is a quantum chemistry software developed by scientists to be used by scientists. It is not primarily a commercial product and it is not sold in order to produce a fortune for its owner (the Lund University). The authors have tried in *MOLCAS* to assemble their collected experience and knowledge in computational quantum chemistry. *MOLCAS* is a research product and it is used as a platform by the Lund quantum chemistry group in their work to develop new and improved computational tools in quantum chemistry. Most of the codes in the software have newly developed features and the user should not be surprised if a bug is found now and then.

The basic philosophy behind *MOLCAS* is to develop methods that will allow an accurate *ab initio* treatment of very general electronic structure problems for molecular systems in both ground and excited states. This is not an easy task. Our knowledge about how to obtain accurate properties for single reference dominated ground states is today well developed and *MOLCAS* contains a number of codes that can perform such calculations (MP2, CC, CPF, DFT etc). All these methods treat the electron correlation starting from a single determinant (closed or open shell) reference state. Such codes are today standard in most quantum chemistry program systems.

However, the basic philosophy of *MOLCAS* is to be able to treat, at the same level of accuracy also, highly degenerate states, such as those occurring in excited states, at the transition state in some chemical reactions, in diradicaloid systems, heavy metal systems, etc. This is a more difficult problem since the single determinant approach will not work well in such cases. The key feature of *MOLCAS* is the multiconfigurational approach. *MOLCAS* contains codes for general and effective multiconfigurational SCF calculations at the Complete Active Space (CASSCF) level, but also employing more restricted MCSCF wave functions (RASSCF). It is also possible, at this level of theory, to optimize geometries for equilibrium and transition states using gradient techniques and to compute force fields and vibrational energies.

However, even if the RASSCF approach is known to give reasonable structures for degenerate systems -- both in ground and excited states -- it is not in general capable of recovering more than a small fraction of the correlation energy. It is therefore necessary to supplement the multiconfigurational SCF treatment with a calculation of the dynamic correlation effects. In the early versions of *MOLCAS*, this was achieved by means of the multireference (MR) CI method. This method has, however, severe limitations in the number of electrons that can be correlated and the size of the reference space. It is not a method that can be used to study excited states of anything but small molecules. But here it has the capacity to produce very accurate wave functions and potential surfaces. The MRCI code of *MOLCAS* is used by many groups for this purpose. Today it is also possible to run the COLUMBUS MRCI code together with *MOLCAS*.

In the years 1988-92, a new method was developed, which can be used to compute dynamic electron correlation effects for multiconfigurational wave functions. It is based on second order perturbation theory and has been given the acronym CASPT2. It was included into the second version of *MOLCAS-5*. From the beginning it was not clear whether the CASPT2 method would be accurate enough to be useful in practice. However, as it turned out it was surprisingly accurate in a number of different types of applications. The CASPT2 approach has become especially important in studies of excited states and spectroscopic properties of large molecules, where no other *ab initio* method has, so far, been applicable. The method is based on second order perturbation theory and has therefore limitations in accuracy, but the error limits have been investigated in a large number of applications. The errors in relative energies are in almost all cases small and the results can be used for conclusive predictions about molecular properties in ground and excited states. The major application areas for the CASPT2 method are potential energy surfaces for chemical reactions and photochemistry. The method is under constant development. Recently, a multistate version, which will allow the simultaneous study of several electronic states, including their interaction in second order. This code is especially useful in cases where two, or more energy surfaces are close in energy. We have for a number of year also tried to develop an analytical CASPT2 gradient code. For different reasons, this work is as yet unfinished. Instead we have in the present version (7) included a numerical procedure, which allows automatic geometry optimization at the CASPT2 level of theory. It is applicable to all states and systems for which the CASPT2 energy can be computed.

The program RASSI has the capacity to compute the interaction between several RASSCF wave functions based on different orbitals, which are in general non-orthonormal (nonorthogonal CI). RASSI is routinely used to compute transition dipole moments in spectroscopy, but can also be used, for example, to study electron transfer or other properties where it might be of value to use localized wave functions.

The size of the systems that can be treated with *MOLCAS* have been limited due to limitations in storing two-electron integrals for large basis set. This limit has now been moved substantially to larger systems by the introduction of a Cholesky decomposition of the two-electron integrals. This feature is introduced in *MOLCAS-7* and can be used for the SCF, CASSCF, CASPT2, RASSI and MP2 codes. It speeds up all calculations by orders of magnitude and extends the size of the basis sets that can be used. The accuracy can be controled by the threshold used in the decomposition. The same approach can be used to generate R/I auxiliary basis sets on the fly, which can then be used, for example to compute energy derivatives at the SCF, DFT; and RASSCF levels of theory.

It should finally be clearly stated that *MOLCAS* is not a black box tool. The user should be an educated quantum chemist, with some knowledge about the different quantum chemical models in use today, their application areas and their inherent accuracy. He should also have a critical mind and not take a printed output for granted without checking that the results are in agreement with his presumptions and consistent with the model he has employed. The skill to use *MOLCAS* effectively does not come immediately, but we have tried to help the user by providing together with this manual a book of examples, which explains how some different key projects were solved using *MOLCAS*. We are sure that the users will find them helpful in their own attempts to master the software and use it in the chemical applications. The MOLCAS group arranges regular MOLCAS workshops, which teaches how to use the software.