Lecture 05

IIT Kanpur July 201
20 followers
Oct 22, 2023
Welcome, to this lecture. This is the 5th lecture of week 1. In the previous lectures,
we have introduced the concept of LP spaces on some appropriate measure space or more
specifically in the case of probability spaces. The conditions that we require are on collection
of measurable functions, we introduce certain kind of integrability assumptions. What we
have also seen is that on this LP spaces, there is a natural norm that comes in. So,
we have verified those properties and then coming from the norm, we have induced a metric
topology on it and we have also shown that in that metric topology, the LP spaces are
complete when we are choosing p between 1 and infinity, 1 included infinity excluded.
So, therefore, LP spaces become Banach spaces. So, in this lecture, we are going to continue
similar discussions, but we will focus for attention to several related spaces. So, without
further delay, let us move on to the slides. So, as just pointed out, we have already shown
that this LP spaces which are real valued measurable functions with certain integrability
assumptions, this collection forms a Banach space which essentially means it is a complete
norm linear space. So, let us move on and look at the first related space that we are
going to discuss.
So, from the previous one, we will make a small change. Consider the extended real line.
This is essentially the set of real numbers together with plus and minus infinity and
as soon as you include plus and minus infinity, there is a usual extension of the algebra
that we know. So, you have to again define a real number plus infinity as infinity and
so on. So, therefore, that is the standard arithmetic that you extend and therefore,
you get the relevant algebra. So, then consider extended real valued measurable
functions now with the same kind of integrability conditions. So, now, we are continuing with
this measure space. On top of the measure space, we are looking for collection of functions
which are now extended real valued. Now, with the extended real valued setup, we also have
the appropriate Borel sigma field and all that. So, you can talk about measurability
of functions and then for such functions, we also want this integrability assumptions
that essentially describes this collection. This collection of functions which we now
denote by Lp mu R bar. So, we carry this value space on in the notation to clarify the distinction
with the previous Lp spaces. Now, here as considered in our discussions
earlier, we will continue with this choice of p and we will see the need for this soon.
Now, if you take any extended real valued such an integrable function, then what happens
is that as a consequence of the integrability assumption which is again that the integration
of mod f to the power p should be finite. So, under that integrability assumption, what
we are going to get is that the set of points where your function takes values plus and
minus infinity that should have measure 0. Now, you see that if this set of values or
set of points were to have positive measure, then this integral could not have been finite.
So, this automatically follows from the integrability assumption that this set has to have 0 measure.
Consequently the function which we are starting by assuming that it is extended real valued,
but now we are saying that by the integrability assumption the set of points where it takes
plus and minus infinity as its value that should have measure 0. Therefore, our final
conclusion is that any such function must be essentially real valued almost everywhere.
So, with almost the full set of domain points, what you end up having is that the function
must be taking real numbers as its values. Now, we continue using our earlier identifications
with mu almost everywhere equivalence. So, this we have discussed earlier in remark 1
in the setup of the LP spaces that we have discussed. Now, we may continue to use the
similar almost everywhere equivalence and therefore, identify each such function in
this new LP space coming from the old LP spaces. So, let me repeat we are going to take a function
in this extended real valued setup, but then by our recent understanding that all such
functions actually becomes real valued and therefore, this function can be identified
inside the usual LP because you have the integrability condition.
Now, consider this again this norm function introduce the similar notation, we look at
the pth root of that integral that we assume to be finite. Therefore, if you now repeat
the earlier arguments as we have done for the usual LP spaces, we will be able to conclude
that this LP space which is consisting of this extended real valued functions must first
of all be closed under scalar multiplication, then closed under vector addition meaning
addition of functions. Therefore, as a general consequence of those
results you will get that this new LP space becomes a linear space, but together with
this norm function you again verify the similar properties and show that this gives you a
norm on this linear space. So, therefore, this space becomes a norm linear space and
finally, you follow the similar argument as we discussed in the previous lecture that
as soon as you have a norm you discuss the metric topology on top of that and check the
convergences of Cauchy sequences. And hence if you follow that similar line
of arguments you will end up with the fact that these LP spaces again form Banach spaces.
Again this is crucial that we continue to choose the number p the parameter p from 1
to infinity 1 included infinity excluded. But now let us move on to a similar setup
with another space, but now we are going to look at a more general class which is now
complex valued. So, our usual LP spaces were real valued functions, but now we are going
to consider classes of functions which are now complex valued. But again on the complex
numbers you can discuss about the Borel sigma field and consequently you can talk about
measurable functions complex valued measurable functions.
Now you look at mod of f which is the absolute value for complex numbers consider mod of
f raised to the power p and ask for the integrability condition once more. So, that gives you a
new class of functions which is complex valued measurable functions with a appropriate integrability
condition. But then again we follow the same analysis as we have discussed so far. And
again we can show that this new space of complex valued measurable functions with an appropriate
integrability conditions forms a Banach space. And as you expect the corresponding norm is
given by the pth root of this integral that we assume to be finite.
So, let me clarify once more. So, we would like to verify first that this space is closed
under scalar multiplication, then we will verify vector addition which is addition of
functions. So, from that that general natural consequence as it will follow that it will
form a linear space. On that linear space you verify that this norm function is a satisfies
the properties of norm. So, this space becomes a non-linear space and coming from the norm
you talk about the metric topology verify the convergence of Cauchy sequences and end
up showing that these spaces are also Banach spaces.
But this complex valued functions have some very nice interesting properties which we
will now focus on. So, take any such complex valued function, then point wise you can look
at the real part of that value and the imaginary part of that value. So, let me clarify if
I have a point in the domain call it small omega, then f of omega is a complex number
and therefore, point wise you can talk about the real part of f of omega and imaginary
part of f of omega. So, therefore, you have this decomposition
and therefore, you end up having for each point small omega a real number which is the
real part of the function and a purely imaginary number which is the imaginary part of the
function. Therefore, you end up having associating two omega two real numbers one is real part
of f another is imaginary part of f and therefore, you have these two real valued functions.
This decomposition that we obtained gives you the decomposition of this complex valued
a real part and a imaginary part. Now, as a general consequence of this construction
what you will have is that this real part of f and imaginary part of f both are now
measurable functions defined on this measure space that we have started off with and these
are now real valued. Moreover, you also have by construction modulus of real part of f
is less or equal to absolute value of f and imaginary part of f is less equals to absolute
value of f. This follows from the identification that
the absolute value of f of omega the square of that is exactly equal to real part of f
of omega square of that plus the imaginary part of f of omega the square of that. If
you square it you will get the absolute value of f of omega squared and a natural consequence
of that is this inequality that we obtain. But, then once you have this inequalities
a straightforward calculation involving integrability will show you that real part of f raised to
the power p is less or equals to mod of to the power p and therefore, this integral must
be finite. By a similar reasoning you will have imaginary part of f modulus of that raised
to the power p that is now a real valued function corresponding to that we are verifying this
integrability condition and this integrability condition is satisfied.
Therefore, this real part of f and imaginary part of f are real valued measurable functions
defined on this measure space that you are working with and hence what you end up having
is that each of these real part of f and imaginary part of f are actually coming from usual L
p space real valued p integrable functions that we considered earlier. We may therefore,
write this earlier decomposition in this space decomposition format. So, earlier decomposition
that we have seen was for the function we split it into two real valued functions, but
then we have it in more general settings in small f is arbitrary we end up having this
identification that L p mu c is actually I should write it as L p of mu here essentially
with real values which we do not write and here also we write L p of mu with real number
which we usually do not write. So, therefore, correction will be made in
the lecture notes that will be supplied to you at the end. So, what we are saying is
that all the complex valued measurable functions that we are considering with the appropriate
integrability conditions is getting split into these two parts. So, in the our usual
L p space that we have discussed. So, this essentially tells you that if you understand
the usual L p spaces then you can talk about the complex valued measurable functions with
the integrability conditions that we considered and also for the extended real valued cases
that we also have seen we have essentially identified with the usual L p space because
by the integrability assumption the functions were taking plus and minus infinity values
with measure 0. So, let us just go back and highlight that point once more. So, that we
have seen for the extended real valued function case the set of points where the function
was taking plus and minus infinity was actually a set of measure 0.
So, therefore, if we understand the real valued measurable functions the with appropriate
integrability condition they these L p spaces well enough then we are essentially understanding
these two new spaces that we have just discussed. So, let us move on.
Now, we introduce a related concept called essential supremum of a function. This concept
will be useful because we want to extend the set of p that we can consider. So, remember
that we have been working with p between 1 and infinity 1 included infinity excluded,
but right now we are going to introduce a new concept which is this essential supremum
which will let us define a concept or define a space corresponding to the value p equals
to infinity. So, let us see this new concept. So, given any measurable functions real valued
measurable function on this measure space that we are considering define the essential
supremum to be infimum over a certain set of values. What we want is that we are looking
for extended real numbers small c such that this inequality is satisfied that f is dominated
by this constant almost everywhere. So, therefore, what we are asking is that outside a set of
measure 0 you should essentially have f dominated by a constant. So, now, collect all such constant
c and ask for the infimum of this quantity. Remember that since we are allowing the extended
real line in particular c equals to plus infinity this inequality is always satisfied because
we are considering f to be real valued. So, therefore, this set of points that we are
considering here is non-empty because plus infinity is always an element here, but then
it might also include some other real number c and for which this inequality is satisfied.
And therefore, what we will look at the end is the infimum of such a set and that infimum
is what we are going to call as the essential supremum of f written as S sup by notation.
So, this essential supremum now leads us to this new concept. So, we look at this new
space of real valued measurable functions once more, but with a new condition that the
essential supremum of mod of f should be finite. And that essential supremum of mod
of f we associate this new number or new notation which is the infinity norm of f or the supremum
norm of essential supremum norm of f. We are going to discuss now properties of this L
infinity space that we have just introduced together with the norm function that we are
bringing in. So, the first observation that we make is
that the mod f function f is a real valued measurable function. So, therefore, modulus
of f is a non-negative measurable function this is true point wise. So, therefore, you
have this general inequality. Therefore, as a consequence of that if you follow the definition
of essential supremum you will have immediately that essential supremum of mod of f which
is the infinity norm of f must be non-negative this is just by construction. Consequently
what you are having is that this mapping that we have defined earlier from the L infinity
space for each function it associates this number, but this number now is a non-negative
real number. So, that is by construction and this number is taking values in possibly including
0 up to infinity, infinity excluded. So, therefore, that is the structure of this
infinity norm map, but let us look at this further properties in the light of our earlier
discussion. So, the first thing that we will do is that we will look at the almost everywhere
equivalence again. If it so happens that the L infinity norm of this function f is 0, then
immediately if you follow the that we have been discussing so far, then you will immediately
get that the function f must be 0 almost everywhere. This follows as I said with follows
by the similar arguments that we did for the usual LP spaces.
And therefore, as a consequence of that we will again be able to talk about the equivalence
relation on this collection of functions. And therefore, this will give you a collection
of equivalence classes up to this almost everywhere equalities, but then again to reduce the notational
difficulty as we have done before for the usual LP spaces, we will follow the similar
notational terminology that we work with this equivalence classes only, but do not stress
on the distinguishing an equivalence class and a representative function from that class.
So, essentially we will think of this equivalence classes as functions inside this L infinity
space only. So, this will reduce our notational difficulties, but let us move on and we will
see that this again gives us similar properties that we have seen for the usual LP spaces.
So, start with L infinity function now. So, you are assuming that essential supremum of
modulus of f should be finite and also choose a real number alpha, this will act as a scalar.
So, we are now asking for the scalar multiplication property. Then for any non-negative real number
c, observe that modulus of f is less equals to c almost everywhere, this happens if and
only if modulus of alpha f is less equals to mod alpha times c almost everywhere. This
follows by standard verification. And therefore, as soon as the L infinity norm of f is finite,
therefore you will have the essential supremum of mod of f is finite, you can immediately
show that the essential supremum of modulus of alpha f is also finite.
And hence, as a consequence of that alpha times f that is now a new real valued measurable
function that satisfies the relevant properties and it is a now a element of that a new L
infinity space that we have introduced. And moreover, as a consequence of this equivalence
that we have just proved, it immediately follows that the L infinity norm of alpha times f
is equal to modulus of alpha times the L infinity norm of f. So, this is following
our usual expectations and therefore, we end up having this scalar multiplication property
for the L infinity space. And more generally, if you follow similar arguments by similar
straightforward arguments, you will end up having the additivity property that given
two functions from the L infinity space, we can show that addition will give you again
a real valued measurable function such that essential supremum of modulus of f plus g
is finite and therefore, f plus g is a element of this L infinity space.
Moreover, you can again verify that the L infinity norm of f plus g is less or equal
to L infinity norm of f plus L infinity norm of g. And this actually gives you the triangle
inequality for the L infinity norm. And therefore, together with all these properties, you will
end up having the first that L infinity space is a real vector space. And moreover, this
new function the L infinity norm that we have been calling it, it is actually a norm on
this space L infinity. That L infinity norm first of all it identifies with 0, if and
only if the function is 0 under the mu almost everywhere equivalence, then there is this
scalar multiplication property that we verified. And finally, the triangle inequality holds
for this norm function. So, we end up with the fact that L infinity
space together with this L infinity norm that we have been discussing forms a norm linear
space. And therefore, you end up having again a metric topology. And you can ask for the
similar properties about convergences of Cauchy sequences. And similar to the discussion that
we perform for the usual LP spaces, we will look at Cauchy sequences, but now in this
L infinity norm. And then once you have such a Cauchy sequences, you can justify this point
wise convergences and end up having a limit function, which is again can be shown to belong
to L infinity. So, therefore, you can establish the completeness
of this L infinity space. And consequently, you end up having that L infinity space together
with this norm topology that we have this metric topology coming from the norm, it forms
a Banach space. So, that is a complete norm linear space. So, we have seen several related
spaces related to the usual LP spaces that we have considered earlier.
But now we are going to look at a relevant functional analytic property of this LP spaces,
which are called dual spaces. Dual spaces are collections of continuous linear functionals
on the given space. Remember, as soon as you have this topology on this LP spaces,
you can talk about linear functionals on top of it and you can talk about whether this
linear functionals are continuous or not. Now, if you have this collection of continuous
linear functionals by construction itself, you can show that this forms again a nice
Banach space. So, first you have to show it forms a vector space, then with the dual norm
that we usually define from functional analysis, you end up having a complete norm linear space
the for the dual space structure. Therefore, this Banach spaces which are formed
of this continuous linear functionals, they in some sense capture certain important properties
of this usual LP spaces. Now, let us restrict our attention to P between 1 and infinity,
1 excluded infinity also excluded. So, let us start with this situation, then what can
be shown by standard arguments that the dual space of this LP actually is the LQ space,
where Q comes from this relation that whenever you choose P between 1 and infinity, strictly
between 1 and infinity, then as soon as you choose the Q with this relation. So, therefore,
for that Q, if you look at the LQ space that can be identified as the dual of LP.
So, this is a general nice structure that comes from the LP spaces, but now we can ask
what happens to the excluded value of P. So, so far for LP spaces we have included 1, but
in this specific discussion we have started by excluding the value 1. So, let us look
at the L1 space and ask for the dual structure. Now, the dual space of L1 actually turns out
to be identified with L infinity. So, this is also a nice property that L infinity can
be identified with the dual of L1. Now, again you can ask that in this lecture
we have also seen the P equal to infinity case which led us to this L infinity space.
So, what is the dual of this L infinity? Now, as it happens that in general this dual of
L infinity will contain L1 and typically be bigger than that, but the description of that
in general is difficult because it will also depend on the structure of the underlying
measure. So, to avoid the technicalities in this course we are not going to discuss
about the dual space of L infinity. So, in this lecture we have seen several related
spaces to the usual LP spaces and we ended up discussing this dual spaces of LP.
So, we stop here and in the next lecture we are going to see some problems based on the
lectures or based on the material that we have covered in this week and hopefully those
problems will help you understand the material better. So, see you next time.
Thank you.
