Our main result is as follows. Recall that there exists a unique mapping
${u}_{0}\mapsto S{u}_{0}(t,x)$
, defined for
${u}_{0}\in {C}^{\infty}$
, which for all sufficiently large
$s$
extends to a uniformly continuous mapping from
${H}^{s}(\mathbb{T})$
to
${C}^{0}\left(\right[0,\infty ),{H}^{s}(\mathbb{T}\left)\right)\cap {C}^{1}\left(\right[0,\infty ),{H}^{s2}(\mathbb{T}\left)\right)$
, such that
$S{u}_{0}$
is a solution of the modified Cauchy problem NLS* .
${C}^{\infty}(\mathbb{T})$
is of course a dense subset of
${\mathcal{\mathscr{H}}}^{s,p}$
for any
$p\in [1,\infty )$
.
Theorem 1.1.
For any
$p\in [1,\infty )$
, any
$s\ge 0$
, and any
$R<\infty $
, there exists
$\tau >0$
for which the solution mapping
$S$
extends by continuity to a uniformly continuous mapping from the ball centered at
$0$
of radius
$R$
in
${\mathcal{\mathscr{H}}}^{s,p}(\mathbb{T})$
to
${C}^{0}\left(\right[0,\tau ],{\mathcal{\mathscr{H}}}^{s,p}(\mathbb{T}\left)\right)$
.
For the unmodified equation this has the following consequence. Denote by
${H}_{c}^{0}={H}_{c}^{0}(\mathbb{T})$
the set of all
$f\in {H}^{0}$
such that
$\parallel f{\parallel}_{{L}^{2}}=c$
. Denote by
${S}^{\prime}{u}_{0}$
the usual solution [
2]
of the unmodified Cauchy problem NLS with initial datum
${u}_{0}$
, for
${u}_{0}\in {H}^{0}$
.
Corollary 1.2.
Let
$p\in [1,\infty )$
and
$s\ge 0$
. For any
$R<\infty $
there exists
$\tau >0$
such that for any finite constant
$c>0$
, the mapping
${H}_{c}^{0}\ni {u}_{0}\mapsto {S}^{\prime}{u}_{0}$
is uniformly continuous as a mapping from
${H}_{c}^{0}$
intersected with the ball centered at
$0$
of radius
$R$
in
${\mathcal{\mathscr{H}}}^{s,p}$
, equipped with the
${\mathcal{\mathscr{H}}}^{s,p}$
norm, to
${C}^{0}\left(\right[0,\tau ],{\mathcal{\mathscr{H}}}^{s,p}(\mathbb{T}\left)\right)$
.
The unpublished result of the author and Erdogan shows that for initial data in
${L}^{2}$
, for which the solution is known to exist globally in time,
$\parallel u\left(t\right){\parallel}_{{\mathcal{\mathscr{H}}}^{p}}\le C\parallel {u}_{0}{\parallel}_{{\mathcal{\mathscr{H}}}^{p}}$
uniformly for all
$t\in [0,\infty )$
, provided that
$\parallel {u}_{0}{\parallel}_{{\mathcal{\mathscr{H}}}^{p}}$
is sufficiently small. This result, once published, will combine with Theorem 1.1 to yield global wellposedness for sufficiently small data.
The following result quantifies the relation between the nonlinear evolution
NLS* ond the corresponding linear Cauchy problem
$$\begin{array}{c}\{\begin{array}{cc}& i{v}_{t}+{v}_{xx}=0\\ & v(0,x)={u}_{0}\left(x\right).\end{array}\end{array}$$ 
(1.2)

Proposition 1.3.
Let
$R<\infty $
and
$p\in [1,\infty )$
. Let
$q>p/3$
also satisfy
$q\ge 1$
.
Then there exist
$\tau ,\varepsilon >0$
and
$C<\infty $
such that for any initial datum
${u}_{0}$
satisfying
$\parallel {u}_{0}{\parallel}_{{\mathcal{\mathscr{H}}}^{p}}\le R$
, the solutions
$u=S{u}_{0}$
of NLS* and
$v$
of 1.2 satisfy
$$\begin{array}{c}\parallel u(t,\cdot )v(t,\cdot ){\parallel}_{{\mathcal{\mathscr{H}}}^{q}}\le C{t}^{\varepsilon}\text{for all}t\in [0,\tau ].\end{array}$$ 
(1.3)

Here
$u$
the solution defined by approximating
${u}_{0}$
by elements of
${C}^{\infty}$
and passing to the limit. Thus for
$p>1$
the nonlinear terms are in a sense smoother than the linear evolution.
Our next result indicates that the function
$u(t,x)$
defined by the limiting procedure of Theorem 1.1 is a solution of the differential equation in a more natural sense than merely being a limit of smooth solutions. Define Fourier truncation operators
${T}_{N}$
, acting on
${\mathcal{\mathscr{H}}}^{s,p}(\mathbb{T})$
, by
$\widehat{{T}_{N}f}\left(n\right)=0$
for all
$\leftn\right>N$
, and
$=\widehat{f}\left(n\right)$
whenever
$\leftn\right\le N$
.
${T}_{N}$
acts also on functions
$v(t,x)$
by acting on
$v(t,\cdot )$
for each time
$t$
separately. We denote by
$S\left({u}_{0}\right)$
the limiting function whose existence, for nonsmooth
${u}_{0}$
, is established by Theorem 1.1 .
Proposition 1.4.
Let
$p\in [1,\infty )$
,
$s\ge 0$
, and
${u}_{0}\in {\mathcal{\mathscr{H}}}^{s,p}$
. Write
$u=S\left({u}_{0}\right)$
. Then for any
$R<\infty $
there exists
$\tau >0$
such that whenever
$\parallel {u}_{0}{\parallel}_{{\mathcal{\mathscr{H}}}^{s,p}}\le R$
,
$\mathcal{N}u(t,x)=\left(\rightu{}^{2}2\mu \left(\rightu{}^{2}\left)\right)u$
exists in the sense that
$$\begin{array}{c}{lim}_{N\to \infty}\mathcal{N}\left({T}_{N}u\right)(t,x)\text{exists in the sense of distributions in}{C}^{0}\left(\right(0,\tau ),{\mathcal{D}}^{\prime}(\mathbb{T}\left)\right).\end{array}$$ 
(1.4)

Moreover if
$\mathcal{N}\left(u\right)$
is interpreted as this limit, then
$u=S\left({u}_{0}\right)$
satisfies NLS* in the sense of distributions in
$(0,\tau )\times \mathbb{T}$
.
More generally, the same holds for any sequence of Fourier multipliers of the form
$\widehat{{T}_{\nu}f}\left(n\right)={m}_{\nu}\left(n\right)\widehat{f}\left(n\right)$
where each sequence
${m}_{\nu}$
is finitely supported,
${sup}_{\nu}\parallel {m}_{\nu}{\parallel}_{{\ell}^{\infty}}<\infty $
, and
${m}_{\nu}\left(n\right)\to 1$
as
$\nu \to \infty $
for each
$n\in \mathbb{Z}$
; the limit is of course independent of the sequence
$\left({m}_{\nu}\right)$
. Making sense of the nonlinearity via this limiting procedure is connected with general theories of multiplication of distributions [1] ,[6] , but the existence here of the limit over all sequences
$\left({m}_{\nu}\right)$
gives
$u$
stronger claim to the title of solution than in the general theory.
Unlike the fixed point method, our proof yields no uniqueness statement corresponding to these existence results. But this failing is unavoidable; for all
$p>2$
, solutions of the Cauchy problem in the class
${C}^{0}\left(\right[0,\tau ],{\mathcal{\mathscr{H}}}^{p})$
, in the sense of Proposition 1.4 , are not unique [
4]
.
1.5 Method
Define the partial Fourier transform
$$\begin{array}{c}\widehat{u}(t,n)=\left(2\pi {)}^{1}{\int}_{\mathbb{T}}{e}^{inx}u\right(t,x)dx.\end{array}$$ 
(1.5)

Our approach is to regard the partial differential equation as an infinite coupled nonlinear system of ordinary differential equations for these Fourier coefficients, to express the solution as a power series in the initial datum
$$\begin{array}{c}\widehat{u}(t,n)={\sum}_{k=0}^{\infty}{\hat{A}}_{k}\left(t\right)(\widehat{{u}_{0}},\cdots ,\widehat{{u}_{0}})\end{array}$$ 
(1.6)

where each
${\hat{A}}_{k}\left(t\right)$
is a bounded multilinear operator^{2 }
from a product of
$k$
copies of
${\mathcal{\mathscr{H}}}^{s,p}$
to
${\mathcal{\mathscr{H}}}^{s,p}$
, to show that the individual terms
${\hat{A}}_{k}\left(t\right)(\widehat{{u}_{0}},\cdots ,\widehat{{u}_{0}})$
are welldefined, and to show that the formal series converges absolutely in
${C}^{0}(\mathbb{R},{\mathcal{\mathscr{H}}}^{s,p})$
to a solution in the sense of 1.4 . The case
$s\ge 0$
follows from a very small modification of the analysis for
$s=0$
, so we discuss primarily
$s=0$
, indicating the necessary modifications for
$s>0$
at the end of the paper.
The analysis is quite elementary, much of the paper being devoted to setting up the definitions and notation required to describe the operators
${\hat{A}}_{k}\left(t\right)$
. A single number theoretic fact enters the discussion: the number of factorizations of an integer
$n$
as a product of two integer factors is
$O\left({n}^{\delta}\right)$
as
$n\to \infty $
, for all
$\delta >0$
; this same fact was used in a more sophisticated way by Bourgain [
2]
.
The author is grateful to J. Bourgain, C. Kenig, H. Koch, and D. Tataru for invitations to conferences that stimulated this work, and to Betsy Stovall for thorough proofreading of the manuscript.
2 A system of coupled ordinary differential equations
2.1 General discussion
Define
$$\begin{array}{c}\sigma (j,k,l,n)={n}^{2}{j}^{2}+{k}^{2}{l}^{2}.\end{array}$$ 
(2.1)

It factors as
$$\begin{array}{c}\sigma (j,k,l,n)=2(nj)(nl)=2(kl)(kj)\text{provided that}jk+l=n\text{}.\end{array}$$ 
(2.2)

Written in terms of Fourier coefficients
${\widehat{u}}_{n}\left(t\right)=\widehat{u}(t,n)$
, the equation
$i{u}_{t}+{u}_{xx}+\omega \left(u{}^{2}2\mu (\leftu{}^{2}\right)\right)u=0$
becomes
$$\begin{array}{c}i\frac{d{\widehat{u}}_{n}}{dt}{n}^{2}{\widehat{u}}_{n}+\omega {\sum}_{jk+l=n}{\widehat{u}}_{j}\overline{{\widehat{u}}_{k}}{\widehat{u}}_{l}2\omega {\sum}_{m}{\widehat{u}}_{m}{}^{2}{\widehat{u}}_{n}=0.\end{array}$$ 
(2.3)

Here the first summation is taken over all
$(j,k,l)\in {\mathbb{Z}}^{3}$
satisfying the indicated identity, and the second over all
$m\in \mathbb{Z}$
. Substituting
$$\begin{array}{c}{a}_{n}\left(t\right)={e}^{i{n}^{2}t}\widehat{u}(t,n),\end{array}$$ 
(2.4)

2.3 becomes
$$\begin{array}{c}\frac{d{a}_{n}}{dt}=i\omega {\sum}_{jk+l=n}^{*}{a}_{j}{\overline{a}}_{k}{a}_{l}{e}^{i\sigma (j,k,l,n)t}i\omega {a}_{n}{}^{2}{a}_{n}.\end{array}$$ 
(2.5)

where the notation
${\sum}_{jk+l=n}^{*}$
means that the sum is taken over all
$(j,k,l)\in {\mathbb{Z}}^{3}$
for which neither
$j=n$
nor
$l=n$
. This notational convention will be used throughout the discussion.
The effect of the term
$2\omega \mu \left(\rightu{}^{2})u$
in the modified differential equation NLS* is to cancel out a term
$2i\omega \left({\sum}_{m}\right{a}_{m}{}^{2}){a}_{n}$
, which would otherwise appear on the righthand side of 2.5 .
Reformulated as an integral equation,
2.5 becomes
$$\begin{array}{c}{a}_{n}\left(t\right)={a}_{n}\left(0\right)+i\omega {\sum}_{jk+l=n}^{*}{\int}_{0}^{t}{a}_{j}\left(s\right){\overline{a}}_{k}\left(s\right){a}_{l}\left(s\right){e}^{i\sigma (j,k,l,n)s}dsi\omega {\int}_{0}^{t}\left{a}_{n}\right(s\left){}^{2}{a}_{n}\right(s)ds.\end{array}$$ 
(2.6)

However, in deriving 2.6 from 2.5 , we have interchanged the integral over
$[0,t]$
with the summation over
$j,k,l$
without any justification. 2.6 is fully equivalent to
$$\begin{array}{c}\widehat{u}(t,n)=\widehat{{u}_{0}}\left(n\right)i{n}^{2}{\int}_{0}^{t}\widehat{u}(s,n)ds+i\omega {\sum}_{jk+l=n}^{*}{\int}_{0}^{t}\widehat{u}(s,j)\overline{\widehat{u}(s,k)}\widehat{u}(s,l)dsi\omega {\int}_{0}^{t}\left\widehat{u}\right(s,n\left){}^{2}\widehat{u}\right(s,n)ds.\end{array}$$ 
(2.7)

Substituting for
${a}_{j},{a}_{k},{a}_{l}$
in the righthand side of 2.6 by means of the equation itself yields
$$\begin{array}{cc}{a}_{n}\left(t\right)& ={a}_{n}\left(0\right)+i\omega {\sum}_{jk+l=n}^{*}{a}_{j}\left(0\right){\overline{a}}_{k}\left(0\right){\overline{a}}_{l}\left(0\right){\int}_{0}^{t}{e}^{i\sigma (j,k,l,n)s}dsi\omega \left{a}_{n}\right(0\left){}^{2}{a}_{n}\right(0){\int}_{0}^{t}1ds\end{array}$$ 
(2.8)

$$\begin{array}{cc}& +\text{additional terms}.\end{array}$$  
$$\begin{array}{cc}& ={a}_{n}\left(0\right)\left(1i\omega t\left{a}_{n}\right(0){}^{2}\right)+\frac{1}{2}\omega {\sum}_{jk+l=n}^{*}\frac{{a}_{j}\left(0\right){\overline{a}}_{k}\left(0\right){a}_{l}\left(0\right)}{(nj)(nl)}({e}^{i({n}^{2}{j}^{2}+{k}^{2}{l}^{2})t}1)\end{array}$$  
$$\begin{array}{cc}& +\text{additional terms.}\end{array}$$  
$$\begin{array}{}\end{array}$$  
We recognize
$1i\omega t\left{a}_{n}\right(0){}^{2}$
as a Taylor polynomial for
$exp(i{a}_{n}\left(0\right){}^{2}t)$
, but for our purposes it will not be necessary to exploit this by recombining terms, and in particular we will not exploit the coefficient
$i$
which makes this exponential unimodular.
2.2 A sample term
One representative additional term is
(2.9)
(
i
ω
)
4
∑
j
1
−
j
2
+
j
3
=
n
*
∑
m
1
1
−
m
2
1
+
m
3
1
=
j
1
*
∑
m
1
2
−
m
2
2
+
m
3
2
=
j
2
*
∑
m
1
3
−
m
2
3
+
m
3
3
=
j
3
*
∫
0
≤
r
1
,
r
2
,
r
3
≤
s
≤
t
a
m
1
1
(
r
1
)
a
¯
m
2
1
(
r
1
)
a
m
3
1
(
r
1
)
a
¯
m
1
2
(
r
2
)
a
m
2
2
(
r
2
)
a
¯
m
3
2
(
r
2
)
a
m
1
3
(
r
3
)
a
¯
m
2
3
(
r
3
)
a
m
3
3
(
r
3
)
e
i
σ
(
j
1
,
j
2
,
j
3
,
n
)
s
e
i
σ
(
m
1
1
,
m
2
1
,
m
3
1
,
j
1
)
r
1
e
−
i
σ
(
m
1
2
,
m
2
2
,
m
3
2
,
j
2
)
r
2
e
i
σ
(
m
1
3
,
m
2
3
,
m
3
3
,
j
3
)
r
3
d
r
1
d
r
2
d
r
2
d
s
.
Substituting via 2.6 for each coefficient
$a$
yields a main term
$$\begin{array}{c}(i\omega {)}^{4}{\sum}_{({m}_{k}^{i}{)}_{1\le i,k\le 3}}^{*}\mathcal{\mathcal{I}}(t,\left({m}_{k}^{i}{)}_{1\le i,k\le 3}{)}^{3}{\prod}_{i,j=1}{a}_{{m}_{j}^{i}}^{*}\right(0)\end{array}$$ 
(2.10)

plus higherdegree terms, where the superscript
$*$
indicates here that the sum is taken over only certain
$({m}_{k}^{i}{)}_{1\le i,k\le 3}\subset {\mathbb{Z}}^{9}$
(more precisely, over most of a copy of
${\mathbb{Z}}^{8}$
affinely embedded in
${\mathbb{Z}}^{9}$
),
${a}_{{m}_{j}^{i}}^{*}\left(0\right)={a}_{{m}_{j}^{i}}\left(0\right)$
if
$i+j$
is even and
$=\overline{{a}_{{m}_{j}^{i}}\left(0\right)}$
if
$i+j$
is odd, and
$$\begin{array}{c}\mathcal{\mathcal{I}}(t,({m}_{k}^{i}{)}_{1\le i,k\le 3})={\int}_{0\le {r}_{1},{r}_{2},{r}_{3}\le s\le t}{e}^{i\theta (t,s,{r}_{1},{r}_{2},{r}_{3},\{{m}_{j}^{i}:1\le i,j\le 3\left\}\right)}d{r}_{1}d{r}_{2}d{r}_{2}ds,\end{array}$$ 
(2.11)

with
$$\begin{array}{c}\theta (t,s,{r}_{1},{r}_{2},{r}_{3},({m}_{j}^{i}{)}_{1\le i,j\le 3})=\sigma ({j}_{1},{j}_{2},{j}_{3},n)s+{\sum}_{i=1}^{3}(1{)}^{i+1}\sigma ({m}_{1}^{i},{m}_{2}^{i},{m}_{3}^{i},{j}_{i}){r}_{i};\end{array}$$ 
(2.12)

here
${j}_{1},{j}_{2},{j}_{3},n$
are defined as functions of
$\left({m}_{j}^{i}\right)$
by the equations governing the sums in 2.9 . Continuing in this way yields formally an infinite expansion for the sequence
$\left({a}_{n}\right(t){)}_{n\in \mathbb{Z}}$
in terms of multilinear expressions in the initial datum
$\left({a}_{n}\right(0\left)\right)$
. This expansion is doubly infinite; the single (and relatively simple) term 2.10 is for instance an infinite sum over most of a copy of
${\mathbb{Z}}^{8}$
for each
$n$
.
The discussion up to this point has been purely formal, with no justification of convergence.
In the next section we will begin to describe the terms in this expansion systematically.
3 Trees
On a formal level
$a\left(t\right)=\left({a}_{n}\right(t){)}_{n\in \mathbb{Z}}$
equals an infinite sum
${\sum}_{k=1}^{\infty}{A}_{k}\left(t\right)\left(a\right(0),a(0),a(0),\cdots )$
where each
${A}_{k}\left(t\right)$
is a sum of finitely many multilinear operators, each of degree
$k$
. We now describe a class of trees which will be used both to name, and to analyze, these multilinear operators.
Definition 3.1.
A tree
$T$
is a finite partially ordered set with the following properties:

(1)
Whenever
${v}_{1},{v}_{2},{v}_{3},{v}_{4}\in T$
and
${v}_{4}\le {v}_{2}\le {v}_{1}$
and
${v}_{4}\le {v}_{3}\le {v}_{1}$
, then either
${v}_{2}\le {v}_{3}$
or
${v}_{3}\le {v}_{2}$
.

(2)
There exists a unique element
${v}_{0}\in T$
satisfying
${v}_{0}\ge v$
for all
$v\in T$
.

(3)
Each
$v\in T$
has either three children, or no children;
$w$
is said to be a child of
$v$
if
$w<v$
and if there exists no
$u\in T$
satisfying
$w<u<v$
.

(4)
For each
$v\in T$
there is given an element of
$\{\pm 1\}$
, denoted
${\pm}_{v}$
.
Definition 3.2.
Elements of
$T$
are called nodes. A terminal node is one with zero children. The maximal element of
$T$
is called its root node. For any
$u\in T$
,
${T}_{u}=\{v\in T:v\le u\}$
is a tree, with root node
$u$
.
${T}^{\infty}$
denotes the set of all terminal nodes of
$T$
, while
${T}^{0}=T\backslash {T}^{\infty}$
denotes the set
${T}^{0}$
of all nonterminal nodes. The three children of any
$v\in {T}^{0}$
are denoted by
$(v,1),(v,2),(v,3)$
.
The number
$\leftT\right$
of nodes of a tree is of the form
$1+3k$
for some nonnegative integer
$k$
.
The number of terminal nodes is then
$$\begin{array}{c}\left{T}^{\infty}\right=1+2k=\frac{2}{3}\leftT\right+\frac{1}{3}.\end{array}$$ 
(3.1)

Definition 3.3.
An ornamented tree is a tree
$T$
together with the following additional structure:

(1)
Associated to each node
$v\in T$
is copy of
$\mathbb{Z}$
, indexed by the variable
${j}_{v}$
.

(2)
There is given a partition of the set of all nonterminal nodes of
$T$
into two disjoint classes, called simple nodes and general nodes. Terminal nodes are neither simple nor general.

(3)
For each nonterminal node
$v\in {T}^{0}$
, and each
$i\in \{1,2,3\}$
such that the child
$(v,i)$
is nonterminal, there is given a coefficient
${\varepsilon}_{v,i}\in \{1,0,1\}$
.

(4)
Associated to each node
$v\in T$
is a
$\mathbb{Z}$
valued function
${\rho}_{v}$
of
$\mathbf{j}=({j}_{u}{)}_{u\in T}$
, defined by
$$\begin{array}{c}{\rho}_{v}(\mathbf{j})=0\text{if}v\in {T}^{\infty}\text{}\end{array}$$ 
(3.2)

and
$$\begin{array}{c}{\rho}_{v}(\mathbf{j})=\sigma ({j}_{(v,1)},{j}_{(v,2)},{j}_{(v,3)},{j}_{v})+{\sum}_{i=1}^{3}{\varepsilon}_{v,i}{\rho}_{(v,i)}\text{if}v\in {T}^{0}\text{}.\end{array}$$ 
(3.3)

${\rho}_{v}(\mathbf{j})$
actually depends only on
$\{{j}_{u},{\varepsilon}_{u,i}:u\le v\}$
. We will use the symbol
$T$
to denote both the ornamented tree and the underlying tree, and will often write
${\rho}_{v}$
instead of
${\rho}_{v}(\mathbf{j})$
.
Definition 3.4.
Let
$T$
be a tree.
$\mathcal{J}\left(T\right)\subset {\mathbb{Z}}^{T}$
denotes the set of all
$\mathbf{j}=({j}_{v}{)}_{v\in T}$
satisfying the restrictions
$$\begin{array}{cc}& {j}_{v}={j}_{(v,1)}{j}_{(v,2)}+{j}_{(v,3)}\text{for every}v\in {T}^{0}\text{}\end{array}$$ 
(3.4)

$$\begin{array}{cc}& \{{j}_{v},{j}_{(v,2)}\}\cap \{{j}_{(v,1)},{j}_{(v,3)}\}=\varnothing \text{for every general node}v\in {T}^{0}\text{}\end{array}$$ 
(3.5)

$$\begin{array}{cc}& {j}_{v}={j}_{(v,i)}\text{for all}i\in \{1,2,3\}\text{for every simple node}v\in {T}^{0}\text{}.\end{array}$$ 
(3.6)

$$\begin{array}{}\end{array}$$  
Definition 3.5.
Let
$T$
be any tree.
${\sigma}_{w}:\mathcal{J}\left(T\right)\to \mathbb{Z}$
denotes the function
${\sigma}_{w}(\mathbf{j})=0$
if
$w$
is terminal, and
${\sigma}_{w}(\mathbf{j})={j}_{w}^{2}{j}_{(w,1)}^{2}+{j}_{(w,2)}^{2}{j}_{(w,3)}^{2}$
if
$w$
is nonterminal.
Let
$\delta ,{c}_{0}>0$
be sufficiently small positive numbers, to be chosen later. The following key definition involves these quantities.
Definition 3.6.
Given an ornamented tree
$T$
and
$\mathbf{j}\in \mathcal{J}\left(T\right)$
, we say that a node
$v\in T$
is frozen if
$v$
is nonterminal and
$$\begin{array}{c}\left{\rho}_{v}\right(\mathbf{j}\left)\right\le {c}_{0}\left\sigma \right({j}_{(v,1)},{j}_{(v,2)},{j}_{(v,3)},{j}_{v}){}^{1\delta}.\end{array}$$ 
(3.7)

If
$v$
is not frozen, then
$v$
is said to be alive. A nonterminal node
$v$
is said to be exceptional if
${\rho}_{v}(\mathbf{j})=0$
.
Whether
$v$
is frozen depends on the values of
${j}_{u}$
for all nodes
$u\le v$
, as well as on
${\varepsilon}_{u,i}$
for all nonterminal
$u\le v$
, not merely on the structure of
$T$
; a nonterminal node will be frozen for some
$\mathbf{j}$
, but alive for others. Thus it would be more felicitous to say that a pair
$(v,\mathbf{j})$
is frozen, rather than a node
$v$
.
Exceptional nodes are of course frozen. If
$v\in {T}^{0}$
is a general node all three of whose children of
$v$
are terminal, then
$v$
cannot be exceptional, for
${\rho}_{v}=\sigma ({j}_{(v,1)},{j}_{(v,2)},{j}_{(v,3)},{j}_{v})=2({j}_{v}{j}_{(v,1)})({j}_{v}{j}_{(v,3)})$
cannot vanish, by 3.5 . But if
$v$
has at least one nonterminal child, then nothing prevents
${\rho}_{v}$
from vanishing, and if
$v$
is a simple node all of whose children are terminal, then
$v$
is certainly exceptional.
Definition 3.7.
A weathered ornamented tree
$(T,{T}^{\prime})$
is an ornamented tree
$T$
together with a subset
${T}^{\prime}\subset {T}^{0}$
and the collection
$$\begin{array}{c}\mathcal{J}(T,{T}^{\prime})=\{\mathbf{j}\in \mathcal{J}(T):\text{}v\in T\text{is frozen if and only if}v\in {T}^{\prime}\text{.}\}\end{array}$$ 
(3.8)

4 Multilinear operators associated to trees
Definition 4.1.
Let
$T$
be any tree, and let
$t\in \mathbb{R}$
. The associated tree coefficients are
$$\begin{array}{c}{\mathcal{\mathcal{I}}}_{T}(t,\mathbf{j})={\int}_{\mathcal{\mathcal{R}}(T,t)}{\prod}_{u\in {T}^{0}}{e}^{{\pm}_{u}i\omega {\sigma}_{u}(\mathbf{j}){t}_{u}}d{t}_{u}\end{array}$$ 
(4.1)

where
$$\begin{array}{c}\mathcal{\mathcal{R}}(T,t)=\left\{\right({t}_{u}{)}_{u\in {T}^{0}}:0\le {t}_{u}\le {t}_{{u}^{\prime}}\le t\text{whenever}u\le {u}^{\prime}\}.\end{array}$$ 
(4.2)

The following upper bounds for the coefficients
${\mathcal{\mathcal{I}}}_{T}(t,\mathbf{j})$
are the only information concerning them that will be used in the analysis.
Lemma 4.1.
Let
$T$
be any tree, and let
$\mathbf{j}\in \mathcal{J}\left(T\right)$
. Then for all
$t\in [0,1]$
,
$$\begin{array}{cc}& \left{\mathcal{\mathcal{I}}}_{T}\right(t,\mathbf{j}\left)\right\le {t}^{\left{T}^{0}\right}\end{array}$$ 
(4.3)

$$\begin{array}{cc}\text{and}& \left{\mathcal{\mathcal{I}}}_{T}\right(t,\mathbf{j}\left)\right\le {2}^{\leftT\right}{\sum}_{\left({\varepsilon}_{u,i}\right)}{\prod}_{w\in {T}^{0}}\langle {\rho}_{w}(\mathbf{j}){\rangle}^{1}.\end{array}$$ 
(4.4)

$$\begin{array}{}\end{array}$$  
The notation
$\langle x\rangle $
means
$(1+x{}^{2}{)}^{1/2}$
. The sum here is taken over all of the
${3}^{\left{T}^{0}\right}$
possible choices of
${\varepsilon}_{u,i}\in \{0,1,1\}$
; these choices in turn determine the quantities
${\rho}_{w}$
. This lemma will be proved in § 6 .
Definition 4.2.
Let
$T$
be any ornamented tree. The tree operator
${\mathcal{S}}_{T}\left(t\right)$
associated to
$T$
is for each
$t\in \mathbb{R}$
the multilinear operator that maps
$({x}_{v}{)}_{v\in {T}^{\infty}}$
, where each
${x}_{v}$
is a sequence of complex numbers, to the sequence of complex numbers
$$\begin{array}{c}{\mathcal{S}}_{T}\left(t\right)\left({x}_{v}{)}_{v\in {T}^{\infty}}\right(n)={\sum}_{\mathbf{j}\in \mathcal{J}\left(T\right):{j}_{{v}_{0}}=n}{\mathcal{\mathcal{I}}}_{T}(t,\mathbf{j}\left){\prod}_{w\in {T}^{\infty}}{x}_{w}\right({j}_{w}).\end{array}$$ 
(4.5)

5 Formalities
With all these definitions and notations in place, we can finally formulate the conclusion of the discussion in § 2 ; proofs will be supplied later.