The aim of theoretical physics is derived from a small amount of basic principles (such as minimum working principles). In the field of software design, it is full of "best practices" in the total grinding forest. Too many rules will only mean there is no rule. The status quo in the software design is indicated that this area is still in a very young stage, and more knowledge should be borrowed from other fields. Some specific analysis techniques in the software have been explained in the BLOG in front of me. But how to effectively apply these analysis techniques, we also need some guiding theoretical framework. If the software design is placed in a broader system engineering, a suitable dominant principle should be the minimum complex principle. That is, everything we do during the software analysis, whether object-oriented analysis, workflow-based analysis, etc., its purpose is to minimize the complexity of system build. As long as it is possible to effectively reduce the complexity of the system, the method used and the established model is good. The complexity is the only measure, rather than whether it is in line with someone's point of view, and whether popular technologies are used (of course, popping technology often means reduced complexity with other system interactions).
Li Xiaolong said: Simple is beautiful. But is the easiest thing is the best? The scientific community has a negative conclusion. A very profound basic fact that complexity is that complexity is classified, and different complex hierarchies have essential differences, and a lower complexity method can solve a higher complex level. So Einstein said: make everything as Simple as Possible, but not simpler. The existence of complexity column means that as we get more information or because the system itself has developed, when the model of the system we established along the complex level column, we have proposed solve the problem The solution must also make a corresponding evolution. That is, we provide some fixed solutions, but must be a complete strategy (STRATEGY). This way we are not isolated on a question, but to see its past, now and the future.
When analyzing, we all know "from the general to special, from special to general", but how to scientifically, have steps to do? I learned a basic analysis framework from the BBGKY HIERARCHY of plasma physics: level Column Theory. This theory first defines a priority model: gas consists of N identical atoms, and its kinetics are described by n mutually coupled Newton mechanics equations. The second step of theory is to start from the most extreme simplification, assuming that there is no interaction between all atoms, ie, assuming the atoms are independent of each other. It will get the VLASOV equation. The third step of the theory is gradually increased. When the gas density is high, it must be considered two or two collisions between the molecules to obtain the Boltzman equation. When we need to consider higher order interactions, we get the associated motion equation, so continued to build an infinite level.
Abstract, this theory contains the following:
Establish the most widely used model M0 that describes all situations, the model defines the boundaries of the system
Establish a simplest model MN containing minimal assumptions, the model as the basis of solving and basic reference.
Establish a model level column that gradually increases from the complexity of Mn to M0, ensuring that there is a valid modeling of the system at the level of each complexity, and when solving high-order problems, you can refer to lower Solution of the Problem.
How to make a choice from this infinite level column, must be determined according to the specific application scenario, Vapnik describes this solution strategy in its statistical learning theory:
Integrated cost = Dialect between training data and model prediction model itself
Learning = minimize [comprehensive price]
That is, selecting the simplest in the scheme that can solve the problem.
