DIFFERENTIALS OF COMPLEX INTERPOLATION PROCESSES
JESÚSM. F. CASTILLO
This paper contains an
expanded
version of mytalk at the RIMS ConferenceThe research of
geometric
structures inQuantum
information based onOperator
Theory
andrelatedtopics.
I would liketothank theorganizers
forgiving
metheopportunity
toparticipate
atthe conference andespecially
toProf. MuneoCho formaking
itpossible.
1. THECLASSICALTHEORY REVISITED
Let me
explain
first what adifferentialprocessis,
howthey
appearand what dothey
mean. Assumeonehasa space3of Banachspacevalued functions definedonsome
parameter
space F; for whichwe can assumethat is a Banachspace itself when endowed withanaturalnorm\Vert\cdot\Vert_{S} (say,
thesupremumonF).
Evaluationata
point
$\theta$\in Fproduces
aBanachspaceX_{ $\theta$}=\{f( $\theta$) : f\in \mathfrak{F}\}
,whichisthequotient
spaceof3endowedwith the naturalquotient
norm\displaystyle \Vert x\Vert_{ $\theta$}=\inf\{\Vert f\Vert : f( $\theta$)=x\}
. If$\delta$_{ $\theta$}
denotes the evaluationmap thenX_{ $\theta$}=$\delta$_{ $\theta$}(\mathfrak{F})
. Since$\delta$_{ $\theta$} is anopenmapping,
infact
$\delta$_{ $\theta$}(B_{\mathfrak{F}})
=B_{X_{ $\theta$}}
,given
x \in X_{ $\theta$}, an $\epsilon$‐extremalf_{x}
is anelementf_{x}
\in S sothatf_{x}( $\theta$)=x
and such that\Vert f_{x}\Vert \leq(1+ $\epsilon$)\Vert x\Vert_{ $\theta$}.
A differential process is a
correspondence
x \rightarrowf_{x}'( $\theta$)
. Of course that one hasassumed that function of3aredifferentiable. Much trickieristhe
point
of wheref_{x}'( $\theta$)
lies.To
clarify this,
let usconsider theperfect
situationwhere a differentialprocessappears:
complex interpolation.
Recall that aninterpolation
methodmeanstohaveaspaceparameterFsothat
given
two(in
thesimplest
version)
BanachspacesA,
Bone can
assign
toeach $\theta$\in F aBanach spaceX_{ $\theta$}
that isintermediate between Aand B inthesensethat whenanoperatorsendsA\rightarrow Aand B\rightarrow B
automatically
sends X_{ $\theta$} \rightarrow X_{ $\theta$}. Thereare many
interpolation
methods: real methods(several),
minimal
methods,
orbitmethod,
methodsspecific
for Köthe spaces We focusnow onthe
complex
method.We describeit in the
following
way: Assume that bothA,
B canbe embeddedinto a
larger
Banachspaceand consideras3
theso called Calderonspaceof holo‐morphic
A+B valued functions defined onthe unitcomplex
strip S=\{z
: 0 \leq Rez\leq1\}
such thatf(it)
\in A andf(1+it)
\in B. A few technicalproperties
are
usually
added.Then,
for each inner point $\theta$ \in S, thespaceX_{ $\theta$}
=$\delta$_{ $\theta$}(S)
is aninterpolation
spacebetween A and B.Let us present a
simple
and naturalexample:
PickA=P_{1}
andB=\ell_{\infty}
. Inter‐polation
atz=1/2
produces
thespacep_{2} by
the Riesz‐Thorin theorem:i.e.,
eachholomorphic
function that takes valuesin\ell_{1}
at0andtakes valuesin\ell_{\infty}
at1 takes values in P_{2} at1/2.
What is the differential process here?Well, given
x\in\ell_{2}
anextremal canbe
given
by
f_{x}(z)
=x^{2(1-z)}
since—wesimplify
abitjust
picking
0and 1 to represeńt
points
it and 1+it-f_{x}(0)
=x^{2} \inP_{1}
andf_{x}(1)
= 1 \in\ell_{\infty}.
Hence differentiationon thepositive cone
produces
theprocess:The research has been supported in part by project MTM2013‐45643‐C2‐1‐P and Ayudaa
x\rightarrow-2x\log|x|.
This has to be understood as the map that associated to the sequence x the
sequence
(-2x(n)\log|x(n)|)_{n}
What is themeaning
ofthismap?
Moreprecisely,
Whatisthe
meaning
of thisnon‐linear, non‐continuous,
andofcoursenotevenwell defined froml_{2} intoP_{2}map?
Well, given
a Banach space X,non‐linear,
non‐continuous maps defined on Xbut
taking
values \mathrm{i}\mathrm{n}\backslash somebigger
space(from
now onsimply
referred to as thecloud)
have theirplace
in Banach spacetheory. Indeed,
let $\Omega$ : X \rightarrow 0 be one of those maps. We will ask it acouple
of conditions: to behomogeneous
( $\Omega$( $\lambda$ x) = $\lambda \Omega$(x))
and toverify
that for any twopoints
x, y \in X theCauchy
difference$\Omega$(x+y)- $\Omega$(x)- $\Omega$(y)\in X
. Inotherwords,
themap $\Omega$cantake valueswhereveritwants, but the
Cauchy
differences fallin X. With suchamapone canform the derived space
d_{ $\Omega$}X=\{(w, x) \in \mathrm{O}\times X : x\in X;w- $\Omega$ x\in X\}
endowed with the norm
\Vert(w, x)\Vert_{ $\Omega$}
=\Vert w- $\Omega$ x\Vert+\Vert x\Vert
.Actually,
\Vert. \Vert_{ $\Omega$}
isnotanorm,
only
aquasi‐norm
since it fails thetriangle inequality.
Butin most casesit is
equivalent
to a norm—theone
having
as unit ball the closed convexhull of the unit ball of\Vert
.\Vert_{ $\Omega$}-\mathrm{s}\mathrm{o}
we cansimplify
andjust
call it a norm The space d_{ $\Omega$}X contains thesubspace
\{(x, 0) x\in X\}
isometric to Xand thequotient
spaced_{ $\Omega$}X/\{(x, 0) : x \in X\}
isagain
isometric to X. This situation is well known inhomological algebra:
An exact sequence 0 \rightarrow \mathrm{Y} \rightarrow X\rightarrow Z\rightarrow 0, whereY,
Z areBanachspaces and thearrows are
(bounded)
operators is adiagram
inwhich the kernel of eacharrow coincides with theimage
of thepreceding
one.By
the openmapping
theorem this means that \mathrm{y} is(isomorphic to)
asubspace
ofX in such a waythat thecorresponding quotient
X/Y
isisomorphic
to Z. The spaceX isusually
calledatwistedsumofY and X.Thus,
there is anexact sequenceof Banachspaces0 \rightarrow X \rightarrow d_{ $\Omega$}X \rightarrow X \rightarrow 0
and the space d_{ $\Omega$}X is a twisted sum of X. Twisted sums have been studied
long
since. Thestarting problem
isprobably
that ofdeciding
if a twisted sumofHilbert spaces must itself be a Hilbert space. The first
negative
answer camefrom
Enflo,
Lindenstrauss and Pisier[13]
using
anad‐hocconstruction;
but it was Kalton[14]
who showed that all twisted sums of twoquasi‐Banach
spacesY,
Xare
actually
inducedby
aquasi‐linear
map $\Omega$ : X \rightarrow Y. Recall that a map $\Omega$ : X\rightarrow Y iscalledquasi‐linear
ifit ishomogeneous
and thereis aconstant Msuchthat
\Vert $\Omega$(u+v) - $\Omega$(u)
- $\Omega$(v)\Vert
\leqM\Vert u+v\Vert
for all u,v \in X. Infact,
there isa
correspondence
(see
[11,
Theorem 1.5.\mathrm{c}, Section1.6])
between exact sequences0 \rightarrow
Y\rightarrow $\theta$\rightarrow X\rightarrow 0
of Banach spacesand aspecial
kind ofquasi‐linear
maps $\omega$ : X \rightarrow \mathrm{Y}, called z‐linear maps, which are thosequasi‐linear
mapssatisfying
\displaystyle \Vert $\omega$(\sum_{i=1}^{n}u_{i})-\sum_{i=1}^{n} $\omega$(u_{i})\Vert
\leqM\displaystyle \sum_{i=1}^{n}\Vert u_{i}\Vert
for all finite sets u_{1},...,u_{n} \in X. \mathrm{A}quasi‐linear
map $\Omega$ : X\rightarrow Y induces theexactsequence0\rightarrow Y\rightarrow jY\oplus_{ $\Omega$}X\rightarrow^{p}X\rightarrow
0 inwhichY\oplus_{F}X denotes thevector space Y\times Xendowed with the
quasi‐norm
\Vert(y, x)\Vert_{ $\Omega$}
=\Vert y- $\Omega$(x)\Vert+\Vert x\Vert
. Theembedding
isj(y)
=(y, 0)
while thequotient
map is
p(y, x)
=x. When $\Omega$ is z‐linear,
thisquasi‐norm
isequivalent
to a norm[11,
Chapter
1].
On the otherhand,
theprocess toobtaina z‐linearmap out from anexact sequence 0\rightarrow\mathrm{Y}\rightarrow i
$\theta$
\rightarrow^{\mathrm{q}}X
\rightarrow 0 ofBanach spaces is thefollowing:
get ahomogeneous
bounded selectionb:X\rightarrow $\theta$
for thequotient
map q, and then alinear
P:X\rightarrow $\phi$
selection for thequotient
map. Then$\omega$=i^{-1}(b-P)
isa z‐linearmap X\rightarrow \mathrm{Y}. Thecommutative
diagram
obtained
by taking
as T:$\theta$\rightarrow \mathrm{Y}\oplus_{ $\Omega$}X
the operatorT(x)
=(x-\ell qx, qx)
showsthat the upper and lower exact sequencesare
equivalent.
Twoquasi‐linear
mapsF,
G : X\rightarrow Y aresaid tobeequivalent,
denoted F\equiv G, ifthe difference F-Gcanbewrittenas B+L,whereB : X\rightarrow Y isa
homogeneous
bounded map(not
necessarily
linear)
and L:X\rightarrow Y isa linearmap(not
necessarily
bounded).
In[18]
Kalton and Peck show that in theparticular
case of\ell_{p}
spaces(more
generally,
Banach spaces with an unconditionalbasis)
it ispossible
togive
anexplicit quasi‐linear
map$\Omega$_{p}
:p_{p}\rightarrow P_{\mathrm{p}}
by
meansof$\Omega$_{p}(x)=x\displaystyle \log\frac{|x|}{||x\Vert_{p}}
whenxisa
finitely supported
sequence, and with theunderstanding
that\log 0=0.
And thismeans that on theunitsphere
\Vert x\Vert_{2}
= 1the Kalton‐Peckmap$\Omega$_{2} is, up to a -2
factor,
the differential processcorresponding
to the naturalinterpolation
scale
(l_{1}, \ell_{2})
by
thecomplex
method.Tohave a better
understanding
of this connection betweeninterpolation
scales and differential processes it willhelp
us todisplay
a few elements ofhomological
algebra
inBanachspacetheory.
Twoexact sequences0\rightarrow \mathrm{Y}\rightarrow X_{1}\rightarrow Z\rightarrow 0 and 0\rightarrow Y\rightarrow X_{2}\rightarrow Z\rightarrow 0 areequivalent
if there existsanoperator T:X_{1}\rightarrow X_{2} such that thefollowing diagram
commutes:The classical 3‐lemma
(see [11,
p.3])
shows that Tmust be anisomorphism.
An exactsequence istrivial ifandonly
ifitisequivalent
to0\rightarrow Y\rightarrow Y\times Z\rightarrow Z\rightarrow 0,
where\mathrm{Y}\times Z isendowedwiththeproduct
norm. In thiscase wesaythat theexactsequence
splits.
Ofcoursetwoexact sequences
induced
by
twoquasi‐linear
maps$\Omega$,
$\Psi$ areequivalent
if andonly
if $\Omega$ and $\Psi$ areequivalent.
Given an exact sequence 0 \rightarrow \mathrm{Y} \rightarrow X \rightarrow Z \rightarrow 0 with associated
quasi‐linear
map Fandanoperator $\alpha$ : Y\rightarrow Y',thereisa commutative
diagram
The lower sequence is called the
push‐out
sequence, its associatedquasi‐linear
map is
equivalent
to $\alpha$\circ F, and the space POiscalled thepush‐out
space. WhenF is z
‐linear,
sois $\alpha$\circ F.The
push‐out technique
willclearly
showusthe connection betweendifferentialprocessesand
quasi‐linear
maps.Indeed,
thecomplex interpolation
methodworks,
asweindicated
above, by considering
the exact sequence0 \rightarrow \mathrm{k}\mathrm{e}\mathrm{r}$\delta$_{ $\theta$} \rightarrow \mathfrak{F} \rightarrow^{$\delta$_{ $\theta$}} X_{ $\theta$} \rightarrow 0
Now,
abasicprinciple
in[5,
Theorem4.1]
establishesthat,
evenif functions of Scanhave their derivativestaking
valuesin thecloud,
the functions of\mathrm{k}\mathrm{e}\mathrm{r}$\delta$_{ $\theta$}
take their valueat $\theta$precisely
inX_{ $\theta$}
;thus,
if$\delta$_{ $\theta$}'
denotestheoperatorf\rightarrow f'( $\theta$)
,theresultabovesaysthat
$\delta$_{ $\theta$}'
:\mathrm{k}\mathrm{e}\mathrm{r}$\delta$_{ $\theta$}\rightarrow X_{ $\theta$}
isalinear continuousoperator. To formulate thisinthe
homological language,
observe thatwehavealready
thesetting
forapush‐out
diagram
(2)
So,
it turnsoutthat ifonehasacomplex interpolation
scale(X_{0}, X_{1})
andinter‐polates
at $\theta$ and obtainsthe space X_{ $\theta$}then the methodyields
atwistedsumofX_{ $\theta$}.
To knowexactly
whichone onejust
hastofollowthediagram:
ifB_{ $\theta$}, L_{ $\theta$}
arehomo‐geneusbounded and linear selctions for
$\delta$_{ $\theta$}
then thequasi‐linear
mapassociated totheupper sequence is
B_{ $\theta$}-L_{ $\theta$}
andconsequently
thequasi‐linear
mapassociatedtothe lower
pushLout
sequenceis$\delta$_{ $\theta$}'(B_{ $\theta$}-L_{ $\theta$})
.In other
words,
up toalinear map it isthe differentialprocess$\delta$_{ $\theta$}'B_{ $\theta$}.
Kalton
[15, 16]
started thestudy
of differentialprocessesand didatremendously
deep
work in the contextof \mathrm{K}\ddot {}\mathrm{t}\mathrm{h}\mathrm{e} functionspaces. X over a measurespace( $\Sigma$, $\mu$)
.As a
particular
case of which are the Banach spaces with a 1‐unconditional ba‐sis. We denote
by L_{0}
the space ofall $\mu$‐measiưable functions and this isgoing
to beour cloud in this context. A centralizer on X is a
homogeneous
L_{0}‐valuedmap $\Omega$ : X \rightarrow
L_{0}
such that\Vert $\Omega$(ax)-a $\Omega$(x)\Vert_{X}
\leqC\Vert x\Vert_{X}\Vert a\Vert_{\infty}
for all a \in L_{\infty}and x \in X. This notion coincides with Kaltons notion of
strong
centralizerintroduced in
[15].
Centralizers arisenaturally
inacomplex interpolation
scheme,in whichthe
interpolation
scale ofspaces share a commonL_{\infty}
‐module structure:in such\cdot
case, the space\mathcal{H} also
enjoys
thesameL_{\infty}‐module structure inthe form(u\cdot f)(z)
=u\cdot f(z)
. In thisway, the fundamental sequence of theinterpolation
scheme 0\rightarrow
\mathrm{k}\mathrm{e}\mathrm{r}$\delta$_{ $\theta$}
\rightarrow \mathcal{H} \rightarrow X_{ $\theta$} \rightarrow 0 is an exact sequence in the category of L_{\infty}-modules. In aninterpolation
schemestarting
with acouple
(X_{0}, X_{1})
of Köthe functionspaces, the map$\Omega$_{ $\theta$}=$\delta$_{ $\theta$}'B_{ $\theta$}
isacentralizeronX_{ $\theta$}.
Thus,
to some extent, it can be said that the centralizer is the core part of thequasi‐linear
map, orthat thequasi‐linear
map is alinearperturbation
of the centralizer.Moreover,
centralizerson Köthe functionspaces arequasi‐linear
maps[15,
Lemma4.2].
In additiontothis,
Kaltonproved
in[15,
Section4]
that everyself‐extension ofaKöthefunctionspaceXis
(equivalent to)
theextensioninducedby
acentralizeronX. Ofcoursethat twocentralizers$\Omega$_{1},
$\Omega$_{2}onXareequivalent
ifand
only
iftheinduced exact sequencesareequivalent,
whichhappens
ifandonly
therefore sense to define a centralizer $\Omega$ on X to be bounded when there exists
a constant C > 0 so that
\Vert $\Omega$(x)\Vert_{X}
\leqC\Vert x\Vert_{X}
for all x \in X; which inparticular
meansthat$\Omega$(x)\in X
for all x\in X. Two centralizers$\Omega$_{1},
$\Omega$_{2}aresaidtobeboundedly
equivalent
when$\Omega$_{1}-$\Omega$_{2}
isbounded.We have the
following outstanding
result of Kalton[16,
Theorem7.6]:
Theorem 1.1. LetX be a
separable superreflexive
Köthefunction
space. Thenthere exists a constantc
(depending
on the concavityof
aq‐concaverenorming
of
X)
such thatif
$\Omega$ \uparrow s areal centralizeronX with$\rho$( $\Omega$)\leq c
, then(1)
There is apairof
Köthefunction
spacesX_{0},
X_{1} such thatX=\{X_{0},
X_{1})_{1/2}
and$\Omega-\Omega$_{1/2}
is bounded.(2)
The spacesX_{0},
X_{1} areuniquely
determinedup toequivalent renorming.
An
example
is inorder:taking
thecouple
(P_{1}, \ell_{\infty})
, the mapB(x)
=x^{2(1-z)}
isa
homogeneous
bounded selection for the evaluation map$\delta$_{1/2}
:\mathcal{H}\rightarrow P_{2}
;hence theinterpolation procedure yields
thecentralizer-2$\Omega$_{2};while thecouple
(\ell_{p}, l_{p^{*}})
yields
-2(\displaystyle \frac{1}{p}-\frac{1}{p^{*}})$\Omega$_{2}
. Aswe seethetwocentralizersarethesameuptothe scalarfactor.That scalar factor howevercannot beoverlooked since it
actually
determines the endpoints X_{0}, X_{1}
intheinterpolation
scale.2. PERSPECTIVES
The research program in which \mathrm{I}'\mathrm{m} involved
contemplates
to advance in theunderstanding
of Kalton;s theorem. The basicquestion
canbestatedas:Problem 1. Determine towhat.extent Kaltonis theorem works for
general
inter‐polation
scales.A second line of research focusesonthe
underlying
connectionbetween Kalton\dot{J}\mathrm{s} theorem andthetheory
oftwistedsums. Observe that it iscontained in Kaltons theorem that the associated centralizer isbounded if andonly
if the two extremesof the scale coincide. In other
words,
theequality
notion thatcorresponds
todifferential processes seems to be that of bounded
equivalence
because one gets thattheinducedexact sequence is trivial ifandonly
if the associated centralizerisbounded. But that is not the notionthat appears via
homology:
the inducedsequence istrivial if and
only
if the associated exact sequence isequivalent
to0;i.e., thesum ofabounded
plus
alinearmap. So thequestion
isProblem 2. Does thereexistaKalton\mathrm{s}theorem valid for standard
equivalence
ofmaps
(bounded
plus
linear)
instead of boundedequivalence?
A sentence in
[6,
p.364]
suggests apositive
answer for Banach spaces withunconditional basis:
If
(Z_{0}, Z_{1})
are twosuper‐reflexive
sequence spaces andZ_{ $\theta$}
=[Z_{0}, Z_{1}]_{ $\theta$}
for
0 < $\theta$ < 1 is the usualinterpolation
spaceby
the Calderonmethod,
one can
define
a derivativedX_{ $\theta$}
which is a twisted sumX_{ $\theta$}\oplus_{ $\Omega$}X_{ $\theta$}
whichsplits
if
andonly if
Z_{1} =wZ_{0} for
someweight
sequence w =(w(n))
wherew(n)
>0
for
all n. These remarksfollow easily
from
the methodsof
[16].
Infact,
anappeal
to the methods of[4]
allows one to prove it. Here is an sketch ofproof.
Let(X_{0}, X_{1})
be aninterpolation couple
ofsuperreflexive
Banachspaceshaving
a commonunconditional basis(e_{n})
and let 0< $\theta$<1. If$\Omega$_{ $\theta$}istrivial thenX_{1}=wX_{0}
forsome
weight
sequence.Proof.
We need to show first that ifthe associated centralizer $\Omega$_{ $\theta$} is trivial then there is a functionf
\inP_{\infty}
sothat$\Omega$_{ $\theta$}(x)
—fx
is bounded.Indeed,
if there is a linearmap Lsothat$\Omega$_{ $\theta$}-Ltakesvalues inX_{ $\theta$}andisboundedthere. Thetechniques
in[8]
show that aftersomeaveraging
it ispossible
togetalinear map $\Lambda$ such thatevery unit ofP_{\infty}. From whereit follows that
$\Lambda$(ax) =a $\Lambda$(x)
forevery a\in P_{\infty}. Itisthenastandard fact that $\Lambda$must have the form
$\Lambda$(x)=fx
forsome functionf.
Andsinceone canassume$\Omega$_{ $\theta$}(e_{n})=0
,necessarily f\in\ell_{\infty}
. Therestissimple, just
pick
w=e^{f} and observe that the centralizerassociated tothecouple
(X_{0}, wX_{0})
isboundedly equivalent
to$\Omega$_{ $\theta$}
.Thus, by
Kaltonisuniqueness
theoremX_{1} =wX_{0}. \squareTurinng
tospecific properties,
thegeneral
questionis:Problem 3. Obtain conditions to detect that the differential process $\Omega$_{ $\theta$} has a
certainproperty.
The first property to look at is
triviality,
and thisbrings
us back toproblem
2. The next
interesting
property issingularity
(a
quasi‐linear
map is said to besingular
when no restriction to an infinite dimensionalsubspace
istrivial);
the Kalton‐Peckmap$\Omega$_{2} issingular.
Athorough study
onsingularity
canbe found in[8].
Super‐singularity
has however been consideredin[10].
After Problem 3
they
comequestions involving
thestability
of theproperties
of the differential process, a
problem
for which severalpartial
results have beenobtainedin
[12, 9].
Precisely:
Problem4. Local
stability.
Assume that forsome $\theta$the differential process$\Omega$_{ $\theta$}
istrivial
(in
somesense).
Does thereexistaneighborhood
V of $\theta$sothat $\Omega$_{ $\nu$}is also trivial for all $\nu$\in V?Problem 5. Global
stability.
Assume that for some $\theta$ the differentialprocess$\Omega$_{ $\theta$}
istrivial(in
somesense),
Doesthereit follows that $\Omega$_{ $\nu$}is trivial for all\mathrm{v}? Sinceinterpolation
methodsthere aremany, the next set ofquestions
isProblem6. Other
interpolation
methods. Set and solve theproblems
above for otherinterpolation
methods.In
particular,
the paper[3]
clearly
shows thathomological techniques provide
a natural frameto iterate differential
complex
processes. Inparticular,
to obtain twistedsumsof twistedsums, and so on. It is natural toaskProblem6. Iteration. Doesasimilariteration process existfor other
interpola‐
tionmethods?
REFERENCES
[1] J.Bergh,J.Löfström, Interpolationspaces. An introduction. Springer‐Verlag,(1976).
[2] $\Gamma$.Cabello,\mathrm{J}.\mathrm{M}. $\Gamma$.Castillo,S.Goldstein,JesúsSuarez, Twistingnoncommutative L_{p}\rightarrowspaces,
Adv. inMath.294 (2016)454—488.
[3] F.Cabello, J.M.F.Castillo, N.J.Kalton, Complex interpolationandtwisted twisted Htlbert spaces.Pacific J. Math. (2015)276(2015)287‐ 307.
[4] $\Gamma$. CaUello Sánchez, \mathrm{J}.\mathrm{M}. $\Gamma$. Castillo, J. Suárez, Onstrictly singularnonlinearcgntralizers,
Nonlinear Anal.75 (2012),3313‐3321.
[5] M.J. Carro, J. Cerdá and J. Soria, Commutators andinterpolation methods, Ark. Mat. 33
(1995)199‐216.
[6] P.G.Casazza,N.J.Kalton, Unconditional basesandunconditionalfinite‐dimensionaldecom‐
positionsinbanachspaces, IsraelJ. Math. 95(1996), 349‐373.
[7] J. M. F. Castillo, Simple twist of K., inKalton Selectavol.2., pp. 251‐254, Contemporary Mathematicians,Birkhauser (2016).
[8] J.M.F. Castillo, V. Ferenczi, M. Gonzalez, Singular exact sequencesgenerated by complex
interpolation,Trans.Amer. Math. Soc. (2016) (in press).
[9] J.M.F.Castillo,W.Correa,V. Ferenczi,M.Gonzalez, Stabilitypropertiesoftwistedsumsof
Banach spacesgenerated by complex interpolations, to appear.
[10] J.M.F. Castillo,W. Cuellar,V. Ferenczi,Y.Moreno, Complexstructures ontwisted Hilbert spaces, to appear inIsrael J.Math.
[11] \mathrm{J}.\mathrm{M}. $\Gamma$.Castillo,M.González, Three‐space problemsinBanachspacetheory, SpringerLecture
[12] M. Cwikel, B. Jawerth, M.Milman, R. Rochberg, Differentntestirnates and commutators ininterpolation theory. In AnalysisatUr.bana II,London MathematicalSociety, Lecture NoteSeries, (E.R.Berkson,N.T.Peck,and J. Uhl,Eds Vol. 138,pp. 170‐220,Cambridge Univ.Press,Cambridge, 1989.
[13] P. Enflo, J. Lindenstrauss and G. Pisier, On the three‐space problem forHilbert spaces, Math.Scand.36 (1975), 199‐210.
[14] N.J.Kalton, Thethree‐space problem for locally boun\grave{d}edF‐spaces,Compo. Math. 37(1978), 243‐276.
[15] N. J.Kalton, Nonlinearcommutatorsxninterpolation theory, Mem. Amer.Math.Soc.,385
(1988).
[16} N.J.Kalton, Differentials of complex interpolationprocessesforKöthefunctionspaces, Trans. Amer. Math. Soc.333(1992)479‐529.
[17] N.J.Kalton,S.Montgomery‐Smith, Interpolation ofBanachspaces,Chapter36inHandbook oftheGeometryof Banachspaces, W.B. Johnsonand J. Lindenstrauss eds. pp. 1131‐1175.
[18] N.J. Kaltonand N.T.Peck, Twistedsumsofsequence spaces and thethreespaceproblem,
Trans. Amer. Math. Soc.255(1979)1‐30.
[19] R.RochbergandG.Weiss, Derivativesof analytic families ofBanach spaces, Ann.of Math. 118(1983)315‐347.
DEPARTAMENTO DE MATEMÁTICAS, UNIVERSIDAD DE EXTREMADURA, AVDA. DE ELVAS \mathrm{s}/\mathrm{N}, 06011BADAJOZ