Why aquaporin dose not permeate proton?Two mechanisms concievable for proton conduction.

Diffusion as a hydronium ion. proton jump (Grotthuss mechanism)

Enzymatic reaction to decompose cellulose into sugarhydrolysis reactionWater is one of reacting species (substrate)The position of water molecule in the reactionPocket is essential.

The finding so far is a big step toward final solution, but not quite enough to predict entire enzymatic reaction.

Following problems have not been done yet.

To realize the free energy profile, the electronic structure should be calculated along the reaction coordinate. 3D-RISM/FMO

Dynamics of protein should be done in order to take into account the protein fluctuation.

3D-RISM/MD Huge amount of calculation should be made.Key words,3D-FFT and Eigen-value-problem

3D-RISM Theory3D-RISM equationHNC ClosureSolute-solvent interaction potentialConvolution integralJust a multiplicationin the Fourier space3D-DF

Flow chart1.Potential parameter for solute and solvent molecules2. Calculate the interaction potential energy3. Initial value of 4. Convert by 3D-FFT5. Solve 3D-RISM in k-space6. Inverse transform of c by 3D-FFT7. Solve HNC eq. to get 8. Go back to 4 if is not converged9. Calculate 3D-distribution function from and cINPUT:Potential Parameters of solute and solvent moleculeStructure of solute and solvent moleculeOUTPUT:Distribution functions of solvent moleculeSolvation free energy3D-RISM equationClosure equationInteraction Potential inverse 3D-FFT3D-FFT

Electronic structure in solution3D-RISM-FMOCombined 3D-RISM/FMO calculationSolve solvent distribution and electronic structure self-consistentlyBottle neckSolvated Fock and electro-static potential easy to parallelizeFMO3D-RISM 3D-FFT

FMO and Solvated Fockand potential) is most expensive, but those can be readily parallelized. fragment SCF3D-RISMConverge?UV potentialSolvated FockFragment pair SCFSolvated Fockfor fragment pairFast eigen-value-problem solverIs essential!

Fluctuation of protein in soultion3D-RISM/MDDescribe sovent with 3D-RISM, while move solute with MD.Solvent is always equilibrium to the solute structure.Bottle neck necessary to accelerate 3D-FFTGradient redily parallelizedMD()3D-RISMgradientpotential

3D-RISM/MDSimulation of protein in solvent atoms (protein)Solvent (waterelectrolytes, etc.)0.5If one use SR110004node/64core(2Tflops)320sec/iteration(3D-RISM,62%: gradient,38%: others, 0%)1ns with time step 1fs, multi time step3 yearsIf one use ( times faster than SR11000), 1.5 hoursIf this became reality, the protein folding can be done.

Thank you for your attention.We do not simulate the earth, but try to save it from the energy and environmental crisis. But, in order make it reality, we need 3D-FFT andeigen-value-problem solver well tuned for the nextgeneration super-machine.