Ionization potential of aluminum clusters J. Akola, H. Ha ¨ kkinen,* and M. Manninen Department of Physics, University of Jyva ¨skyla ¨, P.O. Box 35, FIN-40351 Jyva ¨skyla ¨, Finland Received 2 February 1998 Structure, electronic structure, and ionization potential of aluminum clusters of 2–23 atoms are studied with a total energy method based on the density-functional theory. The calculated adiabatic ionization potentials agree remarkably well with the data from threshold photoionization measurements. The analysis of results gives insight into hybridization effects in the smallest clusters as well as reveals certain clusters that exhibit a clear jellium-type shell structure. An explanation of the experimental results in the size region of 12–23 atoms is given in terms of coexisting, competing icosahedral, decahedral, and fcc-based clusters. S0163-18299800228-8 Measurements of the ionization potential IPof small metal clusters, i.e., the energy needed for removal of an elec- tron from the cluster, yield valuable information on the elec- tronic structure. 1 In the crudest level, the observed trends can often be understood by considering a simple model of the cluster as a conducting sphere, in which there are two con- tributions to the ionization potential: the binding energy of electron to the sphere analogous to the work function W of a metal surfaceand the electrostatic contribution due to the charging energy of the cluster ion. In fact, the size-evolution apart from the well-known shell effects 1,2 of measured ion- ization potentials of simple metal clusters seem to follow nicely an average trend given by V i ( R ) =W+e 2 / R , where R is the cluster radius and =0.5 comes from the classical charging energy of a sphere of radius R . Similarly, the elec- tron affinity, the energy gained by attaching an electron to the cluster, is seen to follow a trend A e ( R ) =W-e 2 / R , with =in the classical consideration. The model fulfills the obvious limit V i =A e W as R . With quantum cor- rections to the parameters and the measured difference V i -A e is reproduced reasonably well. 3 Small aluminum clusters seem however to behave in a way that is not consistent with the above model of a metallic sphere. 4,5 There is an initial rise of IP up to N =4, i.e., N e =12, N e being the number of valence electrons. This is not explainable by the jellium model. Furthermore, strong devia- tions from the e 2 / R behavior are seen up to N 20, and even beyond. The probable explanation for the behavior of IP for small N is an incomplete s - p hybridization, whence for larger clusters there certainly are strong shell effects aris- ing from electronic or atomic structure. 6 In this context it is interesting to note that the mass spectra in the relatively small size range a few hundred atomscan be explained by octahedral growth pattern, indicating that already in this size range the cluster prefers the bulk fcc symmetry. 7,8 In this work we have studied systematically the ionization potential of small Al clusters in the size range of 2–23 atoms by an ab initio total energy method. 9 By analyzing the degree of s - p hybridization for the smallest clusters and the com- patibility of the jellium-type shell structure for the larger ones this work is complementary to the previous semiempir- ical or ab initio calculations. 10–13 Particularly regarding the ionization potential this calculation is the most systematic one reported to date, to our knowledge. As a by-product the calculations also yield information on the geometry of the ground states of neutral and charged clusters, and some of the isomers of the neutral ones. Their effect on the measured IP will be discussed. Our results suggest an aspect that the oscillation of the measured threshold ionization in the size range of 12–23 atoms 4 is due to competition and coexistence of icosahedral, decahedral, and fcc-based structures. The calculations are done using the BO-LSD-MD Born- Oppenheimer local-spin-density molecular dynamics method devised by Barnett and Landman, fully documented in Ref. 9. In the BO-LSD-MD method one solves for the Kohn-Sham KSone-electron equations using a suitable parametrization for the local spin-density approximation to calculate the exchange-correlation partfor the valence elec- trons of the system corresponding to a given nuclear configu- ration of the classical ions. From the converged solution the Hellmann-Feynmann forces on ions can be calculated, which together with the classical Coulomb repulsion between the positive ion cores determine the total forces on ions, accord- ing to which one can perform structural optimizations or classical molecular dynamics for the ions. The current imple- mentation uses plane waves combined with fast Fourier transform techniques as the basis for the one-electron wave functions and norm-conserving, nonlocal, separable 14 pseudopotentials by Troullier and Martins 15,16 to describe the valence-electron–ion interaction, and the LSD parametriza- tion by Vosko, Wilk, and Nusair. 17 Here we wish to stress that the method does not apply any of the standard supercell techniques in calculating the total energy of a finite system. This is an important aspect pertinent to this study: the ion- ization potential is a straightforward difference in the total energies of the neutral and charged cluster, without or with the relaxation of the charged cluster to evaluate the vertical or adiabatic IP, respectively. We give the ground-state structures and ionization poten- tials for Al–Al 7 in Table I. Table II shows IP’s for Al 12 – Al 23 . Both tables show the experimental data by Schriver et al., 4 and our adiabatic values are compared to the experimental threshold ionizationdata as well as to some earlier calculations in Fig. 1. For each cluster both neutral and ionwe found the ground state to have the minimum total spin i.e., S =0 and S =1/2 for a cluster with even and PHYSICAL REVIEW B 15 AUGUST 1998-I VOLUME 58, NUMBER 7 PRB 58 0163-1829/98/587/36014/$15.00 3601 © 1998 The American Physical Society