First-Principles Theory of Hydrogen Diffusion in Aluminum Hakan Gunaydin1, Sergey V. Barabash2, K. N. Houk1,3, and V. Ozoliņš*2,3 1 Department of Chemistry and Biochemistry, University of California, Los Angeles, California, 90095 2 Department of Materials Science and Engineering, University of California, Los Angeles, California, 90095 3 California NanoSystems Institute, University of California, Los Angeles, California 90095 E-mail: vidvuds@ucla.edu Quantum effects on H diffusion Tunneling corrections for the hydrogen diffusion in Al in the temperature range of our study (650 to 850 K) were computed using the Wigner[1] and Eckart[2] tunneling methods. We used the values of the activation energy and imaginary frequency at the transition state for a TdOh jump computed by Wolverton, Ozolins and Asta in Ref. [3]. The computed corrections to the classical tunneling rate are given in Fig. S1, which shows that the inclusion of tunneling corrections increases the rate of hydrogen diffusion in Al by about 1 to 3 percent at the temperature range of our simulations. This is consistent with the results of other first-principles studies, which typically find that quantum corrections become very small at temperatures above 400 K.[4-7] Figure S1. Tunneling contributions to the diffusion rate of H in Al computed with the Wigner and Eckart methods. S1 More elaborate quantum tunneling methods have been developed by Fermann and Auerbach [6], who include the effect of quantization of vibrational energy levels. Fermann and Auerbach also introduce the concept of crossover temperature used to separate the quantum from the classical regime. The crossover temperature refers to a temperature below which the tunneling corrections to the rate of H diffusion cannot be neglected in comparison with the classical hopping dynamics. These authors show that even in the case of covalently bound hydrogen (binding energy of ~1 eV and an imaginary frequency of ~1500i cm-1), the crossover temperature is only 368 K, above which tunneling corrections do not affect the rate of H migration significantly. In addition, Fermann and Auerbach show that above the crossover temperature their derived tunneling correction gives results that are nearly identical to the Wigner tunneling correction. Determination of Hydrogen Residence Times The diffusion coefficient of H in Al was calculated using the residence time method. Within this approach, one needs to determine the average time that H spends in the Oh and Td sites; these residence times were obtained directly from the computed ab initio molecular dynamics (AIMD) trajectories. At each time step, the hydrogen was assigned either to a Td or an Oh site based on distance criteria. When the hydrogen-Td site distance was smaller than the hydrogen-Oh site distance, hydrogen was considered to reside at the Td site, and when the hydrogen-Oh distance was smaller than the hydrogen-Td distance, hydrogen was considered to reside at the Oh site. The residence time histograms obtained at 800 K are given in Fig. S2. As seen from Fig. S2, hydrogen residence times in the Td site are much longer than the hydrogen residence times in the Oh site. The longer average residence time in the Td site corresponds to a higher binding energy of hydrogen in the Td site in comparison with the Oh site. S2 Figure S2. Oh (left panel) and Td (right panel) residence times obtained from AIMD simulations at T=800 K. Yellow lines represent the Poisson fits to the data. Uncorrelated H jumps at 800 K To ascertain the possibility that successive TdOh and OhTd jumps are uncorrelated, we calculated the probability for the hydrogen to return to the same site after two consecutive jumps at T=800 K. According to our AIMD simulations, these probabilities are 5/50 and 12/48 for jumps originating from the Td and Oh sites, respectively. The fact that these values are very close to the theoretical probabilities of 1/8 and 1/4 suggests that two consecutive jumps are uncorrelated and the diffusion of H in Al can be calculated assuming that the direction of each TdOh (OhTd) jump is independent of the preceding OhTd (TdOh) jump. Timeline for the diffusion of H in Al Figure S3 shows the residence times that were obtained for the diffusion of H in Al at T=800 K. The blue strides correspond to Td site occupations and yellow strides correspond to Oh site occupations. As seen from Fig. S3, residence times are uncorrelated. Figure S3. Timeline for the Td site and Oh site occupations for H atom at T=800 K. Yellow time strides correspond to Oh site occupations and blue time strides correspond to Td site occupations. The total length of this simulation is 20 ps. S3 Calculating Diffusivity from Residence Times To calculate the diffusion coefficient from the measured AIMD residence times, we use the fact that the Oh sites form a simple face-centered cubic lattice and that diffusion away from any given Oh site necessarily goes through one of the 12 neighboring Oh sites (see Fig. S4). Figure S4. A schematic diagram of Td Oh jumps. Oh site is shown green, and the Td site is red. The twelve nearest-neighbor Oh sites are on the edges of the cubic unit cell. Therefore, one can use the standard expression for diffusion on cubic lattices, D 2 6 t NN , where a 2 is the separation between nearest-neighbor sites and t NN is the average time for a hydrogen to jump from an Oh site to any one of the neighboring Oh sites via one or more intermediate residences on Td sites. The average Oh to Oh jump time t NN is calculated as follows. Let’s assume that the average residence times are t1 and t2 for Oh and Td sites, respectively, and that the probability to have a residence time, t, in the Oh site follows the Poisson distribution with an average t1: O (t) h 1 t t1 e . t1 (1.1) A hydrogen atom in an Oh site can jump to any one of the eight surrounding Td sites with a probability distribution given by Eq. (1.2) below. The probability to make a jump at time t from an Oh site to a chosen neighboring Td site is: pOh Td (t) O (t) h 8 , (1.2) where the factor 8 accounts for the number of nearest-neighbor (NN) Td sites. A hydrogen atom in a Td site has four jump possibilities to any one of the surrounding Oh sites. The corresponding S4 jump and residence probabilities for the Td site can be obtained from equations analogous to Eqns (1.1) and (1.2) above. In particular, the jump and residence probabilities for the Td site are given by pTd Oh (t) T (t ) d T (t) d 4 , 1 t t2 e , t2 (1.3) (1.4) where the factor 4 in Eq. (1.3) accounts for the fact that each Td site has 4 neighboring Oh sites. The H atom in an Oh site can return to the same Oh site after two jumps or diffuse to one of the neighboring Oh sites at a distance a 2 away from the original Oh site (where a is the lattice parameter). The probability that two jumps will take hydrogen back to the original Oh site is 8 (4 * 8) 1 4 . Hence, the probability distribution for the diffusion of hydrogen to a NN_Oh site via a two jump route (Oh Td NN_Oh) can be obtained from: (1.5) where the factor 24 accounts for 24 different routes of reaching a NN_Oh site via one of the NN Td sites. Eq. (1.5) gives the total probability for a hydrogen atom to move from an Oh site to a NN_Oh site in time interval t via one intermediate residence on a Td site. The total probability or reaching the NN_Oh site is a sum over all possible pathways of diffusing into a NN_Oh site. First, the hydrogen can jump to a NN_Oh site via the Oh Td NN_Oh route considered above. Second, the hydrogen can reach a NN_Oh site via a four-jump route, Oh Td Oh Td NN_Oh , i.e., it can jump into one of the Td sites, then back into the original Oh site, then to one of the neighboring Td sites, followed by a jump to a NN_Oh site. The probability distribution for this four-jump process is given by: t t t 0 1 2 p4 (t) 192 d1 d 2 d 3 pOh Td (1 )pTd Oh ( 2 1 )pOh Td ( 3 2 )pTd Oh (t 3 ) , (1.6) where is the time of the first OhTd jump, 2 is the time of the return TdOh jump, and 3 is the time of the second Oh Td jump. The factor 192 accounts for the fact that there are 8 ways to perform the first two jumps which bring hydrogen back to the same Oh site, and 24 ways to get out of it after two more [see Eq. (1.5)]. S5 The third contribution comes from a 6 jump route (Oh Td Oh Td Oh Td NN_Oh): t t t t t 0 1 2 3 4 p6 (t) 8 8 24 d 1 d 2 d 3 d 4 d 5 pOh Td ( 1 )pTd Oh ( 2 1 ) (1.7) pOh Td ( 3 2 )pTd Oh ( 4 3 )pOh Td ( 5 4 )pTd Oh (t 5 ) The total probability to jump out of the Oh site into a NN_Oh site in time t is the sum over all these processes: p NN (t ) p2 n (t ) . (1.8) n 1 Integrals like (1.6) and (1.7) were computed analytically by using Mathematica. The total jump probability satisfies p NN (t)dt 1 . (1.9) 0 The mean time required to jump to a NN_Oh site is given by t NN tpNN (t) dt 0 4 t1 t2 . 3 (1.10) Since the Oh sites form a simple FCC lattice, the diffusivity can be can be calculated from the conventional expression, D 2 6 t NN for random walk on cubic lattices, resulting in the following expression: (1.11) which was used to convert from the residence times to diffusion coefficients in Table I of the manuscript. Trapping of Hydrogen in Vacancies and Non-Vacancy Trap Sites The effect of H trapping in vacancies and non-vacancy trap sites can be quantified by using a grand-canonical formalism. We assume that the concentration and spatial distribution of vacancies remain constant (and uniformly disordered) due to the slow kinetics of vacancy diffusion under typical experimental conditions. These assumptions hold during typical lowtemperature high-fugacity charging with hydrogen (such methods are commonly used to increase H concentration before hydrogen desorption experiments measuring the hydrogen diffusivity) S6 and for hydrogen evolution experiments due to the much faster diffusion rates of hydrogen in comparison with vacancy diffusion. For instance, in Ref. [10] aluminum samples were charged over a time period of 48 hours at 90 oC. The rate of Al self-diffusion at 90 oC is approximately 10-19 m2/s, whereas the rate of diffusion of H in Al at 90 oC is 10-10 m2/s. This corresponds to a penetration depth/diffusion-mean-distance of ~200 nm for aluminum vacancies and ~1 cm for hydrogen atoms during the charging process. Similarly, hydrogen desorption typically occurs over short time spans (on the order of minutes) during which the vacancy diffusion, with a diffusion-mean-distance of a few tens or hundreds of Angstroms, can be safely neglected. Figure S5 shows the relative rates of H and vacancy diffusion in Al: the diffusion of H is many orders of magnitude faster than the diffusion of Al vacancies under both charging and desorption conditions. Therefore, under the conditions relevant to the discussion vacancies do not reach thermal equilibrium. Figure S5. The relative diffusivities of H and Al vacancies. The presence of trap sites such as vacancies, dislocations and grain boundaries at low H concentrations can significantly alter hydrogen diffusivity in metals. In the case of H in Al, the calculated hydrogen-vacancy binding energies are higher than the calculated hydrogen diffusion barriers, and we expect that vacancies and other trap sites will have a significant contribution to the effective diffusion coefficient. In order to describe the hydrogen distribution between the lattice and trap sites quantitatively, we adopt a grand-canonical formalism for a solid in equilibrium with a reservoir of hydrogen at a given chemical potential, . The total concentration of hydrogen in the solid, CH, can be S7 controlled by varying the hydrogen chemical potential (CH increases with increasing ). We assume that hydrogen in the trap sites is always in local equilibrium with hydrogen in the lattice, and therefore the whole system can be characterized by using the same value of chemical potential. The validity of this assumption was checked by estimating the hydrogen residence time in a vacancy using the calculated H-vacancy binding energies and diffusion barriers from Ref. [3], and was found to be on the order of microseconds, i.e. much faster than the time scale of diffusion experiments. Strict mathematical criteria for the validity of the assumption of local equilibrium have been derived by Oriani [8], and they appear to hold for most systems and experimental conditions of practical interest. We write the total hydrogen concentration as a sum of H content in the lattice (CL) and H content in the vacancies and non-vacancy trap sites (Ctrap) CH CL Ctrap . (2.1) Following Oriani’s treatment of hydrogen diffusion in a lattice with trap sites, we assume that only the lattice hydrogen contributes to the measured hydrogen flux (i.e. vacancy diffusion with trapped hydrogen is much slower and can be neglected on the time scale of typical experiments): J DL CL , (2.2) where DL is the lattice diffusion coefficient without traps. Most experiments do not measure DL directly since they have no means of measuring the gradient in the concentration of lattice hydrogen. Instead, they measure the relation between the gradient of the total hydrogen concentration [given by Eq. (2.1)] and the flux: J DC H , (2.3) where D is the apparent diffusivity. Due to the local equilibrium condition between the lattice hydrogen and hydrogen in the trap sites, some hydrogen is released from (absorbed by) the trap sites upon decrease (increase) in the lattice concentration CH due to compositional fluxes. This process changes the concentration of the lattice hydrogen and decreases the concentration gradient C L in Eq. (2.2), leading to a decrease in the apparent diffusivity. Using the assumption of local equilibrium, the gradient of the total hydrogen concentration can be related to the gradient of the lattice concentration by the following relation: S8 C trap C H C L 1 , C L (2.4) where the derivative in the parentheses describes the release and absorption of hydrogen in the trap sites in response to a change in the lattice concentration due to hydrogen fluxes. Introducing the chemical potential , using the mathematical identity Ctrap CL 1 C Ctrap , and L combining it with Eqs. (2.2) and (2.3), we obtain that the experimentally measured apparent diffusivity D is related to the true lattice diffusivity DL by the following relation: D DL 1 C Ctrap 1 L . (2.5) Since the hydrogen population is an increasing function of the chemical potential, the denominator in Eq. (2.5) is always larger than 1, and the apparent diffusivity is smaller than the lattice diffusivity. Only in the case when all traps are empty (very small binding energies) or completely saturated (high binding energies or very high lattice concentrations and correspondingly high ) does the observed diffusivity equal the lattice diffusivity. From Eq. (2.5) it is seen that D is an implicit function of the temperature and local hydrogen concentration in the lattice (which in turn is determined by the local value of the chemical potential). To evaluate this dependence, we turn to constructing a model for H trapping using the first-principles calculated hydrogen-vacancy binding energies from Refs. [3] and [9]. We consider a system with N lattice sites and a uniform vacancy concentration CV, which does not change appreciably during the charging and discharging processes due to the relatively slow rate of vacancy diffusion. We account for non-vacancy trap sites (dislocations, grain boundaries, etc.) by introducing an effective concentration CD of such sites. In what follows, we assume that the concentration of vacancies, trap sites and lattice hydrogen is sufficiently small (on the order of 10-4; similar to the values reported by Young and Scully [10]) to be treated in the ideal solution approximation where all lattice sites are independent and hydrogen atoms do not interact with each other. The total concentration of hydrogen is given by a sum of lattice hydrogen (occupying Td and Oh sites), H in vacancies ( C VH ) and H in other non-vacancy traps sites ( C HD ): S9 C H CTd COh CHV C HD . (2.6) The total free energy of the system is given by G kBT lnQ , (2.7) where the statistical partition function is Q Q Q Q QTd 2N N NCV Oh NCD V D . (2.8) Here, QTd , QOh , QV and QD are partition functions for the tetrahedral Td and octahedral Oh lattice sites, vacancies, and other trap sites, respectively. These can be expressed in terms of the site energies, temperature, and hydrogen chemical potential as follows: E QTd 1 STd e Td , (2.9) E QOh 1 SOh e Oh , 12 QV 1 SnH-Ve (2.10) EnH V n , (2.11) n1 QD 1 SD e ED . (2.12) The energies of hydrogen in the Td and Oh sites, ETd, and EOh, are measured relative to fcc Al, while the hydrogen-vacancy energy EnH-V is measured with respect to fcc Al with a vacancy (i.e., the vacancy formation energy is not included in EnH-V ). Sum over n in Eq. (2.11) accounts for the possibility that up to 12 hydrogen atoms can be housed in one vacancy. Non-vacancy trap sites are described by using an effective hydrogen absorption energy ED. In all cases, the reference state for hydrogen is the energy of an H2 molecule at T=0 K. The chemical potential therefore includes all effects due to pressure, temperature and charging conditions. Symmetry factors STd, SOh, SnH-V and SD account for the degeneracy of each type of site, e.g. S=1 for the Oh and Td sites, while S1H-V=8 since the hydrogen ion can be in any of the eight neighboring Td sites next to a vacancy. Fractional occupations of each type of site can be found by taking the derivatives of the free energies with respect to the chemical potential: STd e Td 1 GTd fTd 2N T QTd E 1 GV 1 fV NCV T QV S10 12 , nS n1 (2.13) H-V H-V En n e , (2.14) with similar expressions for the other hydrogen sites. Hydrogen concentrations are found by multiplying these occupations with the number of sites of each type: CTd 2 fTd , COh fOh , CHV CV fV , and CHD CD fD . These formulae are used to convert from the difficult-to-measure chemical potential in local equilibrium to the total hydrogen concentration. Finally, we evaluate the derivatives entering the denominator of the expression for the apparent diffusivity, Eq. (2.5). By taking the derivative of Eq. (2.13), we obtain CL 2 2 2 fTd fOh , T where fTd 2 fTd 2 fTd 2 (2.15) is the variance of the site occupations. Similarly, for the trap sites we obtain Ctrap 2 2 CV fV CD fD , T (2.16) and the expression for the apparent diffusivity becomes D DL . CV fV C D f D 2 2 1 (2.17) 2 fTd 2 fOh 2 We evaluated the derived expressions using the first-principles data taken from Ref. [9] and [3]. The value for hydrogen binding to non-vacancy trap sites is not known from calculations, and therefore we use an experimental estimate of 0.273 [10]. The symmetry factors for most configurations (except the most highly symmetric ones, such as corresponding to n=1, 2, 6 and 12 H atoms in a vacancy) are guessed with an upward bias. We checked that the results are not particularly sensitive to the precise values of these guesses, since the energies of these nonsymmetric configurations are far from the convex hull in the energy-versus-occupation plot (see Fig. 2 in Ref. [9]), and therefore occur with negligibly small probabilities. References [1] [2] [3] [4] E. Wigner, Z. Phys. Chem. B19, 203 (1932). C. Eckart, Phys. Rev. 35, 1303 (1930). C. Wolverton, V. Ozolins, and M. Asta, Phys. Rev. B 69, 144109 (2004). B. Bhatia, and D. S. Sholl, Phys. Rev. B 72, 224302 (2005). S11 [5] [6] [7] [8] [9] [10] P. G. Sundell, and G. Wahnstrom, Phys. Rev. B 70, 081403 (2004). J. T. Fermann, and S. Auerbach, J. Chem. Phys. 112, 6787 (2000). D. E. Jiang, and E. A. Carter, Phys. Rev. B 70, 064102 (2004). R. A. Oriani, Acta Metallurgica 18, 147 (1970). G. Lu, and E. Kaxiras, Phys. Rev. Lett. 94, 155501 (2005). G. A. Young, and J. R. Scully, Acta mater. 46, 6337 (1998). S12