Advertisement
English transcript
Hello every one, we will continue our lecture series on this optimal control Guidance and
estimation course.
We have finished as six lectures already and we all to do already, get some over view of
calculus of variations is scalar. I mean, scalar variables what I am going to do here,
in this particular lecture is to review that briefly, what we discuss in the last class
followed by its extension to vector. I mean vector consumption all that multi dimensional
things. This is very similar to scalars anyway.
So, first thing is summary of what we discuss in the last lecture, the calculus of variations
for scalar problems especially.
So, this is the concept that we discussed, we discussed something like on one side is
a function and its increment and the other side is a functional and its increment. And
both the appear to be quite a similar provided the x x is turns out to be like, in this particular
case it is free variable time whereas, in this it is the x of t basically. So, it you
can interpret is something like a function of functions sort of things and that is, that
how? Where this? The this delta x can be a variation with respect to time. Then if it
varies with respect to time then, how this J varies? That is ultimately the implication
basically. Then we discussed many things about an increment of a function is this delta f
approximated is d f similarly, increment is delta J approximated is del J.
Slight notification change this is d f was it is delta J sort of thing. So, just to may
say that this is a variations.
Then we discuss about differential of a function and variation of a functional, think like
that and we carried out this algebra. And using this taylor series, we learned it up
there kind of expression for delta f and then it turns out that, when delta t goes to zero
and we learned of something like d f. So, in general you can tell d f is some thing
like f dot delta t in general basically. So, very similar concept you can do this way also
here and you can take first variation, second variation thing like that, the first variation
is can be computed this way. And as I told last class somebody does not tell where the
default variation means, first variations can be computed this way.
We also gave examples, how to compute? And think like that in the last class actually.
So, then coming to the boundary conditions we discussed two class of boundary conditions,
and the fixed end point problems, and free end point problems. Fixed end point problem
can be kind of fixed both in time and state our time in the free variable; sorry this
time is the free variables and x is the dependent variable. It can be fixed at both sense, t
zero at t zero x can be fixed at t of x can, x can also be fixed or it can also be something
like free end point problem, where the end point can be completely free or it can it
can be required to lie on a specific curve and I gave you an example of satellite launch
and all that on there. When you launch a satellite does not matter
where you join there we from there onwards it will keep on staying in the orbit anyway.
So, all that it matters is us to leave it somewhere in the orbit or orbitral condition
actually. So, that is that is how? This problems are we find at way.
So, then optimum of a functional we also discussed and we had this formal definition that functional
is said to have a relative optimum at x star of t remember is not a particular point, but
x star itself is a function of time. So, it is functional is said to relative optimum
at x star of t. If the; if there exists some epsilon is greater than zero such that the
function I mean this x of t will satisfy this kind of a condition. Whenever it satisfies
this then you have this increment of J should have the same sign.
That means if we are talking about the minimum then this delta J, J of x minus J of x star
should always be positive, no matter what kind of variations we talk about? If whatever
variation you have either up or down, whatever it is actually, irrespect to whatever variation
you have delta J will keep on the thing positive, that mean J of x star is suppose to have a
local minimum. Once similarly, if delta J is less than zero then J of x star is suppose,
to have a local maximum actually. And obviously, if the relationships are satisfied arbitrary
large value for I mean is satisfied for arbitrary large value of epsilon, then J of x star is
said to have a global optimum actually.
And we also discussed about this variation concept on only x star represents to be an
optimum path, then all that we are talking about is neighborhood path of that actually.
So, whatever neighbors we are getting that kind of that kind of a thing we are talking
about actually.
So, for then we discussed about some fundamental theorem of calculus of variation and it turns
out that, for x star of t to be a candidate optimum the following condition has to hold
good. That means, the first variation has to be zero for all admissible delta x. This
delta x turns out to be the difference between these two actually. So, for all admissible
delta x, your delta J has to be equal to zero, that happens to be a necessary condition.
And a sufficiency condition happens to be second variation should be greater than zero
for a minimum or second variation should be less than zero for a maximum. And when it
I mean this concept of kind of you can think about peri parallel to the static optimization,
what had the ideas of calculus of variation of other.
We also discuss this nice beautiful little fundamental lemma, then it will stole that
if for every continuous function g of t this equation satisfied, this integral equation
is satisfied. Where the variation delta x is continuous in t, then the I mean the only
condition that is required is delta x has to be continued that all. Then it turns out
that g of t has to be 0 and the entire in to. So, that is that is nice property and
then many many times we will use it also. And this was a very simple proof also we took
at a interval and then in that interval, we took the function everywhere 0, but in that
it is not 0. And we can always construct a delta x, which
is again non 0 in that particular interval and this integral happens to be non 0. If
you do that then this count I mean may thus the kind of I mean counter arguments sort
of things. So, were; So, taking help of that we could. So, that it is it cannot happen.
So, method proof by contradicts and all that. So, then we landed up with the conclusion
that, g of t has to be 0 it is looks as a very simple theorem, but there is lot of great
integrations later.
Whenever conditions of optimality then we defined the problem that we have to optimize
this kind of a function, where t zero and t f both are fixed and all that you to do
is to make sure that the first variation is zero. And after some algebra the, that I mean
using this partial fraction of, I mean partial fraction of this integral and all that. We
could able to show that, it get ultimately leads to this two conditions and the first
condition is called E L equations or Euler Lagrange Equation and the second one is Transversality
Condition, which leads to this boundary condition sort of things.
And we also noted that part of this equation might be already satisfied by problem formulation,
you have the word if here x zero is fixed it is not a free, then d d x not is already
zero. So, this condition you already satisfied actually, whatever is not satisfied you have
to, you can extract from there. Whatever you satisfied you already satisfied any way. So,
even though these equation is valid for the entire path from t not to t f, but this conditions
are valid for t zero and t f only basically. There is a differential equation from this
will give a corresponding differential equation as we give a corresponding boundary condition
sort of things. So, using this two we suppose to solve the optimal I mean dynamic optimization
problems.
So, now this quickly this prove that that followed you something like this it has the
delta L we took at and delta L is by definition like this. And we expanded using taylor's
series the this first term, first term in derivative and then either terms. And then
if you take in the limit delta L will be become delta L, this delta L and then that turns
out to be like that. So, this delta J can be approached matter something like this and
then this particular path I mean this second part of the integral.
You can kind of integrate by parts. So, then do this I mean algebra and then kind of with
this kind of conclusion. This part being like that and then you can put it by here and then
ultimately it will list to that. So, if this has to be zero, then this has to be zero and
that has to be zero and that is how? We got the; this two condition of optimality that
I was talk actually. So, the same trick will all good for vector algebra also, that is
why I thought of kind of reversing if again.
So, this ultimately leads to this two condition and condition one must be satisfied regardless
of the boundary condition or end condition. But second part of the equation; that means,
this equation we already, we satisfied for the problem specification. So, the amount
of extra information contained by this equation, that is with the boundary condition specified
actually so; that means, depending on the specified boundary conditions here to extract
additional information from this condition.
In general we, this was in restrictive sense we try to expand that and then in general
we lended up with the, this kind of thing I mean this general transversality condition
sort of things. When t this variation of t was also allowed and other variation of initial
time and final time that was allowed actually, where you start? Where you stop? Is up to
that designers also any way, sometimes the problem when you start? And when you stop?
That can be 3 also in a way. If you bring that into a account then this
leads to something like this and then we discussed about various special cases, case one, case
two I think up to case five and are somewhere different different conditions. That means,
if both the end point are fixed; that means, both time and x naught at t zero and t f are
all fixed and they does not; obviously, give any additional information, because every
delta x zero delta x of delta t naught delta t f everything was to zero.
But if t naught and t f are fixed, but x naught and x f are naught fixed then obliviously
this part is also zero, delta t f and delta t naught both are zero will end up with only
this part of the equation. Similarly, if t naught and x naught are fixed; that means,
t f and x are both are free, then you can consider only the t f part of the equations
and then you lend up to something like this. And similarly, if you depending on, what is
fixed? And What is free? We kind of take the other part into account because whatever is
given, if it satisfies some part of the equation automatically, we considered that is I mean
just taken sort of thing. So, this is how we extract this information
contain from transversality condition. Ultimately there the key idea is the number of differential
equations and number of un condition would be same, and then it depends on how many free
variables are there with us? And how many extra information is can we derive from this
equation actually? So, depending on in example and all will talk later, it will be more and
more clear how to be; how to use it actually?
So, the very special case at the end, when t naught and x naught is fixed and t f and
x f are constraint to lie on a given curve. Then you to go back to that, they because
it is fixed anyway so, this t naught x naught does not give any extra information. But t
f x f we cannot take that is zero, this part of it delta x f and delta t f cannot be zero,
but they are suppose to satisfies certain constraint equation, So, this constraints
equation can be derived something like this, because except x is a function of eta t. So,
from that delta x f is nothing but that so, you can substititute that and then land up
with this and where delta t f is not really zero, because it is this is constraints to
lie on a given curve; that means, your final time is kind of free actually.
That is not zero the coefficient as to be zero that of end up with this addition equation,
which will be which is needed for this kind of problem definition actually.
Then we also discuss little bit on second variation and the second variation we expanded
the second term of the taylor series and then let of I mean kind of observed that, this
lands up this gives us this hexane matrix sort of ideas.
And then to define this pi, pi let that matrix evaluated at the optimum point and if happens
to be a pdf matrix it is the minimum and if happens to be a negative different matrix,
it happens to be maximum. So, first what you have to do in again problem is to evaluate
this matrix phi first and then evaluate whether that the phi is a positive definite matrix
or not actually or negative definite matrix. If neither of the above then further math
is obviously, required and like static optimization, if you remember? We have this third term fourth
term anything like that and; obviously, those things are beyond this beyond this scope of
this course. And also remember unlike static optimization
in this particular thing x star and x star dot both are time dependent, they are not
really a specific point value. So, they are x star is a optimal solution which is a function
of time basically, thus the path trajectory that we are looking for. So, this what you
will evaluate phi is not really a constant matrix, but it is a function of time. So,
when you talk about whether this matrix is positive or negative is to remains positive
or negative for all time actually. So, for in the entire interval this matrix
should turns out to be a positive different matrix. That means, one easy way of doing
that is, just evaluate the eugene values kind of symbolically as the function of time and
then plot it in this interval actually if they all remain in positive then you have
done sort of thing. Also remember as a last comment that is just that we are source of
that is we are just talking about is valid only for free optimization problem. The moment
you bring in some sort of a state equation constraint or any other algebraic constraint
I think like tha,t then if the constraint is active then; obviously, this is the condition
does not actually. So, this is what we discussed in the last
class. How about extending some of these concepts for the vector problem? Because that is where
we are have a main interest to like whenever state variables are a typically of one dimensional
control is n dimensional thing like that actually.
So, what I mean the consider the same problem, but in multiple dimension and to begin with
we consider the problem without any constraint.
So, just an I mean what we see here is very similar concepts the only difference will
be this small x will be replaced by capital N know, where this capital X contains n dimension
I mean n components, so, x 1 x 2 up to x n actually. And obviously, again the similar
sort of ideas will take t zero and t f both are fixed and our objective is to make sure
that delta J goes to zero for arbitrary delta X t. So, again the similar algebra it will
do that in a second, but using a similar algebra it will lend up a very very similar local
equation. But remember these two equations now, are
not exactly sending what we have sent before. Especially this particular equation contains
n equations really, del L by del X one minus d by d t del L by del X one dot equal to zero,
then you can substitute by x 2 x t 3 like that. So, it really contains about n equations
actually. About an here, transpose is also must because this itself is a vector and that
is also a vector. So, what we are talking here, is very closely what we have done in
scalar problem, but not exactly one to one sort of thing. And proof also will this particular
condition derivation proof will also be various very close, but we just be careful about here
algebra that is what.
Because vector matrix algebra is not very similar to scalar algebra, that is the only
thing actually. So, what we are looking for is delta J? The has to go to zero and delta
J is something like this and obviously, the second term again is a problem. So, we expand
it try to kind of simplifying the second problem and second problem this delta X dot is nothing,
but d by d t of delta X by definition. And now we integrate by parts and then you just
tell this is the first term this is second term. So, you will keep that is first term
into integration of the second term, which is delta x minus in differential of the first
term in case like this is one. So, this is how it is think must perhaps,
this is like this is one term I missing there probably may be this is, this is the transpose
here. This is just note the transpose I mean this a because this is again, these vector
metric algebra this has to be noted actually, because we just cannot do cannot refer to
lose this transpose and all that. So, this is the vector and that is the vector ultimately
multiplication is to be a scalar actually. So, this is the how it is and then we clear
with this, with this same thing I mean we this, this term lended up with something like
this, this leads to that. So, you take this term and substitute it back and this expression
and then see that this, this these two terms and then substitute that and then lend up
this with something like this actually. Exactly same just that if this transpose cannot
be omitted and then left cannot go to right think like that actually the multiplication
out of transposes to be emplaced think like that way.
So, here one more transpose is missing looks like. So, let me correct that, this is where
I think probably here anyway. So, this how algebra proceeds and then same thing it goes
to zero. So, obliviously this coefficient has to go to zero. So, this first term will
give us this E-L equation.
And the second term will give us the transversality or boundary condition basically. So, in any
given problem we have it is avoided d t to satisfy both the conditions and carry out
the necessity algebra, which will make sure that the these condition are satisfied.
And again extending the transversality condition we can go for this similar ideas del L by
del X transpose delta X something like that. And special case is very similar to the scalar
cases these are the cases that I am talking actually. These are the cases that I am talking;
that means, either fixed end point or partly fixed of the t naught and t f both are fixed
or the states are free like that each other. So, that kind of conditions we are talking
here and everything will be very very parallel to what we have discussed there.
Such that it you have to be slightly careful about here algebra actually alright. So, this
is the condition that is there on now, that is all about free optimization which is very
parallel to the scalar cases. Now, about constraint optimization so, that is more important because
many of our problems will invariably have an equality constraint in the form of state
equation later. So, we are is more interested with this constraint of variational problems
and that to with multiple dimension and how does it go? Even very close to this static
optimization sort of concepts here.
So, what you are looking for the problem is to optimize this cost function, where are
the cost functional J, which is; which can be a non-linear function cotaining both X
X dot as well as time. And it is subject to this equality constraint and if you notice
a little bit carefully and suppose, the X is really the state of the problem state vector
of the problem. Then this constraint is nothing but state equation, sort of thing because
normally we have X dot equal to normally we have s
…
estimation course.
We have finished as six lectures already and we all to do already, get some over view of
calculus of variations is scalar. I mean, scalar variables what I am going to do here,
in this particular lecture is to review that briefly, what we discuss in the last class
followed by its extension to vector. I mean vector consumption all that multi dimensional
things. This is very similar to scalars anyway.
So, first thing is summary of what we discuss in the last lecture, the calculus of variations
for scalar problems especially.
So, this is the concept that we discussed, we discussed something like on one side is
a function and its increment and the other side is a functional and its increment. And
both the appear to be quite a similar provided the x x is turns out to be like, in this particular
case it is free variable time whereas, in this it is the x of t basically. So, it you
can interpret is something like a function of functions sort of things and that is, that
how? Where this? The this delta x can be a variation with respect to time. Then if it
varies with respect to time then, how this J varies? That is ultimately the implication
basically. Then we discussed many things about an increment of a function is this delta f
approximated is d f similarly, increment is delta J approximated is del J.
Slight notification change this is d f was it is delta J sort of thing. So, just to may
say that this is a variations.
Then we discuss about differential of a function and variation of a functional, think like
that and we carried out this algebra. And using this taylor series, we learned it up
there kind of expression for delta f and then it turns out that, when delta t goes to zero
and we learned of something like d f. So, in general you can tell d f is some thing
like f dot delta t in general basically. So, very similar concept you can do this way also
here and you can take first variation, second variation thing like that, the first variation
is can be computed this way. And as I told last class somebody does not tell where the
default variation means, first variations can be computed this way.
We also gave examples, how to compute? And think like that in the last class actually.
So, then coming to the boundary conditions we discussed two class of boundary conditions,
and the fixed end point problems, and free end point problems. Fixed end point problem
can be kind of fixed both in time and state our time in the free variable; sorry this
time is the free variables and x is the dependent variable. It can be fixed at both sense, t
zero at t zero x can be fixed at t of x can, x can also be fixed or it can also be something
like free end point problem, where the end point can be completely free or it can it
can be required to lie on a specific curve and I gave you an example of satellite launch
and all that on there. When you launch a satellite does not matter
where you join there we from there onwards it will keep on staying in the orbit anyway.
So, all that it matters is us to leave it somewhere in the orbit or orbitral condition
actually. So, that is that is how? This problems are we find at way.
So, then optimum of a functional we also discussed and we had this formal definition that functional
is said to have a relative optimum at x star of t remember is not a particular point, but
x star itself is a function of time. So, it is functional is said to relative optimum
at x star of t. If the; if there exists some epsilon is greater than zero such that the
function I mean this x of t will satisfy this kind of a condition. Whenever it satisfies
this then you have this increment of J should have the same sign.
That means if we are talking about the minimum then this delta J, J of x minus J of x star
should always be positive, no matter what kind of variations we talk about? If whatever
variation you have either up or down, whatever it is actually, irrespect to whatever variation
you have delta J will keep on the thing positive, that mean J of x star is suppose to have a
local minimum. Once similarly, if delta J is less than zero then J of x star is suppose,
to have a local maximum actually. And obviously, if the relationships are satisfied arbitrary
large value for I mean is satisfied for arbitrary large value of epsilon, then J of x star is
said to have a global optimum actually.
And we also discussed about this variation concept on only x star represents to be an
optimum path, then all that we are talking about is neighborhood path of that actually.
So, whatever neighbors we are getting that kind of that kind of a thing we are talking
about actually.
So, for then we discussed about some fundamental theorem of calculus of variation and it turns
out that, for x star of t to be a candidate optimum the following condition has to hold
good. That means, the first variation has to be zero for all admissible delta x. This
delta x turns out to be the difference between these two actually. So, for all admissible
delta x, your delta J has to be equal to zero, that happens to be a necessary condition.
And a sufficiency condition happens to be second variation should be greater than zero
for a minimum or second variation should be less than zero for a maximum. And when it
I mean this concept of kind of you can think about peri parallel to the static optimization,
what had the ideas of calculus of variation of other.
We also discuss this nice beautiful little fundamental lemma, then it will stole that
if for every continuous function g of t this equation satisfied, this integral equation
is satisfied. Where the variation delta x is continuous in t, then the I mean the only
condition that is required is delta x has to be continued that all. Then it turns out
that g of t has to be 0 and the entire in to. So, that is that is nice property and
then many many times we will use it also. And this was a very simple proof also we took
at a interval and then in that interval, we took the function everywhere 0, but in that
it is not 0. And we can always construct a delta x, which
is again non 0 in that particular interval and this integral happens to be non 0. If
you do that then this count I mean may thus the kind of I mean counter arguments sort
of things. So, were; So, taking help of that we could. So, that it is it cannot happen.
So, method proof by contradicts and all that. So, then we landed up with the conclusion
that, g of t has to be 0 it is looks as a very simple theorem, but there is lot of great
integrations later.
Whenever conditions of optimality then we defined the problem that we have to optimize
this kind of a function, where t zero and t f both are fixed and all that you to do
is to make sure that the first variation is zero. And after some algebra the, that I mean
using this partial fraction of, I mean partial fraction of this integral and all that. We
could able to show that, it get ultimately leads to this two conditions and the first
condition is called E L equations or Euler Lagrange Equation and the second one is Transversality
Condition, which leads to this boundary condition sort of things.
And we also noted that part of this equation might be already satisfied by problem formulation,
you have the word if here x zero is fixed it is not a free, then d d x not is already
zero. So, this condition you already satisfied actually, whatever is not satisfied you have
to, you can extract from there. Whatever you satisfied you already satisfied any way. So,
even though these equation is valid for the entire path from t not to t f, but this conditions
are valid for t zero and t f only basically. There is a differential equation from this
will give a corresponding differential equation as we give a corresponding boundary condition
sort of things. So, using this two we suppose to solve the optimal I mean dynamic optimization
problems.
So, now this quickly this prove that that followed you something like this it has the
delta L we took at and delta L is by definition like this. And we expanded using taylor's
series the this first term, first term in derivative and then either terms. And then
if you take in the limit delta L will be become delta L, this delta L and then that turns
out to be like that. So, this delta J can be approached matter something like this and
then this particular path I mean this second part of the integral.
You can kind of integrate by parts. So, then do this I mean algebra and then kind of with
this kind of conclusion. This part being like that and then you can put it by here and then
ultimately it will list to that. So, if this has to be zero, then this has to be zero and
that has to be zero and that is how? We got the; this two condition of optimality that
I was talk actually. So, the same trick will all good for vector algebra also, that is
why I thought of kind of reversing if again.
So, this ultimately leads to this two condition and condition one must be satisfied regardless
of the boundary condition or end condition. But second part of the equation; that means,
this equation we already, we satisfied for the problem specification. So, the amount
of extra information contained by this equation, that is with the boundary condition specified
actually so; that means, depending on the specified boundary conditions here to extract
additional information from this condition.
In general we, this was in restrictive sense we try to expand that and then in general
we lended up with the, this kind of thing I mean this general transversality condition
sort of things. When t this variation of t was also allowed and other variation of initial
time and final time that was allowed actually, where you start? Where you stop? Is up to
that designers also any way, sometimes the problem when you start? And when you stop?
That can be 3 also in a way. If you bring that into a account then this
leads to something like this and then we discussed about various special cases, case one, case
two I think up to case five and are somewhere different different conditions. That means,
if both the end point are fixed; that means, both time and x naught at t zero and t f are
all fixed and they does not; obviously, give any additional information, because every
delta x zero delta x of delta t naught delta t f everything was to zero.
But if t naught and t f are fixed, but x naught and x f are naught fixed then obliviously
this part is also zero, delta t f and delta t naught both are zero will end up with only
this part of the equation. Similarly, if t naught and x naught are fixed; that means,
t f and x are both are free, then you can consider only the t f part of the equations
and then you lend up to something like this. And similarly, if you depending on, what is
fixed? And What is free? We kind of take the other part into account because whatever is
given, if it satisfies some part of the equation automatically, we considered that is I mean
just taken sort of thing. So, this is how we extract this information
contain from transversality condition. Ultimately there the key idea is the number of differential
equations and number of un condition would be same, and then it depends on how many free
variables are there with us? And how many extra information is can we derive from this
equation actually? So, depending on in example and all will talk later, it will be more and
more clear how to be; how to use it actually?
So, the very special case at the end, when t naught and x naught is fixed and t f and
x f are constraint to lie on a given curve. Then you to go back to that, they because
it is fixed anyway so, this t naught x naught does not give any extra information. But t
f x f we cannot take that is zero, this part of it delta x f and delta t f cannot be zero,
but they are suppose to satisfies certain constraint equation, So, this constraints
equation can be derived something like this, because except x is a function of eta t. So,
from that delta x f is nothing but that so, you can substititute that and then land up
with this and where delta t f is not really zero, because it is this is constraints to
lie on a given curve; that means, your final time is kind of free actually.
That is not zero the coefficient as to be zero that of end up with this addition equation,
which will be which is needed for this kind of problem definition actually.
Then we also discuss little bit on second variation and the second variation we expanded
the second term of the taylor series and then let of I mean kind of observed that, this
lands up this gives us this hexane matrix sort of ideas.
And then to define this pi, pi let that matrix evaluated at the optimum point and if happens
to be a pdf matrix it is the minimum and if happens to be a negative different matrix,
it happens to be maximum. So, first what you have to do in again problem is to evaluate
this matrix phi first and then evaluate whether that the phi is a positive definite matrix
or not actually or negative definite matrix. If neither of the above then further math
is obviously, required and like static optimization, if you remember? We have this third term fourth
term anything like that and; obviously, those things are beyond this beyond this scope of
this course. And also remember unlike static optimization
in this particular thing x star and x star dot both are time dependent, they are not
really a specific point value. So, they are x star is a optimal solution which is a function
of time basically, thus the path trajectory that we are looking for. So, this what you
will evaluate phi is not really a constant matrix, but it is a function of time. So,
when you talk about whether this matrix is positive or negative is to remains positive
or negative for all time actually. So, for in the entire interval this matrix
should turns out to be a positive different matrix. That means, one easy way of doing
that is, just evaluate the eugene values kind of symbolically as the function of time and
then plot it in this interval actually if they all remain in positive then you have
done sort of thing. Also remember as a last comment that is just that we are source of
that is we are just talking about is valid only for free optimization problem. The moment
you bring in some sort of a state equation constraint or any other algebraic constraint
I think like tha,t then if the constraint is active then; obviously, this is the condition
does not actually. So, this is what we discussed in the last
class. How about extending some of these concepts for the vector problem? Because that is where
we are have a main interest to like whenever state variables are a typically of one dimensional
control is n dimensional thing like that actually.
So, what I mean the consider the same problem, but in multiple dimension and to begin with
we consider the problem without any constraint.
So, just an I mean what we see here is very similar concepts the only difference will
be this small x will be replaced by capital N know, where this capital X contains n dimension
I mean n components, so, x 1 x 2 up to x n actually. And obviously, again the similar
sort of ideas will take t zero and t f both are fixed and our objective is to make sure
that delta J goes to zero for arbitrary delta X t. So, again the similar algebra it will
do that in a second, but using a similar algebra it will lend up a very very similar local
equation. But remember these two equations now, are
not exactly sending what we have sent before. Especially this particular equation contains
n equations really, del L by del X one minus d by d t del L by del X one dot equal to zero,
then you can substitute by x 2 x t 3 like that. So, it really contains about n equations
actually. About an here, transpose is also must because this itself is a vector and that
is also a vector. So, what we are talking here, is very closely what we have done in
scalar problem, but not exactly one to one sort of thing. And proof also will this particular
condition derivation proof will also be various very close, but we just be careful about here
algebra that is what.
Because vector matrix algebra is not very similar to scalar algebra, that is the only
thing actually. So, what we are looking for is delta J? The has to go to zero and delta
J is something like this and obviously, the second term again is a problem. So, we expand
it try to kind of simplifying the second problem and second problem this delta X dot is nothing,
but d by d t of delta X by definition. And now we integrate by parts and then you just
tell this is the first term this is second term. So, you will keep that is first term
into integration of the second term, which is delta x minus in differential of the first
term in case like this is one. So, this is how it is think must perhaps,
this is like this is one term I missing there probably may be this is, this is the transpose
here. This is just note the transpose I mean this a because this is again, these vector
metric algebra this has to be noted actually, because we just cannot do cannot refer to
lose this transpose and all that. So, this is the vector and that is the vector ultimately
multiplication is to be a scalar actually. So, this is the how it is and then we clear
with this, with this same thing I mean we this, this term lended up with something like
this, this leads to that. So, you take this term and substitute it back and this expression
and then see that this, this these two terms and then substitute that and then lend up
this with something like this actually. Exactly same just that if this transpose cannot
be omitted and then left cannot go to right think like that actually the multiplication
out of transposes to be emplaced think like that way.
So, here one more transpose is missing looks like. So, let me correct that, this is where
I think probably here anyway. So, this how algebra proceeds and then same thing it goes
to zero. So, obliviously this coefficient has to go to zero. So, this first term will
give us this E-L equation.
And the second term will give us the transversality or boundary condition basically. So, in any
given problem we have it is avoided d t to satisfy both the conditions and carry out
the necessity algebra, which will make sure that the these condition are satisfied.
And again extending the transversality condition we can go for this similar ideas del L by
del X transpose delta X something like that. And special case is very similar to the scalar
cases these are the cases that I am talking actually. These are the cases that I am talking;
that means, either fixed end point or partly fixed of the t naught and t f both are fixed
or the states are free like that each other. So, that kind of conditions we are talking
here and everything will be very very parallel to what we have discussed there.
Such that it you have to be slightly careful about here algebra actually alright. So, this
is the condition that is there on now, that is all about free optimization which is very
parallel to the scalar cases. Now, about constraint optimization so, that is more important because
many of our problems will invariably have an equality constraint in the form of state
equation later. So, we are is more interested with this constraint of variational problems
and that to with multiple dimension and how does it go? Even very close to this static
optimization sort of concepts here.
So, what you are looking for the problem is to optimize this cost function, where are
the cost functional J, which is; which can be a non-linear function cotaining both X
X dot as well as time. And it is subject to this equality constraint and if you notice
a little bit carefully and suppose, the X is really the state of the problem state vector
of the problem. Then this constraint is nothing but state equation, sort of thing because
normally we have X dot equal to normally we have s
…