The Newton-Raphson method works well if the initial geometry is close to a stationary point, but may not work well if it is not.

 

Also, technically the method may be difficult to use for some types of problem.

 

Technical difficulties arise because of the need to calculate the Hessian, which is a potentially very large matrix (no. of variables squared)

 

·   calculation of the second derivatives can be demanding (requiring a lot of computer time)

·   storage can be difficult (a lot of computer memory required)

·   Hessian must be inverted (diagonalized), problem for more than 1000 variables (1000 ´ 1000 matrix), i.e. approx. 300 atoms


Can avoid calculating the Hessian matrix at every step by

·   using a guessed matrix (e.g. the identity matrix)

·   by using a Hessian update scheme (constructs an approximate Hessian based on e.g. the gradient and coordinates at current and new points) - various update schemes have been developed

 

Calculating the Hessian at a lower level of theory can also reduce the demands of the calculation:

 

For example, can calculate a more approximate Hessian by ‘molecular mechanics’, which could be used in a geometry optimization with an ‘ab initio’ electronic structure method.  (Calculation of the Hessian matrix for an ab initio method could be very time-consuming).

 

Next: how to specify the geometry in a calculation