what is mathematical modeling(什么是数学建模)

This blog is only for my personal review of mathematical modeling module.（lecture notes and personal understandings）

1. Why mathematical models?

This is all about complexity. Problems in science and engineering tend to be complex, usually involves a large number of entities, the relationships among entities could also be nightmare. We are craving for a tool that enable us to go more deeper into the problems to get more insight, or on the other hand , learn more about the future with existing data. Modeling allow us to tackle such complexity, or in another words, to simplify problems and make them tractable.

2. What is system?

system is as an entity or collection of entity whose properties we want to study. System can be defined alternatively as an assemblage of objects joined in some regular interaction or interdependence. As a modeler, what are the factors we are interested? We need to understand what is the system of interest and how does it work. What causes changes in the system and how sensitive the system is to these changes.

3. What is model and mathematical model?

Marvin Minsky defines a model as “To an observer B, an object $A^{\ast }$ is a model of object A to the extent that B can use $A^{\ast }$ to answer questions of interest about A”.Two interesting implications are worth-mentioning. Firstly, this definition implies that a model is supposed to serve one specific problem or one series of problem. Secondly, model vary as observer changes.

Apparently, the purposes of modeling are:

• To answer the question.
• To solve the problem.

Mathematical models inherit the same properties of general models, but they are built by mathematical concepts and equations. The reason why we need mathematical model is that we have to make systems observable and measurable. They map relationships between system input and output.

Modeler are always looking for an easy way out! Never overcomplicate problems! So what is a good model? The best model is the simplest one that still allows us to understand the system and solve problems related to it. Because if a model is overcomplicated, it is hard to capture more insight whereas a model is not able to describe the problem we are dealing with if it is too simple.

How do people simplify a complex problem? How do we tackle complexity on daily bases? We tend to extract the information that we believe is vital and selectively ignore the others. The same idea is used in simplification for mathematical models. A mathematical model usually takes the shape of {S,Q,M}, S is the system,or a reduced system. Q is a set of questions that we are trying to solve by using the models. M represents a set of solustions (equtions) that can solve Qs.

Two points should be emphasized here are that the order of {S,Q,M} follows chronology and the model has to be linked to its purposes.

I came cross a interesting question earlier today. “What is the relationship between mathematical models and linear regression?” It is a stupid question though. From my point of view, a mathematical model is a general term for combination among algorithms to solve a specific problem, whereas linear regression itself is an algorithm, but it could be a complete mathematical model if it is able to solve a problem without combining other algorithms.

4. Classification of models.

Models can be classified differently based on their characteristics, there are many kinds of mathematical models:

starting with most general classifications:
1.Black-Box Model
when using this type of model, we cannot get any insight of the model. Only thing we know is that we feed our inputs into the model and we get corresponding outputs. We have no idea what happend in the process, we have no ideas what have our inputs gone through to get this output.
Just like when we are trying to use ‘SVM’ in ‘Sklearn’ to fit the dataset of ‘iris’. We just use a existing model to fit our data and we get 100% classification rate. We don’t bother how Lagrange multiplier is used and how it manages to find the optimal.

2.White-Box Model
we know the input-output mapping and how the relationships or parameters govern the output. In another words, we know 100% what is going on in the process, we know every detail. We know how and why that our inputs are transformed into outputs.

3. Grey-box Model
It is at the middle of two extremes mentioning above. Grey-box model is where we start with an unknown process(it is safe to say it starts at a black-box level) and as we carry out more and more experiments(simulations), the details in the process are gradually getting transparent.In another words ,we learn more and more relationships between input and output.

Mathematical model can be categorised more specifically.

• Phenomenological model and Mechanistic model
  The formal defination of the former is:

  A mathematical model (S, Q, M) is called phenomenological if it was constructed based on experimental data only, using no a priori information about S.

  The major problem of this kind of model is that “you don’t know what you don’t know”,meaning you don’t know if you are missing any vital parameters.

• Mechanistic model
  Unlike the former one, when using this type of model, “you know what you should kown”. Formal definiation is :

  A mathematical model (S, Q, M) is called mechanistic if some of the statements in M are based on a priori information about S.

  Models can also be grouped base on time and space.

• Static and dynamic models
  A mathematical model is a dynamic model if at least one of its parameters depends on time. Otherwise, if any of a system parameter not depend on time, we call that is a static model.

• lumped and distributed model
  a model can be considered as a distributed model if any of its parameter depends on location. And lumped otherwise.

5. How to build a model?

There are several steps we have to follow to build a valid model.
Figure 1 graphically shows the steps we need to go through:

1. Problem identification.
   Apparently, the first step is to set a or multiple goal(s) that our model needs to serve.
   What we should take into consideration when building a model?
   1.What is (are) the purpose(s) and objective(s) of model we are building?
   2. What is the system?
   3. What information do we have so far. Does it describe system?
   4. what is the criterion that we use to measure outcomes.

2. System Analysis
   At step of system analysis, we need to focus on data. Are data we have reliable? Are they sufficient? Do we need to collect more data? After we can draw a “layout” of the model, we need to think about how to simplify the model. Simplification is the key! Because it is hopeless to model all the factors in a complex system. The more factors involved, the more combinations will be. Consequently, more experiments, simulations, trials we have to do to guarantee the accuracy. It would require much more recouses and much more time-comsuming.

3. Formulate Mathematical Model.
   After we understand what is the situation we are dealing with and how much we understand the current data. We then look for the simplest model first. The simplest model is the one that involves the least parameters.

4. Obtain Mathematical Solutions.
   Use numerical method to solve the model. If the simplest one is good to solve the Qs, then mission accomplished. But if it doesn’t, we tend to complicate our model by gradually adding more related parameters until it satisfies the purposes.

   5.Interpret The Model
   Examine the results. Figure out the robustness of the model.

   6.Compare With Reality.
   How does it perform on the test data? Does solutions make scence? can it be simplified?

6. Examples

Following is an example that my lecuturer used in the lecture.

Suppose we want to model the population of red fox in London. We want know how that population change monthly. Following are some information about red fox in London.
Initial number of foxes in London: 150
monthly reproduction rate: r=0.25, monthly death rate: d= 0.1

What is S ,Q,M.

My solution:
S:
The entire system should involve red foxes and london.

Q:
How many red foxes will have in London in month N?

M：
F(n)= F(n-1)*(1+r-d), where F() is the population of foxes in month n.