Slide 1

Slide 1 text

SKETCHES AT THE FRONTIERS OF COMPUTATIONAL CHEMISTRY ROHIT GOSWAMI Created: 2020-11-19 Thu 05:54 1

Slide 2

Slide 2 text

HELLO GROUP! Find me here: Who? Rohit Goswami MInstP AMIChemE AMIE Doctoral Researcher, University of Iceland, Faculty of Physical Sciences https://rgoswami.me 2

Slide 3

Slide 3 text

BASIC CONCEPTS 3

Slide 4

Slide 4 text

SCALING CONCERNS A horse in motion is too quick to see each moving foot The foot still moves! Sampling is intimately linked to the event considered A measure of rarity 4

Slide 5

Slide 5 text

Objects ∞ ways of cataloging these Classical Objects Quantum Objects Optical Effects (relativity) MODELING CATEGORIES Celestial Bodies Homogenous Bodies People Proteins and Macromolecules Atoms and molecules Electrons and Nucleons F = (mv) d dt HΨ = ι dΨ dt L = L₀ 1 − v² c² − − − − − − √ 5

Slide 6

Slide 6 text

Reasonable Assumptions SEPARATION OF CONCERNS The motion of Halley’s comet does not affect a boulder System size de nition The spinning of the Earth does not invalidate cartesian trajectories for small objects on earth Magnitude can split objects Nucleons are larger than electrons 6

Slide 7

Slide 7 text

POTENTIAL ENERGY SURFACES Electrons move on a potential energy surface de ned by the slower nuclei Adiabatic assumption One electron surfaces Born-Oppenheimer assumption Electrons and nuclei are independent 7

Slide 8

Slide 8 text

Quantum Chemistry Approaches Considers electrons separately Is energy optimal in a sense …. many things happen Quickly becomes unfeasible for larger quantities Density Functional Theory Considers only the density of electrons TWIN APPROACHES TO CHEMICAL SYSTEMS [kohnNobelLectureElectronic1999a] 8

Slide 9

Slide 9 text

RESEARCH PROBLEM? Are the predictions of each method signi cantly different? 9

Slide 10

Slide 10 text

WHAT? A random effects model, since the elements will have correlated results , = ϵⱼ + [n(r)] − ∫ (r)n(r)dv − ∫ E dft ∑ j E xc v xc 1 2 n(r)n(r') |r − r'| = ⟨i|h|i⟩ + [ii|jj] − [ij|ji] E HF ∑ i 1 2 ∑ ij = μ + αᵢ + bⱼ + ϵᵢⱼ y model b ∼ N(0, σ ) ² B ϵ ∼ N(0, σ²) 10

Slide 11

Slide 11 text

METHOD Add a grad student! 11

Slide 12

Slide 12 text

INTRODUCING WAILORD Inherits from ORCA 12

Slide 13

Slide 13 text

Removes the human element Validation of inputs and outputs is automated Integrates with SLURM by jobscript generation Flexible and provides graphing utilities and units Lets us focus on the questions SOME FEATURES linDat = waio.orca.genEBASet(Path("bui linDat["deviation"] = esr.magnitude - linDat["perc_error"] = ( linDat.deviation / linDat.angle.ap ) 13

Slide 14

Slide 14 text

BIKESHEDDING AND ELEGANCE Taking a CS hammer to a QC nail grammar_xyz = Grammar( r""" meta = natoms ws coord_block ws? natoms = number coord_block = (aline ws)+ aline = (atype ws cline) atype = ~"[a-zA-Z]" / ~"[0-9]" cline = (float ws float ws float) float = pm number "." number pm = ~"[+-]?" number = ~"\\d+" ws = ~"\\s*" """ ) 11 C -0.180226841 0.360945118 C -0.180226841 1.559292118 C -0.180226841 1.503191118 N -0.180226841 0.360945118 C -0.180226841 -0.781300882 C -0.180226841 -0.837401882 H -0.180226841 0.360945118 H -0.180226841 2.517950118 H -0.180226841 2.421289118 H -0.180226841 -1.699398882 H -0.180226841 -1.796059882 14

Slide 15

Slide 15 text

WATER AND ANOMALIES 15

Slide 16

Slide 16 text

Rare phase transitions Many phases Topological analysis SCME elegantly deals with long range electrostatics Needs SOAP or other ML for shorter ranges Needs to be upscaled OVERVIEW 16

Slide 17

Slide 17 text

CATALYSIS AND MACHINE LEARNING 17

Slide 18

Slide 18 text

Saddle points on potential energy surfaces indicate transition states Finding these is computationally intensive Important for catalysis Slow implementation in MATLAB Being reworked in C++ with SURFSara, NL Part of REAXPRO SCALING GPR-NEB 18

Slide 19

Slide 19 text

FUTURE DIRECTIONS 19

Slide 20

Slide 20 text

Unfortunately very hush- hush at the moment MAGMAL 20

Slide 21

Slide 21 text

PERSONAL GOALS Strengthen the foundations of QC Bring modern comp sci. and stats to bear Better administration of garpur Plus some functional programming paradigms for users! ………… 21

Slide 22

Slide 22 text

ACKNOWLEDGEMENTS Advisor Prof. Hannes Jonsson Committee Dr. Elvar Örn Jónsson, Prof. Egill Skúlason Also Family, Lab members, Everyone here 22

Slide 23

Slide 23 text

THE END 23

Slide 24

Slide 24 text

BIBLIOGRAPHY Kohn, Nobel Lecture: Electronic Structure of Matter\textemdash Wave Functions and Density Functionals, Reviews of Modern Physics, 71(5), 1253-1266 . . . [kohnNobelLectureElectronic1999a] link doi 24

Slide 25

Slide 25 text

THANKS! 25