Lifting theory

Lifting theory was first introduced in a series of pioneering papers by John von Neumann.[1] A lifting on a measure space (E,\mathfrak F,\mu) is a linear and multiplicative inverse


T:L^\infty(E,\mathfrak F,\mu)\to \mathcal L^\infty(E,\mathfrak F,\mu)

of the quotient map


[\ ]:\mathcal L^\infty(E,\mathfrak F,\mu)\to L^\infty(E,\mathfrak F,\mu)\;\;,\;
         f\mapsto [f].\;

In other words, a lifting picks from every equivalence class \underline{}[f] of bounded measurable functions modulo negligible functions a representative
— which is henceforth written \underline{}T([f]) or \underline{}T[f] or simply \underline{}Tf
in such a way that


T(r[f]%2Bs[g])(p)=rT[f](p) %2B sT[g](p)\;\;\forall p\in E\;,r,s\in \mathbf R\;,

T([f]\times[g])(p)=T[f](p)\times T[g](p)\;\;\forall p\in E\;;
\mathrm{one\ requires\ also\ that\ }T[1]=1\;.

Liftings are used to produce disintegrations of measures, for instance conditional probability distributions given continuous random variables, and fibrations of Lebesgue measure on the level sets of a function.

Contents

Existence of liftings

Theorem. Suppose (E,\mathfrak F,\mu) is complete.[2] Then (E,\mathfrak F,\mu) admits a lifting if and only if there exists a collection of mutually disjoint integrable sets in \mathfrak F whose union is E.
In particular, if (E,\mathfrak F,\mu) is the completion of a sigma-finite[3] measure or of an inner regular Borel measure on a locally compact space, then (E,\mathfrak F,\mu) admits a lifting.

The proof consists in extending a lifting to ever larger sub-sigma-algebras, applying Doob's martingale convergence theorem if one encounters a countable chain in the process.

Strong liftings

Suppose now that (E,\mathfrak F,\mu) is complete and that  \underline{}E comes equipped with a completely regular Hausdorff topology \tau\subset\mathfrak F such that the union of any collection of negligible open sets is again negligible – this is the case if (E,\mathfrak F,\mu) is sigma-finite or comes from a Radon measure. Then the support of  \mu, \mathrm{supp}(\mu), can be defined as the complement of the largest negligible open subset, and the collection \underline{}C_b(E,\tau) of bounded continuous functions belongs to  \mathcal L^\infty(E,\mathfrak F,\mu).

A strong lifting for  \underline{}(E,\mathfrak F,\mu) is a lifting 
T:L^\infty(E,\mathfrak F,\mu)\to \mathcal L^\infty(E,\mathfrak F,\mu)
such that  \underline{}T\phi=\phi on \mathrm{supp}(\mu) for all  \underline{}\phi\in C_b(E,\tau). This is the same as requiring that[4]  TU\ge \big(U\cap\mathrm{supp}(\mu)\big) for all open sets  \underline{} U\in\tau.

Theorem. If (\mathfrak F,\mu) is sigma-finite and complete and  \tau has a countable basis then  (E,\mathfrak F,\mu) admits a strong lifting.

Proof. Let  \underline{}T_0 be a lifting for  (E,\mathfrak F,\mu) and  \{U_n\;:\;n=1,2\ldots\} a countable basis for  \tau. For any point p in the negligible set N:=\bigcup{}_n\left\{p\in \mathrm{supp}(\mu): (T_0U_n)(p)<U_n(p)\right\} let  \,T_p be any character[5] on L^\infty(E,\mathfrak F,\mu) that extends the character  \underline{}\phi\mapsto\phi(p) of  \underline{}C_b(E,\tau). Then define, for  \underline{}p\in E and  [f]\in L^\infty,

 
\big(T[f]\big)(p):=\left\{
\begin{array}{lr}
 \big(T_0[f]\big)(p)&\mathrm{\ if\ }p\notin N\\
 \;\;\;\;T_p[f]&\mathrm{\ \ \ if\ }p\in N\;.\\
\end{array}
\right.

 T is the desired strong lifting.

Application: disintegration of a measure

Suppose  (E,\mathfrak F,\mu) and  (X,\mathfrak X,\nu) are sigma-finite measure spaces ( \mu,\nu positive) and \pi:E\to X is a measurable map. A disintegration of \;\mu along \;\pi\; with respect to \;\nu\; is a slew  X\ni x\mapsto \lambda_x of positive sigma-additive measures on  (E,\mathfrak F) such that (1)  \lambda_x is carried by the fiber \;\pi^{-1}(\{x\})\; of  \,\pi over  \,x:


\{x\}\in\mathfrak X\;\;\mathrm{ and }\;\;
\lambda_x\left((E\setminus \pi^{-1}(\{x\})\right)=0
\;\;\;\;\forall x\in X

and (2) for every  \mu-integrable function \;f,


\int_E f(p)\;\mu(dp)=
\int_X \left(\int_{\pi^{-1}(\{x\})}f(p)\,\lambda_x(dp)\right) \;\nu(dx)
     (*)

in the sense that, for \nu-almost all  x\in X\;, f is  \lambda_x-integrable, the function  x\mapsto \int_{\pi^{-1}(\{x\})} f(p)\,\lambda_x(dp) is  \nu-integrable, and the displayed equality (*) obtains.

Disintegrations exist in various circumstances, the proofs varying but almost all using strong liftings. Here is a rather general result. Its short proof gives the general flavor.

Theorem. Suppose E is a polish[6] space and X a separable Hausdorff space, both equipped with their Borel sigma-algebras. Let \mu be a sigma-finite Borel measure on E and \;\pi:E\to X\; an \mathfrak F,\mathfrak Xmeasurable map. Then there exists a sigma-finite Borel measure \nu on X and a disintegration (*).
If \mu is finite, \nu can be taken to be the pushforward[7] \pi_\ast\mu, and then the \lambda_x are probabilities.

Proof. Because of the polish nature of E there is a sequence of compact subsets of E that are mutually disjoint, whose union has negligible complement, and on which \,\pi is continuous. This observation reduces the problem to the case that both E and X are compact and \,\pi is continuous, and \nu=\pi_\ast\mu. Complete \mathfrak X under \nu and fix a strong lifting T for (X,\mathfrak X,\nu). Given a bounded \mu-measurable function \;f, let \lfloor f\rfloor denote its conditional expectation under \pi, i.e., the Radon-Nikodym derivative of[8] \pi_\ast(f\mu) with respect to \nu=\pi_\ast\mu. Then set, for every x\in X, \lambda_x(f):=T(\lfloor f\rfloor)(x)\;. To show that this defines a disintegration is a matter of bookkeeping and a suitable Fubini theorem. To see how the strongness of the lifting enters, note that


\lambda_x(f\cdot\phi\circ\pi)=\phi(x) \lambda_x(f) \;\;\;\;
\forall x\in X\;,\;\phi\in C_b(X)\;,\;f\in L^\infty(E,\mathfrak F,\mu)\;

and take the infimum over all positive \phi\in C_b(X) with \,\phi(x)=1; it becomes apparent that the support of \lambda_x lies in the fiber over x.

References

  1. ^ Bellow, Alexandra (1969), Topics in the theory of lifting, Springer, Berlin, Heidelberg, New York
  2. ^ A subset N\subset E is locally negligible if it intersects every integrable set in \mathfrak F in a subset of a negligible set of \mathfrak F. (E,\mathfrak F,\mu) is complete if every locally negligible set is negligible and belongs to \mathfrak F.
  3. ^ i.e., there exists a countable collection of integrable sets – sets of finite measure in \mathfrak F – that covers the underlying set E.
  4. ^ U,\mathrm{supp}(\mu) are identified with their indicator functions.
  5. ^ A character on a unital algebra is a multiplicative linear functional with values in the coefficient field that maps the unit to 1.
  6. ^ A separable space is polish if its topology comes from a complete metric. In the present situation it would be sufficient to require that E is Suslin, i.e., is the continuous Hausdorff image of a polish space.
  7. ^ The pushforward \pi_\ast\mu of \mu under \pi, also called the image of \mu under \pi and denoted \pi(\mu), is the measure \nu on \mathfrak X defined by \nu(A):=\mu\big(\pi^{-1}(A)\big)\;,\;A\in\mathfrak X.
  8. ^ \;f\mu is the measure that has density \;f with respect to \;\mu

See also