Amalendu Chandra. Department of Chemistry and Computer Centre.

Transcription

1 Molecular Simulations and Amalendu Chandra Department of Chemistry and Computer Centre IIT Kanpur

2

3 Computer Centre HPC Facility at CC Old machines: Two linux clusters with 146 dual-cpu compute nodes (292 cores) connected over Gigabit network. Environment - Parallel as well as sequential on open source OS New machines: An HPC facility of about 3000 cores with high speed interconnect GPU servers, high-end workstations

4 Why HPC? To solve complex problems in science and engineering Higher resolution simulations for longer time Sometimes experiments cannot be done!! Computational experiments can be used to simulate extreme conditions Vast expertise in Numerical Methods

5 Applications Hardware Basic science HPC Support Systems Visualization Numerical algorithms A large number of faculty members across various disciplines are involved in computing

6 HPC Multiscale, Adaptive Finite Element Methods using Domain Decomposition Flow Past Bodies with Complex Geometry and Corners Flow Induced Vibrations Analysis of Aircraft Structures Virtual Reality Computational Chemistry Nanoblock Self Assembly Molecular Simulation (Molecular Dynamics & Monte Carlo Methods) Statistical Thermodynamics Geometric Optimization of Large Organic Systems Electronic Structure Calculations Aggregation and Etching Quantum Simulations Thin Film Dynamics Optical / EM Field Calculations Parallel Spectral Element Methods Large Eddy Simulation of Turbulence Vortex Dominated Flows and Heat Transfer Pseudo-spectral Turbulence Simulations Geo-seismic Prospecting Enhanced Oil Recovery Stress Analysis and Composite Materials Vibration and Control Semiconductor Physics, Feynman Integrals Thermal and Hydraulic Turbomachinery Numerical Weather Prediction Turbulence Modelling through RANS Neural Networks Impurities in Anti-Ferro Magnets Raman Scattering Spin Fluctuation in Quantum Magnets Robotics Multi-Body Dynamics Computer Aided Tomography Nuclear Magnetic Resonance

7 The New HPC Setup The Main Cluster: 260 nodes Dual proc; Nehalem Quad core Smaller Test Clusters Dual proc, Nehalem Quad core Servers Nehalem Quadcore/GPU HPC Disk 100 TB storage Infiniband Network (40 Gbps) Visualization Lab High end graphics W/S

8 System Integration Connection with IITK network Linux cluster (260 nodes) GB switches compute nodes Mstr Mgmt Mgmt Mgmt Comp Comp Comp GB switch IB switch layer servers (Multi-node) Switch Smaller Test Clusters comp Comp compute nodes Comp Storage 100 TB disk 8 GB switch

9 New HPC Facility at IITK The integrated facility has a total of 372 nodes and a performance of ~ 30 TF. Ranks 369 globally DST, IITK

10 Goals: High-end research on different areas of computational science and engineering Computational Mechanics Computational Materials Computational Chemistry and Biology

11 A HUB for Collaborative Research HCRI, Alld Alld U Delhi U BHU SGPIMS, LKW AMU Aligarh HPC Centre IITK CDRI, LKW JNU IIIT, Alld MNNIT Allahabad Lucknow Univ Kanpur Univ + HBTI

12 Training and workshops Visitors program Summer schools/workshops International/national conferences HPC users meeting Future plans Academic Programs on HPC in Science and Engineering

13 Molecular lar Simulations

14 Molecular Simulations: Real System Experiment Experimental Results Test of Model Model System Simulation Essentially Exact Results for Model Test of Theory Theory Theoretical Results

15 Basic methods: Molecular l dynamics: Generate configurations from dynamical evolution of atoms Monte Carlo: Generate configurations using random numbers Molecular Dynamics How do you get V? Construct Global potential energy surface from QM calculations. If 10 points are used along each degree of freedom Total number of calculation would be ~ 10 3N!

16 Reduction of dimensionality Write the full many-body potential in the following form The one-body term can be set to zero. Ignore 3 and higher order terms. 3 2 = Pair interaction potential. Find it from quantum electronic structure calculations. For better results, 2 effective 2 Empirical pair potential

17 Pair Potential Approach Solves dimensionality bottleneck problem in constructing the global potential energy surface. But 1. Fails to describe processes where electronic degrees of freedom play active roles. 2. Same pair potential is used in all thermodynamic conditions (usually not accurate) 3. Cooperative or many-body effects can be important Systems where these are not issues Use empirical pair potentials (MM) => Classical simulations Empirical Force Fields

18 Ignores electronic degrees of freedom Calculates energy as a function of nuclear positions only Many of the molecular force fields in use today can be interpreted in terms of a relatively simple five-component picture of the intra and inter-molecular l interactions. ti V( r N ) = Bond stretching + Bond bending + Bond rotation (torsion) + Non-bonded interactions Electrostatic van der Waals Ref. A. R. Leach, Molecular Modelling, Addison Wesley Longman (1998)

19 Simple water models The simple water models use between 3-5 interaction sites and a rigid water geometry SPC SPC/E TIP3P BF TIP4P ST2 r(oh), Å HOH,deg A 10 3,, kcal Å 12 /mol C,Kcal Å 6 /mol q(o) q(h) q(m) r(om), Å Dipole moment D For another successful model see: A. Chandra and T. Ichiye, J. Chem. Phys. 111, 2701 (1999)

20 Ab initio Molecular Simulations Empirical pair potentials 1. Fail to describe processes where electronic degrees of freedom play active roles. 2. Same pair potentials are used in all thermodynamic conditions (usually not accurate) 3. Cooperative or many-body effects can be important Systems where these are significant issues, calculate the fullmany-body potential from quantum mechanical calculation by considering the system at the level of electrons, protons, neutrons (QM). Ab i iti i l ti > 10 5 ti i Ab initiosimulations => 10 5 times more expensive computationally

21 Ab initio Molecular Dynamics : Forces are calculated from electronic structure calculation that are performed on-the-fly as the MD trajectory is generated. Car-Parrinello Method: R. Car and M. Parrinello, Phys. Rev. Lett., 55, 2471 (1985)

22 Hybrid Quamtum-Classical Molecular Simulations Classical simulations using empirical pair potentials are computationally efficient But 1. Fails to describe processes where electronic degrees of freedom play active roles. 2. Same pair potential is used in all thermodynamic conditions (usually not accurate) 3. Cooperative or many-body effects can be important Where these are not issues Use classical simulations (MM) Otherwise, use ab initio simulations (QM) Systems having both issues & not issues QM + MM

23 Hybrid Quantum-Classical (QM/MM) Method S I (QM) O (MM) I-Inner subsystem, O-Outer subsystem

24 The QM/MM Energy Schemes Subtractive QM/MM scheme sub EQM / MM ( S) EMM ( S) EQM ( I L) EMM ( I L) Additive QM/MM scheme add EQM / MM EMM ( O) EQM ( I L) EQM MM ( I, O) E QM-MM ( I,O) term defines a particular QM/MM method vdw el b E ( I, O) E E E QM MM QM MM QM MM QM MM Bonding Van der Waals Electrostatic Interaction Interaction Interaction

25 Summary

26 Some Applications Hd Hydrogen bond dfluctuations in water and aqueous solutions A. Chandra and coworkers, Phys. Rev. Lett. (2000), J. Chem. Phys. (2008), J. Phys. Chem. A (2008)

27 Vibrational spectral diffusion in water Frequency vs hydrogen bond distance (pure D 2 O) B.S. Mallik, A. Semparithi and A. Chandra, J. Phys. Chem. A (2008) 16 P4 processors used for 3 months for 100 ps trajectory

28 Dynamics after hole creation: Excitation in blue/red Calculated Calculated Experiments: G.M. Gale et al PRL, 82, 1068 (1999)

29 Supercritical water, ammonia and solutions: ScH 2 O 673K/ K/ K/ K/ A.K Soper et al., JCP, 106, 249 (1996)

30 Supercritical water B.S. Mallik and A. Chandra J. Phys. Chem. A (2008)

31 Excess electrons and metal atoms in liquids and clusters consisting of water and ammonia molecules Pratihar and Chandra, J. Chem. Phys. (2007, 2008, 2010), J. Phys. Chem. A (2010).

32 Ab initio MD of excess electron in water cluster at 150K S. Pratihar and A. Chandra, J. Phys. Chem. A (2010, in press).

33 Photoelectron spectra (calculated) Experimental M. A. Johnson et. al., J. Phys. Chem. A 2005, 109, 7896

34 Metallic lithium-ammonia solutions

35 Metallic lithium-ammonia solution Calculations took 2.5 years using 16 processors at IITK and 24 processors in Bochum, Germany A. Chandra and D. Marx A. Chandra and D. Marx Angew Chemie Int Ed (2006)

36 Proton transfer in water

37 D H3 O + = 9.31x10-5 cm 2 /s Mobility of H id O + in liquid water D H2 O = 2.30x10 cm /s Mobility of OH - in liquid water D OH_ = 5.3x10-5 cm 2 /s A.Chandra, M. Tuckerman and D.Marx, Phys. Rev. Lett. (2007); Chem. Rev. (2010). Also see, M.E. Tuckerman, D.Marx and M. Parrinello, Nature, 417, 925 (2002)

38 Proton transfer in water monolayers and chains How do the mechanism and rate of proton transfer in one and two dimensions differ from those in liquid water? A. Bankura and A. Chandra (to be published)

39 Confinement induced water structures inside single walled carbon nanotubes d = 8.1 Å d = 10.8 Å d = 13.6 Å d = 16.3 Å d = 19.0 Å

40 Ab initio molecular dynamics of water-ccl 4 interface The hydrogen bonding environment of interfacial water molecules can be very different from bulk water.

41 Vibrational Power Spectrum

42 The probability distribution of the nearest H and Cl distances at interface

43 Some outstanding problems > Chemical reactions at surfaces and interfaces > Control of chemical reactions under extreme conditions Reaction pathways of heterogeneous catalytic and enzymatic reactions Quantum effects in chemical processes in large macroscopic systems Real time quantum dynamics in many-body quantum potentials at finite temperature

44 Support from: DST, CSIR, BRNS, MCIT, INSA, AvH, IITK