Topological characterization of adsorption phenomena using multibody potential expansions B. Ganapathysubramanian and Prof. Nicholas Zabaras Materials Process Design and Control Laboratory Sibley School of Mechanical and Aerospace Engineering 188 Frank H. T. Rhodes Hall Cornell University Ithaca, NY 14853-3801 Email: zabaras@cornell.edu URL: http://mpdc.mae.cornell.edu CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory OVERVIEW 1. Problem statement 2. Multibody expansions: Representing the PES 3. Constructing the Multi body expansions: Large dimensions, interpolation and the Smolyak algorithm 4. Simple problems in adsorption 5. Coupling MBE with a Grand Canonical simulator 6. Towards topological design 7. Conclusions CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Alternate means of energy production “Catching up energy production with energy demand (is) one of the top 10 problems for the next 50 years” – Prof Smalley Among the most promising means is through fuel cells. Chemical reaction or combustion produces heat and electricity with high efficiency Anode: 2H24H++4e- Cathode: O2 +4H+ +4e-2H2O Cell: 2H2(g) + O2(g) 2H2O(l) Advantages: High efficiency Fuel can be obtained from sources other than petroleum CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Alternate means of energy production for mobile applications Chemical reaction or combustion produces heat and electricity with high efficiency Major issue is the onboard storage of the fuel (hydrogen) Need to store atleast 4 kg of hydrogen for commercial usage of hydrogen1. Many techniques investigated: Most promising is the physisorbtion of hydrogen onto metallic and metallic-hydride surfaces L. Schlapbach, A. Züttel, Hydrogen-storage materials for mobile applications, NATURE 414 (2001) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Towards designing materials with enhanced adsorption Platinum based surfaces have large potential to adsorb hydrogen Recent developments have shown that alloying platinum with metals like Bi and Rb produce cheaper surfaces with similar properties This is the first aspect of designing materials for enhanced adsorption behavior Top layer Adsorption is essentially a surface phenomena Can the surface be designed to enhance adsorption? Research shows that certain surfaces and topological characteristics improve coordination of hydrogen Q. Wang, J. K. Johnson, Optimization of Carbon Nanotube Arrays for Hydrogen Adsorption, J Phys. CHem B CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Issues with modeling adsorption on metallic surfaces Hydrogen molecule or hydrogen atom? Hydrogen molecule, trajectory and velocity of approach is important for chemisorption. Recently shown that scattering of H2 is electronically adiabatic 1. Accurate potential energy surface to find adsorption sites Quantum delocalization effects: hydrogen appears to be smeared out on the surface Medium range effects due to smearing 1. P.Nieto, et. al, Reactive and Nonreactive Scattering of H2 from a Metal Surface Is Electronically Adiabatic, Science (2006) 312. 86 - 89 CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Modeling adsorption on metallic surfaces To take into account the quantum effects need an essentially ab-initio approach. Various studies have been performed that investigate the adsorbtion of hydrogen on metallic (specifically Pt) surfaces in a quantum mechanical framework In the context of designing topological features one needs to necessarily model larger scale structures (~O(μm)) Need a abinitio level accurate strategy that can model large structures in a computationally tractable way 1. Watson G et. al, A comparision of the adsorption and diffusion of hydrogen on the {111} surfaces of Ni, Pd, and Pt from density functional theory calculations, Journal of Physical Chemistry 105, 4889-4894 (2001) 2. G. Källen, G. Wahnström, Quantum treatment of H adsorbed on a Pt(111) surface, Phys Rev B 65 (2001) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Multi-body expansion Total energy 1,2 Position and species Symmetric function Total energy is the sum of energies of higher and higher levels of interaction • • • • • All degrees of freedom included No relaxations needed Needs a database of calculations, regression schemes required Periodicity is not required (large cell, one k-point calculation) Can predict energies over several different lattices 1. R Drautz, M Fahnle, J M Sanchez, General relations between many-body potentials and cluster expansions in multicomponent systems, J. Phys.: Condens. Matter 16 (2004) 3843–3852 2. J W Martin, Many-body forces in metals and the Brugger elastic constants, J. Phys. C, 8 (1975) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Multi-body expansion Need to find a representation for these functions Inversion of potentials: Going from energies to potentials, Mobius transformation EL is found from ab-initio energy database CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Multi-body expansion: Simple examples E0 = V 0 E1(X1) = V (1)(X1) + V0 E2(X1,X2) = V (2)(X1,X2) + V (1)(X1) + V (1)(X2) + V0 Inversion of potentials Evaluate (ab-initio) energy of several two atom structures to arrive at a functional form of E2(X1,X2) V (2)(X1,X2) = E2(X1,X2) - (E1(X1) + E1(X2) – E0) 1 3 2 CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Multi-body expansion: link to other Hamiltonians • All potential approximations can be shown to be a special case of multi-body expansion – Embedded atom potentials CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Comparison with the Cluster Expansion Method • • • • Only chemical degrees of freedom Relaxed calculation required but only a few calculations required Periodic lattice only Results are obtained from superstructures of parent lattice Multi-body expansion • • • • • 1. 2. 3. All degrees of freedom included No relaxations needed Needs a database of calculations, regression schemes required Periodicity is not required (large cell, one k-point calculation) Can predict energies over several different lattices Sanchez and de Fontaine, 1981 Sanchez, et al, Generalized Cluster Description of Multicomponent Systems, Physica A 128 (1984) Connolly,Williams, Density-functional theory applied to phase transformations in transition-metal alloys Phys Rev B, 27 (1983 ) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Multi-body expansion = ∑ + ∑ +∑ +… Total energy represented as hierarchical sum of isolated clusters of atoms - No periodicity - Fully transferable - No relaxation necessary Two issues to be taken care of: 1) How to construct each of these multi body potentials? 2) When to stop the expansion? CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Constructing the multi-body potentials Approximate the n-body potential as a polynomial in the corresponding dimension Use the theory of interpolation to find these polynomials Compute energies of a finite number of n-atom isolated clusters using ab-initio methods and fit the polynomials to these energies Well established theory to find the ‘best approximating polynomial’: again two issues: which polynomial to choose and which points to sample at? Very simple for two-body interactions Enforcing symmetry and reducing the dimensions, this becomes a one dimensional function Just sample at roots of the chebyshev polynomial Have rigorous bounds on the quality of the interpolant generated Becomes more complicated for higher body potentials CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory High dimensional surfaces 4 As the number of atoms in the n-body potential increases, the dimensionality of the n-body potential increases. ‘Curse of dimensions’ comes into play very quickly 1 2 3 Have to approximate high dimensional surfaces accurately Cannot utilize a tensor product space! Come up with intelligent schemes to sample from the hypersurface Dimension points Multi body expansions not a new theory. One of the standing mathematical problems in representation potential energy surfaces- Roszak & Balasubramanian J. Math Chem (1994) Techniques devised for representing the PES: but specific to dimension and could not be generalized to higher body interaction 1 50 2 2500 4 6.25e6 8 3.9e13 16 1.52e27 Murrell & Varandas, Molecular Physics (1986), Salazar, Chem Phys Let (2002), Wu et.al PCCP (1999), Aquilanti et.al, PCCP (2000), Ischtwan & Collins, J. Chem Phys (1993), Schatz, Rev. Mod. Phy (1989), Becker & Karplus, J. Chem Phys (1997) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory SMOLYAK ALGORITHM LET OUR BASIC 1D INTERPOLATION SCHEME BE SUMMARIZED AS Ui( f ) a xi X i x i f ( xi ) IN MULTIPLE DIMENSIONS, THIS CAN BE WRITTEN AS (U i1 U id )( f ) xi1 X i1 xid X id (axi1 axid ) f (x i1 , , x id ) TO REDUCE THE NUMBER OF SUPPORT NODES WHILE MAINTAINING ACCURACY WITHIN A LOGARITHMIC FACTOR, WE USE SMOLYAK METHOD U 0 0, i U i U i 1 , i i1 Aq ,d ( f ) Aq 1,d ( f ) (i1 id id )( f ) i q IDEA IS TO CONSTRUCT AN EXPANDING SUBSPACE OF COLLOCATION POINTS THAT CAN REPRESENT PROGRESSIVELY HIGHER ORDER POLYNOMIALS IN MULTIPLE DIMENSIONS A FEW FAMOUS SPARSE QUADRATURE SCHEMES ARE AS FOLLOWS: CLENSHAW CURTIS SCHEME, MAXIMUM-NORM BASED SPARSE GRID AND CHEBYSHEV-GAUSS SCHEME CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory SMOLYAK ALGORITHM Extensively used in statistical mechanics Uni-variate interpolation Provides a way to construct interpolation functions based on minimal number of points Ui( f ) a xi X i x i f ( xi ) Multi-variate interpolation (U i1 Univariate interpolations to multivariate (U i1 U id )( f ) interpolations xi1 X i1 U id )( f ) xid X id (axi1 (axi1 axid ) f (x , , x id ) xi1 X i1 xid iX id 1 Smolyak interpolation U 0 0, i U i U i 1 , Accuracy the same as tensor product i i1 Aq ,d ( f ) Aq 1,d ( f ) (i1 id id )( f ) i q D = 10 Within logarithmic constant Increasing the order of interpolation increases the number of points sampled CORNELL U N I V E R S I T Y ORDER SC FE 3 1581 8000 4 8801 40000 5 41625 100000 Materials Process Design and Control Laboratory SMOLYAK ALGORITHM: REDUCTION IN POINTS For 2D interpolation using Chebyshev nodes Left: Full tensor product interpolation uses 256 points Right: Sparse grid collocation used 45 points to generate interpolant with comparable accuracy Results in multiple orders of magnitude reduction in the number of points to sample D = 10 For multi-atom systems, sample all combinations of atoms (eg. E(A-A-A), E(A-A-B), E(A-B-B),E(B-B-B) and construct interpolants. CORNELL U N I V E R S I T Y ORDER SC FE 3 1581 1000 4 8801 10000 5 41625 100000 Materials Process Design and Control Laboratory ADAPTIVE SPARSE GRID COLLOCATION The conventional sparse grid method treats every dimension equally. Functions may have widely varying characteristics in different directions (discontinuities, steep gradients) or the function may have some special structure (additive, nearly-additive, multiplicative). The basis proposition of the adaptive sparse grid collocation is to detect these structures/behaviors and treat different dimensions differently to accelerate convergence. Must use some heuristics to select the sampling points. Such heuristics have been developed by Gerstner and Griebel Have to come up with a way to make the Smolyak algorithm treat different dimensions differently. Generalized Sparse Grids: Convention sparse grids imposes a strict admissibility condition on the indices. By relaxing this to allow other indices, adaptivity can be enforced. Admissibility criterion for a set of indices S. where ej is the unit vector in the j-th direction 1. T. Gerstner, M. Griebel, Numerical integration using sparse grids, Numerical Algorithms, 18 (1998) 209–232. CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory MINIMAL CLUSTER REPRESENTATION Specification of clusters of various order by position variables Cluster size Cluster specifier Dimensionality 2 R12 1 3 R12, R23, R31 3 4 R12,R23,R34,R41,R42 ,R31 6 M R12,R23,R34,R41… 3M-6 4 4 5 a 1 2 3 b b 1 2 3 5 a CORNELL U N I V E R S I T Y Improving the computational efficiency by reducing the problem dimension • Convex hull technique to represent all atoms in the positive z-direction • Use independent coordinates to represent the cluster geometry Materials Process Design and Control Laboratory Constructing the multi-body potentials •Needs the least number of ab initio calculations toconstruct the potential, Energy •Provides capabilities to hierarchically improve the quality of interpolation using the previous interpolant, •Can be made to adaptively sample the different dimensions to further reduce the computational requirements Position atoms accuracy tensor sparse 3 10-6 66049 1537 4 10-5 1.9x1019 0.6x106 5 2x10-5 5.4x1033 20x106 CORNELL U N I V E R S I T Y •Completely independent of the number of dimensions of the problem. •Provides a way of constructing fully–transferable ab initio based potentials. Materials Process Design and Control Laboratory Abinitio computation of the energies • Executables – Cluster coordinates – Energy interpolation – Batch input for PWSCF – Read energies from PWSCF – Energy calculation CORNELL U N I V E R S I T Y • Plane-wave electronic density functional program ‘quantum espresso’ (http://www.pwscf.org) calculations are used to compute energies given the atomic coordinates and lattice parameters. •These calculations employ LDA and use ultra-soft pseudopotentials. • Single k-point calculations were used for isolated clusters, the cell size was selected so that the effect of periodic neighbors are negligible. •For multi-component systems, a constant energy cutoff equal to cutoff for the "hardest" atomic potential (e.g. B in B-Fe-Y-Zr) is used. MP smearing (ismear=1, sigma=0.2) is used for the metallic systems. Materials Process Design and Control Laboratory Selection of order of expansion = ∑ + ∑ +∑ +… Two issues to be taken care of: 1) How to construct each of these multi body potentials? 2) When to stop the expansion? Work of B.Paulus 1,2 show that the computed energy oscillates between even and odd number of expansion terms, asymptotically converging to the exact energy Stop the expansion when energy is accurate enough Energies (En) calculated from an n-body expansion correct energy 1. B. Paulus et. al, The convergence of the ab-initio many-body expansion for the cohesive energy of solid mercury Phys. Rev. B 70, 165106 (2004) 2. B. Paulus, The method of increments -- a wavefunction-based ab-initio correlation method for solids, Phys Rep 428 (2006) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Computation of MBE energy filters + + + .. Weighted MBE CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory Selection of order of expansion Weighted 4th order MBE Weighted 2nd order MBE True energies True energies Weighted 3rd order MBE Weighted MBE expansion coefficients are fitted using 12 atom cluster energies and the results are presented for a 16 atom cluster. CORNELL U N I V E R S I T Y True energies Materials Process Design and Control Laboratory Platinum clusters 16 atom FCC cluster Weighted MBE 4th order EM ( X 1 , X 2 ,.., X M ) 0.5884 E2 ( X 1 , X 2 ,.., X M ) 0.3014 E3 ( X 1 , X 2 ,.., X M ) 0.0353E4 ( X 1 , X 2 ,.., X M ). 4 Number of isolated cluster calculations 120 4 560 4 1820 Depth of interpolation Actual energy + + Lattice parameter Energy minima • Coefficients obtained using an 8 atom cluster energies at different lattice parameters CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory MULTIBODY EXPANSION ALGORITHM Interpolation algorithm Generate atom positions in interpolant space II. Computation step Given a phase structure Create ab-initio energy database Interpolation algorithm Select the max. number of terms in expansion I. Database step Build database of interpolants Compute E from Interpolation function Transform to interpolant space Multibody energy summation Decompose to two-atom, three atom etc. positions CORNELL U N I V E R S I T Y Energy of phase structure Materials Process Design and Control Laboratory LINKING THE MULTIBODY EXPANSION TO OTHER SOFTWARE Multi Body Expansio (MBE) The multibody expansion software written in C++ Two parts: potential generation & energy computation Energy computation part is the Hamiltonian Molecular dynamics- LAMMPS Large-scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) is a classical molecular dynamics (MD) code developed by S. Plimpton et. al (Sandia national lab) http://lammps.sandia.gov/pictures.html#twin Directly linked energy computation part in LAMMPS with MBE Useful for molecular dynamics and energy minimization Monte Carlo- MCCCS Towhee Monte Carlo for Complex Chemical Systems (MCCCS) developed by M. G. Martin, J. I. Siepmann et. al. Available at http://towhee.sourceforge.net/ Fortran based code. Linked Towhee and MBE using a library Performs a variety of calculations in all ensembels CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory APPLICATION TO SURFACE PHENOMENA: Ex 1 Predict the most stable adsorption site of hydrogen on metallic surfaces Test for FCC Platinum Depending on the surface there are multiple adsorption sites Many investigations performed using EAM and other semi-emperical models These predict the binding sites fairly accurately Try to predict favorable binding sites and energies using MBE FCC (100) CORNELL U N I V E R S I T Y FCC (110) FCC (111) Materials Process Design and Control Laboratory APPLICATION TO SURFACE PHENOMENA: Ex 1 Test for FCC(111) Generate a 5x5x5 atom cell of Pt Coordination number is 9 Position of hydrogen atom varied along the first primitive cell The potential energy surface is constructed Standing problems in surface chemistry Compare’s extremely well with the abinitio based results of Kallen et.al1 TOP FCC BRIDGE HCP -0.410 -0.455 -0.404 -0.420 1 G.Kallen,G.Wahnstrom, Quantum treatment of H on a Pt(111) surface, Phys Rev B, 65 (2001) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory APPLICATION TO SURFACE PHENOMENA: Ex 1 The atomic potential energy surface (APES) computed from ab-initio techniques First step towards efficient , quick computation of the PES Computational cost Minimum energy surface of H on Pt(111) Plot of minimum energy in z direction for the primitive cell Highly anharmonic potential energy surface FCC->HCP (55 meV), FCC->Top (160 meV) MBE: ~ 10 minutes DFT: ~ days FCC site Confined to fcc-hcp-fcc valleys From ref 1 1. G.Kallen,G.Wahnstrom, Quantum treatment of H on a Pt(111) surface, Phys Rev B, 65 (2001) 2. S.C.Badescu et al, Energetics and Vibrational states for Hydrogen on Pt(111), PRL 88 (2002) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory APPLICATION TO SURFACE PHENOMENA: Ex 1 Predict the most stable adsorption site of hydrogen on metallic surfaces Test for FCC Platinum= (1 0 0) surface Ab initio studies reveal Hallow > Bridge > Top sites Bridge -49.5522 Ry CORNELL U N I V E R S I T Y Hallow -49.81611 Ry Top -49.22788 Ry Materials Process Design and Control Laboratory SEARCHING FOR GROUND STATE CONFIGURATIONS Design of nanostructures: multiple applications like design of memories for data storage Adatoms on surfaces Need to establish which configuration of adatoms can stabilize the surface the most Previously done using abinitio calculations (problems of periodicity and long range interactions) Recently done using a modified cluster expansion method Apply multibody expansion to this problems Take FCC(111) surface. Stable configuration should be a Pt(111) (2x1) H adatom configuration 1. Drutz, Singer, Fahnle, PHYSICAL REVIEW B 67 (2003) 035418 2. Sluiter, Kawazoe, PHYSICAL REVIEW B 68 (2003) 085410 CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory SEARCHING FOR GROUND STATE CONFIGURATIONS Finding the stable structure: 1. Consider a super cell (3x3x3 cell) 2. Place n number of hydrogen atoms on the surface 3. Apply periodic boundary conditions 4. Displace hydrogen atoms to get minima Pt(111) (2x1) H adatom configuration Can link to various other software Minimization found using LAMMPS (Large-scale Atomic/Molecular Massively Parallel Simulator) The Multi body expansion converted into a library Library included into the makefile of LAMMPS Can directly run a variety of Molecular dynamics and minimization scenarios CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory TOPOLOGICAL OPTIMIZATION: A SIMPLE EXAMPLE The adsorption behavior of the atom not only depends on the chemistry of the local surface but also depends on the topology of the global surface The availability of an efficient, computationally tractable method of finding the interaction energy between a large set of atoms paves the way for topological design of surfaces Surface characterization: Roughness: Small scale perturbations to the surface Representing roughness: Roughness represented by two components: PDF of a point above a datum z and the correlation between two points (ACF) ACF depends on the processing methodology, ex shot peening, sand blasting and milling Determine suitable components Determine best surface PDF is usually assumed to be a Gaussian 1 Q. Wang, J. K. Johnson, Optimization of Carbon Nanotube Arrays for Hydrogen Adsorption, J Phys. CHem B CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory TOPOLOGICAL OPTIMIZATION: A SIMPLE EXAMPLE Can consider topological optimization in a functional framework Define a cost functional. Here taken to be the fraction of available sites occupied This cost functional is defined in terms of the topology. Simplest case of roughness is a sinusoidal wave Start from a random configuration, compute the cost functional and minimize the cost functional The cost functional here depends on the frequency of oscillations of the surface How to compute the fraction of available sites occupied? CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory TOPOLOGICAL OPTIMIZATION: A SIMPLE EXAMPLE Compute the fraction of available sites occupied using Monte Carlo methods The Multibody expansion (MBE) provides a accurate PES of the adsorbate Coupling this energy descriptor with a validated Monte Carlo simulator Monte Carlo for Complex Chemical Systems (MCCCS) Towhee is very suitable for such a task Coupled the library of MBE to towhee software Platinum (111) surface Surface dimensions= 0.28 μm x 5.6 nm Total number of Pt atoms in the simulation= 9600 Perform Grand canonical ensemble Monte Carlo to model adsorption Minima reached in 5 iterations Each Monte Carlo simulation=100000 steps Time taken for one iteration= 8 hours Temperature = 300K, Pressure = 10 bar CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory TOPOLOGICAL OPTIMIZATION: A SIMPLE EXAMPLE 4 Atom distribution profile 3.5 3 2.5 2 1.5 1 0.5 0 Convergence history of topological optimization 0 0.25 0.5 0.75 Normalized distance 1 PDF of the adsorbate distribution Atoms beneath the first layer: leads to embrittlement For this simple case the wavelength of the optimal surface is 0.71 μm CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory A LOOK AHEAD …. The next step is to utilize more realistic representations for the surface: -Spline representations and Bezier curves - Larger number of atoms - Must relax the surface atoms also, can analyze the effects of the adsorbate as well as embrittlement effects Analysis/Design of multiple component surfaces: - Platinum + Bismuth predicted to have good adsorption behavior Determine suitable components Determine best surface - Can optimize surface and chemistry to inhibit one type of material and enhance another (prevent CO poisoning ) CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory AND A LOOK BEHIND 1) Represented the energy of a set of atoms as a hierarchical sum of isolated clusters of atoms: The multi body expansion (MBE) 2) Provided a methodology to compute these high dimensional surfaces using sparse grid techniques: Smolyak theorem, adaptive sparse grid methods 3) Coupled the multibody potential framework to several publicly available molecular dynamics and Monte Carlo software 4) Computed the atomic potential energy surface of H adsorption on Pt to high accuracy 5) Applicability of the MBE to finding the ground state stable configurations 6) Laid the groundwork for functional topological optimization of surfaces towards enhancing adsorption with a simple example B. Ganapathysubramanian and N. Zabaras, "Topological characterization of adsorption phenomena using multi-body potential expansions", in preparation. V. Sundararaghavan and N. Zabaras, "Many-body expansions for computing stable structures of multi-atom systems", Physical Reviews B, submitted CORNELL U N I V E R S I T Y Materials Process Design and Control Laboratory