Tutorial 3 -  Linear Algebra 1

generic ECE - U. Houston

                                                                                      Han Q. Le (c) copyrighted

page 1              page 2               page 3

3  General matrix dot product

3.1 Example: hypothetical portfolios

Consider that there are more than 1 client, and the shares are as follow:

In[158]:=

Now, how do we calculate the portfolio for each of them?

In[163]:=

As shown, it is straightforward: the share data are now a matrix of 3 columns for 3 clients (imagine, a brokerage may have millions of clients, and how big their share data matrix is).

This is known as matrix product ot matrix multiplication and matrix-vector dot product is just a special case of matrix dot product.

Review the Matrix Dot Product App again (see 2.4 above):

3.2 Matrix dot product associative property

What if we have this “cascade” problem:

Three bakeries purchase certain quantities of flour and eggs per week as shown below. Given the prices each week, what is the cost of each bakery each week? This is a straight forward matrix multiplication (dot).

Data matrices

Out[92]=

Exercise 3.1

Three bakeries purchase certain quantities of flour and eggs per week as shown. Given the prices each week, what is the cost of each bakery each week?

Answer

This is a straight forward matrix multiplication

In[93]:=

Out[93]=

In[100]:=

Out[100]=

Exercise 3.2

But now, we can get more into the details. There are 5 types of cakes, the quantity of flour and eggs for each type is shown below

In[166]:=

Out[168]=

flour quantity	0.	0.3	0.1	0.5	0.4
egg quantity	0.75	0.25	0.5	0.2	0.6

And the daily quantities of each cake produced by the bakeries are as follow:

In[169]:=

Out[170]=

bakery A	bakery B	bakery C
55	70	40
80	50	40
30	45	30
25	40	30
60	80	50

What are the daily ingredient demands of the bakeries?

Answer

This is what gives the net four and egg demands of the bakeries each week:

In[171]:=

Out[171]=

This is what we did:

In[172]:=

Out[173]=

Then, we multiply the price matrix to the demand matrix to get the cost matrix for the 3 bakeries:

In[95]:=

Out[95]=

End exercise

In other words: costmatrix = pricematrix . demandmatrix
                       costmatrix = pricematrix . (ingredientmatrix .  bakeryproductionmatrix)
                         4 × 3                 4×2           (      2×5                               5×3        )

However, can we do this?
                       costmatrix = (pricematrix . ingredientmatrix)  .  bakeryproductionmatrix

because pricematrix . ingredientmatrix = cakeunitcostmatrix
and the total cost can also be obtained by multiplying cake unit cost with the cake production matrix. This is the cake unit cost

Exercise 3.3

Find the weekly cost of each type of cake, based on the weekly price of the ingredients and the ingredient requirement of each cake type.

Answer

In[85]:=

Out[85]=

flour ($) eggs ($) week 1 34. 33. week 2 31. 30.5 week 3 37.5 35.5 week 4 30. 38.

0. 0.3 0.1 0.5 0.4 flour quantity 0.75 0.25 0.5 0.2 0.6 egg quantity

In[177]:=

Out[178]=

cost of each cake each week

week 1 13.875 9.875 11. 12.45 18.1 week 2 16.5 12.55 13.35 16.15 22.6 week 3 12.375 10.125 10.25 13.3 17.9 week 4 15.75 9.9 12.05 11.95 18.8

End exercise

If we know the weekly ingredient cost of each type of cake, and we know how many cakes of each type each bakery makes daily, then, can we find the net weekly ingredient costs of the bakeries? This is the next exercise

Exercise 3.3

Find the weekly ingredient costs of the bakeries, based on the weekly ingredient cost of each type of cake, and the quantities of cakes of the bakeries.

Answer

We can just multiply this with a # of cakes produced by each bakery per week to get the cost

In[179]:=

Out[180]=

End exercise

Clearly:  
              

In other words, there are more than one way to cut a cake. We can calculate the total flour and egg demand of each bakery per week, and multiply the ingredient prices to get the cost
                                            OR
we can multiply the ingredient amount of each cake with the weekly ingredient price to get the weekly cost of each cake, and multiply to the production volume of the cakes.
Both give the same result.

Mathematically, what it means is that for 3 matrices A, B, C:
                                 (A . B) . C = A . (B . C)
This is known as associative property. From now on, we don’t have to use parentheses when multiplying many matrices together.

In[141]:=

Out[145]=

A.B		C		(A.B).C
	.		=
A		B.C		A.(B.C)
	.		=

Exercise 3.4

Do a demonstration similar to the above to show the associative property of dot product. It is OK to copy and paste the code. But you must choose the range of random integer and the matrix dimensions according to this:
Range of random integer {-r, r} where r is the 2nd largest digit of your Student ID, but not smaller than 3. For example 1094724 is your ID. The second largest digit is 7 (9 is the largest).
For matrix dimensions, we need this: {m, n},{n, p},{p, q}
Let m=3rd largest digit of your ID, but not smaller than 3, n=Min[8, birth month +2], p=Max[3,Floor[birth date/4.]], q=Max[2,Ceiling[last digit birth year/2]].
If you cannot calculate, use the App below (but you should verify to understand and know why the results are what they are).

In[203]:=

Out[203]=

Answer

In[204]:=

Out[208]=

A.B		C		(A.B).C
	.		=
A		B.C		A.(B.C)
	.		=

3.3 Transpose theorem

One more frequently-used property is:
                                       

In[154]:=

Out[157]=

A	B	A.B

Exercise 3.5

Do a demonstration of the transpose theorem like above, but matA and matB are based on your range and dimension from Exercise 3.4.

Answer

In[209]:=

Out[212]=

A	B	A.B

End exercise

3.4  Serious and common errors

In[9]:=

Out[9]=

In[10]:=

Out[11]//MatrixForm=

Shoud we do vt3 . matx3 or matx3 . vt3?

In[12]:=

Out[12]=

Error because we are doing this:

The number of row must match with the number of column. We should do this instead:

In[13]:=

Out[13]=

The rule is simple:

In[14]:=

Out[14]=

If we have a vector of length 5, it should be on the left side of the matrix. If length 3, it should be on the right side. The actual orientation, row or column doesn’t matter, because Mathematica knows that if it is on the left, it is automatically row, and column if on the right. Hence:

In[15]:=

Out[15]=

But the above is not the worst error, because it tells us if there is an error. This is actually worse:

In[16]:=

Out[17]//MatrixForm=

now

In[18]:=

Out[18]=

In[19]:=

Out[19]=

The above are totally different, but there is no execution error warning. This type of hidden bug is far more serious. It is up to the programmer to determine which one is correct.

Exercise 3.6

Generate an {m,n} matrix and an{n,p} of your choice, such that n≠m≠p.
Try to dot {n,p} to {m,n} and observe error. Then, do the dot the correct order to obtain answer.

Answer

3.5  Inverse matrix and determinant

Consider this problem:

the items purchased are

Consider that the cost ($) for the above purchase is:

What are the prices of apple, orange, and pear? This is the equation:

How do we solve?

In[39]:=

Out[39]=

The linear algebraic way to do it is to take the inverse matrix:
                        
    
By definition:
          
Hence:          
        
                                 

In[40]:=

Out[40]=

indeed, these are the prices.

Exercise 3.7

Create 2 equations with 2 unknowns, use inverse matrix to solve it.

Answer

Problem 3.8

Consider this circuit:

Solve for its response. Find  steady-state voltage for harmonic and of arbitrary input frequency, amplitude, and relative phase.

Answer given

The mesh-current equations of the Laplace-transformed circuit is:

where is the initial current in inductor, and is the initial voltage of capacitor.

Define     
We can write the Laplace-transformed mesh current simply:
                                       
we can find although it is rather messy:

In[21]:=

Out[23]//TraditionalForm=

It is much easier to look at with substituted numerical values for the components

In[2]:=

Out[4]//TraditionalForm=

Out[6]//TraditionalForm=

Out[8]//TraditionalForm=

Out[11]//TraditionalForm=

Out[12]=

Out[13]=

The roots indicate the transient responses, which can be overdamped, underdamped, or critically damped.

If we are interested only steady state solutions, we can ignore all the transient terms. Start with:

In[264]:=

Out[267]//TraditionalForm=

Then, because we care only for steady-state solutions, we can also ignore and make all the initial conditions such as current in inductors and voltage in capacitors to be zero. Then:
          
The mes current vector is obtained by Inverse-Laplace transforming

Demo of numerical calculation

End exercise

Can we take the inverse of this?

In[46]:=

Out[47]//MatrixForm=

In[48]:=

Out[48]=

Exercise 3.9

Generate an non-square {m,n} matrix, denote it A. Take the inverse of A and see what happens.
Then define a new matrice:                     and         
Which one, P or Q has inverse, which one doesn’t, and can you explain or guess why one does and the other doesn’t?

Answer

End exercise

By its definition, only square matrix can be inversed.

Note, by theorem, if  , where I is the identity matrix, then:
                  ,              ,     and    
These theorems simply follow the associative and transpose properties. For example:
Given A, let and be such that:  A . = I  and   = I
Then,             .  (A . ) = . I  =
By associative:       . A) . =  
                                    I .   =    or   =  

One important quantity of a square matrix is determinant. In Mathematica, Det[] gives the determinant of a matrix.

Demo: statistical property (Gaussian -> Laplacian distribution) of det of random matrix

In[85]:=

Back to:  page 1       page 2