Movatterモバイル変換


[0]ホーム

URL:


Wikipedia

M. Riesz extension theorem

For more theorems that are sometimes called Riesz's theorem, seeRiesz theorem.

TheM. Riesz extension theorem is atheorem inmathematics, proved byMarcel Riesz[1] during his study of theproblem of moments.[2]

Formulation

edit

LetE{\displaystyle E}  be arealvector space,FE{\displaystyle F\subset E}  be avector subspace, andKE{\displaystyle K\subset E}  be aconvex cone.

Alinear functionalϕ:FR{\displaystyle \phi :F\to \mathbb {R} }  is calledK{\displaystyle K} -positive, if it takes only non-negative values on the coneK{\displaystyle K} :

ϕ(x)0forxFK.{\displaystyle \phi (x)\geq 0\quad {\text{for}}\quad x\in F\cap K.} 

A linear functionalψ:ER{\displaystyle \psi :E\to \mathbb {R} }  is called aK{\displaystyle K} -positiveextension ofϕ{\displaystyle \phi } , if it is identical toϕ{\displaystyle \phi }  in the domain ofϕ{\displaystyle \phi } , and also returns a value of at least 0 for all points in the coneK{\displaystyle K} :

ψ|F=ϕandψ(x)0forxK.{\displaystyle \psi |_{F}=\phi \quad {\text{and}}\quad \psi (x)\geq 0\quad {\text{for}}\quad x\in K.} 

In general, aK{\displaystyle K} -positive linear functional onF{\displaystyle F}  cannot be extended to aK{\displaystyle K} -positive linear functional onE{\displaystyle E} . Already in two dimensions one obtains a counterexample. LetE=R2, K={(x,y):y>0}{(x,0):x>0},{\displaystyle E=\mathbb {R} ^{2},\ K=\{(x,y):y>0\}\cup \{(x,0):x>0\},}  andF{\displaystyle F}  be thex{\displaystyle x} -axis. The positive functionalϕ(x,0)=x{\displaystyle \phi (x,0)=x}  can not be extended to a positive functional onE{\displaystyle E} .

However, the extension exists under the additional assumption thatEK+F,{\displaystyle E\subset K+F,}  namely for everyyE,{\displaystyle y\in E,}  there exists anxF{\displaystyle x\in F}  such thatyxK.{\displaystyle y-x\in K.} 

Proof

edit

The proof is similar to the proof of theHahn–Banach theorem (see also below).

Bytransfinite induction orZorn's lemma it is sufficient to consider the case dim E/F=1{\displaystyle E/F=1} .

Choose anyyEF{\displaystyle y\in E\setminus F} . Set

a=sup{ϕ(x)xF, yxK}, b=inf{ϕ(x)xF,xyK}.{\displaystyle a=\sup\{\,\phi (x)\mid x\in F,\ y-x\in K\,\},\ b=\inf\{\,\phi (x)\mid x\in F,x-y\in K\,\}.} 

We will prove below that<ab{\displaystyle -\infty <a\leq b} . For now, choose anyc{\displaystyle c}  satisfyingacb{\displaystyle a\leq c\leq b} , and setψ(y)=c{\displaystyle \psi (y)=c} ,ψ|F=ϕ{\displaystyle \psi |_{F}=\phi } , and then extendψ{\displaystyle \psi }  to all ofE{\displaystyle E}  by linearity. We need to show thatψ{\displaystyle \psi }  isK{\displaystyle K} -positive. SupposezK{\displaystyle z\in K} . Then eitherz=0{\displaystyle z=0} , orz=p(x+y){\displaystyle z=p(x+y)}  orz=p(xy){\displaystyle z=p(x-y)}  for somep>0{\displaystyle p>0}  andxF{\displaystyle x\in F} . Ifz=0{\displaystyle z=0} , thenψ(z)>0{\displaystyle \psi (z)>0} . In the first remaining casex+y=y(x)K{\displaystyle x+y=y-(-x)\in K} , and so

ψ(y)=caϕ(x)=ψ(x){\displaystyle \psi (y)=c\geq a\geq \phi (-x)=\psi (-x)} 

by definition. Thus

ψ(z)=pψ(x+y)=p(ψ(x)+ψ(y))0.{\displaystyle \psi (z)=p\psi (x+y)=p(\psi (x)+\psi (y))\geq 0.} 

In the second case,xyK{\displaystyle x-y\in K} , and so similarly

ψ(y)=cbϕ(x)=ψ(x){\displaystyle \psi (y)=c\leq b\leq \phi (x)=\psi (x)} 

by definition and so

ψ(z)=pψ(xy)=p(ψ(x)ψ(y))0.{\displaystyle \psi (z)=p\psi (x-y)=p(\psi (x)-\psi (y))\geq 0.} 

In all cases,ψ(z)>0{\displaystyle \psi (z)>0} , and soψ{\displaystyle \psi }  isK{\displaystyle K} -positive.

We now prove that<ab{\displaystyle -\infty <a\leq b} . Notice by assumption there exists at least onexF{\displaystyle x\in F}  for whichyxK{\displaystyle y-x\in K} , and so<a{\displaystyle -\infty <a} . However, it may be the case that there are noxF{\displaystyle x\in F}  for whichxyK{\displaystyle x-y\in K} , in which caseb={\displaystyle b=\infty }  and the inequality is trivial (in this case notice that the third case above cannot happen). Therefore, we may assume thatb<{\displaystyle b<\infty }  and there is at least onexF{\displaystyle x\in F}  for whichxyK{\displaystyle x-y\in K} . To prove the inequality, it suffices to show that wheneverxF{\displaystyle x\in F}  andyxK{\displaystyle y-x\in K} , andxF{\displaystyle x'\in F}  andxyK{\displaystyle x'-y\in K} , thenϕ(x)ϕ(x){\displaystyle \phi (x)\leq \phi (x')} . Indeed,

xx=(xy)+(yx)K{\displaystyle x'-x=(x'-y)+(y-x)\in K} 

sinceK{\displaystyle K}  is a convex cone, and so

0ϕ(xx)=ϕ(x)ϕ(x){\displaystyle 0\leq \phi (x'-x)=\phi (x')-\phi (x)} 

sinceϕ{\displaystyle \phi }  isK{\displaystyle K} -positive.

Corollary: Krein's extension theorem

edit

LetE be areallinear space, and letK ⊂ E be aconvex cone. Letx ∈ E/(−K) be such thatR x + K = E. Then there exists aK-positive linear functionalφE → R such thatφ(x) > 0.

Connection to the Hahn–Banach theorem

edit

The Hahn–Banach theorem can be deduced from the M. Riesz extension theorem.

LetV be a linear space, and letN be a sublinear function onV. Letφ be a functional on a subspaceU ⊂ V that is dominated byN:

ϕ(x)N(x),xU.{\displaystyle \phi (x)\leq N(x),\quad x\in U.} 

The Hahn–Banach theorem asserts thatφ can be extended to a linear functional onV that is dominated byN.

To derive this from the M. Riesz extension theorem, define a convex coneK ⊂ R×V by

K={(a,x)N(x)a}.{\displaystyle K=\left\{(a,x)\,\mid \,N(x)\leq a\right\}.} 

Define a functionalφ1 onR×U by

ϕ1(a,x)=aϕ(x).{\displaystyle \phi _{1}(a,x)=a-\phi (x).} 

One can see thatφ1 isK-positive, and thatK + (R × U) = R × V. Thereforeφ1 can be extended to aK-positive functionalψ1 onR×V. Then

ψ(x)=ψ1(0,x){\displaystyle \psi (x)=-\psi _{1}(0,x)} 

is the desired extension ofφ. Indeed, ifψ(x) > N(x), we have: (N(x), x) ∈ K, whereas

ψ1(N(x),x)=N(x)ψ(x)<0,{\displaystyle \psi _{1}(N(x),x)=N(x)-\psi (x)<0,} 

leading to a contradiction.

References

edit

Sources

edit
  • Castillo, Reńe E. (2005),"A note on Krein's theorem"(PDF),Lecturas Matematicas,26, archived fromthe original(PDF) on 2014-02-01, retrieved2014-01-18
  • Riesz, M. (1923), "Sur le problème des moments. III.",Arkiv för Matematik, Astronomi och Fysik (in French),17 (16),JFM 49.0195.01
  • Akhiezer, N.I. (1965),The classical moment problem and some related questions in analysis, New York: Hafner Publishing Co.,MR 0184042

[8]ページ先頭

©2009-2025 Movatter.jp