topological vector spaces

References

Summary

Initial Topologies

Lemma

If {f_i : X \to X_i} are linear then initial top is TVS
If X_i are locally convex then so is X

Lemma

If {f_i: X \to X_i} gives initial top then
g:A \to X is cont iff g f_i cont \forall i
Moreover, initial top on Y wrt g is the same as wrt {g f_i}

Lemma: Subfamily

If F \sube G is a subfamily of generating maps then
I_F \sube I_G
where I denotes the initial topology.

The topology from G has as open sets all gi-1(U) where U is open For those f ∈ F, their open sets will also be open in G. So G is finer

Reverse

In order for the reverse to hold, given an open set UG in IG, we need we need an open set UF in IF with UF ⊆ UG

Frobenius Algebra

In order to remove <-,z> from our generating set, we need

Given any U and {x | <x,z> ∈ U} there is some w and V with {x | <x,w> ∈ V} or {x | <w, x> ∈ V}

We can always restrict to a basis. So say we look at {x | |<x,z>| < r}

We can take <-, λ z> for *any λ*  ≠ 0 and look at { x | |<x,λ z>| < |λ| r }

That's actually very helpful, because it means that we can take as our generating set a kind of projectivization

But we can actually just reduce to the basis elements So we pick a basis {bi} (possibly infinite-dimensional)

Write z = α1 b(i1) + … + αn b(in) Then |<x, z>| ≤ |α1| |<x, b(i1)>| + … + |αn| |<x, b(in)>|

Now the α and n only depend on z and are thus fixed. So we just take the intersection of the open sets defined by each { x | |<x,b(ik)>| < r / (n αk) }

Strong/Weak Topologies

Weak Topology

Def : Given any TVS V, we can define an initial topology generated by all of the *continuous linear functionals*
This is called the *weak topology* of V.

If we want to talk about topological properties with respect to this topology, we'll call them

Prop: There are no new continuous linear functionals

Let <-,->: V × W → k be a bilinear form (not-necessarily non-degenerate) Let V have the initial topology generated by all <-,w>: V → k Let φ : V → k be a cont linear functional Then there is some w ∈ W such that φ = <-,w>

Is it true that if we take <-,->: V × V → k and both <-,x> and <x,-> as generating the initial topology, we don't get any more continuous linear functionals?

If we take that topology, we've added extra generating functions, which will make more maps continuous, and thus

Frobenius Algebra

So a Frobenius algebra F will have its Frobenius topology, generated by the <x,-> and <-,x>. But it will also have a weak and strong topology.

But by the above proposition, F has no continuous linear functionals that aren't of the form <-,x>

Strong Topology

Semi-norms

Def : p(x+y) \le p(x) + p(y)
      p(\lambda x) = |\lambda|p(x)      

To be a norm, we also need p(x) = 0 implies x = 0

Lemma: Absolute value doesn't change anything

Given linear functionals f_i : X \to k, the initial topology generated is the same as the initial topology generated by |f_i|: X \to R

Basically you use the fact that the absolute value | ċ | : k → R is continuous for both k = C and R. So you use the fact that the initial top on X → V → Vi is the same as the initial top from X → Ǧiven any bilinear form <-,->: V × W → k The initial topologies generated by <-,W>: V → k and <V,-> : W → k are also generated by post-composing with the absolute value

Frobenius Algebra

Any unital Frobenius algebra will have its topology generated simply by ε since

Semi-norm Initial Topology

The SN topology for a family of seminorms is the initial topology with respect to them This is not necessarily a TVS structure, since the seminorms aren't linear maps and don't necessarily come from linear maps

Special Subsets

Absorbing

Def : Let U \sube V
For all x there is some \alpha > 0 with x \in \lambda U for all |\lambda| \geq \alpha

This says that given any element x, if we expand U, it will eventually contain x.

Frobenius Algebras

U = { x | <x,x> }

Balanced

B is balanced if λ B ⊆ B for all |λ|  ≤ 1

Frobenius Algebras

Barreled

B is a barrel if it is

  1. Convex
  2. Balanced
  3. Absorbing
  4. Closed

A barrelled space is a TVS for which every barrel is a nbh of zero.

Frobenius Algebras

Bounded

B is bounded if for all zero nbh U, there is a λ such that B ⊆ λ U This is equivalent to for all zero nbh U there is an α with B ⊆ λ U for all |λ| ≥ α

In other words, every zero nbh absorbs B

The finer the topology, the fewer bounded sets (because they have to satisfy more conditions)

In a LCTVS, B is bounded iff every countable subset A ⊆ B has p(A) bounded for all p

Prop

Continuous functions between TVS map bounded to bounded

Prop

For an initial topology generated by fi: V → Vi, B is bounded iff fi(B) is bounded ∀ i

Prop

B is bounded iff for any sequence bn ∈ B and scalars λn  → 0, we have (λn bn)  → 0

Prop

Let <E,F> be a dual pair A polar A^ˆ is absorbing iff A is bounded

Prop: Preserving Boundedness

If B is bounded then so is

  1. The convex hull (in a LCTVS)
  2. The closure
  3. Balanced hull
  4. Absolutely convex closed hull (in a LCTVS)

Frobenius Algebras

(Totally) Bounded / Precompact

B is precompact if for any nbh of zero, U, there is a finite set of translations of U, {ai + U} which covers B

Prop

Closed under

  1. Convex hull (LCTVS)
  2. Closed absolutely convex hull (LCTVS)

Prop

Compact iff precompact and complete

Closed

A linear function on a TVS is cont iff ker f is closed.

Convex

C is convex if λ x + (1 - λ)y ∈ C for all x,y ∈ C, 0 ≤ λ  ≤ 1

Prop: LCTVS, you get convexity for free on bounded sets

In any LCTVS, convex hulls of bounded sets are bounded

If C is convex then so is both Int(C) and Cl(C)

Frobenius Algebras

  1. Prop

    Let C be convex in a LCTVS, then C is closed iff C is weakly closed

Polar

Topological Vector Spaces

Proposition: Continuous Operations

Let V be a TVS

x \to x + y is a homeo
x \to \lambda x is a homeo
\lambda \to \lambda x is cont

every (not nec open) nbh of zero is absorbing
every nbh U has some W with W + W \sube U

For the absorbing one, we take the map f: λ → λ x For any nbh of zero, U, and any x, we take the preimage of this map, f-1(U). Now λ ∈ f-1(U) iff λ x ∈ U iff x ∈ λ-1 U. Since f(0) = 0 ∈ U, we can take a small ball in f-1(U) around 0, with radius less than 1. Call the radius α

Then you're inside this ball iff |λ| < α

The last one follows from continuity of addition In particular, we're cont at (0,0) in V × V 0 + 0 = 0, so we take a nbh around 0, U, preimage from addition to get a nbh of (0,0). Restricting to a basic nbh we have A1 × A2 with A1 + A2 ⊆ U Now set W = A1 ∩ A2

Locally Convex TVS

Continuity

If V's topology is defined by a family of seminorms, then
x_\alpha \to 0 (as a net) iff each p(x_\alpha) \to 0

Given two LCTVS a map between them f: V → W is cont iff for each seminorm pW, there is a finite number of qVi and a constant αP such that pW(f(x)) ≤ αP (qV1(x) + … + qVn (x)) for all x

Frobenius Algebra

The topology on F × F is induced by the projections followed by the pairing So (x,y)  →  < x, z>, <z,x>, <y,z> or <z,y> for a fixed z.

So the pairing is jointly continuous iff there is some C such that The map |<x,y>| ≤ C (q1(x,y) + … + qn(x,y)) for all x,y

Frobenius Algebra

The topology induced by <-,-> : F × F → k has as a basis of open sets for the origin

{x | |<x,bj>|, |<ai, x>| < r} where {ai, bj} is a finite set of elements of F

In other words, you take a finite number of the generating seminorms, and some radius

Set Types

For any seminorm, {x | p(x) < 1} and {x | p(x)  ≤ 1} are

  1. Convex
  2. Absorbing
  3. Balanced

B is bounded iff each p_α(B) is bounded in R

Locally Bounded Maps

A locally bounded map f: V → W is one which sends bounded sets to bounded sets.

Continuity

A linear map between TVS f: V → W is cont iff it is cont at any single point

A continuous linear map is uniformly continuous.

A map f: V → k is cont iff there is a cont seminorm on V s.t. |f|≤ p

A map f: V → k is cont iff the kernel is closed

Frobenius Algebra

<-,-> : F × F → k

Convergence

Let V be a LCTVS defined by seminorms {p_α} A net (x_α) → x iff p(x_α - x)  → 0 for all p

Universal Constructions

Products

Arbitrary products of LCTVS are still LCTVS. The seminorms are the projections followed by the original seminorms

∏_α V_α → V_β → k