Matrix calculus differentiation

Although the mathematical methods in this book include algebra, calculus, differential equations, matrix, statistics, and numerical analyses, students with background in algebra and calculus alone are able to understand most of the contents. In addition, since simple models are presented before more complex models and additional parameters are added gradually, students should not worry about the difficulties in mathematics. 

Mathematics drives all aspects of chemical engineering. Calculations of material and energy balances are needed to deal with any operation in which chemical reactions are carried out. Kinetics, the study dealing with reaction rates, involves calculus, differential equations, and matrix algebra, which is needed to determine how chemical reactions proceed and what products are made and in what ratios. Control system design additionally requires the understanding of statistics and vector and non-linear system analysis. Computer mathematics including numerical analysis is also needed for control and other applications. 

The evaluation of matrix elements for exphcitly correlated Gaussians (46) and (49) can be done in a very elegant and relatively simple way using matrix differential calculus. A systematic description of this very powerful mathematical tool is given in the book by Magnus and Neudecker [105]. The use of matrix differential calculus allows one to obtain compact expressions for matrix elements in the matrix form, which is very suitable for numerical computations [116,118] and perhaps facilitates a new theoretical insight. The present section is written in the spirit of Refs. 116 and 118, following most of the notation conventions therein. Thus, the reader can look for information about some basic ideas presented in these references if needed. 

It is known from matrix differential calculus that for a matrix variable X and a constant matrix C the following is true ... 

Here, we will need some simple facts from matrix differential calculus. If X is a matrix variable and (3 is a parameter that X depends on, then... 

We begin with the derivative of the secular equation with respect to energy eigenvalues. For some background on matrix differential calculus, see the Refs. 116 and 117. 

Matrix elements are scalar-valued matrix functions of the exponent matrices Lk- Therefore, the appropriate mathematical tool for finding derivatives is the matrix differential calculus [116, 118]. Using this, the derivations are nontrivial but straightforward. We will only present the final results of the derivations. The reader wishing to derive these formulas, or other matrix derivatives, is referred to the Ref. 116 and references therein. 

The gradients of the molecular integrals with respect to the nonlinear variational parameters (i.e., the exponential parameters Ak and the orbital centers Sk) were derived using the methods of matrix differential calculus as introduced by Kinghom [116]. It was shown there that the energy gradient with respect to all nonlinear variational parameters can be written as... 

J. R. Magnus and H. Neudecker, Matrix Differential Calculus with Applications in Statistics and Econometrics, Wiley, Chichester, 1988. 

The matrix/vector form of k, eqn.(34), allows us to exploit the powerful matrix differential calculus, described by Kinghorn[ll], for deriving elegant and easily implementable mathematical forms for integrals and their derivatives required in variational calculations. Alternatively, k can be written purely in terms of the vector variable r,... 

The power of the matrix differential calculus is immediately apparent when one actually computes an analytic gradient for a matrix function. The ease with which results are obtained and the concise compact form of the results seems almost miraculous at times. When the derivatives presented here where first formulated, the results were so surprising that numerical conformation was performed immediately. All of the following matrix derivatives have been confirmed by finite differences term by term on random matrices. 

D.B. Kinghorn, Explicitly Correlated Gaussian Basis Functions Derivation and Implementation of Matrix Elements and Gradient Formulas Using Matrix Differential Calculus, Ph.D. dissertation, Washington State University, 1995. 

Numerical methods include those based on finite difference calculus. They are ideally suited for tabulated experimental data such as one finds in thermodynamic tables. They also include methods of solving simultaneous linear equations, curve fitting, numerical solution of ordinary and partial differential equations and matrix operations. In this appendix, numerical interpolation, integration, and differentiation are considered. Information about the other topics is available in monographs by Hornbeck [2] and Lanczos [3]. 

A. Analytic geometry C. Matrix operations E. Vector analysis G. Differential calculus... 

Similar to the calculus of functions, we can differentiate matrices by differentiating the elements of the matrix in each case. The usual designation for the derivative of a square matrix Y is dY/dx if the matrix elements are functions of the scalar x. The differential coefficient of a product is also similar to that of a product of scalar functions however, the order of the matrices in the product must be maintained. Therefore,... 

It is the nature of the subject that makes its presentation rather formal and requires some basic, mainly conceptual knowledge in mathematics and physics. However, only standard mathematical techniques (such as differential and integral calculus, matrix algebra) are required. More advanced subjects such as complex analysis and tensor calculus are occasionally also used. Furthermore, also basic knowledge of classical Newtonian mechanics and electrodynamics will be helpful to more quickly understand the concise but short review of these matters in the second chapter of this book. 

Here we show how the homogeneous system of linear ordinary differential equations, Eq. (VIII.6) can be solved by well-known methods of calculus (26,54). In the most general case Eq. (VIII.6) can be rewritten in matrix notation as... 

---