Posts tagged Note

Summer Note 2-The Adjoint Method

0

The outline can be listed below:
1. The bounded linear operator, and the continuity definition of a linear operator, which is equivalent to: bounded or continuous at the origin.
2. The dual space, which is defined as a real bounded linear function.  The adjoint operator definition:
Let [latex]T:X \to Y[/latex] be a bounded linear operator between Banach spaces X and Y.  The adjoint operator [latex]T^{*} : X^{*} \to Y^{*}[/latex] can be constructed as: [latex]T^{*} (F)(x)=F(T(x))[/latex] for all [latex]x\in X, F\in Y^*[/latex]
Because every bounded linear operator over a Banach space X can be reconsidered as inner product:
[latex]F(x)=<x,y>[/latex]for all [latex]x\in X[/latex] and [latex]||F||=||y||[/latex] Here y is only linked with F.
So it is a comparatively simple step to define the inner product form of an adjoint operator [latex]y^*[/latex] instead of the function form[latex]T^*[/latex]:
[latex]F_T(x):=<T(x),y>=<x,y^*>[/latex] for all [latex]x\in H[/latex]
3. The differentiability definition in a Banach Space:

Gateaux differentiable at [latex]x_0[/latex]

[latex]DT(x_0;h):=\lim_{\epsilon \to 0} \frac{T(x_0+\epsilon h)-T(x_0)}{\epsilon}[/latex]

Frechet differentiable:

[latex]\lim_{||\delta x||\to 0}\frac{||T(x_0+\delta x)-T(x)-T’(x_0)\delta x||}{||\delta x||}=0[/latex]

They are equal when the two equations exist on all [latex]x,\delta x \in X[/latex]

The partial Frechet derivative [latex]D_x(x_0,y)[/latex]of an operator:

[latex]\lim_{||\delta x|| \to 0}\frac{||U(x_0+\delta x,y)-U(x_0,y)-D_x(x_0,y)(\delta x)||}{||\delta x||}=0[/latex]

Because the differentiability is defined by limit equations, the notion of the chain rule applies to operators that are continuously Frechet differentiable.

Then it comes the optimal question, I had the problem on figuring out the two functions given: [latex]J(\rho,\lambda),E(\rho,\lambda)[/latex].  [latex]J(\rho,\lambda)[/latex]is the cost function and [latex]E(\rho,\lambda)=0[/latex] is the solution of [latex]\frac{\partial \rho(x,t)}{\partial t}+\frac{\partial}{\partial x}(v(\rho)\rho(x,t))=0[/latex] with conditions.

Because [latex]\rho[/latex] is a function of [latex]\lambda[/latex], [latex]J(\rho,\lambda)[/latex] can be seen as [latex]j(\lambda)[/latex].  Then the book gives 3 conditions assumed, which assume everything is continuously Frechet differentiable.  I have some problems here.  I can’t understand how he find these 3 assumptions.  Hope it will be clear when I reading the calculation part.

I should look up the details on Lagrangian operator.  The only thing I know in calculating minimum is Lagrange multipliers.  Which is said not work during the semiconductor factory problem.  I will update the Algorithm For Solving when I am ready.

Summer Note 1 – Mathematical Models for Control Systems

1

It has become a problem that nowadays control system is getting complex. To describe, calculate, and optimize production, 3 classical mathematical models are introduced.

1.Discrete event simulation: very accurate but hard to calculate complex models. Probability distribution calculation needed. Distribution functions calculability limited by sample accessibility.

2.Queueing networks: fairly accurate (using average data) but hard to calculate complex models. Only works with steady models (time infinite).

3.Fluid Models (ODE): [latex]\frac{dq(t)}{dt}=\lambda(t)-\sigma(t)[/latex]
It describes queue change rate is influx rate minus outflux rate.
It is easy to calculate. But it uses average data (no possibility included), which may lead to trivial results. And [latex]\lambda(t)[/latex] and [latex]\sigma(t)[/latex] are not easy to find.

So the author introduced a new model: Continuum Model
It uses a totally new view. The model doesn’t put too much attention on the physical position. A new attribute is introduced for each item, stage x, from 0% to 100%. Then the completion of a item can be involved into mathematical calculation.
The core equation is:
[latex]\frac{\partial \rho (x,t)}{\partial t}+\frac{\partial}{\partial t}(v(\rho)\rho (x,t))=0[/latex]
The function [latex]\rho (x,t)[/latex]is the density, when the stage is x and time is t.

It took me some time to understand the equation. I think it is just like the flood in Yangtze River now. The height of water is the function [latex]\rho (x,t)[/latex], over position x and time t.
The equation is setup up like this because the change of [latex]\rho (x,t)[/latex] over time(x fixed) is just like the change of the height of water. It is caused by the different velocity of water over position x. The different velocity is describe as the [latex]-\frac{\partial}{\partial t}(v(\rho)\rho (x,t))[/latex].
Here the author uses [latex]v(\rho)[/latex], which describe moving speed of 1 item (just like 1 water molecule in Yangtze River).

The velocity [latex]v(\rho)[/latex] is defined by:  [latex]v(\rho)=\frac{v_{max}}{1+\int_0^{1}\rho (s, t) ds}[/latex]

or by [latex]v(\rho)=v_{max}(1-\frac{\int_0^{1}\rho (s,t) ds}{L_{max}})[/latex].  Here [latex]L_{max}[/latex] is the max load.

I haven’t looking up the article where these two definition come from.  But it can be concluded from experience I think.

In this model, the changing chain: influx–WIP–velocity–outflux

BOP

2

The balance of Payments

These are the accounts used in the balance of payments

  • Current Account
  1. Net export/imports of goods (balance of trade)
  2. Net exports/imports of services
  3. Net income (investment income from direct and portfolio investment plus employee compensation)
  4. Net transfers (sums sent home by migrants and permanent workers abroad, gifts, grants, and pensions)
  • Capital Account
  1. Capital transfers related to the purchase and sale of fixed assets such as real estate
  • Financial Account
  1. Net foreign direct investment
  2. Net portfolio investment
  3. Other financial items
  • Net Errors and Omissions
  1. Missing data such as illegal transfers
  • Reserves and Related Items
  1. Changes in official monetary reserves including gold, foreign exchange and IMF position.
Go to Top