LECTURE 5, 18.155, 22 SEPTEMBER 2011 (1) Last time I discussed the pointwise convergence properties of an absolutely summable sequence in Lp (Rn ). Namely, that if P 0 n φj ∈ Cc (R ) is such that kφj+1 − φj kLp < ∞ then except j≥1 on P a set, E, of measure zero, which means not ‘very large’, |φj+1 (x) − φj (x)| < ∞. So, on the complement of this set we j≥1 can define LIM{φj } = f (x) = limj→∞ φj (x). For definiteness we set f (x) = 0 otherwise, but it is not supposed to matter. Here is where we want to get to today, maybe only for p = 1 : s(Cc0 (Rn ); k · kLp ) (1) p LIM / Lp (Rn ) /S n L (R ) isometric isomorphism p n L (R )/a.e. Still have to do some work to check this, in particular to see that this limiting function, f, actually determines the ‘ideal element’ of Lp (Rn ) to which it corresponds. What does this really mean? Well, we need to show that if two absolutely summable sequences have ‘the same pointwise limit’ then they define the same element in Lp (Rn ). Proposition 1. If two sequences φj and φ0j in Cc0 (Rn ) are each absolutely summable in L1 and are such that φ0j (x) − φj (x) → 0 on Rn \ F where F is a set of measure zero, then kφ0j − φj kL1 → 0. (2) Proof. It is somewhat easier to work with series here instead of sequences, so set ψ1 = φ1 , ψ10 = φ01 , ψj = φj − φj−1 , ψj0 = φ0j − φ0j−1 , j ≥ 2. Then our assumptions are that X j (kψj kL1 + kψj0 kL1 ) < ∞, N X (ψj − ψj0 )(x) = 0 on Rn \ F. j=1 The conclusion is just a statement about the difference so consider ψ̃j = ψj − ψj0 . To get rid of F we can choose another 1 2 LECTURE 5, 18.155, 22 SEPTEMBER 2011 absolutely summable (in L1 ) series ηj ∈ Cc∞ (Rn ) such that P |ηj (x)| = ∞ on F. Then consider the intertwined sequence k ψ̃k µj = ηk −η k if j = 3k − 2 if j = 3k − 1 if j = 3k. for k ≥ 1. This is absolutely summable in L1 since all three sequences are. Moreover X X |µj (x)| < ∞ =⇒ |ηk (x)| < ∞ j k =⇒ x ∈ F =⇒ X µj (x) → 0 as N → ∞, j<N since the partial sums of this series alternate between N P φ̃k (x)) k=1 (2) and this plus µk (x) which tends to zero. So, this is what we know, that ∞ X X |µj (x)| < ∞ =⇒ µj (x) = 0. j j=1 Now fix (a large) N and consider the difference (3) Φk (x) = | X j≤N µj (x)| − N +k X |µj (x)|, x ∈ Rn . j=N +1 As k increases, this is a decreasing sequence of continuous functions of compact support – since the sum subtracted is increasing in size. There are two cases – where the condition on the left in (2) holds and when it doesn’t. If it holds then the limit of Φk (x) as k → ∞ is negative, since the right side of (2) converges to zero. On the other hand if the condition in (2) does not hold then Φk (x) → −∞. So, if we put Ψk (x) = (Φk (x))+ – just take the positive part – then Ψk (x) is a decreasing sequence of non-negative continuous functions which all vanish outside a fixed compact set (any set outside which the first vanishes) and which tends pointwise to 0. LECTURE 5, 18.155, 22 SEPTEMBER 2011 3 One of the basic properties of compact spaces is that this implies that Ψk → 0 uniformly. Indeed, take δ > 0 and look at the sets Kk = {x ∈ Rn ; Ψk (x) ≥ δ}. By continuity these are closed and hence compact. By monotonicity they are decreasing, Kk+1 ⊂ Kk , and by pointwise convergence to 0 the total intersection in empty. It follows that some finite intersection is empty and uniform convergence to zero follows. Hence from the properties of Riemann integrals, Z Ψk (x) → 0 and so going back to Φk ≤ Ψk we conclude that Z NX Z X +k |µj (x)| ≤ 0. µ| − lim | k→∞ j≤N j=N +1 The limit here is finite by the absolute summability, so what we have shown, by applying the triangle inequality, is that X X k µj (x)kL1 ≤ kµj kL1 . j≤N j>N This however implies what we want, that lim kφj − φ0j kL1 = 0. N →∞ (3) Now, let us try to put things together a little in this case where p = 1. Simply define X (4) L1 (Rn ) = f : Rn −→ C; ∃ φj ∈ Cc0 (Rn ), kφj+1 − φj kL1 < ∞, j n f (x) = lim φj (x) on R \ E for E of measure zero . j→∞ We can call a sequence φj as given here an approximating sequence (of continuous functions) in L1 for f. On this space the integral is well-defined: (5) Lemma 1. If f ∈ L1 (Rn ) then Z Z f = lim φj j is independent of the choice of approximating sequence. 4 LECTURE 5, 18.155, 22 SEPTEMBER 2011 (4) Proof. If φj and φ0j are two approximating sequences for f in the sense of (4) then we know that φj (x)−φ0j (x) → R0 off a set of measure zero so Proposition (1) shows that limj→∞ |φj −φ0j | = 0. It follows that Z Z Z 0 | φj − φj | ≤ |φj − φ0j | → 0 so the integral of f is well-defined. (5) Lemma 2. If f ∈ L1 (Rn ) then |f | ∈ L1 (Rn ). Indeed, if φj is an approximating sequence for f then |φj | is an approximating sequence (which is why I switched to continuous rather than smooth functions to avoid an approximation argument here) for |f | – it converges to |f (x)| off a set of measure zero and is absolutely summable by the reverse triangle inequality Z Z ||φj+1 (x)| − |φj (x)|| ≤ |φj+1 (x) − φj (x)|. R Thus |f | = kf kL1 is well-defined and is in fact a seminorm on L1 (Rn ). (There are details to check here). (6) Theorem 1. The map sending an absolutely summable sequence in Cc0 (Rn ) to its ‘limit’ given as the limit whenever the pointwise sequence is absolutely summable and zero otherwise, projects to an isometric isomorphism L1 (Rn ) −→ L1 (Rn )/a.e. where equality almost everywhere means taking the quotient with respect to Z 1 n {f ∈ L (R ); |f | = 0} = {f ∈ L1 (Rn ); f = 0 a.e.}. CAUTION: Before taking the quotients this – i.e. LIM in (1) – is not a linear map. (7) Proof. The one thing to check here is that the null space of the seminorm on L1 (Rn ) consists precisely of those f which vanish off a set of measure zero. To see this we check what is really its ‘completeness’ with respect to the seminorm. Thus, suppose that fj ∈ L1 (Rn ) is an LECTURE 5, 18.155, 22 SEPTEMBER 2011 (6) 5 absolutely summable series with respect in the sense that X kfj kL1 < ∞. j Choose approximating sequences for each fj , φj,k . Then, for each j, choose Nj so P large that ψj,1 = φj,Nj +1 is such that kfj − −j ψj,1 kL1 < 2 , and kψj,p kL1 ≤ 2−j where ψj,p = φj,Nj +p − p>1 ψj,Nj +p−1 . This is possible by the absolute summability and then P the double series kψj,p kL1 is absolutely convergent – any fik,p P nite sum is bounded above kfj kL1 + 2. So any (exhaustive) j ordering of this double sequence (such as the usual diagonal exhaustion) gives an absolutely summable series in L1 . It follows P that the double series |φj,p (x)| < ∞ off a set of measure zero j,p P and hence that fj (x) < ∞ off a set of measure zero since this j P is the sum of the absolutely convergent double series ψj,p (x). k,p 1 n R So, suppose we have an element f ∈ L (R ) which satisfies |f | = 0, where this integral is by definition the limit of an approximating sequence. We can apply the discussion above with fj = |f | for all j, since this satisfies (6). So, j|f (x)| → 0 off a set of measure zero, which means that f (x) = 0 off a set of measure zero! So, this establishes that S, defined as the null space of the seminorm consists precisely of the subset of L1 (Rn ) of elements vanishing almost everywhere. Everything is now pretty straightforward. The projected map from absolutely summable sequences to L1 (Rn )/a.e. can be seen to be linear by adding approximating sequences, to be surjective essentially by definition (and of course the existence of the limiting f ’s) and to have null space precisely S by what we have just shown. (8) What about the case p > 1? We can just proceed in the same way, defining X Lp (Rn ) = f : Rn −→ C; ∃ φj ∈ Cc∞ (Rn ), kφj+1 − φj kLp < ∞, j n f (x) = lim φj (x) on R \ E for E of measure zero . j→∞ Now, if f ∈ Lp (Rn ) it follows that |f |p ∈ L1 (Rn ) and then R 1 we can define the seminorm kf kLp = ( |f |p ) p . Indeed, if φj is 6 LECTURE 5, 18.155, 22 SEPTEMBER 2011 an approximating sequence with respect to the Lp norm, then ψj = |φj (x)|p → |f (x)|p almost everywhere. In fact, the ψj form an absolutely summable sequence in L1 since Z kψj+1 − ψj kL1 = ||ψj+1 |p − |ψj |p | Z ≤ C (|ψj+1 |p−1 + kψj |p−1 )|ψj+1 | − |ψj || p/q ≤ C 0 (sup kψj kLp kψj+1 − ψj kLp . j The null space of this is again precisely the space of functions in L1 (Rn ) which vanish off a set of measure zero and then Lp (Rn )/S is a concrete completion of Cc∞ (Rn ) with respect to the Lp norm.