Author Archives: Naveen Agrawal

HCN to CNH isomerization

Author: Naveen Agrawal

Introduction:  Nudged Elastic Band calculation is one of the methods to search for a transition state which considers a chain of images attached through elastic bands between reactant and product.[1] This chain should ideally represent a set of states along the reaction path after the NEB( Nudged Elastic Band) calculation has converged to the minimum energy path. However,we have to guess these set of images to initialize the search. One way to generate these first set of images is a linear interpolation of atomic coordinates between the reactant and product. This scheme provides a faster way to create intermediate states when the reaction itself does not involve a complex rearrangement of atoms, I am presenting a case of isomerization where using linear interpolation creates chemically unreasonably intermediate states and initializing the NEB calculation with these images would not lead to desired reaction trajectory. Specifically for the isomerization of HCN, the linear interpolation leads to unreasonable geometric configurations as described in figure 2 which would be unreasonably difficult or rather impossible to optimize to desired reaction path as described in figure 4.

 Methods: DFT (Density Functional Theory) calculations were performed within the Vienna Ab initio simulation package (VASP, version 5.4.4), using the periodic supercell approach [2]. The projector augmented wave (PAW) method was used for electron-ion interactions [3,4]. The Perdew-Burke-Ernzerhof functional described the electron-electron exchange and correlation energies [5]. A plane wave basis set is used with an energy cutoff of 450 eV. For geometry optimization, the convergence criteria of the forces acting on atoms were 0.02 eV Å-1, while the energy threshold-defining self-consistency of the electron density was 10-5 eV. A cubic unit cell with 20A dimension was used for periodic supercell calculations. Considering that it is a unimolecular rearrangement and the electronic states should be discreet and localized therefore only gamma point was included for the Brillouin zone sampling. Zero Point Vibrational energy (ZPVE) corrections were included in the electronic energy calculated through VASP. Transition state search was performed using the Climbing Image Nudge Elastic band Method (CI-NEB). CI-NEB is an improved version of NEB method which improves the definition of reaction coordinate near the highest energy image.[6]

Geometry optimization of Isomers and linear interpolation: HCN and CNH molecules are optimized using periodic supercell approach as mentioned in the previous section. The optimized structures are linear and the distinction between the two structures is the position of hydrogen atom as attached to Carbon or Nitrogen. The optimized structures as obtained are shown in Figure 1.

Figure 1: Reactant and Product state for HCN to CNH isomerization

A linear interpolation of geometric co-ordinates between the reactant and product state resulted in a series of chemically unreasonable intermediate states as shown in Figure 2. Linear interpolation prompts the transfer of hydrogen  from carbon to the nitrogen through the carbon-nitrogen bond itself which would basically optimize to dissociated atoms due to huge repulsion during the NEB calculation.

Figure 2: Reaction trajectory through linear interpolation from 00 to 05

Transition state search and confirmation: A more reasonable set of intermediate images are considered as presented in Figure 3 as an initial guess to the Climbing Image Nudge Elastic Band (CI-NEB) method. These images were created based on nature of the geometry for the transition state presented in the literature [7] allowing the reasonable transition of hydrogen through C-H bending.

Figure 3: Initial guess of intermediate images for NEB calculation

This sequence of intermediate states resulted into a converged NEB calculation with image 03 as the transition state as indicated by the change of sign of tangent forces along the reaction path and the converged normal forces perpendicular to the path for all the images to be less than 0.02 eV/A as can be seen in the attached snapshot of summary of VASP output attached in table below. Ideally the tangent forces should change sign across the transition state only once, however in the calculation summary the initial images show very small positive sign. This is possibly due to the close proximity of the image 01 and 02 to the reactant state, and the gradient of energy at such small energy differences (0.02 eV) is probably not accurate. However, it is still highly likely that indeed 03 is the transition state as it is the maxima along a reasonable reaction trajectory as indicated by figure 4 and table below. The relative energy of the highest energy image  is 1.99 eV (activation barrier) without ZPVE corrections. I also increased the number of intermediate images to seven while gradually bending the C-H bond to allow better sampling of reaction path and it was possible to see a slightly better transition to the steep barrier of 1.99 eV as can be seen in the attached animation in appendix.

Image numberEnergy relative to initial stateTangent forcesNormal forces
00
10.0228620.00040.008946
20.0225670.005150.007189
31.992472-0.0060650.011774
40.7722631.0239810.019798
50.616651

 

Figure 4: Converged NEB reaction trajectory

A transition state should only have one imaginary normal mode of vibration and should lead to the product or reactant state considered.To confirm that 03 indeed is a valid transition state, a vibrational analysis was performed using IBRION =5 in VASP (finite difference energy derivative with harmonic approximation).Vibrational analysis revealed a high frequency unstable normal mode (negative curvature of energy surface) for the 03 image as shown in the first animation below. There were two additional low frequency mode obtained from vibrational analysis which represented rotation of C-N bond and are very weakly associated with reaction coordinate of H shuttle through rotation of C-H bonds.  Therefore, the activation barrier and the transition state geometry are precise within the convergence bounds of calculation mentioned.

Conclusion: HCN to CNH isomerization involves shuttling of hydrogen atom through bending of C-H bond in the transition state as can be seen in Figure 5 and Figure 4. Linear interpolation between the reactant and the product state provides very unreasonable guess for the reaction trajectory because of the linear geometry of both product and reactant state. Allowing the C-H bending in initial guess images provides a well converged NEB calculation with 1.84 eV of kinetic barrier with ZPVE corrections included. Previous theoretical studies are in agreement with the barrier calculated. [8]The presence of first order saddle point was confirmed with vibrational analysis. 

References:

1. Henkelman, G., Jóhannesson, G., & Jónsson, H. (2002) (pp. 269-302). Springer, Dordrecht.

2. Kresse, G. and Furthmüller, J.Physical review B54(16), p.11169.

3. P. E. Blochl, Phys Rev B, 1994, 50, 17953-1797

4. G. Kresse.et.al Phys Rev B, 1999, 59, 1758-1775

5. J. P. Perdew. et.al , Phys Rev B, 1992, 46, 6671-6687.

6. Henkelman.et.al  , The Journal of chemical physics, 2000,113(22), 9901-9904.

7. Koch.et.al., The Journal of Physical Chemistry C, 2007,111(41), 15026-15033.

8.Gong.et.al, The Journal of chemical physics,2005, 122(14), 144311.

Appendix:

Perferred binding site for atomic oxygen on Pt (111)

Author: Naveen Agrawal

Introduction: Platinum is a model metal system for catalytic reactions. This computational study focuses to determine the preferred binding site for the atomic oxygen. In addition, the effect of increasing the coverage of oxygen on the adsorption energy of oxygen has been evaluated. Platinum has been also studied as a model electrocatalyst for a variety of electrochemical reactions, especially the Pt (111) facet which is the most stable facet thermodynamically. However, experiments and computational studies indicate the presence of atomic oxygen on the metal surface at oxidation potentials as low as 0.85 V (RHE) [1]. This study will show the trend in the adsorption affinity of atomic oxygen as we increase the coverage which could be considered as a consequence of higher oxidation potentials.

Methods: DFT (Density Functional Theory) calculations were performed within the Vienna Ab initio simulation package (VASP, version 5.4.4), using the periodic supercell approach [2].  The projector augmented wave (PAW) method was used for electron-ion interactions [3,4]. The Perdew-Burke-Ernzerhof functional described the electron-electron exchange and correlation energies [5]. A plane wave basis set is used with an energy cutoff of 450 eV. For geometry optimization, the convergence criteria of the forces acting on atoms were 0.05 eV Å-1, while the energy threshold-defining self-consistency of the electron density was 10-5 eV. 5-layer asymmetric atomic slab (with bottom 3-layer fixed to represent the bulk with lattice constant 3.97 Angstrom optimized with 11X11X11 k-points grid and 450 eV cutoff) with 3X3 unit cell and 15 Angstrom vacuum were considered to observe the effect of different coverage of atomic oxygen on the top layer. Correction to total electronic energy in VASP environment was performed using the method of Bengtsson to remove spurious periodic slab dipole interactions [6]. Previous studies have suggested that even 4 -layer atomic slab with bottom 2-layer fixed are enough to represent the bulk and surface features of the Pt (111) system [7].

The binding energy of atomic oxygen on Pt(111) was calculated w.r.t. a single oxygen molecule in vacuum as illustrated by equation 1.

\begin{equation}E_{binding}=\frac{E_{surf+O}-\frac{n*E_{O_2(g))}}{2}-E_{surf}}{n}\end{equation}

Where \(E_{binding}\) is the binding energy per atom of oxygen on Pt(111), \(E_{surf+O}\) is energy for the adsorbate surface system, \(E_{O_2(g)}\) is the energy of oxygen molecule in vacuum evaluated in a 20 A cubic unit cell in VASP with unrestricted spin, \(E_{surf}\) is the energy of Pt(111) slab as described above, and \({n}\) is number of atomic oxygen adsorbed onto Pt(111) slab.

Convergence tests: Convergence tests were performed to evaluate the optimal k-points grid, energy cutoff, and the vacuum thickness for the periodic supercell. The energy tolerance considered for these convergence tests was 0.015 eV.  As the studied unit-cell was a 3X3 and 5-layer atomic slab, the k-points grids considered were (MxMx1), Where M ranges from 3 to 7. We need only unit discretization of the grid in Z direction because the Z-direction supercell dimension is almost 3 times higher than X and Y. The relative energy is plotted in Figure 1(a) across the irreducible k-points obtained through different grids. Similar convergence tests were performed to obtain the optimal energy cutoff and the vacuum thickness for the asymmetric 5-layer atomic slab as shown in the figure below. All the convergence tests were performed with a single oxygen present on the FCC site of Pt (111) 3X3 unit cell as shown in Figure 1(d). Based on the convergence tests k-points grid of 6X6X1 with 54 irreducible k-points, 450 eV energy cutoff, and 15 A of vacuum slab were determined to be suitable within the tolerance specified.

Figure 1: Convergence tests for k-points, energy cutoff, and Vacuum layer thickness for a 5-layer 3X3 unit cell of Pt (111)

Preferred binding site of atomic oxygen and coverage effect: Binding energy relative to atop  (lowest binding energy site) is plotted against different binding configurations possible on Pt(111) for atomic oxygen in Figure 3. Based on the relative binding energy obtained, FCC site which is a three-fold hollow site on Pt (111) as shown in Figure 3 has been determined to be the optimal binding site for the atomic oxygen.

Figure 2: Different possible binding configurations for atomic oxygen on Pt(111)

Figure 3: Preferred binding for atomic oxygen

 

Figure 5 shows that the increased coverage on the fcc site reduces the per-atom binding energy of oxygen on Pt(111). This effect can be visualized from Figure 4, where increasing coverages can lead to metal atom sharing between the adsorbed oxygen atoms possibly reducing the binding per oxygen. In addition, the co-adsorbates of similar nature also have unfavorable dipole-dipole repulsion even without metal atom sharing which causes a reduction in overall binding at higher coverages.

Figure 4: 1/9 to 1/3 ML of oxygen on FCC sites for Pt(111)

 

Figure 5: Effect of coverage of atomic oxygen on binding energy

Conclusion: DFT calculations performed in this study agree with the earlier studies that preferred binding site of oxygen on Pt (111) is fcc (3-fold hollow site). The binding energy difference between hcp and fcc sites matches with earlier theoretical studies and experimental observations [8]. Adsorption energy of atomic oxygen decreases as we increase the coverage on the fcc sites which is consistent with experiments [9].

References:

  1. Kondo.et.al. The Journal of Physical Chemistry C29, 16118-16131.
  2. Kresse, G. and Furthmüller, J.Physical review B54(16), p.11169.
  3. P. E. Blochl, Phys Rev B, 1994, 50, 17953-17979
  4. G. Kresse.et.al Phys Rev B, 1999, 59, 1758-1775
  5. J. P. Perdew. et.al , Phys Rev B, 1992, 46, 6671-6687.
  6. Bengtsson, Physical Review B59(19), 12301.
  7. Kokalj.et.al. Journal of Physics: Condensed Matter39, 7463.
  8. Yeo, Y. Y., L. Vattuone.et.al, The Journal of chemical physics 106, no. 1 (1997): 392-401.
  9. Gu, Z.et.al, The Journal of Physical Chemistry C111(27), 9877-9883.

Lattice parameter for Hf and comparison to experimental measurements

Author: Naveen Agrawal

Introduction: Hafnium is a chemical element with atomic number 72 and electronic configuration [Xe] 4f145d26s2. Hafnium has attracted great technological interest in nuclear science because of its exceptional corrosion resistance and high thermal neutron capture cross-section. Hafnium has been found to form an HCP like crystal as shown in the figure below structure with c/a ratio of 1.58 experimentally[1]. This study involves the Density Functional Theory calculations to predict the optimum lattice parameter for Hf using the Density Functional Theory with plane-waves basis set program ‘CASTEP’ [4] available in the Materials Studio.

 

Hf crystal structure and the lattice parameters

Methods: CASTEP is used to perform plane-wave based electronic structure calculations which in general requires the selection of certain input parameters such as KPOINTS and ENCUT to optimize the computational effort and the accuracy of electronic structure results. In addition, we choose GGA (Generalized Gradient Approximation) based PBE ( Perdew Burke Ernzerhof) exchange-correlation functional  [2,3] and ultra-soft pseudo-potential with core radius 2.096 Bohrs (1.109 Angstrom) generated  with panel of 26 valence electrons (4f14 5s2 5p6 5d2 6s2) [5]. We first optimize the KPOINTS for an energy cutoff of 500 eV and later do the convergence for energy cut-off (ENCUT) which determines the maximum energy plane-wave included in the solution. These convergence tests and further calculations were performed with the SCF (Self-consistent-field) cycle convergence criteria of 2E-06 eV per atom, using finer criteria improved the accuracy of total energy insignificantly w.r.t to the desired accuracy for the study (1 meV). Optimized ENCUT and KPOINTS are used consistently for determining the optimum lattice parameters through variation in lattice constant ‘a’ for a chosen c/a ratio. Later, we compare the respective minimums obtained at different c/a ratios to determine the most energetically favorable lattice parameters.

 

KPOINTS convergence: Hafnium crystal structure (HCP) and lattice parameters (c/a = 1.5811  and a =3.194 A) imported from Materials Studio library were used for initial convergence testing of KPOINTS and ENCUT. Selection of KPOINTS grid was made in such a way to keep the density of KPOINTS in reciprocal space is uniform. Specifically, for Hafnium the number of KPOINTS were kept in the inverse ratio of unit cell vector length or reciprocal cell vector lengths. For the considered lattice, the ratio of KPOINTS to the nearest integers in X and Y directions to Z directions were kept equal to 1.58 approximately. Based on the above criteria, following grids as shown in the table below were considered for testing.

KPOINT gridIrreducible KpointsTotal energy per atom in eVRelative energy to the energy for highest KPOINTS in meV
5x5x310-7866.6951.504
7x7x416-7866.6942.952
9x9x536-7866.700-2.623
11x11x764-7866.6961.004
14x14x9120-7866.6970.000

KPOINTS convergence for the considered Hf lattice w.r.t relative total energy determined w.r.t to energy at highest number of KPOINTS (a =3.194 Angstrom and c/a = 1.581) with ENCUT = 500 eV

The figure above shows the convergence of total energy per atom in eV relative to same determined at the highest number of KPOINTS against irreducible KPOINTS available for the considered grid of KPOINTS. With the tolerance of 1meV, KPOINTS grid of 11X11X7 seems to be a reasonable choice.

ENCUT convergence: ENCUT( Energy cutoff) is the kinetic energy of the highest kinetic energy plane wave that needs to be considered to obtain a converged solution such that any higher energy cutoff would not lead to an energy difference for the tolerance considered (1meV).

ENCUT convergence for the considered Hf lattice (a= 3.194 Angstrom and c/a = 1.581) and optimized KPOINTS (11x11x7)

The figure above shows the variation in the total energy per atom in eV for the different values of ENCUT considered. Based on the desired tolerance, an ENCUT of 500 eV was determined to be optimum.

Optimization of lattice parameters: As Hafnium is an HCP metal, it needs two lattice constants ‘c’ and ‘a’ to specify the crystal structure. I chose to systematically vary both for a chosen c/a to find the minimum energy configuration for the considered c/a. I also compared the minimums obtained for different c/a ratios to find the most favorable set of lattice parameters.

 

Lattice parameter optimization based on cohesive energy per atom with KPOINTS grid (11x11x7) and ENCUT = 500 eV

The figure above shows the variation in cohesive energy per atom relative to the global minimum obtained in eV for different sets of lattice parameters. The systematic variation in lattice constant ‘a’ for a chosen c/a led to different minimum energy configurations indicated by solid black curve. Based on the cohesive energies per atom minimums obtained at different c/ratios, c/a =1.583 with a =3.2011 Angstrom were determined to be optimum lattice parameters.

c/aa in AngstromCohesive energy per atom in eVRelative cohesive energy per atom in meV (millielectronvolt)
1.5903.1968-8.845900.0749
1.5853.1997-8.845980.0019
1.5833.2011-8.845980.0000
1.5813.2024-8.845960.0124
1.5783.2045-8.845910.0641
1.5753.2065-8.845830.1453
1.5703.2097-8.845620.3547

Conclusion: Optimum Lattice parameters obtained through the DFT calculations (a = 3.2011 Angstrom  and c/a = 1.583) using the CASTEP code with the tolerance of  1meV in the total energy came in close agreement with the experiments ( c/a =1.58, a =3.1964 Angstrom)[1].  However, it should be noted the energetic difference between several minimums obtained are of a lesser order than the tolerance specified, therefore, stricter convergence criteria should be useful to resolve the energy scale to such order confidently.

References:

[1] http://periodictable.com/Elements/072/data.html

[2] J.P. Perdew, K. Burke, M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett., 77 (1996) 3865-3868.

[3] J.P. Perdew, J.A. Chevary, S.H. Vosko, K.A. Jackson, M.R. Pederson, D.J. Singh, C. Fiolhais, Atoms, Molecules, Solids, And Surfaces – Applications of theTHE Generalized Gradient Approximation for Exchange and Correlation, Phys. Rev. B, 46 (1992) 6671-6687.

[4] J. Clark Stewart, D. Segall Matthew, J. Pickard Chris, J. Hasnip Phil, I.J. Probert Matt, K. Refson, C. Payne Mike, First principles methods using CASTEP,  Zeitschrift für Kristallographie – Crystalline Materials, 2005, pp. 567.

[5] http://www.physics.rutgers.edu/~dhv/uspp/