Advertisement
English transcript
Hello everybody; we will continue with our lectures on optimal guidance, I mean control
guidance and estimation course, we are in lecture number 2. The very first class, we
saw some, some, overview and then some motivations of this course and all, all. This particular
lecture will have some overview of state space approach and little bit over matrix theory.
So, these are the two things that we, we, used it very expensively here in this course.
So, we thought we will just have an overview of that. Assume that you must have already
taken some course on, on, this subject or this is just to kind of a refresh your knowledge
on that.
So, we start with state space representative, very basic ideas and all that. So, when you
talk about any system, I mean dynamical systems, we typically talk about three types of variables;
one is input variable and the other one is output variable, the third one is state variable,
and I kind of discuss that in the first lectures was also, and typically the input variable
can be classified into two classes; one is manipulative, which is control variable or
either it or it can be non manipulative which is noise, and remember, both are input to
the system any way. And then, considering output variable, that can be defined as either
the variables of interest or that can be like some measured variable or calculated variable.
So, these two may or may not be same. This is what the diagram it tells. See in other
words, your performance output can be y and that you want to drive it to some desired
value Y star about the, where as the measurement output can be z, and that may, that may go
into the controller and then controller is can take place actually. So, it may be so
happening that part of the y variables or part or full of the, I mean all of the y variable
can be part of z and vice versa probably, but they need not be same actually.
So, all these things you can be talked about in a standard state space class and all that,
but anyways coming back, there is also some class, I mean this variables of great importance
is state variable, and that is defined as minimum set of parameters which completely
summarize this system status, and more definitions, more implications and think like that can
be found in a standard text book from on systems theory, and any modern control book will kind
of give you more insides with the example and all actually. But we understand what it
is. It is a, what it what we mean by state variable, and that means, as a minimum number
of variables, so which can describe the system state, and obviously, remember that you cannot
take more than the state variable that is required, or you cannot take less than the
state variable that is required either actually.
So, that is, those things are kind of non. Now continuing further, we have a broadly
two classes of system that we discuss about here, and one is non-linear systems and other
one is linear system actually. When we talk about non-linear system, we represent the
state equations this way, X dot equal f of XU. Typically time variable t is not their,
where X dot represents derivative with respect to time, and remember, state variable dimensional
state is R N; that means N number of states are there. Control can be RM; control belongs
to RM; that means U can be of m dimensional and the output that you are, that of interest
can be of P dimensional actually. Now, also you, this R, letter R represent serial numbers,
so we are typically deal with all real number here. And but in non-linear systems as we
know, are typically it is little bit difficult to handle. So, typical to analyze synthesize
and all that. So, most of the time, you way linearize the
system about some operating point and all that, and then land up with some representations
like that, where X dot equals a X plus BU, Y equals to CX plus DU, but remember, this
X what you see in non-linear system and this X what you see in linear system. The meaning
wise they are different. When you talk about state of non-linear system, actually the two
states, whereas the state of linearize system. Most of the linear systems are anyway linearize.
Linearize state and controls are typically deviation variables actually; that means X
represent delta X and U represents delta U implicitly here. And delta X can be the deviation
of the state, from its separating point and delta U can be the deviation of control from
its operating control and all that actually, so those kind of definitions are available.
So, in this particular course, we will deal both with linear system as well as non-linear
system, I mean we cannot, we are not talking about only one class of thing; deal with both
of the things together actually.
And coming to the state variable selection, typically the number of state variable, that
is the order of the system is equal to the number of independent energy storage elements,
but there are exceptions as well. You can see couple of examples, in probably in some
standard text books like and all that will give you that actually. But is there any restriction
on the selection of the state variables? I mean the answer is yes, and the very primary
requirement is, all the state variables are to be linear independent, and then all of
them must collectively describe the system completely, that is, these are the two things
that state variable should have actually. Now, coming to the linearization of non-linear
systems, see you remember that, that is what I told you here; the linear systems are typically
linearization of the non-linear systems about some operating point actually. So, how is
it done typically, I mean how is it done? Let us see that.
The problem statement here is given a non-linear system like this; X dot equal f of XU. Can
you derive an approximate linear system about in operating point X0 U0, the approximate
linear system can be described something like this actually. And also remember, when you
talk about operating point, typically the miss number is, it is an equilibrium point,
but it need not be true actually. So, operating point, by definition is a point through which
the system trajectory passes; that means even it is passing through a tangent trajectory
and all that actually, that can be a potential operating point, and so that is how sometimes
we will linearize the system about a trajectory; continuously we keep on getting a new, new,
new, linear, I mean linearize systems; that means the A and B matrix become time bearing
and all actually, if you do that. But if it is a equilibrium point, that is
your operating point and then you linearize the system, then you land up with and be constant
matrix, and this constant matrices that is why you will get into this also called linear
time in variant system and all that actually. But remember that the operating point is a
point, through which the system trajectory passes, that is the definition actually. And
obviously, if equilibrium point is also a point through which systems trajectory passes,
because if you just table system, ultimately the trajectory will come and sit there, so
that is also a possibility of an operating point actually. So anyway, these are the ideas
there. The operating point we noticed where, even though we represent the same state, notations
state in control notation X and U. When we talk about linear systems, by definition this
state is actually deviation state, and this state is actually deviation control.
So, proceeding further, how do you get it, that is through the Taylor series expansion
and all that. So here, we have a system, non-linear system, where we have a reference point here
something like X not U not. So, if you put the X equal to X not plus delta X and U equal
to U not plus delta U, and then, that then this side of the story f of X X U can be represented
like this, and then you use the Taylor series expansion. When you can expand it something
like a first term, a constant term, followed by two linear terms in, in, state in control,
then this added terms. So if you, and also remember, the operating point is a point through
which the system trajectory passes; that means this X not U not satisfy the differential
equation; that means, if I substitute X 0 dot, X not dot, then that is nothing but f
of X not U not. So, this is what I exploited here. Remember, this side of the story is
like this. So, about that side of the story will be something like that, X dot equal to
X 0 dot plus delta X dot.
So, that is nothing but equal to this right hand side, neglecting the higher order terms,
that is approximately equal, and then we take, this two will cancel out, because this is
what the operating point is all about. This is, this two will go, and then you land up
with this X not equal to. Then you land up with this X not equal to, I mean delta X dot
equal to a times delta X plus B times delta U. But, instead of writing down all times
delta X and delta U, you think that; we implicitly redefine it as delta X equal to X, and delta
U equal to U is U. Then this leads to X dot equals to AX plus BX. That is how this AX
and BU represent this delta X and delta U, but also I mean if X not U not happens to
be 0 0, then obviously this delta X equal to X and then delta U is also equal to U basically
so; they are corollaries and all that, and remember, A and B will be defined in terms
of this Jacobean matrices that way. So, we have this state equation with the system model
is available to you. So, expand this partial derivatives that way,
and then put it them in a matrix form and evaluated the point of X not and U not itself.
That is what you will do for A and B matrix. A happens to be NY N square matrix and B will
happen to be NYM non square matrix in general basically. Now, if you land up with some system
equation like that and try to carry out lot of algebra and then optimal control synthesis
and think like that. Obviously, we have to deal with this matrices and we. So, we have
lot of requirement for matrix and algebra on the way actually, and also we will deal
with, even if you deal with non-linear system, you have to deal with gradient matrices and
all that, gradient vectors something like that. So, we certainly need the concept from
matrix theory actually. Let us have a quick overview of matrix theory in general, and
then we can from next class onwards, we will go to optimization theory or little bit overview
of numerical methods as well, before you go to optimization theory and all that actually.
Anyway, so coming, coming to the review of matrix theory, so, let us we start with simple
definitions. We all understand what is meant by matrices. So, I am not going to that level.
It is a collection of numbers in an array and thing like that, but coming to some definitions,
standard definition, symmetric matrix is represented as something like this. When A is equal to
A transpose, then obviously A satisfies the property of symmetric. Now, when determinant
of A becomes 0, and here we assume that A is a square matrix. Then we talk about well
the matrix is singular matrix actually; that means inverse does not exit actually, then
what is inverse. Inverse of a matrix A and B, I mean if B is an inverse of A, then this
satisfies the property; A B is equal to B I is nothing but identity basically. So, that
is how this forms the definition, and then, as a standard result a inverse turns out to
be advent of a divided by determinant A. And again I emphasize here that do not get confused
with what is definition and what is standard result actually.
As far as definition is concerned, I will put this is a definition; AB equal to BA equal
to identity, and lot of this theorems, proof and then general results and all will be very
easy to deal do, using this definition AB equal to BA equal to identity. However if
you really want to compute an inverse for a matrix and all, probably that is a, that
is a standard results that we typically use also. Even though it is not very computationally
efficient way of computing A inverse by the way. So, that is the difference, so it happens
to be like one of the same. In other words, both are if and only if condition sort of
things. So, sometimes people get confuse that A inverse definition is this A adjacent of
A divided by determinant of A, but in definition sense, this is the definition and this happens
to be a standard result actually. Now, the next concept is, what is called as orthogonal
matrix. An orthogonal matrix happens to be something like, orthogonal matrix happens
to be something like this way. So, if A transpose equal to A transpose A equal to identity,
then it is an orthogonal matrix. And also remember that A times a transpose
and A transpose A are guaranteed to be kind of, I mean square matrix, even though A is
a non square, and top of that, you can also show that A transpose and A transpose both
happens to be something like positive semi matrix, something like that, is not that typical
solve actually. Anyway, in addition to that, if they satisfy this property that A transpose
equal to A transpose A equal to identity, then the matrix A is called as an orthogonal
matrix. And one property of orthogonal matrix is like this; if a matrix A is orthogonal,
then all its columns are suppose to be orthonormal vectors actually to each other. So, that is
a vector sense basically; that means, if you have a A matrix of sum matrix which is given
by some columns and all that way, and then each of the columns that you see here will
be orthogonal, rather normal to each other actually in a vector sense. So, that then
typical example is something like this, this is a very standard what is called as rotational
matrix; that means, if you have a point, then representational turns out to be like this.
If you have something like, I mean this coordinate frame and think like that, you have this X
verses Y coordinated frame, and you have a point where XY is there, and then the question
here is, if the coordinate frame rotates Y angle theta let us say; that means, you have
Y prime and X prime Y prime, and this is again angle theta. Then in the in the new coordinate
frame X prime Y prime, what is the coordinates, I mean the values of this coordinate X prime
Y prime. So then, that is kind of given by this transformation matrix and all actually;
that is why it is called a rotation matrix. If you take a coordinate in the original coordinate
prime and multiply with this, pre-multiply with this matrix and whatever number you get,
that number happens to be your coordinate in the new coordinate frame actually. And
this result I have already told you, that columns of an orthogonal matrix are nothing,
but they are orthogonal to each other in a vector sense actually.
The next concept is Eigenvalues and Eigenvectors and this is how you can think of interpreting
it actually. Now, this is a common mistake that we people take, if we are a eigenvector
then. Well there is a standard equation for that and then I will solve it, and especially
about Eigenvalues, A minus lambda is determinant equal to zero and then I will get values for
lambda’s and that is nothing but Eignevalues and all. So, that is the standard result again,
but the concept can be understood something like this. Let us start with simple understanding
here. So, let us take a point p in a coordinate frame x and y. So, and talk about this let
us say this is something like x1 and x2. So, the coordinate frame happens to be something
like, well whatever this is the, I mean horizontal and that has the vertical coordinate frame
you can think of in general. So, we have a point p and that is x1 and x2 originally,
and we multiply that by A matrix, some matrix A.
Then we will land up with some numbers y 1 and y 2, and that y 1 and y 2 can represent
a point in the coordinate frame something like U. That means, if I join this origin
to this point p, and origin to this point q, then I am actually getting two vectors.
Now the, if you look at it slightly closely, then it turns out that, by simplify, multiplying
this vector x 1 and x 2 by matrix A. What I have done. I have transformed this point
p to q in general, in a way, what I am doing is, I am actually taking this vector, and
then rotating it as well as stretching it. So, there is a stretching and rotation in
operation involve basically. So, any matrix multiplication is, matrix are also called
linear operators by the way. So, any matrix multiplication with a vector will give you
another vector, and that vector need not contain the same magnitude and same direction actually.
So, it can act as something like a stretching operator as well as the rotation operators.
Now, the question here is something like this. It may not happen in general, but suppose,
I have a vector x some direction somewhere, and I multiply that with matrix A, I get another
y which is a line along the x axis, I mean this vector, origin to x vector and it is
align their; that means this matrix is operating only as a stretching vector not as a operating,
not as a rotation operators, but as simply as a stretching operator. So, that means,
it does not affect any other direction. It just takes the vector x and then just stretches
it along further actually, and stretching by definition can means shrinking also, and
so, that is also possible actually. So, that is the question we are asking actually.
So, in other words, does their exits any vector x, some vector x, for a particular matrix
A, so that this operation a times x will give us vector y which is nothing but a stretched
value of vector x basically, that is all we are talking. So, in other words, if you think
about that, that is how this all starts actually. So, you are asking something some equation
like, what are asking. If y is nothing but I can interpret that y equal to nothing but
lambda x sort of thing, but y is also x remember that. So, what I am getting here is, x equal
to lambda x. That is how the eigen of, I mean Eigen of, Eigen, Eigen, eigenvalue and eigenvector
equation is define actually. So, y is nothing but lambda x; that means that is a stretching
operator, and y by definition is a times x any way. So, that is how we get A into lambda
x actually.
So, that is how you get it, but then the solution, the question is when does, when can we have
a solution to these equation, and that solution to this equation obviously turns out that,
if that is true, and that has to be true as well; lambda I minus A equal to, I mean lambda
I minus A by whole multiply with x equal to zero, and if you really need to have a nontrivial
solution, that means remember this equation is invariably true, if you have 0 0 0 0 as
X vector; that means it is a very trivial thing, we are not moving anything away from
the region actually. So, other than that, if you take any other thing, then these two
have a meaningful solution and nontrivial solution. We should have this relationship
satisfied. That is a standard result from linear system equations actually. That is
how this equation comes here actually. So, this is, if you take this equation, remember
determinant is a scalar operator. So, you get one equation, but then, this can
be the Nth order algebraic equation; that means, you can have N solutions and all that
actually. So, that is how you get, if the matrix is of dimension NYN, then you land
up with something like n equations, I means it is Nth order equation, and hence you will
have Eigenvalues and because this coefficient will be really numbers. You will land up with
either numbers or solutions or complex conservative solution. You cannot have a single complex
number running out actually. So, that is again result actually. Where do we use these concepts,
by the way, how do you evaluate Eigenvectors and all. Once you want to evaluate the Eigenvalues
like this, you go back and put it there. Remember, it does not give you some sort of like a unique
solution. It may give you a something like in other constant equation,
…
guidance and estimation course, we are in lecture number 2. The very first class, we
saw some, some, overview and then some motivations of this course and all, all. This particular
lecture will have some overview of state space approach and little bit over matrix theory.
So, these are the two things that we, we, used it very expensively here in this course.
So, we thought we will just have an overview of that. Assume that you must have already
taken some course on, on, this subject or this is just to kind of a refresh your knowledge
on that.
So, we start with state space representative, very basic ideas and all that. So, when you
talk about any system, I mean dynamical systems, we typically talk about three types of variables;
one is input variable and the other one is output variable, the third one is state variable,
and I kind of discuss that in the first lectures was also, and typically the input variable
can be classified into two classes; one is manipulative, which is control variable or
either it or it can be non manipulative which is noise, and remember, both are input to
the system any way. And then, considering output variable, that can be defined as either
the variables of interest or that can be like some measured variable or calculated variable.
So, these two may or may not be same. This is what the diagram it tells. See in other
words, your performance output can be y and that you want to drive it to some desired
value Y star about the, where as the measurement output can be z, and that may, that may go
into the controller and then controller is can take place actually. So, it may be so
happening that part of the y variables or part or full of the, I mean all of the y variable
can be part of z and vice versa probably, but they need not be same actually.
So, all these things you can be talked about in a standard state space class and all that,
but anyways coming back, there is also some class, I mean this variables of great importance
is state variable, and that is defined as minimum set of parameters which completely
summarize this system status, and more definitions, more implications and think like that can
be found in a standard text book from on systems theory, and any modern control book will kind
of give you more insides with the example and all actually. But we understand what it
is. It is a, what it what we mean by state variable, and that means, as a minimum number
of variables, so which can describe the system state, and obviously, remember that you cannot
take more than the state variable that is required, or you cannot take less than the
state variable that is required either actually.
So, that is, those things are kind of non. Now continuing further, we have a broadly
two classes of system that we discuss about here, and one is non-linear systems and other
one is linear system actually. When we talk about non-linear system, we represent the
state equations this way, X dot equal f of XU. Typically time variable t is not their,
where X dot represents derivative with respect to time, and remember, state variable dimensional
state is R N; that means N number of states are there. Control can be RM; control belongs
to RM; that means U can be of m dimensional and the output that you are, that of interest
can be of P dimensional actually. Now, also you, this R, letter R represent serial numbers,
so we are typically deal with all real number here. And but in non-linear systems as we
know, are typically it is little bit difficult to handle. So, typical to analyze synthesize
and all that. So, most of the time, you way linearize the
system about some operating point and all that, and then land up with some representations
like that, where X dot equals a X plus BU, Y equals to CX plus DU, but remember, this
X what you see in non-linear system and this X what you see in linear system. The meaning
wise they are different. When you talk about state of non-linear system, actually the two
states, whereas the state of linearize system. Most of the linear systems are anyway linearize.
Linearize state and controls are typically deviation variables actually; that means X
represent delta X and U represents delta U implicitly here. And delta X can be the deviation
of the state, from its separating point and delta U can be the deviation of control from
its operating control and all that actually, so those kind of definitions are available.
So, in this particular course, we will deal both with linear system as well as non-linear
system, I mean we cannot, we are not talking about only one class of thing; deal with both
of the things together actually.
And coming to the state variable selection, typically the number of state variable, that
is the order of the system is equal to the number of independent energy storage elements,
but there are exceptions as well. You can see couple of examples, in probably in some
standard text books like and all that will give you that actually. But is there any restriction
on the selection of the state variables? I mean the answer is yes, and the very primary
requirement is, all the state variables are to be linear independent, and then all of
them must collectively describe the system completely, that is, these are the two things
that state variable should have actually. Now, coming to the linearization of non-linear
systems, see you remember that, that is what I told you here; the linear systems are typically
linearization of the non-linear systems about some operating point actually. So, how is
it done typically, I mean how is it done? Let us see that.
The problem statement here is given a non-linear system like this; X dot equal f of XU. Can
you derive an approximate linear system about in operating point X0 U0, the approximate
linear system can be described something like this actually. And also remember, when you
talk about operating point, typically the miss number is, it is an equilibrium point,
but it need not be true actually. So, operating point, by definition is a point through which
the system trajectory passes; that means even it is passing through a tangent trajectory
and all that actually, that can be a potential operating point, and so that is how sometimes
we will linearize the system about a trajectory; continuously we keep on getting a new, new,
new, linear, I mean linearize systems; that means the A and B matrix become time bearing
and all actually, if you do that. But if it is a equilibrium point, that is
your operating point and then you linearize the system, then you land up with and be constant
matrix, and this constant matrices that is why you will get into this also called linear
time in variant system and all that actually. But remember that the operating point is a
point, through which the system trajectory passes, that is the definition actually. And
obviously, if equilibrium point is also a point through which systems trajectory passes,
because if you just table system, ultimately the trajectory will come and sit there, so
that is also a possibility of an operating point actually. So anyway, these are the ideas
there. The operating point we noticed where, even though we represent the same state, notations
state in control notation X and U. When we talk about linear systems, by definition this
state is actually deviation state, and this state is actually deviation control.
So, proceeding further, how do you get it, that is through the Taylor series expansion
and all that. So here, we have a system, non-linear system, where we have a reference point here
something like X not U not. So, if you put the X equal to X not plus delta X and U equal
to U not plus delta U, and then, that then this side of the story f of X X U can be represented
like this, and then you use the Taylor series expansion. When you can expand it something
like a first term, a constant term, followed by two linear terms in, in, state in control,
then this added terms. So if you, and also remember, the operating point is a point through
which the system trajectory passes; that means this X not U not satisfy the differential
equation; that means, if I substitute X 0 dot, X not dot, then that is nothing but f
of X not U not. So, this is what I exploited here. Remember, this side of the story is
like this. So, about that side of the story will be something like that, X dot equal to
X 0 dot plus delta X dot.
So, that is nothing but equal to this right hand side, neglecting the higher order terms,
that is approximately equal, and then we take, this two will cancel out, because this is
what the operating point is all about. This is, this two will go, and then you land up
with this X not equal to. Then you land up with this X not equal to, I mean delta X dot
equal to a times delta X plus B times delta U. But, instead of writing down all times
delta X and delta U, you think that; we implicitly redefine it as delta X equal to X, and delta
U equal to U is U. Then this leads to X dot equals to AX plus BX. That is how this AX
and BU represent this delta X and delta U, but also I mean if X not U not happens to
be 0 0, then obviously this delta X equal to X and then delta U is also equal to U basically
so; they are corollaries and all that, and remember, A and B will be defined in terms
of this Jacobean matrices that way. So, we have this state equation with the system model
is available to you. So, expand this partial derivatives that way,
and then put it them in a matrix form and evaluated the point of X not and U not itself.
That is what you will do for A and B matrix. A happens to be NY N square matrix and B will
happen to be NYM non square matrix in general basically. Now, if you land up with some system
equation like that and try to carry out lot of algebra and then optimal control synthesis
and think like that. Obviously, we have to deal with this matrices and we. So, we have
lot of requirement for matrix and algebra on the way actually, and also we will deal
with, even if you deal with non-linear system, you have to deal with gradient matrices and
all that, gradient vectors something like that. So, we certainly need the concept from
matrix theory actually. Let us have a quick overview of matrix theory in general, and
then we can from next class onwards, we will go to optimization theory or little bit overview
of numerical methods as well, before you go to optimization theory and all that actually.
Anyway, so coming, coming to the review of matrix theory, so, let us we start with simple
definitions. We all understand what is meant by matrices. So, I am not going to that level.
It is a collection of numbers in an array and thing like that, but coming to some definitions,
standard definition, symmetric matrix is represented as something like this. When A is equal to
A transpose, then obviously A satisfies the property of symmetric. Now, when determinant
of A becomes 0, and here we assume that A is a square matrix. Then we talk about well
the matrix is singular matrix actually; that means inverse does not exit actually, then
what is inverse. Inverse of a matrix A and B, I mean if B is an inverse of A, then this
satisfies the property; A B is equal to B I is nothing but identity basically. So, that
is how this forms the definition, and then, as a standard result a inverse turns out to
be advent of a divided by determinant A. And again I emphasize here that do not get confused
with what is definition and what is standard result actually.
As far as definition is concerned, I will put this is a definition; AB equal to BA equal
to identity, and lot of this theorems, proof and then general results and all will be very
easy to deal do, using this definition AB equal to BA equal to identity. However if
you really want to compute an inverse for a matrix and all, probably that is a, that
is a standard results that we typically use also. Even though it is not very computationally
efficient way of computing A inverse by the way. So, that is the difference, so it happens
to be like one of the same. In other words, both are if and only if condition sort of
things. So, sometimes people get confuse that A inverse definition is this A adjacent of
A divided by determinant of A, but in definition sense, this is the definition and this happens
to be a standard result actually. Now, the next concept is, what is called as orthogonal
matrix. An orthogonal matrix happens to be something like, orthogonal matrix happens
to be something like this way. So, if A transpose equal to A transpose A equal to identity,
then it is an orthogonal matrix. And also remember that A times a transpose
and A transpose A are guaranteed to be kind of, I mean square matrix, even though A is
a non square, and top of that, you can also show that A transpose and A transpose both
happens to be something like positive semi matrix, something like that, is not that typical
solve actually. Anyway, in addition to that, if they satisfy this property that A transpose
equal to A transpose A equal to identity, then the matrix A is called as an orthogonal
matrix. And one property of orthogonal matrix is like this; if a matrix A is orthogonal,
then all its columns are suppose to be orthonormal vectors actually to each other. So, that is
a vector sense basically; that means, if you have a A matrix of sum matrix which is given
by some columns and all that way, and then each of the columns that you see here will
be orthogonal, rather normal to each other actually in a vector sense. So, that then
typical example is something like this, this is a very standard what is called as rotational
matrix; that means, if you have a point, then representational turns out to be like this.
If you have something like, I mean this coordinate frame and think like that, you have this X
verses Y coordinated frame, and you have a point where XY is there, and then the question
here is, if the coordinate frame rotates Y angle theta let us say; that means, you have
Y prime and X prime Y prime, and this is again angle theta. Then in the in the new coordinate
frame X prime Y prime, what is the coordinates, I mean the values of this coordinate X prime
Y prime. So then, that is kind of given by this transformation matrix and all actually;
that is why it is called a rotation matrix. If you take a coordinate in the original coordinate
prime and multiply with this, pre-multiply with this matrix and whatever number you get,
that number happens to be your coordinate in the new coordinate frame actually. And
this result I have already told you, that columns of an orthogonal matrix are nothing,
but they are orthogonal to each other in a vector sense actually.
The next concept is Eigenvalues and Eigenvectors and this is how you can think of interpreting
it actually. Now, this is a common mistake that we people take, if we are a eigenvector
then. Well there is a standard equation for that and then I will solve it, and especially
about Eigenvalues, A minus lambda is determinant equal to zero and then I will get values for
lambda’s and that is nothing but Eignevalues and all. So, that is the standard result again,
but the concept can be understood something like this. Let us start with simple understanding
here. So, let us take a point p in a coordinate frame x and y. So, and talk about this let
us say this is something like x1 and x2. So, the coordinate frame happens to be something
like, well whatever this is the, I mean horizontal and that has the vertical coordinate frame
you can think of in general. So, we have a point p and that is x1 and x2 originally,
and we multiply that by A matrix, some matrix A.
Then we will land up with some numbers y 1 and y 2, and that y 1 and y 2 can represent
a point in the coordinate frame something like U. That means, if I join this origin
to this point p, and origin to this point q, then I am actually getting two vectors.
Now the, if you look at it slightly closely, then it turns out that, by simplify, multiplying
this vector x 1 and x 2 by matrix A. What I have done. I have transformed this point
p to q in general, in a way, what I am doing is, I am actually taking this vector, and
then rotating it as well as stretching it. So, there is a stretching and rotation in
operation involve basically. So, any matrix multiplication is, matrix are also called
linear operators by the way. So, any matrix multiplication with a vector will give you
another vector, and that vector need not contain the same magnitude and same direction actually.
So, it can act as something like a stretching operator as well as the rotation operators.
Now, the question here is something like this. It may not happen in general, but suppose,
I have a vector x some direction somewhere, and I multiply that with matrix A, I get another
y which is a line along the x axis, I mean this vector, origin to x vector and it is
align their; that means this matrix is operating only as a stretching vector not as a operating,
not as a rotation operators, but as simply as a stretching operator. So, that means,
it does not affect any other direction. It just takes the vector x and then just stretches
it along further actually, and stretching by definition can means shrinking also, and
so, that is also possible actually. So, that is the question we are asking actually.
So, in other words, does their exits any vector x, some vector x, for a particular matrix
A, so that this operation a times x will give us vector y which is nothing but a stretched
value of vector x basically, that is all we are talking. So, in other words, if you think
about that, that is how this all starts actually. So, you are asking something some equation
like, what are asking. If y is nothing but I can interpret that y equal to nothing but
lambda x sort of thing, but y is also x remember that. So, what I am getting here is, x equal
to lambda x. That is how the eigen of, I mean Eigen of, Eigen, Eigen, eigenvalue and eigenvector
equation is define actually. So, y is nothing but lambda x; that means that is a stretching
operator, and y by definition is a times x any way. So, that is how we get A into lambda
x actually.
So, that is how you get it, but then the solution, the question is when does, when can we have
a solution to these equation, and that solution to this equation obviously turns out that,
if that is true, and that has to be true as well; lambda I minus A equal to, I mean lambda
I minus A by whole multiply with x equal to zero, and if you really need to have a nontrivial
solution, that means remember this equation is invariably true, if you have 0 0 0 0 as
X vector; that means it is a very trivial thing, we are not moving anything away from
the region actually. So, other than that, if you take any other thing, then these two
have a meaningful solution and nontrivial solution. We should have this relationship
satisfied. That is a standard result from linear system equations actually. That is
how this equation comes here actually. So, this is, if you take this equation, remember
determinant is a scalar operator. So, you get one equation, but then, this can
be the Nth order algebraic equation; that means, you can have N solutions and all that
actually. So, that is how you get, if the matrix is of dimension NYN, then you land
up with something like n equations, I means it is Nth order equation, and hence you will
have Eigenvalues and because this coefficient will be really numbers. You will land up with
either numbers or solutions or complex conservative solution. You cannot have a single complex
number running out actually. So, that is again result actually. Where do we use these concepts,
by the way, how do you evaluate Eigenvectors and all. Once you want to evaluate the Eigenvalues
like this, you go back and put it there. Remember, it does not give you some sort of like a unique
solution. It may give you a something like in other constant equation,
…