Some years ago Kleinman and Bylander [Phys. Rev. Lett. 48, 1425] proposed a fully nonlocal form of norm-conserving pseudopotentials. Its application reducesif compared to other ab initio pseudopotentialsthe computational effort to calculate potential matrix elements. However, if the procedure is not applied cautiously, it can destroy important chemical properties of the atoms. In this paper we identify the origin of this problem, and we give a theorem which tells if a "ghost" state occurs below the zero-node state of the atom. We also show how the difficulties can be avoided, i.e. , how reliable, fully nonlocal, norm-conserving pseudopotentials can be obtained. Some years ago Kleinman and Bylander' (KB) proposed a fully nonlocal form of norm-conserving pseudopotentials. Its application reducesif compared to other ab initio pseudopotentialsthe computational efFort to calculate potential matrix elements. Although the suggestion is intriguing, it has not been applied widely. The reason is that the calculated chemical binding of molecules and solids [e.g., GaAs (Ref. 2)] is sometimes described incorrectly. These problems arise, although for atomic calculations the wave functions and the logarithmic derivatives Di(E) and dDt(E)ldE at the reference energies E& and for r ) r, equal those of the all-electron calculations. Here r, defines the range of "pseudoization" of the ionic potential, and it roughly equals the range of the core electrons.In this paper we analyze the properties of the KB Hamiltonian and explain why it may cause unphysical results. In short, the problem is due to the fact that the KB Hamiltonian does not obey the Wronskian theorem, which implies that atomic eigenfunctions are energetically ordered such that (for a given quantum number l) the energies increase with the number of nodes. As this theorem is not valid for the KB Hamiltonian, it can have eigenstates with nodes even below the zero-node state. Or, the zero-node states may be followed directly by an n 2 node state. Both possibilities will usually prevent an application of these potentials for a reliable description of chemical binding. Below we show how this problem can be avoided, so that no diSculties arise in actual calculations.We take selenium as an example. In Fig. 1 the highest occupied eigenstates (4s and 4p) of the all-electron calculation, the corresponding states of the norm-conserving pseudopotential of Bachelet, Hamann, and Schluter (BHS), and the corresponding results of the BHS-based KB potential are displayed. The two pseudopotential calculations give exactly identical wave functions, which for r & r, also equal those of the all-electron calculation. Figure 2 shows the logarithmic derivatives of the s, p, and d states of the three calculations. In the d results of the all-electron calculation we also see the Se 3d level at -2.01 hartrees, which in the pseudopotential calculations is treated as a core state. We see that at the refer-1.0 0.5 -0.5-0.5 0~e 1~~L -0.5-/ I 0.5 I I I I I I I 7 -0.5--1.0 0 r (a.u. ) FIG. 1. The 4s (sol...