Computer Chemistry

Mario Marsili

• Publisher: CRC Press
• Year: 2017
• Print ISBN: 9780367835491
• eBook ISBN: 9781351367455
• Edition: 1
• Topic: Physical Sciences
• Subtopic: Chemistry
• Index: Chemistry

Computer Chemistry illustrates the methods and philosophies of how a computer can be instructed to "understand" chemical facts, formulas and rules. It focuses on discussions of all of the major sections in both theoretical framework and practical application through examples. It includes the Synthesis Design Systems for the simulation of chemical reactions, the Structure Elucidation Systems for the interpretation of spectral data, the Molecular Modelling Systems for the visualization of chemical structures and the calculation of physico-chemical parameters.

Chapter 1
Introduction

I. Man and Computers

Computers have entered most areas of scientific research, industrial production, and educational activities to such an extent that an impact has even been made on the social life, mental attitude, and the psychology of people. Computers can often replace or support many human activities at low costs: cars are assembled by robots; teachers are substituted by computer programs, experienced instructors by simulators. This has occurred because computers are millions of times faster than man. Speed is the name of the game, and speed means competitiveness on the market, low financial investments, and better overall performance. On the other hand, a certain number of disappearing human activities, obsolete and no longer considered profitable, are transformed into new equivalents under a different perspective: the computer perspective. Somebody who in the past manufactured coil springs for wristwatches is almost no longer required, having been replaced by somebody constructing the integrated circuits on which modern watches rely.

Computers have disclosed new frontiers in medicine, improving diagnostic techniques (e.g., imaging in computerized axial tomography). They have caused a real revolution in data management and communication and allow modeling of extremely sophisticated systems like astrophysical events or weather forecasts.

Computers undoubtedly provide a number of astonishing improvements in several sectors of the modern world, but are at the same time the backbone of modern warfare, which has created the most incredible array of annihilating weapons ever (pattern-recognizing "intelligent" missiles, for example). For the single human, this double-faced process of technological evolution has bloomed into a wealth of new professions, all of them connected to computer science, be it theoretical or applied.

Computers are neither good or bad; a knife is neither good nor bad. Each depends on its use. Philosophical fights are raging everywhere on the role of man in a computer-dominated world in which few selected specialists have the knowledge and the power to press strategic buttons on keyboards, and no final solution is expected soon. The question whether human intuition (in other words, the artistic gift, the invention, the intellectual breakthrough) can be replaced by computer simulation, once computers have enough memory and speed to tackle such problems, is indeed a central question and contains even a touch of moral texture.

If a computer simulation based on artificial intelligence systems leads to some unexpected brilliant scientific discovery, is this the merit of the human programmer or of the "thinking" computer?

Chemistry is no exception within the framework of this discussion. The introduction of computer-assisted research techniques into chemistry over the last 15 years has caused a split pattern of reactions among chemists. Whenever computers have been used in a kind of subordinate, secondary, concealed way, they have been accepted as precious and powerful help. This has especially been the case with regard to chemical information and in analytical chemistry. On the contrary, as soon as computers entered an apparent role of equality with the human chemist in solving problems of a more decisional type, exerting a primary, direct influence on man-tailored research strategies and methods, an evident anxiety arose among traditional-minded chemists. Chemists saw (and still see) their leading role as "masters of the art" endangered by an "idiot made of steel". Grown on a serious misunderstanding of the role of computers in chemistry, this attitude in some cases has led to mental rejection of this new technology at the level of its cultural root. On the other hand, enthusiasts are readily found who expect immediate successful results to a variety of difficult problems, believing that "the computer can do everything." They forget that computers still depend primarily on man's performance.

To understand the reasons for a methodology called computer chemistry, to correctly place it among modern research methods, and to detect its benefits and limitations — these points must be discussed in some depth.

II. Computers in Chemistry

A. Computational Programs

A distinction was postulated above between a direct, or primary, influence of computer action on chemical research and a subordinate, secondary one. Historically this distinction, caused by an independent growth of what is called computer chemistry from other traditional fields of computer applications in chemistry, was rooted in two main facts: the attempt to create computer programs to emulate chemical thinking, and the parallel development of a new, fascinating, and promising branch of computer science, artificial intelligence (AI). AI, which will be discussed later to some extent, is the part of computer science dealing with the computer-generated perception and solution of complex symbol-oriented and semantic problems.

In the early 1970s, chemists were acquainted with a purely numerical use of computers in chemistry. Quantum chemistry and X-ray structure determination were the poles of heaviest exploitation of the fast computational capacity of a computer. In both of these important research fields, the investigator faces such an enormous quantity of bare numbers that their successful treatment would be utterly unfeasible without electronic data processing. The main role of computers in performing these tasks simply consists of managing huge arrays of numbers following a user-implemented, rigid, predetermined prescription. The result of what in a joking manner is termed "number crunching" is in all of these situations a mere numerical result. In other words, the computer delivers a certain number of specific magnitude that interests the user, and the path along which such a number is generated is a one-way road within the codified program. Solving iteratively thousands of Coulomb or exchange integrals and refining Fourier coefficients are examples of such a path. Here the computer follows a fixed scheme of data processing. The final result, for example, could be the energy of some specific electronic state of a given molecule or an array of cartesian coordinates for atoms in a molecule. That is what we expect. The magnitudes of energy and coordinates will change if the investigated substrate is different, but this is obvious. They will also change if a different degree of approximation, refinement, or parameterization is chosen by the user. What does not change is the certainty that some number will come out as the unique result. We might not known in advance what energy value a certain molecule will show at its conformational minimum, but that is the main reason for using a computer: to do the necessary calculations according to user-determined equations which already contain the solution to the problem in all its principles. Due to its advantage in speed, the computer offers a numerical result for final interpretation by man. The program run by the computer contains no alternatives other than to produce quantitative numerical answers of one and the same kind, repetitively, as it has been instructed to do. Truly, there are no alternatives to atomic coordinates for a program that calculates atomic coordinates. The statement "I shall ask the computer to tell me the energy of formation of this molecule" appears to be conceptually and semantically wrong. Justified questioning anticipates the potential existence of an answer; answering demands the a priori existence of choice elements among which a suitable answer can be found.

A quantum mechanical program, once implemented according to a particular approach, is geared in a way as to solely calculate a set of numerical quantities, and it has no choice elements on which to exert any kind of deductive evaluation for constructing an answer. Thus, the actual calculation is just a reproduction of the equations contained in the program, substituting real numbers for symbols: no influence is exerted by the computer on the strategic content of the program, on its structure, or on its meaning, and the computer will not be able to change the structures of the equations themselves during execution. Question and answer are like vectors: each has a magnitude and a direction in space. The direction determines the difference between a vector and a scalar. Selecting a direction (i.e., including deduction in the formulation of a certain answer by considering the nature of the available choice elements) means adding a qualitative element to a purely quantitative response. Calculating orbital energies cannot produce chemical answers within the conceptual framework just expounded because programs tackling these kinds of computational problems yield scalar numbers (e.g., energies) as results. The direction that we miss in such results, which is nothing less than the general structure of the solution scheme, is called the solution model. In lucky cases of a known theory, this direction is known in advance by the investigator and formulated as a sequence of instructions in a computer program. We can finally assert the following:

Assertion I — Computational programs in chemistry rely on predefined solution schemes, the models, which are known in their qualitative essence by the user. The output of such programs is a quantitative response, a scalar, for the model under specific, user-given conditions. The generation of such responses follows a rigid, unbranched, and constant data processing mechanism. No strategy evaluation is involved.

It clearly now appears that computer support in this fashion does not scratch the polished image of any scientist devoting his time to the discovery of fundamental theories or models. He remains master of the situation and welcomes computer aid as a fast and reliable processor of numbers in a kind of subordinate position. In final words, the computer will not teach him anything.

B. Semantic Programs

What would happen to human psychology and to scientific research if a computer started to deliver qualitative answers, to give strategic advice, to propose models, to change the structure of user input equations, or to emulate chemical reasoning?

To do this, a computer perception of quality must be created. Quality involves comparison; comparison involves rules for judgment; using rules involves the capacity of autonomous acting; acting involves effects; effects involve interpretation and ranking, which finally contribute to the establishment of quality. Quality and quantity together build our response vector, the answer.

Computer chemistry started off right at this point: it provided programs, along with the first blooming achievements and concepts in AI, that were able to help chemists discover strategies. These programs had to be organized flexibly enough to deal with varying mechanisms for making choices. This key term requires the questions addressed to the computer to have, in principle, a manifold set of possible outcomes, which undergo evaluation and ranking.

The intrinsically different response vectors may differ in probability (the magnitude of the vector) and in direction (the quality, the conceptual content of the computer-generated solution, the strategic orientation). Such programs are well suited, in general terms, to provide alternative models, thus enhancing knowledge. That is exactly the complementary (not the opposite) situation to computational programs. The latter apply established models, while the former use general rules (empirical or theoretical), to produce models and ranking strategies. For example, calculating the energy in calories that one needs to move one's arm while playing chess (i.e., to pick up a piece, move it to its new position, and lower the arm again) corresponds to the use of a program belonging to the computational class. However, asking the computer that has been "taught" chess rules to predict all possible sequences of moves leading to checkmate, starting from a user-given initial pattern, is an example of the use of programs of the AI class. Here the process of establishing strategies, predicting countermoves, and ranking sequences of moves according to chance of success is the principal feature of such an autodeductive program.

In computer chemistry, chemical rules are transformed into a program inside a computer, making the electronic device look like it is thinking chemically and therefore turning it into a seeming threat, a cold, stainless steel rival of any human chemist. Computer answers of the following kind are common today, and they make the instinctive repulsion among a few, if not justifiable, at least comprehensible; for example, "Your mass spectrum belongs with 96% probability to a molecule with three chlorine atoms," or "There are 24 different reaction routes within an exothermic range of 0 to 10 kcal/mol that can lead to your desired product; I will draw them for you," or "After interpreting all your spectral data, three molecular structures were found compatible and were generated; here they are," or "You don't have to care for the temperature parameter while running your chemical reactor; adjust the pH to 5.5 instead."

These answers clearly go far beyond those to which chemists had been typically accustomed. They offer direct intervention into operational strategy, as well as tactical realization. They lead to a redesign of a certain experimental setup or to a new, unexpected conceptual insight. Thus, a revised model can be developed. We finally can assert the following:

Assertion II — Semantic programs are the core of computer chemistry systems. They are tailored to reproduce schemes of human reasoning — in our case, of chemical thinking. They use chemical rules to treat the strategic, decisional kind of problem. They have a primary influence on subsequent methodologies, the establishment of models, the creation of alternatives, and the intelligent interpretation of data in chemical research.

C. Computer Chemestry and Human Psychology

The first accomplishment that must be fulfilled is the computer perception and recognition of chemical symbols. Our whole comprehension of chemistry is based on a reiterate confluence of symbols and their chemical content in the human brain, where they are perceived and stored. This process, which takes place over all the years of apprenticeship in chemistry, establishes an automatism that elicits all our chemical knowledge if a visual or phonetic stimulation is conveyed to our cerebral chemical data base. For example, if someone is told the word "benzene", he most likely will visualize in his mind the familiar pictorial symbol for benzene; however, at the same time he will subconsciously correlate to it a number of specific features that he knows are hidden somewhat cryptically in the depiction which certainly belong to benzene as a real chemical entity.

The benzene symbol automatically includes the six hydrogen atoms not drawn explicitly, and the ring inside the hexagon is immediately understood as symbolizing six delocalized π electrons. Even the concept of de localization is recalled in the brain and is readily formulated as a (4n + 2)π-electron Hückel rule. This happens at an astonishingly high speed in the human mind. The reason for it is that symbols and their correlated chemical and physical properties are already stored in the brain; they represent our chemical knowledge base. Recalling chemical data (retrieving structural formulas) is a procedure that we do every day while discussing chemistry. A computer does very similar work when used for chemical data retrieval, one of the first applications of computer technology in chemistry. Conceptually, data retrieval is remotely connected to semantic programming, as it generally deals with the matching of input character strings (the name of a molecule, for example) with corresponding strings inside the data base. A relation to truly semantic systems is to be found just in the ability of modern retrieval systems to accept symbols as input, to perform sophisticated logical search and matching operations, and to return the results in an equally sophisticated, symbol-oriented manner. However, no additional original material is generated by the computer during a search session. Autogenous creation of something new must occur by different paths, both in the brain and in computers. Searching for a chemical structure in an array of collected structures stored on some magnetic device can have only one of two possible outcomes: found or not found. In the "not found" situation, the computer cannot augment...

Table of contents

• Cover
• Title
• Copyright
• Dedication
• Preface
• The Author
• Table of Contents
• Chapter 1 Introduction
• Chapter 2 The Computer as a Laboratory
• Chapter 3 Problem Solving and Artificial Intelligence
• Chapter 4 Molecular Modeling
• Chapter 5 Autodeductive Systems for Reaction Kinetics
• Chapter 6 Structure Elucidation Systems
• Chapter 7 Computer Simulation of Organic Reactions
• Appendix
• Index