Anne M. Chaka, Carlos A. Gonzalez, Russell D. Johnson III
Computational Chemistry Group, National Institute of Standards and Technology, 100 Bureau Drive Mail Stop 8380,Gaithersburg, MD 20899-8380
 Hybrid functionals, i.e. those which include some fraction of the exact exchange, have become extremely popular.  There are many demonstrated cases in which the accuracy is a significant improvement over Hartree-Fock and often over MP2, at considerably lower cost. There are also many examples, however, in which hybrid functionals give poor or inconsistent results for seemingly similar systems. NIST has developed a Computational Chemistry Sick List to keep track of systems for which Density Functional Theory or wavefunction methods have found to be unreliable. 
The reason for the variability of the hybrid functionals is not well understood. These functionals have, in general, been parameterized using a specific set of chemical structures and property data, and have not necessarily been validated for transferability to other systems and other properties.  The question remains: Can functionals based on some fraction of the exact exchange be developed for general chemical applications, or are there physical limits to what can be achieved with parameterization? We will present a systematic investigation of hybrid functionals and exchange parameterization with respect to two classes of chemical problems in which hybrid functional have exhibited considerable variability, i.e. reaction barrier heights and singlet-triplet gaps.