1 s2.0 S0168007201000355 Main PDF

Annals of Pure and Applied Logic 111 (2001) 61–113
www.elsevier.com/locate/apal
Logarithmic-exponential series
Lou van den Driesa; ∗;1 , Angus Macintyreb; 2 , David Markerc; 3
a University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
b University of Edinburgh, UK
c University of Illinois at Chicago, USA
Abstract
We extend the *eld of Laurent series over the reals in a canonical way to an ordered di+erential
*eld of “logarithmic-exponential series” (LE-series), which is equipped with a well behaved
exponentiation. We show that the LE-series with derivative 0 are exactly the real constants, and
we invert operators to show that each LE-series has a formal integral. We give evidence for
the conjecture that the *eld of LE-series is a universal domain for ordered di+erential algebra
in Hardy *elds. We de*ne composition of LE-series and establish its basic properties, including
the existence of compositional inverses. Various interesting sub*elds of the *eld of LE-series
are also considered. c 2001 Elsevier Science B.V. All rights reserved.
1. Introduction
The ordered di+erential *eld R((x−1 )) of formal Laurent series in descending powers
of x over R consists of all series of the form
f(x) = ak xk + ak−1 xk−1 + · · · + a1 x + a0 + a−1 x−1 + a−2 x−2 + · · ·

in*nite part of f *nite part of f
with real coe8cients ai . We have x¿R for the ordering, and x = 1 for the derivation.
There is a formal composition operation that assigns to f(x); g(x) ∈ R((x−1 )) with g¿R
an element f(g(x)) ∈ R((x−1 )). However, x−1 has no antiderivative in R((x−1 )). There
is also no reasonable exponential operation de*ned on all of R((x−1 )), since such an
∗ Corresponding author.
E-mail addresses: vddries@math.uiuc.edu (L. van den Dries), angus@maths.ed.ac.uk (A. Macintyre),
marker@math.uic.edu (D. Marker)
1 Partially supported by the National Science Foundation.
2 Partially supported by an EPSRC Senior Research Fellowship.
3 Partially supported by the National Science Foundation and an AMS Centennial Fellowship.
0168-0072/01/$ - see front matter c 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 1 6 8 - 0 0 7 2 ( 0 1 ) 0 0 0 3 5 - 5
62 L. van den Dries et al. / Annals of Pure and Applied Logic 111 (2001) 61–113
operation would have to satisfy exp x¿xn for all n. Exponentiation only makes sense
for the *nite elements of R((x−1 )):
∞
1
exp(a0 + a−1 x−1 + a−2 x−2 + · · ·) = ea0 (a−1 x−1 + a−2 x−2 + · · ·)n
n!
n=0
= ea0 (1 + b1 x−1 + b2 x−2 + · · ·)
for certain reals b1 ; b2 ; : : : .

To remove these defects we extend here R((x−1 )) to an ordered di+erential *eld
R((x−1 ))LE , the *eld of logarithmic-exponential series (or LE-series for short) over
R. Its elements are in*nite series of “LE-monomials” arranged from left to right in
decreasing order and multiplied by real coe8cients, for example
x 2
ee − 3ex + 5x1=2 − log x + 1 + x−1 + x−2 + x−3 + · · · + e−x + x−1 e−x :
The reversed order type of the set of LE-monomials that occur in a given LE-
series can be any countable ordinal. (For the series displayed it is ! + 2.) Such series
occur, for example, in solving implicit equations of the form P(x; y; e x ; ey ) = 0 for
y as x → + ∞, where P is a (nonconstant) polynomial in 4 variables over R, see
[5, 6]. In general, one can view LE-series as a somewhat novel kind of asymptotic
(often divergent) expansions for real functions de*ned near +∞. (The Stirling ex-
pansion for the Gamma function is an especially simple example.) They also arise as
formal solutions to algebraic di+erential equations. Perhaps the most important natural
I
source of LE-series is in the works of Ecalle [8] and Il’yashenko [14] on the Dulac
problem.
We give an explicit construction of R((x−1 ))LE in Section 2, but to give already
now an idea of this remarkable mathematical structure we show here some typical
computations in R((x−1 ))LE :
(1) Taking a reciprocal:
1 1
= = x−1 (1 + xe−x + x2 e−2x + · · ·)
x − x2 e−x x(1 − xe−x )
= x−1 + e−x + xe−2x + · · · :
(2) Formal integration:

x ∞

e
dx = constant + n!x−1−n ex (a divergent series):
x
n=0
(3) Formal composition:

Let f(x) = x + log x and g(x) = x log x. Then
f(g(x)) = x log x + log(x log x)

= x log x + log x + log(log x)
L. van den Dries et al. / Annals of Pure and Applied Logic 111 (2001) 61–113 63
and

log x
g(f(x)) = (x + log x) log(x + log x) = (x + log x) log x + log 1 +
x
∞
n
(−1)n+1 log x
= x log x + (log x)2 + (x + log x)
n x
n=1
∞
(−1)n+1 (log x)n+1
= x log x + (log x)2 + log x + :
n(n + 1) xn
n=1
(4) Compositional inversion:

The series g(x) = x log x has a compositional inverse of the form

x log log x 1
1+F ; ;
log x log x log x
where F(X; Y ) is an ordinary convergent power series in the two variables X and Y
over R. (Cf. [6, Section 4], and see also [1, Chapter 2] for other interesting examples.)
In this paper we concentrate on developing the formalism of LE-series. Section 1 es-
tablishes notation and also contains some useful facts on generalized series *elds k((G))
for which we do not know a convenient reference. In Section 2 we follow in essence
the original method of Dahn and GLoring [3, 4] in constructing the *eld R((x−1 ))LE , but
present it, we hope, in a more intuitive way. (In addition it is notationally more natural
than in our earlier paper [6].) We also show in Section 2 that each LE-series over R
contains only countably many monomials. In Section 3 we de*ne the (formal) deriva-
tive of an LE-series, and prove the nontrivial fact that the derivative of an LE-series
vanishes if and only if the series is a real constant. In Section 4 we show that the
ordered di+erential *eld R((x−1 ))LE satis*es the same kind of di+erential inequalities
as Hardy *elds do, cf. [22a]. (It is an interesting question—which can be made precise
in several ways—how far this analogy really goes.) In Section 5 we prove that each
LE-series has an antiderivative. In Section 6 we de*ne the (formal) composition of
LE-series and establish the expected properties like associativity, Taylor expansion, the
chain rule, and the existence of compositional inverses. In Section 7 we characterize
I
Ecalle’s *eld R[[[x]]] of transseries as a sub*eld of R((x−1 ))LE and establish some of
its properties.
All the above is strictly a one-variable theory, but instead of R, any ordered *eld
with a “good” exponential function can serve as a *eld of coe8cients for a *eld of
LE-series. As this extra generality could be useful in questions about dependence on
parameters and does not cost any extra e+ort, we actually work in this somewhat more
general setting.
It seems that the interest in logarithmic-exponential series has two rather di+erent
origins. Dahn and GLoring [2, 3] were inNuenced by Tarski’s problem about the real
I
exponential *eld, while Ecalle [8] and Il’yashenko [14] encountered LE-series in their
work on the Dulac problem. There are also potential connections with the theory of
surreal numbers of Conway and Kruskal, and “super exact asymptotics”.
The geometric part of Tarski’s problem has been solved in the mean time by Wilkie
[27], but LE-series remain useful for deciding various questions about the functions
de*nable in the real exponential *eld and expansions of it, as we showed in our
previous paper [6] to which the present paper is a sequel. But we make a fresh start
here and only assume familiarity with [6] in a few places.
Notational conventions on ordered sets and groups
Throughout we let m and n range over N = {0; 1; 2; : : :}. By “ordered set” we will
always mean a linearly ordered set. Given an ordered set G and a ∈ G we put
G ¿a := {g ∈ G: g¿a}, G ¡a := {g ∈ G: g¡a},
G ¿a := {g ∈ G: g ¿ a}, G 6a := {g ∈ G: g 6 a}.
If in addition A and B are subsets of G we put
g¿A if g¿a for all a ∈ A,
g¡A if g¡a for all a ∈ A,
A¡B if a¡b for all a ∈ A and b ∈ B.
Let G be an ordered abelian (multiplicative) group. (As usual this includes the
requirement that the ordering is invariant under multiplication: a¡b implies ac¡bc,
for all a; b; c ∈ G.) Given subsets A and B of G we put
AB = A · B := {ab: a ∈ A; b ∈ B},
An := {a1 · · · an ∈ G: a1 ; : : : ; an ∈ A} (with A0 = {1}),

[A] := n An (the submonoid of G generated by A),
A := [A ∪ A−1 ] = the subgroup of G generated by A.
We will deal often with sets A ⊆ G that are reverse well ordered, that is, for which
there is no in*nite increasing sequence g0 ¡g1 ¡g2 ¡ · · · in A. The following, from
[19], is a basic result about such subsets.
Neumann’s Lemma. If A; B ⊆ G are reverse well ordered; so is AB; and for each
g ∈ AB there are only 4nitely many pairs (a; b) ∈ A × B with ab = g. If A ⊆ G ¡1 is
reverse well ordered; then [A] is too; and for each g ∈ [A] there are only 4nitely many
tuples (n; a1 ; : : : ; an ) with a1 ; : : : ; an ∈ A such that g = a1 · · · an .
1. Fields of generalized series
1.1. Let G be a multiplicative ordered abelian group with unit 1G (or just 1 if no
confusion results) and let k be a *eld. We de*ne k((G)) to be the 4eld of series over
k with monomials in G; its elements are the formal sums

s= cg g
g∈G
with coe8cients cg ∈ k, such that Supp s := {g ∈ G: cg = 0} is reverse well ordered

in G.
By Neumann’s Lemma these series can be added and multiplied in the usual way,
making k((G)) into a *eld with sub*eld k (identifying c ∈ k with the series s = c · 1G )
and G into a multiplicative subgroup of k((G))× (identifying g ∈ G with the series
s = 1 · g). The leading coe5cient Lc(s), the leading monomial Lm(s) and the leading

term Lt(s) of a nonzero element s = cg g are given by: Lc(s) = cg0 , Lm(s) = g0 and
Lt(s) = Lc(s)Lm(s) = cg0 g0 , where g0 = max Supp s. We extend the functions Lc, Lm
and Lt to (multiplicative) functions on all of k((G)) by setting Lc(0) = Lm(0) = Lt(0)
= 0 ∈ k((G)). Then Lm is just a nonarchimedean absolute value, except that its values
lie in the ordered set G ∪ {0}, with g¿0 for all g ∈ G, instead of in R¿0 ; in partic-
ular Lm(s + s ) 6 max(Lm(s); Lm(s )) for s; s ∈ k((G)). For each subset S of G we
put
k((S)) := {s ∈ k((G)) : Supp s ⊆ S};
which is a k-linear subspace of k((G)). Note that
k((G)) = k((G ¿1 )) ⊕ k ⊕ k((G ¡1 )) (direct sum of k-linear subspaces)
and that k((G 61 )) = k ⊕ k((G ¡1 )) = {s ∈ k((G)) : Lm(s) 6 1} is a valuation ring of

k((G)) with maximal ideal m := k((G ¡1 )). Each s ∈ k((G))× can be written as s = Lt(s)
(1 + ) with ∈ m. We consider k((G)) as a topological *eld by taking the valuation
topology: the discs D(g) := {s : Lm(s)¡g}, g ∈ G, form a basis of open neighbor-
hoods of 0. If G1 is a subgroup of G, we shall consider k((G1 )) as a sub*eld of k((G))
in the obvious way.
If k is an ordered *eld we make k((G)) into an ordered *eld by declaring a nonzero
element s to be positive if Lc(s)¿0; this ordering extends both the ordering of k and
the ordering of the group G, and the interval topology given by this ordering coincides
with the valuation topology if G = {1}.
1.2. Comment. In the previous papers [5, 6] we started with an additive ordered abelian
group and formed a multiplicative copy t so that → t is an order reversing
isomorphism from the additive group onto the multiplicative group t . The elements

of k((t )) appear then as generalized power series s = a t . In terms of [5, 6], the
support supp s of this series is a subset of rather than of t , and is related to Supp s
as follows:
Supp s = {t : ∈ supp s}:
(The notational di+erence is the capital S of “Support” versus the lower case s of
“support”.) In dealing with logarithmic-exponential series it is awkward to consider
each logarithmic-exponential monomial arti*cially as a “power” t . This was done in
[6] with the unfortunate consequence that in general t = exp( log t), see Remark 2:11
of [6]. When de*ning composition of logarithmic-exponential series this would become
notationally confusing. That is why we have chosen here a better motivated and more
Nexible notation.
1.3. Suppose ai ∈ k((G)) for all i ∈ I for some, possibly in*nite, set I . We say that the

sum i∈I ai exists if the following two conditions are satis*ed:
(a) For each g ∈ G there are only *nitely many i ∈ I with g ∈ Supp ai ;

(b) The union i∈I Supp ai is reverse well ordered in G.
Clearly, if these conditions are satis*ed we can associate with the family (ai ) a well

de*ned element ai ∈ k((G)). (Note that each element a = cg g of k((G)) is indeed

the sum of the family (cg g) in this sense.) We will frequently use the following
rules for manipulating such sums. Let (ai )i∈I , (bj )j∈J and (ci; j )(i; j) ∈ I ×J be families of

elements of k((G)) such that ai , bj and i; j ci; j exist. Then

(1) for each d ∈ k((G)), dai exists, and dai = d ai ,

(2) if I = J , then (ai + bi ) exists, and (ai + bi ) = ai + bi ;

(3) i; j ai bj exists, and i; j ai bj = ( ai ) · ( bj );

(4) if % : I → J is a bijection and ai = b%(i) for all i, then ai = bj ,

(5) j ci; j exists for each i ∈ I , and i( j ci; j ) exists, and

ci; j = ci; j :
i; j i j
These in*nite sums occur everywhere in the subject. For instance, given an ordinary

power series F = !∈Nn a! Y1!1 · · · Yn!n ∈ k[[Y1 ; : : : ; Yn ]] and in*nitesimals 1 ; : : : ; n ∈ m

we have a well de*ned element F(1 ; : : : ; n ) := a! 1!1 · · · n!n of k((G)), since this
in*nite sum exists by Neumann’s Lemma.
We can also use these in*nite sums to introduce e8ciently certain useful operators
on k((G)). Let D be a k-linear subspace of k((G)) such that for each family (ai )i∈I

of elements in D for which ai exists we have ai ∈ D. (Example: D = {s ∈ k((G)) :

Lm s¡A} where A is any subset of G; note: D = k((G)) for A = ∅.) Let us say that
an operator P : D → D is small if there is a reverse well ordered subset R of G ¡1
such that Supp P(s) ⊆ R · Supp s for all s ∈ D (we do not assume P is k-linear or even
additive). An obvious induction on n then shows that the nth iterate Pn of P (with
P0 = I = id D ) satis*es
Supp Pn (s) ⊆ Rn · Supp s
for all s ∈ D. Hence for each sequence (cn ) of coe8cients in k we have a well de*ned
operator

cn Pn : s → cn Pn (s) : D → D;
the existence of the right hand sum being a consequence of Neumann’s Lemma. In
particular, if P is a small operator, then the perturbation I − P of the identity is a

bijection from D onto itself with inverse Pn .

We shall use small operators in Section 5 to integrate logarithmic-exponential series,

and in Section 6 to obtain compositional inverses to them.
1.4. Suppose now that G1 is a convex subgroup of G and G is the internal direct
product of subgroups G1 and G2 , that is G = G1 G2 and G1 ∩ G2 = {1}. Then we have
an isomorphism
k((G)) ∼
= k((G1 ))((G2 ))
of k((G1 ))-algebras given by

 

ag g →  ag 1 g 2 g 1  g 2 : (∗)
g∈G g2 ∈G2 g1 ∈G1
Note that in k((G)) we actually have

 

ag g =  ag1 g2 g1  g2 ;
g∈G g2 ∈G2 g1 ∈G1
where all sums are interpreted according to the notation in 1:3. Note also that if k is
an ordered *eld, then the isomorphism (∗) is order preserving. Thus, there is no harm
in identifying k((G)) with k((G1 ))((G2 )) and we shall do so when convenient. With
this identi*cation we have
k((G)) = A ⊕ B (direct sum of k((G1 ))-linear subspaces);
where A = {s ∈ k((G)) : Supp s¿G1 } and B = {s ∈ k((G)) : Supp s 6 g1 for some

g1 ∈ G1 } = k((G1 ))((G261 )).
1.5. Exponential ordered #elds. An exponential ordered 4eld is a pair (K; E) where
K is an ordered *eld and E : K → K ¿0 (the exponential map) is a strictly increasing
homomorphism from the additive group of K into its positive multiplicative group K ¿0 .
If in addition E(K) = K ¿0 we call (K; E) a logarithmic-exponential ordered 4eld. In
that case the inverse to E is written as log : K ¿0 → K, and one also writes ar :=
E(r log a) for a; r ∈ K, a¿0, and the usual identities hold: ar+s = ar a s , (ab)r = ar br ,
and ars = (ar ) s for a; b; r; s ∈ K with positive a; b.
A pre-exponential ordered 4eld is a quadruple (K; A; B; E) such that K is an ordered
*eld, A an additive subgroup of K and B a convex subgroup of K with K = A ⊕ B, and
E : B → K ¿0 is a strictly increasing homomorphism from the additive group B into the
multiplicative group K ¿0 .
1.6. A basic example. Let k be an ordered exponential *eld with exponential map exp,
and consider the ordered multiplicative group xk with order preserving isomorphism
r → xr from the additive group of k onto xk . This gives rise to the pre-exponential
ordered *eld
(k((xk )); A; B; E)
by setting A := {s ∈ k((xk )) : Supp s¿1}, B := {s ∈ k((xk )) : Supp s 6 1}, and de*ning

E : B → (k((xk )))¿0 by
∞ n

E(a + ) = exp(a)
n!
n=0
for a ∈ k and ∈ m(B) := {s ∈ B : Supp s¡1}. To simplify notation we also write k((xk ))
for the resulting pre-exponential ordered *eld.
Note that upon setting x := x1 we have x¿k, and k((xk )) contains the *eld k(x) of
rational functions in x over k as a sub*eld. We think of an element of k((xk )) as a
series

ar x r
r∈k
(in “descending powers of x”, as one goes from left to right). Put t := x−1 so that
0¡t¡a for all positive a ∈ k and note that k((xk )) contains the *eld k((t)) = k((x−1 ))
:= k((xZ )) of formal Laurent series in t as a sub*eld since Z ⊆ k (as an ordered
subgroup). For us it is more convenient to consider the in*nitely large x as the “inde-
pendent variable” rather than the in*nitesimal t.
1.7. Let (K; A; B; E) be a pre-exponential ordered *eld. We de*ne its 4rst extension
(K ; A ; B ; E ) as follows: take a multiplicative copy E(A) of the ordered additive
abelian group A with order preserving isomorphism EA : A → E(A). Let K := K((E(A))),
A := {s ∈ K : Supp s¿1} and B := {s ∈ K : Supp s 6 1}, so that B is a convex sub-
ring of K . Note that K = A ⊕ B and B = K ⊕ m(B ). We now extend E to
E : B → (K )¿0 by
∞ n

E (a + b + ) = EA (a)E(b)
n!
n=0
for a ∈ A, b ∈ B and ∈ m(B ). One veri*es easily that E extends E and that (K ; A ; B ;

E ) is again a pre-exponential ordered *eld. Note that the map E is de*ned on the
whole *eld K since K ⊆ B .
We de*ne for each n a pre-exponential ordered *eld (Kn ; An ; Bn ; En ): for n = 0
this is just (K; A; B; E) itself, and (Kn+1 ; An+1 ; Bn+1 ; En+1 ) is the *rst extension of
(Kn ; An ; Bn ; En ). This gives an increasing sequence
K = K0 ⊆ K1 ⊆ K2 ⊆ · · ·

of ordered *elds. Let K∞ = Kn , and de*ne E∞ : K∞ → (K∞ )¿0 to be the common
extension of all the En . Then (K∞ ; E∞ ) is clearly an ordered exponential *eld.
2. The #eld of logarithmic-exponential series
From now on in this paper we 4x a logarithmic-exponential ordered 4eld k.
2.1. Starting with the pre-exponential ordered *eld (K; A; B; E) = k((xk )) as in 1:6;
we denote the exponential *eld (K∞ ; E∞ ) that results from the construction in 1:7
by k((x−1 ))E or by k((t))E , and its exponential map E∞ by E, to simplify nota-
tion. We de*ne the increasing sequence G0 ⊂ G1 ⊂ · · · of multiplicative subgroups
of k((t))E by G0 = xk and Gn+1 = Gn E(An ), where (Kn ; An ; Bn ; En ) is as in 1:7. By
induction on n one easily checks that Gn is convex in Gn+1 , Gn ∩ E(An ) = {1} and
Kn = k((Gn )), since k((Gn )) ((E(An ))) = k((Gn+1 )) by the identi*cation in 1:4. Unless
we say otherwise we will consider Kn as the series *eld k((Gn )) over k. Note that
An = {f ∈ Kn : Supp f¿Gn−1 }, which is even true for n = 0 by setting G−1 := {1} ⊆ G0 .
(Similarly, K−1 := k in the following.)

We put G E := Gn , an ordered subgroup of the positive multiplicative group
of the ordered *eld k((t))E . The elements of G E are called E-monomials. Thus we

have the *eld inclusion k((t))E = k((Gn )) ⊂ k((G E )). The inclusion is proper because
n E E 0 n+1
n 1=E (x) ∈ k((G ))\k((t)) , where E (x) := x and E (x) = E(En (x)); to see this,
note that E (x) ∈ Gn and E (x)¿Gn−1 for each n. Note that k((t))E is dense in the
n n
topological *eld k((G E )).
The lemma below follows easily by induction on n and 1.4. We record it for later
use.
2.2. Lemma. (i) k((Gn )) = An ⊕ An−1 ⊕ · · · ⊕ A0 ⊕ k ⊕ mn (a direct sum of k-linear

spaces) where mn is the maximal ideal of k((Gn61 ));
(ii) if an ∈ An ; an−1 ∈ An−1 ; : : : ; a0 ∈ A0 ; r ∈ k and ∈ mn are all positive; then
an ¿an−1 ¿ · · · ¿a0 ¿r¿;
(iii) An ⊕ An−1 ⊕ · · · ⊕ A0 = {f ∈ k((Gn )) : Supp f¿1};
(iv) Gn = xk · E(An−1 ⊕ · · · ⊕ A0 ); and xk ∩ E(An−1 ⊕ · · · ⊕ A0 ) = {1}.
2.3. Lemma. Let f ∈ Kn ; f¿0. Then
f ∈ E(k((t))E ) ⇔ Lm f ∈ E(An−1 ⊕ · · · ⊕ A0 ):
Proof. By Lemma 2.2(iv) we have Lm f = xr E(a) where r ∈ k and a ∈ An−1 ⊕ · · · ⊕ A0 .

Let c = Lc f¿0. Then f = cxr E(a)(1 − ) with ∈ mn . If Lm f ∈ E(An−1 ⊕ · · · ⊕ A0 ),
then r = 0 (by Lemma 2.2(iv)), so
∞ i

f = E a + log(c) − ∈ E(Kn−1 ) ⊆ E(k((t))E ):
i
i=1
Conversely, suppose f = E(g), with g ∈ Kl , l ¿ n. Write g = + + s + , with

+ ∈ Al ⊕ · · · ⊕ A0 , s ∈ k and , ∈ ml . Then
∞

,i
f = E(g) = exp(s)E(+) 1 + :
i!
i=1
Hence Lm f = E(+) = xr E(a), and thus r = 0 by Lemma 2.2(iv).
2.4. Notational conventions about in#nite sums. In accordance with 1.3 we shall write

f= cr x r in K0

to mean that cr ∈ k for all r ∈ k, and that f = cr xr in the sense of the series *eld
k((G0 )). Similarly, if we write

f= fa E (a) in Kn+1 ;
we mean that the sum is over all a ∈ An , that fa ∈ Kn for all a ∈ An , and that f =
fa E(a) in the sense of the series *eld Kn+1 = Kn ((E(An ))) over Kn . Thus in this

case f ∈ Kn+1 , and also f = fa E(a) in the sense of the series *eld k((Gn+1 )), by
1.3 and 1.4. Finally, we write

f= fi in k((t))E

to mean that there exists an n such that all fi ∈ Kn and that f = fi in the sense

of the series *eld Kn = k((Gn )). (In particular f ∈ Kn for this n, and clearly fi is
independent of the particular value of n for which all fi ∈ Kn .)
2.5. Comparison with [6]. The *eld R((t))E only di+ers notationally from the R((t))E
of [6]. To facilitate comparison we specify the isomorphism - (of exponential or-
dered *elds) from our present R((t))E onto the R((t))E of [6]. Under - the present
Kn = R((Gn )) maps onto Kn = R((t n )) of [6], with present Gn , An and Bn mapping
onto t n , Jn and On of [6], respectively. For n = 0 we have K0 = R((xR )) (present),

−r
and K0 = R((t R )) (in [6]) with -( ar xr ) = ar t . Inductively, for f = fa E(a)

in present Kn+1 we put -(f) := -(fa )t −-(a) . For all intents and purposes we may
identify our present R((t))E via - with the R((t))E of [6].
2.6. The substitution map /. To extend k((t))E to a logarithmic-exponential ordered

*eld k((t))LE , we will need to have available a certain map / : k((t))E → k((t))E , which
plays a key role throughout this paper. It embeds k((t))E into itself as an ordered expo-
nential *eld and is the identity on k, in particular, /(E(f)) = E(/(f)) for f ∈ k((t))E .
We think of it as “substituting E(x) for x”, symbolically: /(f(x)) = f(E(x)). (In Sec-
tion 6 we justify this formula when we de*ne composition of LE-series.) The de*nition
of /(f) is by induction:

if f = ar xr in K0 ; then /(f) := ar E(rx);

if f = fa E(a) in Kn+1 ; then /(f) := /(fa )E(/(a)):
To make sense of this one shows simultaneously by induction on n that /(Gn ) ⊆ Gn+1 ,
/(An ) ⊆ An+1 and /(Kn ) ⊆ Kn+1 . (See also [6, Section 2] for details.) The properties of
/ stated above are easily derived, and in fact, one obtains /(Gn ) ⊆ E(An ⊕ · · · ⊕ A0 ).
Thus Lm(/(f)) ∈ E(An ⊕ · · · ⊕ A0 ) for all f ∈ Kn and so, by Lemma 2.3, /(f) ∈
E(k((t))E ) for every positive element f ∈ k((t))E . This means that each
positive element of /(k((t))E ) has a logarithm in k((t))E .

It is easy to see that / also preserves in*nite sums: if f = fi in k((t))E , then

/(f) = /(fi ) in k((t))E .
2.7. The construction of k((t))LE . We construct an increasing chain of exponential or-

dered *elds
k((t))E = L0 ⊂ L1 ⊂ L2 ⊂ · · ·
and isomorphisms 1n from Ln onto k((t))E . Let L0 = k((t))E and let 10 be the identity.
Given Ln and 1n , take Ln+1 ⊃ Ln and an isomorphism of exponential ordered *elds
1n+1 : Ln+1 → k((t))E such that 1n+1 (z) = /(1n (z)) for all z ∈ Ln . So Ln sits inside Ln+1
as /(k((t))E ) sits in k((t))E . Thus by a remark in 2.6 every positive element of Ln has
a logarithm in Ln+1 .

Let k((t))LE = Ln . Clearly k((t))LE is a logarithmic-exponential ordered *eld. We
call k((t))LE the *eld of logarithmic-exponential series over k. We also write it as
k((x−1 ))LE and call its elements LE-series over k.
There will be no harm in denoting the exponential map of k((t))LE by the same
symbol E as the exponential map of k((t))E . Its inverse will be written both as log and
as L. We often use the nth iterates En and Ln of these maps, with
E0 = L0 := identity map on k((t))LE ;
while for n¿0, Ln (z) is only de*ned for z¿En−1 (0). As k((t))LE is a logarithmic-
exponential ordered *eld we have by 1.5 a well de*ned element fg := E(g log f) for
f; g ∈ k((t))LE with f¿0; note that for f = x and g = r ∈ k this agrees, fortunately,
with xr viewed as an element of K0 .
Let ln := Ln (x) ∈ Ln , in particular l0 = x and l1 = log x. The isomorphism 1−1 n from
k((t))E onto Ln maps x to ln , and we want to view it as “substitution of ln for x”, that
is 1−1
n (f(x)) = f(ln ). In Section 6 we will make sense of this last formula.
For each n we have an isomorphism 1−1 n ◦ 1n+1 : Ln+1 → Ln , of exponential ordered
*elds, and these isomorphisms have a unique common extension to an automorphism
of the exponential ordered *eld k((t))LE . This automorphism extends /, and we shall
denote it by / as well. Thus we have an automorphism /k for each (positive or

negative) integer k. Note that /n (z) = 1n (z) for z ∈ Ln . In particular, /−n (x) = Ln x.
2.8. k((t))LE as a #eld of generalized series. It will be convenient to construe Ln as a

*eld of generalized series. Put G E; n := 1−1 E
n (G ), an ordered subgroup of the positive
multiplicative group of Ln . We have an isomorphism

a3 3 → a3 1−1 E
n (3) : k((G )) → k((G
E; n
))
3∈G E 3∈G E; n
of ordered *elds. There is no loss of generality in taking 1−1 n to be the restriction

of this isomorphism to k((t))E , with image Ln ⊆ k((G E; n )), for each n. Consider the
increasing sequence
G E = G E;0 ⊂ G E;1 ⊂ G E;2 ⊂ · · ·

of subgroups of the positive multiplicative group of k((t))LE . Let G LE := G E; n , the
(ordered) group of LE-monomials. Then the identi*cations above give an ordered *eld
inclusion k((t))LE ⊆ k((G LE )). Thus for f ∈ k((t))LE its support Supp f is a reverse well
ordered subset of G LE , and we can speak of the leading monomial Lm (f) of f (in
G LE ∪ {0}), its leading coe5cient Lc (f) (in k), and its constant term (in k), the
latter being the coe8cient of the LE-monomial 1 when f is written as an element of
k((G LE )). We consider k((t))LE as a topological *eld by taking the interval topology
given by the ordering of k((t))LE ; this topology equals the topology induced by the
valuation topology of k((G LE )). An element ∈ k((t))LE is said to be in*nitesimal if
||¡k¿0 .
To make sense of in*nite sums in k((t))LE we put
Lm; n := 1−1
m (Kn ) ⊆ Lm Gm; n := 1−1
and m (Gn ) ⊆ G
E; m

so that Lm; n = k((Gm; n )), and Lm = n Lm; n . We write

f= fi in k((t))LE

to mean that there exist m and n such that all fi ∈ Lm; n , and that f = fi in the

sense of Lm; n = k((Gm; n )). In that case also f ∈ Lm; n , and fi is independent of the

particular values of m and n for which all fi ∈ Lm; n . Note that if f = fi in k((t))LE

then /k (f) = /k (fi ) in k((t))LE for all integers k.
2.9. The valuation on k((t))LE . The leading monomial function Lm behaves like a non-
archimedean absolute value, and the corresponding valuation (in the sense of Krull)
v : k((t))LE → log(G LE ) ∪ {∞}
is de*ned by v(f) := − log(Lm(f)) for f = 0. The value group log(G LE ) is here

ordered as an additive subgroup of k((t))LE . The valuation ring of v is, of course, the
‘closed unit disc’ {f : Lm(f) 6 1}. Note also that the restriction of v to k((t))E has

value group ( n An ) ⊕ k log(x), by Lemma 2.2.
2.10. A countability property of logarithmic-exponential series. Here we show that

each LE-series over R contains only countably many LE-monomials. This ultimately
stems from the well known fact that each well ordered subset of the real line R is
countable, via the following lemma due to Esterle [9].
Lemma. Suppose each (reverse) well ordered subset of the ordered abelian multi-
plicative group G is countable. Then also each (reverse) well ordered subset of the
ordered 4eld R((G)) is countable.
Corollary. Each (reverse) well ordered subset of the ordered 4eld R((t))LE is count-
able; in particular; Supp f is countable for each f ∈ R((t))LE . The cardinality of
R((t))LE is 2ℵ0 .
Proof. The lemma implies by an induction on n that each (reverse) well ordered subset
of the ordered *eld Kn = R((Gn )) is countable, and that Kn has cardinality 2ℵ0 . Thus
R((t))E also has these properties. Since each ordered *eld Ln is isomorphic to R((t))E
it follows that their union R((t))LE inherits these properties as well.
Remark. By a theorem of Harrington and Shelah [13] each well ordered subset of
a Borel order is countable. However, this fact does not contain the corollary above
as a special case: the main result in [12] implies that the underlying ordered set of
R((t))LE is not order embeddable in a Borel order. On the other hand, in Section 7 we
will introduce the logarithmic-exponential ordered sub*eld R((t))LE; ft of R((t))LE whose
elements are the LE-series with “support of hereditarily *nite type”. This substructure
is probably large enough for most purposes, and is a Borel structure. See Section 7
for more details.
2.11. Other approaches. We obtained k((t))LE as a proper sub*eld of the series *eld
k((G LE )). Could we have constructed a similar logarithmic-exponential ordered *eld as
a full series *eld k((G))? The answer is negative: Kuhlmann et al. [16] showed that
the ordered additive group of k((G)) is not isomorphic to its positive multiplicative
group when G = {1}.
On the other hand, Kuhlmann and Kuhlmann [15] construct logarithmic-exponential
ordered *elds as countable unions of *elds of generalized series that are closed under
a natural logarithmic operation at each stage. We have not considered in detail how
their *eld of “exponential-logarithmic” series is related to our *eld of LE-series. While
the construction in [15] may have its virtues, the fact that our series are of the form
/−n (f) ( = f(ln )), where f = f(x) is an exponential series, will be of great use to
us in later sections. This byproduct of the construction routinely reduces problems to
the case of exponential series.
Ressayre [R] proved that every model of the theory of the real exponential *eld is
isomorphic to a “truncation closed” sub*eld of a generalized series *eld over R with
the usual exponentiation on in*nitesimals. This fact can be used to give alternative
proofs for some results in [6] that depend there on LE-series.
Recently, see 7.23, we became aware of Van der Hoeven’s thesis, which contains
also various constructions of logarithmic-exponential series *elds.
3. The derivation on k((t))LE
3.1. We start by de*ning inductively a derivation f → f on k((t))E , to be thought of

as “di+erentiation with respect to x”.

r
For f = ar x in K0 we set f = rar xr−1 . Then the map f → f : K0 → K0

is easily seen to be a derivation. Assume we have de*ned already the derivation

f → f : Kn → Kn . We then extend this map to Kn+1 as follows: for f = a∈An fa E(a)

in Kn+1 we set

f = (fa + fa a )E(a):
a∈An
It is easy to check that then the map f → f : Kn+1 → Kn+1 is again a derivation.
This inductive de*nition provides us with a derivation f → f on k((t))E , as promised.
It has the following strong additivity property.

3.2. Lemma. Suppose fi exists in Kn . Then fi also exists in Kn and ( fi ) =
i∈I
fi .
The (inductive) proof is entirely routine and left to the reader.

For F ∈ k[[X1 ; : : : ; Xm ]] we let @F=@Xi denote the formal partial derivative of F with
respect to Xi .
3.3. Corollary. Let F ∈ k[[X1 ; : : : ; Xm ]] and let 1 ; : : : ; m be in4nitesimals of k((t))E .

Put := (1 ; : : : ; m ). Then
m
@F
F() = ()i :
@Xi
i=1
Proof. Write F = !∈Nm a! X ! , and suppose all i ∈ Kn . Then F() = a! ! in Kn ,

and the desired result is almost immediate from the previous lemma.
3.4. Corollary. If f ∈ k((t))E ; then E(f) = f E(f).
Proof. From the de*nition of the derivation we see that E(a) = a E(a) for a ∈ An . An
easy induction shows that this is also true for a ∈ An ⊕ · · · ⊕ A0 .
Suppose f ∈ Kn . By Lemma 2.2(i) we have f = a+r+ where a ∈ An ⊕· · ·⊕A0 ; r ∈ k

and ∈ Kn is in*nitesimal. Then
∞ i

E(f) = E (a) exp(r) :
i!
i=0

By Corollary 3:3 we have ( i =i!) = i =i!. Thus,
i i
E(f) = a E(a) exp(r) + E(a) exp(r)
i! i!
= (a + )E(f)
= f E(f):
Next, we show that the *eld of constants of our derivation is exactly k. To obtain
this fact we have to prove a little more.
3.5. Lemma. Let f ∈ Kn \{0} have constant term 0; and let g ∈ An \{0}. Then:
(i) f = 0;
(ii) Lm (f =f) = x−1 if n = 0; and Lm (f =f) ∈ Gn−1 if n¿0;
(iii) g ∈ An \{0} if n¿0.
Proof. By induction on n. For n = 0 we have Lm f = xr for some non-zero r ∈ k,

hence Lm f = xr−1 , so that f = 0. This proves the case n = 0. Assume the lemma
holds for a certain value of n, and let f ∈ Kn+1 \{0} have constant term 0. Write

f = a∈An fa E(a) in Kn+1 . Put + := max{a ∈ An : fa = 0}. To show that (i) holds for
our f we distinguish several cases.
Case + = 0: Then

f = f0 + fa E(a);
a∈An ; a¡0
where f0 ∈ Kn \{0} has constant term 0 and

f = f0 + (fa + fa a )E(a):
a∈An ;a¡0
By the inductive assumption f0 = 0, hence f = 0.

Case + = 0: Then

f = (f+ + f+ + )E(+) + (fa + fa a )E(a):
a∈An ; a¡+
We claim that f+ + f+ + = 0, from which it follows in particular that f = 0.

Subcase f+ = 0: Then by the inductive assumption f+ ∈ k and + = 0, hence f+ +
f+ + = f+ + = 0.
Subcase f+ = 0: Then we write f+ = 6 + c where c ∈ k and 6 ∈ Kn \{0} has constant

term 0. Then

6 6+c
f+ + f+ + = 6 + (6 + c)+ = 6 ++ :
6 6
Clearly

6+c
Lm = 1 if Lm 6¿1 or c = 0;
6
¿ 1 if Lm 6¡1 and c = 0:
Hence by the inductive assumption applied to 6 and +:

6 6+c
Lm ¡Lm(+ ) 6 Lm + ;
6 6
and thus f+ + f+ + = 0.

Next we prove (ii) above for our f (with n replaced by n + 1). If + = 0, then

f f0
Lm = Lm ∈ Gn−1 ⊆ Gn
f f0
by the inductive assumption, while if + = 0, then

f f+ + f+ +
Lm = Lm ∈ Gn :
f f+
Next we prove (iii) (with n replaced by n + 1) for g ∈ An+1 \{0}. Note that

g has constant term 0, so that g = 0. Write g = 0¡a∈An ga E(a) in Kn+1 . Then

g = 0¡a∈An (ga + ga a )E(a), and thus g ∈ An+1 .
Remark. Suppose that f ∈ Kn+1 \{0} has constant term 0 and that + = 0, where
+ is de*ned as in the proof of Lemma 3:5. For later use we note that then
Lt(f+ + f+ + ) = Lt(f+ + ) (with the notations of the proof of Lemma 3:5) and thus
Lt f = Lt(f+ + )E(+):
This is clear if f+ = 0. If f+ = 0, then f+ = 6 + c as in the corresponding subcase in the

proof of Lemma 3:5, where it is shown that then Lm (6 =6)¡Lm ((f+ =6)+ ). Hence
Lt(6 =6 + (f+ =6)+ ) = Lt((f+ =6)+ ) and thus Lt(f+ + f+ + ) = Lt(6(6 =6 + (f+ =6)+ )) =
Lt(6)Lt((f+ =6)+ ) = Lt(f+ + ):
As an immediate consequence of Lemma 3:5 we obtain:
3.6. Corollary. If f ∈ k((t))E and f = 0; then f ∈ k.

Our next goal is to extend the derivation to k((t))LE . For this we need the following
lemma, where / : k((t))E → k((t))E is as in 2.6.
3.7. Lemma. Let f ∈ k((t))E . Then

(1) /(f) = /(f )E(x),
m
(2) /m (f) = /m (f ) i=1 Ei (x).
Proof. Identity (1) is an absolutely straightforward induction on n, where f ∈ Kn , and

(2) follows by induction on m from (1).
Let D denote the derivation f → f on k((t))E . To motivate how we extend D to

k((t))LE , consider an element f ∈ Ln , and put f̃ := 1n (f). Thinking of f as f̃(ln ), the
chain rule suggests that we should have
(Df̃)(ln ) 1−1 (D1n (f))

f := (Df̃)(ln ) · ln = n−1 = n n−1 :
i=0 li i=0 li
Thus, we introduce the derivation

1
Dn := n−1 (1−1
n ◦ D ◦ 1n )
i=0 li
on Ln : this is indeed a derivation on Ln , since the conjugate 1−1

n ◦ D ◦ 1n of D is one.
Note that D0 = D. We now verify that Dn+1 extends Dn . Let f ∈ Ln . Then
1 1
Dn+1 f = n 1−1
n+1 D1n+1 (f) = n 1−1
n+1 D(/1n (f))
i=0 li i=0 li
1
= n 1−1
n+1 (/D1n (f) · E(x)) (by Lemma 3:7(1))
i=0 li
1 1
= n 1−1 −1
n (D1n (f)) · ln = n−1 1n (D1n (f))
i=0 li i=0 l i
= Dn f:

The desired extension of D to k((t))LE is the derivation n Dn on k((t))LE , which
we also denote by d=d x. For f ∈ k((t))LE we often write f instead of df=d x. We
summarize some basic facts about this derivation in the following theorem.
3.9. Theorem. Let f ∈ k((t))LE and fi ∈ k((t))LE for all i ∈ I . Then

(1) If f = i∈I fi in k((t))LE ; then also f = fi in k((t))LE .
(2) Let F ∈ k[[X1 ; : : : ; Xm ]] and let 1 ; : : : ; m be in4nitesimals of k((t))LE . Put :=
(1 ; : : : ; m ). Then
m
@F
F() = ()i :
@Xi
i=1
(3) E(f) = f E(f); and (log |f|) = f =f for nonzero f.

(4) If f = 0; then f ∈ k.
m
(5) /m (f) = /m (f ) i=1 Ei (x).
Proof. These results follow easily from 3.2–3.4, 3.6 –3.8. To give an example of
how this goes, let us derive the *rst part of (3). Take n such that f ∈ Ln and put
n−1
a := 1= i=0 li . Then also E(f) ∈ Ln , hence
E(f) = a1−1 −1
n ((1n E(f)) ) = a1n (E(1n f) )
= a1−1
n (E(1n f)(1n f) ) (by Corollary 3:4)
= aE(f)1−1
n (1n f)

(since 1−1
n commutes with E)
= f E(f):
Let us also derive (5) for m = 1, using the same notations:
/(f) = a1−1 −1
n ((1n /f) ) = a1n (/(1n f) )
= a1−1
n (/((1n f) )E(x)) (by Lemma 3:7)
= a/(1−1
n (1n f) )ln−1 (since 1−1
n commutes with /)
= a/(a−1 f )ln−1 = /(f )a/(a−1 )ln−1
= /(f )E(x):
Now an induction on m gives (5).
3.10. Remark. “The derivation of k((t))LE ” always refers to d=d x. There are other
useful derivations on k((t))LE . For example, d=dt := −x2 d=d x is a derivation that would
be natural to consider if we had taken t as the independent variable instead of x. The
properties of d=dt are, of course, immediate consequences of those of d=d x.
3.11. We know from [6, Corollary 2:8] that R((t))LE —more precisely, a certain natural
expansion of it, also denoted by R((t))LE —is an elementary extension of Ran; exp . Let
H (Ran; exp ) be the Hardy *eld of germs at +∞ of the functions f : R → R that are de-
*nable in Ran; exp . Then H (Ran; exp ) is also naturally an elementary extension of Ran; exp ,
in fact, generated as such over Ran; exp by the germ of the identity function, see [5,
Section 5]. Thus, there is a unique elementary Lan; exp -embedding of H (Ran; exp ) into
R((t))LE that sends the germ of the identity function to x. We shall call this the natural
embedding. Note that, as a Hardy *eld, H (Ran; exp ) is in particular a di+erential *eld.
3.12. Corollary. The natural embedding of H (Ran; exp ) into R((t))LE is a di8erential
4eld embedding.
Proof. From [5] we know that the functions de*nable in Ran; exp are given piecewise
by terms in the language Lan; exp; log . The result then follows easily from Theorem 3:9
by induction on terms.
4. Valuation theoretic properties of the derivation
The canonical derivation on a Hardy *eld has some useful valuation theoretic proper-
ties related to L’Hospital’s rule, see [22]. In this section we show that these properties
also hold for k((t))LE . We state the results in terms of the multiplicative “leading
monomial” function instead of the corresponding (additive) valuation.
4.1. Proposition. Let f; g ∈ k((t))LE . Then the following hold:

(1) If Lm f = 1 and Lm g = 1; then Lm f 6 Lm g if and only if Lm f 6 Lm g .
(2) If Lm f¡Lm g = 1; then Lm f ¡Lm g .
(3) If Lm f 6 1; then Lm f ¡1.
(4) If Lm f ¿ 1; then |f |¡|f|1+ for each ∈ k¿0 .
(5) If 0¡Lm f 6 1; then |f |¡|f|1− for each ∈ k¿0 .
Remark. Property (1) implies that the canonical valuation on k((t))LE is a di8erential
valuation in the sense of Rosenlicht [21, Corollary 1], where it is also shown that (2)
and (3) follow from that fact. Properties (4) and (5) are also easily derived, as in [22]
for Hardy *elds.
Property (5), with = 1=2, say, implies that the derivation is a continuous operation
on the topological *eld k((t))LE .
Proof. As already mentioned it su8ces to prove (1) since the other properties then fol-
low as in [21, 22]. Let Lm f = 1 and Lm g = 1. We may also assume f = 0 and g = 0,
since otherwise (1) is trivial. If f; g ∈ K0 the desired result follows from Lm f = x−1
Lm f, and Lm g = x−1 Lm g.
Suppose (1) holds for elements of Kn and let f; g ∈ Kn+1 . By subtracting constant
terms we may as well assume that the constant terms of f and g are zero. Let

f = a∈An fa E(a) in Kn+1 and put + := max{a ∈ An : fa = 0}. Then Lm f = Lm(f+ )

E(+), while, by the remark following Lemma 3:5,

Lm f0 if + = 0;
Lm f =
Lm (f+ + )E(+) if + = 0:

Let g = a∈An ga E(a) in Kn+1 and put 8 := max{a ∈ An : ga = 0}. If + = 8, then

Lm f 6 Lm g if and only if +¡8 if and only if Lm f 6 Lm g .
If + = 8 = 0, then Lm f0 = 1 and Lm g0 = 1. By induction Lm f0 6 Lm g0 if and only
if Lm f0 6 Lm g0 . Thus Lm f 6 Lm g if and only if Lm f 6 Lm g . If + = 8 = 0,
then Lm f 6 Lm g if and only if Lm f+ 6 Lm g+ if and only if Lm f 6 Lm g .
This proves (1) for f; g ∈ k((t))E . The general case easily reduces to this special
case.
4.2. As an application of this result we show that the di+erential equation y + y = 0
has no nontrivial solutions in k((t))LE .
Suppose f ∈ k((t))LE \{0} and f + f = 0. Then
((f )2 + f2 ) = 2f f + 2f f = 0:
Thus, by Theorem 3:9, (f )2 + f2 = c for some c ∈ k¿0 . If Lm f¿1, then f is in*nite
so (f )2 + f2 = c, a contradiction. If Lm f¡1, then, by Proposition 4:1, Lm (f )¡1,
so Lm ((f )2 + f2 )¡1, contradicting c = 0. Thus Lm (f) = 1. Then Proposition 4:1
gives Lm (f )¡1,
√ so that f LE + f = 0, again a contradiction.
Adjoining i = −1 to k((t)) (and extending the derivation) does not change mat-
ters since if f + gi were a solution of y + y = 0 (with f; g ∈ k((t))LE ), then f and g
would also be solutions.
The next result relates the derivation and the ordering.
4.3. Proposition. Suppose f ∈ k((t))LE and f¿k. Then f ¿0.
Proof. We easily reduce to the case that f ∈ k((t))E . The case f ∈ K0 is immediate.
Assuming the result holds for all f ∈ Kn , let f ∈ Kn+1 . After subtracting from f its
constant term we may assume f has constant term 0. Write f as in the proof of
Proposition 4:1 and de*ne + as in that proof. Then by the remark following Lemma 3:5,

Lt f0 if + = 0;
Ltf =
Lt (f+ + )E(+) if + = 0:
Since f+ ¿ 0 and + ¿ 0 we can apply the inductive hypothesis to give the desired
result.
4.4. Corollary. Suppose f ∈ k((t))LE and f¿xk . Then f ¿f1− for each ∈ k¿0 .
Proof. The hypothesis gives log f¿k log x, which implies log f − log x¿k. Hence
f =f¿1=x by Proposition 4:3. Also 1=x¿f− for ∈ k¿0 . Thus f =f¿f− , that is,
f ¿f1− .
Note that Proposition 4:1(4) and Corollary 4:4 together say that the derivative of
any f¿xk is close to f in a certain “relative to f” sense.
Many analogues of results about Hardy *elds can be deduced without any di8culty,
by copying (with minor modi*cations) Rosenlicht’s proofs in [21–24]. Here is one
such result. For f ∈ k((t))LE with f¿k, let the comparability class Cl(f) of f be the
set of all 6 ∈ k((t))LE for which there exist positive r; s ∈ k such that fr 6 6 6 fs .
For f; g ∈ k((t))LE with f; g¿k we set Cl(f)¡Cl(g) if 6¡9 for all 6 ∈ Cl(f) and
for all 9 ∈ Cl(g). This de*nes a linear order on the set of comparability classes.
4.5. Proposition. Suppose f; g ∈ k((t))LE and f; g¿k. Then Cl(f) 6 Cl(g) if and only
if Lm (f =f) 6 Lm (g =g) if and only if Lm (log f) 6 Lm (log g).
5. Integration
In this section we will prove that we can integrate in k((t))LE ; that is, given
g ∈ k((t))LE we can *nd f ∈ k((t))LE with f = g. We call such an f an integral of
g. Indeed we will show that each series in Ln has an integral in Ln+1 . We begin by
considering g ∈ k((t))E and showing that there is only one obstruction to *nding an
integral in k((t))E .
5.1. The residue of a series g ∈ k((t))E is by de*nition the coe8cient of the monomial
x−1 in g; and is denoted by res g.
5.2. Proposition. Let g ∈ k((t))E . Then g has an integral in k((t))E if and only if
res g = 0.
Remark. The residue obstruction is natural since if f ∈ k((t))LE and f = x−1 ; then
f = log x + r for some r ∈ k; and log x does not belong to k((t))E .

5.3. Towards the proof of Proposition 5:2, let g = ar xr ∈ K0 and res g = a−1 = 0.
Then clearly
ar r+1
f := x
r+1
r=−1
satis*es f = g. Next, assume inductively that each element of Kn with residue 0 has
an integral in Kn . Let

g := ga E(a)
a ∈ An

in Kn+1 . Then f = fa E(a) in Kn+1 is an integral of g if and only if fa + fa a = ga

for all a ∈ An . Thus if g has an integral in Kn+1 ; then the equation y + ya = ga is
solvable in Kn ; for each a ∈ An such that ga = 0. Conversely, if each such equation

has a solution fa ∈ Kn ; then f := ga =0 fa E(a) ∈ Kn+1 is an integral of g. For a = 0

the equation y + ya = ga reads y = ga ; and has a solution in Kn by the inductive
assumption. Suppose a ∈ An and a = 0. Then the equation y + ya = ga has the same
solutions in Kn as the equation y−3y = h; where 3 := (−a )−1 and h := (a )−1 ga . Note
that 3; h ∈ Kn ; and that if n = 0; then Lm 3¡x while if n¿0; then 3 is in*nitesimal
with respect to Kn−1 by Lemma 3:5(iii). Therefore the “if ” part of Proposition 5:2 is
a consequence of the following lemma.
5.4. Lemma. Let h; 3 ∈ Kn ; with Lm 3¡x if n = 0; and 3 in4nitesimal with respect to

Kn−1 if n¿0. Then the equation y − 3y = h has a solution in Kn .
Proof. Let P be the operator on Kn de*ned by Py = 3y ; and let I be the identity
operator on Kn . Then the equation y − 3y = h (to be solved for y ∈ Kn ) becomes
(I − P)(y) = h. For n = 0 we view K0 as the series *eld k((G0 )); and we note that then
P is a small operator in the sense of 1.3, since
Supp Py ⊆ (Supp 3) · (Supp y ) ⊆ (x−1 Supp 3) · (Supp y)
for all y ∈ K0 . Thus, by 1.3, the equation (I − P)(y) = h has the solution y =

i
(I − P)−1 (h) = P (h) in K0 . Next, suppose n¿0. Let Supp∗ y denote the support
of an element y ∈ Kn ; where we consider Kn as the series *eld Kn−1 ((E(An−1 ))) over
Kn−1 . Thus Supp∗ y ⊆ E(An−1 ) and Supp∗ y ⊆ Supp∗ y for y ∈ Kn . When Kn is viewed
in this way as a series *eld over Kn−1 ; then P is again a small operator, because
Supp∗ Py ⊆ (Supp∗ 3) · (Supp∗ y)
for all y ∈ Kn . The desired result follows again from 1.3.
5.5. We have now established the “if ” direction of Proposition 5:2. For the “only if ”
direction, suppose g ∈ k((t))E has an integral in k((t))E . Let c := res g. Then g − cx−1
has residue 0; hence has an integral in k((t))E by the “if ” part of Proposition 5:2. Thus,
g − (g − cx−1 ) = cx−1 has an integral in k((t))E ; which implies c = 0 by the remark
following Proposition 5:2. Therefore res g = 0; as desired.
5.6. Theorem. Each g ∈ k((t))LE has an integral.
Proof. By Proposition 5:2 each g ∈ k((t))E = L0 has an integral in L1 . Suppose now

that g ∈ Lm . Then we have for f ∈ k((t))LE :
f = g ⇔ /m (f ) = /m (g)
m

⇔ /m (f) = /m (g) Ei (x) (by Theorem 3:9):
i=1
m m
Since /m (g) i=1 Ei (x) ∈ k((t))E there is y ∈ L1 such that y = /m (g) i=1 Ei (x). Then
f := /−m (y) is an integral of g and belongs to Km+1 .
Since we can take exponentials and integrals we can solve *rst order linear di+er-
ential equations.
5.7. Corollary. If f; g ∈ k((t))LE ; then there is y ∈ k((t))LE such that y + fy = g; and

in the homogeneous case (g = 0) we can take y = 0.
Proof. Let h0 ; h1 ∈ k((t))LE such that h0 = f and h1 = E(h0 )g. Then y = h1 =E(h0 ) is a
solution. If g = 0 we can take h1 = 1 which gives a nontrivial solution of the homoge-
neous equation.
Second order linear di8erential equations
5.8. Corollary 5:7 contains all there is to say about *rst order linear ODEs in one un-
known over k((t))LE . It would be nice to extend this to linear ODEs in one unknown
of any order. By Remark 4:2 the second order linear equation y + y = 0 has no
nontrivial solutions in k((t))LE . The following result completely settles the question
which homogeneous second order linear ODEs in one unknown have nontrivial solu-
tions in k((t))LE . (This is because such an equation u + +u + 8u = 0 (+; 8 ∈ k((t))LE )
transforms into an equation of the form y + fy = 0 by the change of variables u := y
where 0 = ∈ k((t))LE satis*es 2 + + = 0.)
5.9. Theorem. Let f ∈ k((t))LE . The equation y + fy = 0 has a nontrivial solution in
k((t))LE if and only if
1 1 1
f¡ 2
+ 2 2
+ 2
4x 4x (log x) 4x (log x) (log log x)2
2
1
+··· +
4x2 (log x)2 · · · (log : : : log(x))2

n-times
for some n.
5.10. Our present proof of Theorem 5:9 seems too long to us. That is why we do not
present it here, but only give a few indications below. (Details are available from the
authors upon request.) The topic of solvability of ODE’s in R((t))LE is a large one,
and we expect to return to it in a later article with a more de*nitive treatment and
more general results.
In our present proof of Theorem 5:9 we construct solutions explicitly using an ex-
tension of the “method of undetermined coe8cients”. Proofs that equations have no
nontrivial solutions use the valuation theoretic methods as in 4.2.
The precise cut in the ordered di+erential *eld k((t))LE that occurs in Theorem
5:9 has to do with the fact that each positive in*nite element of k((t))LE is bounded
below by log : : : log(x) for some n. There is evidence that the solvability of di+erential

n-times
equations in R((t))LE reNects solvability in Hardy *elds (of germs at +∞) of *nite
rank in the sense of [23]. (In these Hardy *elds each in*nitely increasing element is
also bounded below by some iterate log : : : log(x) of the logarithm. In the Hardy *eld

n-times
setting we let “x” denote the germ at +∞ of the identity function.)
More generally, we conjecture that R((t))LE is a kind of universal domain for ordered
di+erential algebra in Hardy *elds of *nite rank. We hope to return to this important
issue in a later paper. Here we only mention without proof the following exact analogue
of the theorem above for such Hardy *elds.
5.11. Theorem. Let f belong to a Hardy 4eld H of 4nite rank. Then the equation
y + fy = 0 has a nontrivial solution in a Hardy 4eld extension of H if and only if
1 1 1
f¡ + 2 + 2
4x2 4x (log x)2 4x (log x)2 (log log x)2
1
+··· +
4x2 (log x)2 · · · (log : : : log(x))2

n-times
for some n.
This theorem is closely related to the discussion in [7, pp. 78–80] and also to
[25, Theorem 3, part (3)]. (The main point is that a “nonoscillating” solution y of
y + fy = 0 in the sense of [7] gives rise to a solution z := −y =y de*ned on some
interval (a; +∞) of the corresponding *rst order Riccati equation z − z 2 = f. If the
germ of f at +∞ lies in a Hardy *eld, then the germ at +∞ of such a solution of
the Riccati equation lies in a bigger Hardy *eld.)
6. Composition and inversion
6.1. The goal of this section is to introduce the (formal) composition f ◦ g of two series
f; g ∈ k((t))LE with g¿k; and to show that this operation has the expected properties.
This is an elaborate a+air. At the end of this section we comment on how this section
relates to similar material in [8]. In order to state the main results we now introduce
some conventions and notations. We put
k((t))LE
∞ := {g ∈ k((t))
LE
: g ¿ k}:
Throughout this section f; g and h range over k((t))LE ; with g¿k and h¿k. Further
restrictions on f; g and h may be imposed in some particular context.
6.2. Theorem. There is a unique operation
(f; g) → f ◦ g : k((t))LE × k((t))LE

∞ → k((t))
LE
such that
(1) r ◦ g = r for all r ∈ k;
(2) f ◦ x = f and x ◦ g = g;
(3) the map 6 → 6 ◦ g : k((t))LE → k((t))LE is an ordered 4eld embedding;

(4) if f = fi in k((t))LE ; then f ◦ g = fi ◦ g in k((t))LE .

(5) E(f) ◦ g = E(f ◦ g) and /(f) = f ◦ E(x);
(6) (f ◦ g) ◦ h = f ◦ (g ◦ h).
We also establish the chain rule and the existence of compositional inverses:
6.3. Proposition. Let ◦ be the composition operation of the theorem. Then:
(f ◦ g) = (f ◦ g) · g :
Moreover; k((t))LE
∞ is a group under the operation ◦; with identity element x.
Before we de*ne composition we make a few simple observations:
6.4. Lemma. Let ◦ be a composition operation as in the theorem (without assuming

uniqueness). Then
(1) (log f) ◦ g = log(f ◦ g) for f¿0;
(2) xr ◦ g = gr ;
(3) /m (f) ◦ g = f ◦ Em (g) and /−m (f) ◦ g = f ◦ Lm (g).
Proof. If f¿0 we have by (5) of the Theorem:
E((log f) ◦ g) = E(log f) ◦ g = f ◦ g = E(log(f ◦ g));
hence (log f) ◦ g = log(f ◦ g); which is (1). Thus by the Theorem:
xr ◦ g = E(r log x) ◦ g = E((r log x) ◦ g)) = E(r log g) = gr ;
which is (2). Now the *rst identity of (3) follows easily by associativity from the
second identity of (5) in the Theorem. To obtain the second identity of (3) note that
/m (/−m (f)) = f = (f ◦ Lm x) ◦ Em x = /m (f ◦ Lm x);
by (1), associativity, and the *rst identity. Hence /−m (f) = f ◦ Lm x; and thus
/−m (f) ◦ g = (f ◦ Lm x) ◦ g = f ◦ Lm g;
again by (1) and associativity.

6.5. The theorem and the lemma suggest an inductive scheme for de*ning composition.
The three key inductive formulas are as follows:

If f = ar xr in K0 ; then f ◦ g = ar g r :

If f = fa E(a) in Km+1 ; then f ◦ g = (fa ◦ g)E(a ◦ g):
If f = /−m (f̃) with f̃ ∈ k((t))E ; then f ◦ g = f̃ ◦ Lm g:

It is easy to show that the *rst in*nite sum ar gr exists. The di8culty in showing
existence of the second in*nite sum on the right is to control Supp (f ◦ g) in terms of
Supp f and Supp g; during the induction. The next two (easy) technical lemmas will
be useful in this regard. Notation (also used later in this section): for any subset K of
k((t))LE ⊆ k((G LE )) we put

Supp K := Supp s:
s∈K
6.6. Lemma. Let K be a sub4eld of LM; N = k((GM; N )) with k ⊆ K; and let A be a

k-linear subspace of K such that E(+)¿K for all + ∈ A¿0 . (Hence +¿k for all
+ ∈ A¿0 .)
Then there is an injective homomorphism + → u+ from the additive group A into
the multiplicative group GM; N +1 such that for all + ∈ A
(i) E(+) = u+ · y+ with y+ ∈ LM; N ; and Supp y+ ⊆ [(Supp +)61 ] ⊆ Supp K .
(ii) +¿0 ⇒ u+ ¿K.
Proof. Put AM; i := /−M (Ai ); and mM; N := /−M (mN ) so that
k((GM;N )) = AM; N ⊕ AM; N −1 ⊕ · · · ⊕ AM; 0 ⊕ k ⊕ mM;N :
Write each + ∈ A as + = a + r + with a ∈ AM; N ⊕ AM; N −1 ⊕ · · · ⊕ AM; 0 ; r ∈ k and

∈ mN ; and put u+ = E(a) and y+ = exp(r) p =p!. One easily checks that the lemma
holds with these choices of u+ ’s and y+ ’s.
6.7. Lemma. Let A be an ordered abelian (multiplicative) group, C a convex subgroup

of A; and U and V reverse well ordered subsets of A such that
(i) (u1 ¡u2 in U ) ⇒ u2 =u1 ¿C;
(ii) v ∈ V ⇒ v¡C.
Suppose for each u ∈ U and p ∈ N we are given a reverse well ordered set Su;p ⊆ C.
Then the subset

W := Su;p · u · V p
u∈U;p∈N
of A is reverse well ordered. Moreover; for each w ∈ W there are only 4nitely many
tuples (u; p; s; v1 ; : : : ; vp ) with u ∈ U; p ∈ N; s ∈ Su;p and v1 ; : : : ; vp ∈ V such that
w = s · u · v1 · : : : · vp :
Proof. Let UW ; VW and WW be the images of U; V and W under the canonical map from
A onto AW := A=C. Then clearly
p
WW = UW · VW = UW · [VW ]:
p∈N
Using that UW and VW are reverse well ordered subsets of the ordered abelian group AW
and that VW ¡1; we easily obtain the desired result from Neumann’s Lemma.
6.8. We *rst de*ne the formal composition f ◦ g when f belongs to k((t))E . This is
done in stages, the mth stage de*ning f ◦ g for all f ∈ Km .
Instead of f ◦ g we will also write f(g); because we think of f ◦ g as the result of
substituting g for the indeterminate x in the series f = f(x).

Stage 0: Write f ∈ K0 as f = r∈k ar xr (all ar ∈ k), and put

f(g) := ar gr :
For this de*nition to make sense we have to check that the right hand sum ex-
ist in k((t))LE . Suppose g ∈ LM; n = k((GM; n )); and write g = c3(1 + ) with c = Lc(g);
¡1
3 = Lm(g); and ∈ mM; n . Put S := Supp ; so S ⊆ GM; n . Then, with r ∈ k:
r
r r r
g =c 3 p
p
p
which gives Supp gr ⊆ 3r [S]. Since 3¿1 and {r ∈ k : ar = 0} is reverse well ordered,
the set {3r : r ∈ k; ar = 0} is also reverse well ordered (in GM; n ). Thus by Neumann’s

lemma a gr exists in LM; n . The same argument shows that if f = i∈I fi in K0 ; then

r
f(g) = fi (g) in LM; n . This fact is easily seen to imply that the map f → f(g) : K0 →
LM; n is an ordered *eld embedding over k.
Stage m+1: Assume inductively that we have de*ned f(g) as an element of k((t))LE
for all f ∈ Km and all g. We will also assume inductively that for g ∈ LM; n the following
four properties hold.
(1)m For all f ∈ Km we have f(g) ∈ LM; m+n .

If f = i∈I fi in Km , then f(g) = i∈I fi (g) in LM; m+n ;

(2)m The map f → f(g) : Km → LM; m+n is an ordered *eld embedding over k.
We shall denote the image of this map by k((Gm (g))).
(3)m If D ∈ Supp k((Gm (g))), then D¿% for some % ∈ k((Gm (g)))¿0 · Supp g .
(4)m If m = 0 and f ∈ K0 ; ∈ LM; n with Lm()¡Lm(g), then

∞
f(p) (g)
f(g + ) = p in LM;n :
p!
p=0
If m¿0 and f ∈ Km ; ∈ LM; n ; ||¡k((Gm−1 (g)))¿0 , then

∞
f(p) (g)
f(g + ) = p in LM;m+n :
p!
p=0
6.9. To justify the inductive assumption for m = 0 we now show that (1)0 – (4)0 hold
for g ∈ LM; n . The arguments of Stage 0 have already established (1)0 and (2) 0 . For

(3)0 , write g = c · (3 + h) with c = Lc(g) and 3 = Lm(g). Then gr = cr 3r p pr (h=3)p .

Thus for each D ∈ Supp gr there are p ∈ N and D1 ; : : : ; Dp ∈ Supp h ⊆ Supp g such that
D = 3r · (D1 =3) · · · (Dp =3)¿ 12 c−r gr · (D1 =3) · · · (Dp =3) ∈ k((G0 (g)))¿0 · Supp g

which proves (3)0 . For (4)0 , let f = r ar xr in K0 . Then f(p) =p! = r ar pr xr−p ,

hence f(p) (g)=p! = r ar pr gr−p , and thus

f(g + ) = ar (g + )r = ar (g(1 + g−1 ))r
r r

r

r

r −1 p
= ar g (g ) = ar gr−p p
r p
p p
(p; r)∈N×k
f (p)
(g) p
=
p
p!
as desired.
6.10. We are now ready to de*ne f(g) for f ∈ Km+1 . Write f = a∈Am fa E(a) in
Km+1 , and put

f(g) := fa (g)E(a(g)): (∗)
For this de*nition to make sense we have to check that the right hand sum in (∗)
exists in k((t))LE . That is where the four part inductive assumption is essential.
Remark. At the beginning of “stage m + 1” we assumed that f(g) has already been
assigned a value for f ∈ Km . This agrees with (∗): if f ∈ Km is considered as an
element of Km+1 , then only one term in the in*nite sum in (∗) can be nonzero, and
this term equals f(g) as given by the original assignment. So there is no conNict of
notation.
6.11. To complete the induction we have to establish the following:

(I) The in*nite sum in (*) de*ning f(g) for f ∈ Km+1 exists in k((t))LE .
(II) Properties (1)m+1 through (4)m+1 hold for g ∈ LM; n .
6.12. Proof of (I). Let f = a∈Am fa E(a) in Km+1 and let g ∈ LM; n . We will show that

then a fa (g)E(a(g)) exists in LM; m+n+1 . This amounts to checking:

(a) For each 3 ∈ GM; m+n+1 there are only *nitely many a ∈ Am such that 3 ∈
Supp fa (g)E(a(g)).
(b) The subset

Supp fa (g)E(a(g))
a∈Am
of GM; m+n+1 is reverse well ordered.
Suppose for a moment that every monomial of Supp k((Gm (g))) is greater than some
element of k((Gm (g)))¿0 . It is then not hard to verify (a) and (b), by *rst deriving
from this assumption that
a1 ¡ a2 in Am ⇒ Supp fa1 (g)E(a1 (g)) ¡ Supp fa2 (g)E(a2 (g)):
The problem, however, is that Supp k((Gm (g))) may contain monomials that are smaller
than all positive elements of k((Gm (g))). (For example, this happens for m = 0 and
g = x+E(−x) with the monomial E(−x).) In that case there may be interference among
the sets Supp fa (g)E(a(g)) as a ranges over Am , and this causes trouble in verifying (a)
and (b). It is because of this possibility that the following more complicated argument
(using (3)m and (4)m ) seems to be unavoidable.
Write g = h + , where h ∈ LM; n contains those monomials of g that are greater than
some element of k((Gm (g)))¿0 and where , ∈ LM; n contains those monomials of g that
are less than k((Gm (g)))¿0 . We now claim that Lt(f(g)) = Lt(f(h)) for f ∈ Km¿0 . To
see this, note that by (4)m we have
∞
f(p) (g) p
f(h) = f(g − ,) = f(g) + (−1)p , :
p!
p=1
All monomials of the in*nite sum on the right hand side are less than Lm(f(g)), since
f(g) ∈ k((Gm (g)))¿0 and |,|¡k((Gm (g)))¿0 . Thus Lt(f(g)) = Lt(f(h)) as claimed. It
follows that each element of k((Gm (g)))¿0 (no matter how small) is greater than
some element of k((Gm (h)))¿0 , and conversely, that each element of k((Gm (h)))¿0
is greater than some element of k((Gm (g)))¿0 . The last statement implies that |,| ¡
k((Gm (h)))¿0 .
Hence by (4)m we have for a ∈ Am (and h in the role of g):
∞
a(p) (h)
a(g) = a(h) + ,p :
p!
p=1
The in*nite sum in this formula is k-in*nitesimal, and therefore

 
∞
E(a(g)) = E(a(h)) 1 + Fp (a (h); : : : ; a(p) (h)),p  ;
p=1
where Fp (X1 ; : : : ; Xp ) ∈ Q[X1 ; : : : ; Xp ] is a polynomial independent of a and g. We also

let F0 = 1 ∈ Q. Applying (4)m once more we get
∞
fa(p) (h)
fa (g) = ,p
p!
p=0
and multiplying the last two identities gives

∞

fa (g)E(a(g)) = E(a(h)) Ha;p ,p ; (†)
p=0

p
where Ha;p := i=0 (fa(i) (h)=i!)Fp−i (a (h); : : : ; a(p−i) (h)) ∈ k((Gm (h))).
We now apply Lemma 6:6 to the *eld K := k((Gm (h))), with N := m + n and with
A := {a(h) : a ∈ Am }. The lemma tells us that for each a ∈ Am we can write E(a(h)) =
ua ya , where ua ∈ GM; N +1 and ya ∈ LM; N , such that Supp ya ⊆ Supp K and such that
ua1 =ua2 ¿K whenever a1 ¿a2 in Am . Next, we apply Lemma 6:7 with A := GM; N +1 ,
C := convex hull of Supp K in GM; N ; U := {ua : fa = 0}; V := Supp ,, and where for
each u = ua ∈ U and p ∈ N we take
Su;p := Supp Ha;p · Supp ya
which is a reverse well ordered subset of C. Our application of Lemma 6:6 shows
that hypothesis (i) of Lemma 6:7 is satis*ed in this situation. In order to verify that
hypothesis (ii) of Lemma 6:7 is also satis*ed we *rst derive the following:
() Each element of Supp K is greater than some element of K ¿0 .
This reduces to showing that each element of Supp K is greater than some element
of K ¿0 . By (3)m applied to h instead of g this in turn will be the case if each element
of Supp h is greater than some element of K ¿0 . But each element of Supp h is a
product of *nitely many monomials each of which is greater than some element of
k((Gm (g)))¿0 , hence greater than some element of K ¿0 (by an earlier remark), and
thus their product is greater than some element of K ¿0 . This *nishes the proof of ().
Now each element of Supp , is less than k((Gm (g)))¿0 , and thus less than K ¿0 ,
which by () implies that each element of Supp , is less than Supp K , hence less
than C. Thus hypothesis (ii) of Lemma 6:7 is satis*ed in our situation. Note that by
(†) above we have for each u = ua ∈ U :

Supp fa (g)E(a(g)) ⊆ Su;p · u · V p :
p
Hence by Lemma 6:7 we obtain properties (a) and (b) above, which express the

existence of fa E(a(g)) in LM; m+n+1 . This *nishes the proof of (I).
6.13. Proof of (II). The way we established (I) also leads to the *rst part of (1)m+1 ,
and the second part follows in a similar way. Property (2)m+1 follows easily from

the de*nitions and (1)m+1 . To prove (3)m+1 , let f = fa E(a) in Km+1 , and let
D ∈ Supp f(g). Following the proof of (I) and using the notations introduced
there it follows from (†) that D = sua v with s ∈ Supp K ; ua ∈ U , and v ∈ Supp , ⊆
Supp g . Since ua ¿cE(a(h))¿(c=2)E(a(g)) for some c ∈ k¿0 it follows easily from
other facts obtained in the proof of (I) that D is greater than some element of
k((Gm+1 (g)))¿0 · Supp g . This concludes the proof of (3)m+1 .
Now (4)m+1 . Let f ∈ Km+1 and ∈ LM; n with ||¡k((Gm (g)))¿0 . We have to show

that f(g + ) = p (f(p) (g)=p!)p in LM; m+n+1 .

Case 1: Suppose f = E(a) with a ∈ Am . It is a routine exercise to show that then
for each p ∈ N we have
f(p)
= Fp (a ; : : : ; a(p) )E(a);
p!
where Fp (X1 ; : : : ; Xp ) ∈ Q[X1 ; : : : ; Xp ] is the polynomial introduced in the proof of (I).
Thus, arguing as in that proof and using (2)m+1 we have
f(g + ) = E(a(g + ))

 
∞

= E(a(g)) 1 + Fp (a (g); : : : ; a(p) (g))p 
p=1
∞
f(p) (g)
= p :
p!
p=0
Case 2: Suppose f = sE(a) with s ∈ Km and a ∈ Am . By (4)m we have s(g + ) =
(p)
p (s (g)=p!)p , which in combination with Case 1 easily implies the desired result
for f.

General Case: Write f = fa E(a) in Km+1 . Put 6a = fa E(a) for a ∈ Am . Then by

(1)m+1 and case 2

6(p) a (g) p
f(g + ) = 6a (g + ) =
a a p
p!
6(p)
a (g)
= p
a;p
p!

6(p)
a (g) p
=
p a
p!
f(p) (g)
= p
p
p!
provided the sum a;p (6(p)

a (g)=p!)
p
exists in LM; m+n+1 . To verify that this sum
indeed exists we proceed as follows. Note that 6(p) a =p! = sa;p E(a) with sa;p ∈ Km (and
sa;p = 0 if fa = 0). Next write g = h + , exactly as in the proof of (I). As in that proof
we obtain for a ∈ Am and p ∈ N
∞
6(p)
a (g)
= sa;p (g)E(a(g)) = E(a(h)) Ha;p;q ,q
p!
q=0

with Ha; p; q ∈ k((Gm (h))). Hence the above sum a;p will exist in LM; m+n+1 if the sum

Ha;p;q E(a(h)),q p
a;p;q
exists in LM; m+n+1 . To see that the sum we just displayed indeed exists we argue, once
again, as in the proof of (I). We let K; N; A; ua and ya (for a ∈ Am ); A and C be
exactly as in the latter half of that proof. We also put V1 := Supp and V2 := Supp ,
and for u = ua ∈ U and p; q ∈ N we take
Su;p;q := Supp Ha;p;q · Supp ya ;
a reverse well ordered subset of C. Note that with these notations we have
Supp(Ha;p;q E(a(h)),q p ) ⊆ Su;p;q · u · V1p · V2q :
An easy extension of Lemma 6:7 then tells us that the subset

W := Su;p;q · u · V1p · V2q
u∈U;p;q∈N
of A = GM; m+n+1 is reverse well ordered and that for each element w of W there
are only *nitely many tuples (u; p; q; s; v1; 1 ; : : : ; v1; p ; v2; 1 ; : : : ; v2; q ) with u ∈ U; p; q ∈ N,
s ∈ Su; p; q , v1; 1 ; : : : ; v1; p ∈ V1 and v2; 1 ; : : : ; v2; q ∈ V2 such that
w = s · u · v1;1 · · · v1;p · · · v2;1 · · · v2;q :

But that is exactly what we need to conclude that the above sum a; p; q exists. This
*nishes the proof of (II).
6.14. We have now de*ned f ◦ g—also written as f(g)—for f ∈ k((t))E . We re-

mark here that we had no choice in the matter: if ◦ is a composition operation as in
Theorem 6:2 (without assuming uniqueness), then it has to assign to elements f; g with
f ∈ k((t))E the value that the above de*nitions assigned to it, as follows from Lemma
6:4 and the subsequent discussion in 6.5.
Let us record here some properties that follow easily from our inductive de*nition
and proof. These properties will be useful in extending this composition to f ∈ k((t))LE
and in deriving similar properties for this extended composition.
6.15. Lemma. Let f ∈ k((t))E . Then

(1) r ◦ g = r for all r ∈ k;
(2) f ◦ x = f and x ◦ g = g;
(3) the map 6 → 6 ◦ g : k((t))E → k((t))LE is an ordered 4eld embedding;

(4) if f = fi in k((t))E ; then f ◦ g = fi ◦ g in k((t))LE ;

(5) E(f) ◦ g = E(f ◦ g); /(f) ◦ g = f ◦ E(g); and /−n (f) = f ◦ Ln (x);
(6) fs ◦ g = (f ◦ g) s for all s ∈ k; where f¿0;
(7) (f ◦ g) = (f ◦ g) · g .
Proof. Items (1) – (4) have been established during the induction or are immediate
consequences. For (5) – (7) we also argue by induction on f, using (1) – (4) freely.

r
Suppose *rst that f ∈ K0 and write f = ar x . Put f∞ := r¿0 ar xr and (f) :=
r
r¡0 ar x , so that f = f∞ + a0 + (f). Then
(f)k
E(f) = E(f∞ ) · exp(a0 ) · ;
k!
k
hence by (3) and (4)

(f)k ◦ g
E(f) ◦ g = (E(f∞ ) ◦ g) · exp(a0 ) · :
k!
k
Also f ◦ g = (f∞ ◦ g) + a0 + ((f) ◦ g), and hence

((f) ◦ g)k
E(f ◦ g) = E(f∞ ◦ g) · exp(a0 ) · :
k!
k
Now note that f∞ ∈ A0 , so that E(f∞ ) ◦ g = E(f∞ ◦ g) by de*nition; moreover, by

(3) we have (f)k ◦ g = ((f) ◦ g)k for k ∈ N. Hence E(f) ◦ g = E(f ◦ g), which is

the *rst identity of (5). For the second identity we note that /(f) = ar E(rx) so

that /(f) ◦ g = ar E(rg) = f ◦ Eg. For the third identity of (5) we *rst note that
/−n (xr ) = (/−n x)r = (Ln (x))r by an identity observed at the end of (2.7); thus, by (4)

and an identity in (2.8): /−n f = ar (Ln x)r = f ◦ Ln x.

For (6) we assume f¿0. Write f = axb (1 + ) with a; b ∈ k; a¿0 and ∈ m.

Then fs = a s xbs k ( ks ) k , so that fs ◦ g = a s g bs k ( ks )(k ◦ g). On the other hand,

f ◦ g = ag b (1 + ( ◦ g)), and hence ( f ◦ g) s = a s g bs k ( ks )( ◦ g)k . Thus fs ◦ g =

( f ◦ g) s . Now (7):

(f ◦ g) = ar g r = rar gr−1 g = rar gr−1 · g = (f ◦ g) · g :
Next we assume inductively that (5) – (7) hold for all f ∈ Km . Suppose f ∈ Km+1

and write f = fa E(a) in Km+1 . Put f∞ := a¿0 fa E(a) and ( f) := a¡0 fa E(a),
so that f = f∞ + f0 + ( f). Then
(f)k
E(f) = E(f∞ ) · E(f0 ) · ;
k!
k
hence by (3) and (4)

(f)k ◦ g
E(f) ◦ g = (E(f∞ ) ◦ g) · (E(f0 ) ◦ g) · :
k!
k
Also f ◦ g = (f∞ ◦ g) + ( f0 ◦ g) + (( f) ◦ g), and hence

((f) ◦ g)k
E(f ◦ g) = E(f∞ ◦ g) · E(f0 ◦ g) · :
k!
k
Now note that f∞ ∈ Am+1 , so that E( f∞ ) ◦ g = E(f∞ ◦ g) by de*nition; moreover, by

the inductive assumption we have E(f0 ) ◦ g = E(f0 ◦ g), and by (3) we have ( f)k ◦ g =
(( f) ◦ g)k for k ∈ N. Hence E(f) ◦ g = E(f ◦ g), which is the *rst formula of (5). The

second formula of (5) is obtained as follows: /(f) = /(fa )E(/a), hence

/(f) ◦ g = (/(fa ) ◦ g)(E(/a) ◦ g) = (fa ◦ Eg)E(/a ◦ g)

= (fa ◦ Eg)E(a ◦ Eg) = f ◦ Eg:
The third formula of (5) follows from the various inductive assumptions and (4):

/−n (f) = /−n (fa )E(/−n a) = (fa ◦ Ln x)E(a ◦ Ln x)

= (fa ◦ Ln x)(E(a) ◦ Ln x)
= f ◦ Ln x:
For (6) we assume f¿0. Write f = f+ E(+)(1 + ) with + := max{a ∈ Am : fa = 0},

and ∈ mm+1 . Then fs = f+s E(s+) k ( ks )k and f ◦ g = (f+ ◦ g)E(+ ◦ g)(1 + (g)), so
that
s
s s
f ◦ g = (f+ ◦ g)(E(s+) ◦ g) (k ◦ g)
k
k
s
= (f+ ◦ g)s E(s+ ◦ g) ( ◦ g)k
k
k
s
s s
= (f+ ◦ g) (E(+ ◦ g)) ((g)k
k
k
= (f ◦ g)s :

To obtain (7), note that f = ( fa + a fa )E(a) and f(g) = fa (g)E(a(g)). Thus

f(g) = (fa (g) + a(g) fa (g))E(a(g))

= (fa (g)g + a (g)g fa (g))E(a(g))
= f (g)g :
This *nishes the proof of the lemma.
6.16. We are now ready to de*ne f ◦ g for arbitrary f. Suppose that f ∈ Lm . Then
f̃ := /m ( f) ∈ k((t))E , and we put
f ◦ g := /m f ◦ Lm g = f̃ ◦ Lm g: (∗∗)
Note that if m = 0 the right hand side of (∗∗) equals f ◦ g as previously de*ned.
In fact, formula (∗∗) is forced on us by part (3) of Lemma 6:4 and the subsequent
discussion in 6.5. In other words, we have established the part of Theorem 6:2 that
says there is at most one composition operation with the desired properties (1) – (6).
We have to show, of course, that the right hand side of (∗∗) is independent of the
choice of m ( for given f and g). This will follow if we prove that increasing m does
not change the right hand side in (∗∗). Increasing m by 1 has the e+ect of replacing
f̃ by /(f̃). Thus the right hand side in (∗∗) gets replaced by
/(f̃) ◦ Lm+1 g = f̃ ◦ E(Lm+1 g) = f̃ ◦ Lm g;
so this right hand side is indeed invariant when we increase m.

We have now (at last) de*ned unambiguously a composition operation
( f; g) → f ◦ g : k((t))LE × k((t))LE LE
∞ → k((t)) :
It remains to show that this composition satis*es properties (1) – (6) of Theorem 6:2.
6.17. Proof of (1) -- (6) of Theorem 6.2. Let f ∈ Lm and de*ne f̃ as before. We will
freely use Lemma 6:15, which already gives us (1) and the second identity of (2). To
get the *rst identity of (2) we observe, using the third formula of Lemma 6:15(5):
/m (f ◦ x) = /m (f̃ ◦ Lm x) = /m (/−m f̃) = /m f;
hence f ◦ x = f.
Properties (3) and (4) follow from parts (3) and (4) of Lemma 4 and the fact that
the powers of / are ordered *eld embeddings that preserve in*nite sums. The *rst
identity of property (5) is obtained from /m (Ef) = E(/m ( f)) as follows:
E(f) ◦ g = E(f̃) ◦ Lm g = E(f̃ ◦ Lm g) = E(f ◦ g):
The second identity of (5) is obtained as follows:
/(f) = /(f) ◦ x = /m (/(f) ◦ Lm x = /(f̃) ◦ Lm x

= f̃ ◦ E(Lm x) = f̃ ◦ Lm (E(x)) = f ◦ E(x):
To obtain associativity we need the following auxiliary identities:
(log g) ◦ h = log(g ◦ h); and gr ◦ h = (g ◦ h)r for r ∈ k:
The *rst of these follows by applying E to both sides, and using (5), and for the
second we write g r = E(r log g) and apply the *rst identity and (5). We are now ready
to prove (6) for f = ar x in K0 . Then f ◦ g = ar g r , hence by (4) and the second

auxiliary identity:

(f ◦ g) ◦ h = ar (gr ◦ h) = ar (g ◦ h)r = f ◦ (g ◦ h):

Next, suppose (6) holds for all f ∈ Kn , and let f = fa E(a) in Kn+1 . Then f ◦ g =
( fa ◦ g)E(a ◦ g)), and thus

(f ◦ g) ◦ h = ((fa ◦ g) ◦ h)E((a ◦ g) ◦ h)

= (fa ◦ (g ◦ h))E(a ◦ (g ◦ h))
= f ◦ (g ◦ h):
We have now established associativity for f ∈ k((t))E . Using this, and the *rst aux-
iliary identity above we obtain for f ∈ Lm :
(f ◦ g) ◦ h = (f̃ ◦ Lm g) ◦ h = f̃ ◦ ((Lm g) ◦ h)
= f̃ ◦ (Lm (g ◦ h)) = f ◦ (g ◦ h):
This *nishes the proof of (6), and thereby the proof of the theorem.
The following special case of the preservation of in*nite sums under composition is
worth mentioning.
6.18. Corollary. Let F ∈ k[[Y1 ; : : : ; Yn ]]; and let 1 ; : : : ; n be in4nitesimals of k((t))LE .

Then
F(1 ; : : : ; n ) ◦ g = F(1 (g); : : : ; n (g)):
6.19. We now verify the chain rule: ( f ◦ g) = (f ◦ g)g . Using previous notations this
is seen as follows:
(f ◦ g) = (f̃ ◦ Lm g)

= (/m (f) ◦ Lm g)(Lm g) (by 6:15(7))
g
= (/m (f) ◦ Lm g) m i−1 (by 3:9(3))
i=1 L g
m
g
= (/m (f ) ◦ Lm g) Ei (x) ◦ Lm g m i−1 (by 3:9(5))
i=1 i=1 L g
m
g
= (/m (f ) ◦ Lm g) Li−1 (g) m i−1 g
i=1 i=1 L
m m
= (/ (f ) ◦ L g)g
= (f ◦ g)g :
Compositional Inverse
6.20. By property (3) of Theorem 6:2 the map
6 → 6 ◦ g : k((t))LE → k((t))LE
is injective and sends k((t))LE LE
∞ to itself. Thus k((t))∞ is a monoid under composition
with identity element x. It remains to show that this monoid is actually a group. First
some easy observations:
(1) The elements En (x) and Ln (x) are each others inverse, namely
En (x) ◦ Ln (x) = Ln (x) ◦ En (x) = x:
(2) Whenever f ◦ g = x, then f¿k and g ◦ f = x.
Observation (1) is clear. To see why (2) holds, let f ◦ g = x. Since x¿k it follows by
property (3) of Theorem 6:2 that f¿k. By associativity we get
(g ◦ f) ◦ g = g ◦ (f ◦ g) = g ◦ x = x ◦ g
and thus, by the injectivity of the map above, x = g ◦ f, as desired.
The existence of compositional inverses is established in several lemmas which re-
present successive steps in constructing the inverse.
6.21. Lemma. Given any g; there are m and n such that

Lm (x) ◦ g ◦ En (x) = x + ;
where is an in4nitesimal of k((t))E .
Proof. If g ∈ Ln , then g ◦ En (x) = /n (g) ∈ k((t))E , which reduces the proof of the
lemma to the case that g ∈ k((t))E . Actually, we will only assume that the leading
monomial Lm(g) of g belongs to Gm for some m, as this is more convenient in the
inductive argument below. We now proceed by induction on m.
Suppose m = 0. Then g = cxr (1+,) for some c; r ∈ k¿0 and in*nitesimal , ∈ k((t))LE .
Conjugating with L(x) three times in succession has the following e+ect, as one easily
veri*es:
L3 (x) ◦ g ◦ E3 (x) = x +
for some in*nitesimal . Suppose ∈ Lk with k¿0. Then L3 (x) ◦ g ◦ E4 (x) = E(x)(1 +
(E(x))=E(x)), hence L4 (x) ◦ g ◦ E4 (x) = x + 1 with in*nitesimal 1 ∈ Lk−1 . Continuing
this way we obtain
L3+k (x) ◦ g ◦ E3+k (x) = x + k
with in*nitesimal k ∈ L0 = k((t))E . This *nishes the case m = 0.
Suppose that the lemma holds for all g whose leading monomial belongs to Gm .
Let g have leading monomial in Gm+1 \Gm . Write g = 6E(a)(1 + ,) with 0¡6 ∈ Km ;
0¡a∈ Am and in*nitesimal ,. Then
L(x) ◦ g = log(g) = a + log(6) +
for some in*nitesimal . Clearly Lm(L(x) ◦ g) = Lm(a) ∈ Gm . Now apply the inductive
hypothesis to L(x) ◦ g.
6.22. Lemma. Let m 6 n; f ∈ Km and ∈ Kn ; such that Lm ¡x in case m = 0; and

Lm ¡Gm−1 for m¿0. Then f(x + ) ∈ Kn and Lm f(x + ) = Lm f.
Proof. By induction on m. For m = 0 this follows straight from the de*nition in 6.8.
Assume the result holds for all f ∈ Km . Let m + 1 6 n; f ∈ Km+1 and ∈ Kn with
Lm ¡Gm . We have to show that then f(x + ) ∈ Kn . For this we simply consult the
proof of (4)m+1 in 6.13, with g = x. It is shown there that if f = E(a) with a ∈ Am ,
then
 
∞
f(x + ) = E(a) 1 + Fp (a ; : : : ; a(p) )p  ;
p=1
and since the in*nite sum on the right is an in*nitesimal of Kn , we get f(x + ) ∈ Kn ,
and Lm f(x + ) = E(a) = f = Lm f. The general case now follows easily from the
inductive hypothesis and the fact that Lm f(x + ) = Lm((Lm f)(x + )).
6.23. Lemma. Let ∈ Kn and Lm ¡x. Then there exists , ∈ K0 with Lm ,¡x such
that (x + ,) ◦ (x + ) = x + ∗ with Lm ∗ ¡G0 . In particular, (x + ,) ◦ (x + ) = x
if n = 0.
Remark. We do not need to assume here that is in*nitesimal. But if is in*nitesimal,

then clearly any , as in the lemma must also be in*nitesimal.
Proof. Write = 0 +1 with 0 ∈ K0 and Lm 1 ¡G0 . For f ∈ K0 we have by the Taylor
expansion of (4)0 :
f(x + ) = f(x + 0 ) + a
for some a ∈ Kn with Lm a¡G0 . Hence, it su8ces to *nd , ∈ K0 with Lm ,¡x such
that f(x + 0 ) = x for f = x + ,. Thus replacing by 0 , we assume from now on that
∈ K0 . Put D := {, ∈ K0 : Lm ,¡x}. Then we have for , ∈ D
∞

(x + ,) ◦ (x + ) = x + + , + (,(p) =p!)p
p=1
= x + + (I + P )(,);
where I; P : D → D are the operators de*ned by I(,) = , and

∞

P (,) = (,(p) =p!)p :
p=1
Thus to *nd , ∈ D such that (x + ,) ◦ (x + ) = x amounts to solving the equation

(I + P )(,) = − for the unknown , ∈ D. To do this we consider K0 as the series *eld
k((G0 )) over k. Then P is a small operator, because
Supp P (,) ⊆ x−1 [Supp ] · Supp ,
for , ∈ D. Thus by 1.3 we have that , := (I − P + P2 − · · ·)(−) is a well de*ned

element of D and satis*es (x + ,) ◦ (x + ) = x.
6.24. Lemma. Let 0¡m 6 n; and let ∈ Kn with Lm ¡Gm−1 . Then there exists
, ∈ Km with Lm ,¡Gm−1 such that (x + ,) ◦ (x + ) = x + ∗ with Lm ∗ ¡Gm . In
particular, (x + ,) ◦ (x + ) = x if m = n.
Proof. Write = 0 + 1 with 0 ∈ Km and Lm 1 ¡Gm . For f ∈ Km we have by

Lemma 6.22 and the Taylor expansion of property (4)m in 6.8:
f(x + ) = f(x + 0 ) + a
for some a ∈ Kn with Lm a¡Gm . Hence, it su8ces to *nd , ∈ Km with Lm ,¡Gm−1

such that f(x + 0 ) = x for f = x + ,. Thus replacing by 0 , we assume from now
on that ∈ Km . Put D := {, ∈ Km : Lm ,¡Gm−1 }. By Taylor expansion we have for
, ∈ D:
∞

(x + ,) ◦ (x + ) = x + + , + (,(p) =p!)p
p=1
= x + + (I + P )(,)
where I; P : D → D are the operators de*ned by I(,) = , and

∞

P (,) = (,(p) =p!)p :
p=1
Thus to *nd , ∈ D such that (x + ,) ◦ (x + ) = x amounts to solving the equation

(I + P )(,) = − for the unknown , ∈ D. To do this we use once more the method of
small operators.
Let Supp∗ y denote the support of an element y ∈ Km , where we consider Km as
the series *eld Km−1 ((E(Am−1 ))) over Km−1 . Thus Supp∗ y ⊆ E(Am−1 ) and Supp∗ y ⊆
Supp∗ y for y ∈ Km . When Km is viewed in this way as a series *eld over Km−1 , then
P is a small operator, because
Supp∗ P (,) ⊆ S · Supp∗ ,
for , ∈ D, where
∞

S := (Supp∗ )p ⊆ E(Am−1 )¡1 :
p=1
Thus by 1:3 we have that , := (I − P + P2 − · · ·)(−) is a well de*ned element of

D and satis*es (x + ,) ◦ (x + ) = x.
6.25. Corollary. For each in4nitesimal ∈ Kn there is an in4nitesimal , ∈ Kn such that

(x + ,) ◦ (x + ) = x.
Proof. By one application of Lemma 6:23 followed by n applications of Lemma 6:24,

for m = 1; : : : ; m = n successively, we can compose on the left with suitable elements
(x + ,0 ); (x + ,1 ); : : : ; (x + ,n ) with in*nitesimals ,0 ∈ K0 ; ,1 ∈ K1 ; : : : ; ,n ∈ Kn to obtain
the identity x. Lemma 6.22 guarantees that we stay inside Kn during this process.
6.26. Given any g we obtain now as follows an f¿k with f ◦ g = x. First, by

Lemma 6.21 we *nd m; n such that Lm (x) ◦ g ◦ En (x) = x + with in*nitesimal ∈ KN
for some N . Then by Corollary 6:25 we have (x + ,) ◦ Lm (x) ◦ g ◦ En (x) = x for some
in*nitesimal , ∈ KN . Hence, by a remark in 6.20:
En (x) ◦ (x + ,) ◦ Lm (x) ◦ g = x:
6.27. Levels. Level is a coarse notion of “growth rate” introduced by Rosenlicht in [24].
Marker and Miller [17] used LE-series to study levels for functions in H (Ran; exp ). For
completeness, we restate one of their results in the formalism of this paper.
For f ∈ k((t))LE
∞ and k ∈ Z, we say that f has level k, if Lm(L
N +k
( f)) = LN (x))
n n
for some N ∈ N with N + k¿0. (Thus E (x) has level n, and L (x) has level −n.)
6.28. Proposition. Every element of k((t))LE has a unique level. Indeed, if f ∈ Ln and
m is least such that Lm(1n ( f)) ∈ Gm ; then f has level m − n.
This proposition is closely related to Lemma 6.21. (The reader may even use this
lemma to derive the above proposition as an exercise.)
Composition of germs
6.29. The Hardy *eld H (Ran; exp ) has a natural composition operation assigning to
germs 6; 9 ∈ H (Ran; exp ) with 9¿R the germ 6 ◦ 9 ∈ H (Ran; exp ). Each 9 ∈ H (Ran; exp )
with 9¿R has a compositional inverse 6 ∈ H (Ran; exp ): 6¿R and 6 ◦ 9 = 9 ◦ 6 = germ
of the identity function on R. Concerning these operations we have the following—
obviously desirable—result.
6.30. Corollary. The natural embedding of H (Ran; exp ) into R((t))LE respects compo-
sition and compositional inverse.
Proof. To show that compositions 6 ◦ 9 are preserved we use the fact from [5] that
functions de*nable in Ran; exp are given piecewise by terms in the language Lan; exp; log .
This allows us to proceed by induction on terms representing 6, using 6:18 in the main
inductive step. The preservation of compositional inverses follows from the preservation
of composition.
6.31. Remark. The series in the image of the natural embedding above are in some
intuitive sense “convergent” LE-series. It would be desirable to make this more precise
along the lines of [4] by introducing a suitable inductive system of normed subalgebras
of R((t))LE . The problem of composition for such a conjectural notion of convergent
LE-series is already stated at the end of [4].
>
Relation to Ecalle’s transseries
I
6.32. Ecalle states as part of his “Lemme 4:1:2” [8, p. 139] the stability of his “trigYebre”
R[[[x]]] of transseries under composition and compositional inverse. He does not touch
upon the di8culty—mentioned in 6:5 and the beginning of 6:12 above—of making
sense of composition. As to the existence of compositional inverse, Ecalle I gives a
half page sketch which provided the idea for the proof above. As the *rst stage in
constructing the inverse he indicates something close to our Lemma 6.21. The second
stage consists in inverting an operator similar to what we do in the proof of Lemma
6:24. Unfortunately, this operator only makes sense (to us) if Taylor expansion of
f(x + ) as a power series in the in*nitesimal were valid more generally than is
actually the case. It is for this reason that this second stage takes the more complicated
form of Lemmas 6:23, 6:24 and Corollary 6:25 in our proof.
Here is an example to show that a restriction as in 6:8; (4)m is necessary for
Taylor expansion to be possible. Let f(x) := E(x3 ) ∈ K1 and := x−1 , an in*nitesi-

∞ (p)
mal. Then p=0 ( f (x)=p!)p does not exist, since Lm((f(p) (x)=p!)p ) = xp E(x3 )
increases with p.
7. Interesting sub#elds of R((t))LE
7.1. In this section we impose certain natural restrictions on the LE-series consid-
ered, and show that this leads to sub*elds of k((t))LE that share many properties of
k((t))LE and are nicely contained in k((t))LE , for example as truncation closed sub*elds.
We pay special attention to the sub*eld k((t))LE; ft consisting of the LE-series with sup-
port of hereditarily *nite type, as de*ned in 7:10 and 7:11 below. This sub*eld is in
I
essence Ecalle’s trigYebre R[[[x]]] of transseries when k = R, and we prove here in
detail some of its properties stated or hinted at in [8].
7.2. For each m, let Fm be a subset of k[[X1 ; : : : ; Xm ]] such that the subring k[X1 ; : : : ;
Xm ; Fm ] is closed under the operations @=@Xi for i = 1; : : : ; m. Let F := (Fm )m∈N .
We say that a sub*eld K of k((t))LE is weakly F-closed if F(1 ; : : : ; m ) ∈ K whenever
F ∈ Fm and 1 ; : : : ; m are in*nitesimals in K. (NB: we use the modi*er “weakly”
because in [6, 3.2] the de*nition of “F-closed” also requires being real closed.)
We de*ne k((t))LE; F to be the smallest weakly F-closed sub*eld of k((t))LE that
contains k(x), is closed under E, and contains log f whenever it contains f and
f¿0.
7.3. Proposition. The 4eld k((t))LE; F is closed under the derivation, and under com-
position: if f; g ∈ k((t))LE; F and g¿k, then f ∈ k((t))LE; F and f ◦ g ∈ k((t))LE; F .
Proof. Let K be the set of all f ∈ k((t))LE; F such that f ∈ k((t))LE; F . One checks
immediately (using Theorem 3:9) that K is a weakly F-closed sub*eld of k((t))LE; F ,
that K contains k(x), and that E(K) ⊆ K and log(K¿0 ) ⊆ K. Thus K = k((t))LE; F . This
proves the assertion on the derivation. Next, let K∗ be the set of all f ∈ k((t))LE; F such
that f◦ g ∈ k((t))LE; F for all g ∈ k((t))LE; F with g¿k. Again, the results of Section 6
imply easily that K∗ is a weakly F-closed sub*eld of k((t))LE; F which contains k(x),
LE; F
and that E(K∗ ) ⊆ K∗ and log(K¿0 ∗ ) ⊆ K∗ . Hence K∗ = k((t)) , as desired.
7.4. Remark. Note that each positive u ∈ k((t))LE; F has an nth root u1=n = E((log u)=n)
in k((t))LE; F for every n¿0. When k = R and Fm contains all f ∈ R[[X1 ; : : : ; Xm ]] that
converge on some neighborhood of the origin in Rm , for all m, then the arguments of
[5, 2.5 and 2.6] imply that R((t))LE; F is real closed.
Let now Fmconv be the set of all f ∈ R[[X1 ; : : : ; Xm ]] that converge on some neigh-
conv
borhood of the origin in Rm , for all m. Then R((t))LE; F is exactly the image of the
natural embedding of H (Ran; exp ) into R((t))LE that we considered in 3:11 and 6:29.
(This is proved in the same way as Corollary 3:12.) Note that by 6:29 and Corol-
conv
lary 6:30, each series ¿R in R((t))LE; F has its compositional inverse in this *eld
as well.
7.5. Subseries and truncation. Let G be an ordered abelian group, put K := k((G)),

and let f = g∈G ag g ∈ K. Then any set F ⊆ G determines the subseries g∈F ag g

of f (an element of K), and any h ∈ G determines the truncation g¿h ag g of f

(an element of K). We say that a set F ⊆ K is closed under subseries (in K) if f ∈ F
whenever f is a subseries of some element of F. We say that F ⊆ K is truncation
closed if f ∈ F whenever f is a truncation of some element of K. If F ⊆ K is closed
under subseries it is clearly also truncation closed. Note that for each set S ⊆ G the
additive subgroup k((S)) of K is closed under subseries.
We now apply these notions to subsets of the series *eld K = k((G LE )), and just
write ‘closed under subseries’ and ‘truncation closed’ without referring every time to
this ambient series *eld.
By construction all Km are closed under subseries, and so is their union k((t))E .
Hence each Ln is closed under subseries, and thus their union k((t))LE also. The proofs
in [6, Section 3] (where we work over k = R) give the following.
7.6. Proposition. The 4eld k((t))LE; F is truncation closed.
7.7. LE-series with support of hereditarily #nite type. In the case that Fm = k[[X1 ; : : : ;
Xm ]] for all m we shall explicitly describe k((t))LE; F as the set of LE-series with
support of “hereditarily *nite type” (Proposition 7:12). We start with an excursion
concerning subsets of *nite type in ordered abelian groups.
Let G be an ordered abelian (multiplicative) group. A set A ⊆ G is said to be of 4nite
type (in G when we have to specify the ambient group) if there are g; g1 ; : : : ; gk ∈ G
with g1 ; : : : ; gk ¡1 such that A ⊆ g[g1 ; : : : ; gk ]. (Recall that [g1 ; : : : ; gk ] denotes the sub-
monoid of G generated by g1 ; : : : ; gk .) In the following lemma we collect some basic
facts on subsets of G of *nite type.
7.8. Lemma. Subsets of G of 4nite type are reverse well ordered. The union of 4nitely
many subsets of G of 4nite type is of 4nite type. Let A ⊆ G be of 4nite type in G.
Then the following hold:
(1) If A ⊆ G 61 ; then there are a1 ; : : : ; ak ∈ A ¡1 such that A ⊆ [a1 ; : : : ; ak ].
(2) If A = ∅; then A has a largest element max A; and there are a1 ; : : : ; ak ∈ A ¡1
such that A ⊆ (max A)[a1 ; : : : ; ak ].
(3) If H is a subgroup of G and A ⊆ H; then A is of 4nite type in H .
Proof. We leave the *rst two parts of the lemma as easy exercises, and only prove
the three numbered assertions about A ⊆ G of *nite type in G. Let A ⊆ G 61 . Take
g; g1 ; : : : ; gp ∈ G with g1 ; : : : ; gp ¡1 such that A ⊆ g[g1 ; : : : ; gp ]. We equip Np with the
product partial order:
(i1 ; : : : ; ip ) 6 (j1 ; : : : ; jp ) ⇔ i1 6 j1 and : : : and ip 6 jp :
It is well known that each subset of Np has only *nitely many minimal elements, in
i
particular, the set I of minimal elements of {(i1 ; : : : ; ip ) ∈ Np : gg1i1 · · · gpp ∈ A} is *nite.
i1 ip
For each i = (i1 ; : : : ; ip ) ∈ I , put hi = gg1 · · · gp , so that hi 61. Since A ⊆ G 61 it follows
immediately that

A⊆ hi [g1 ; : : : ; gp ] ⊆ [g1 ; : : : ; gp ; (hi )i∈I ]:
i∈I
Simplifying notation, let {b1 ; : : : ; bq } = {g1 ; : : : ; gp } ∪ {hi : i ∈ I }, so that we have

A ⊆ [b1 ; : : : ; bq ]. Put
j
J := set of minimal elements of {(j1 ; : : : ; jq ) ∈ Nq : 1 = bj11 · · · bqq ∈ A }:
j
Thus the set J is *nite. For j = ( j1 ; : : : ; jq ) ∈ J , put aj := b1j1 · · · bqq . It is easy to check
that then A ⊆ [(aj )j∈J ] ⊆ A 61 .
For (2), assume A = ∅. Then A has a largest element max A by the *rst part of the
lemma. Then (max A)−1 A ⊆ G 61 is of *nite type, and the desired result now follows
from (1) applied to (max A)−1 A.
Property (3) is an immediate consequence of (2).
7.9. Remark. Concerning the *rst statement of the lemma, one can easily make this
more precise as follows: the reversed order type of a subset of G of the form
g[g1 ; : : : ; gk ] with g1 ; : : : ; gk ∈ G¡1 is at most !k .
One might ask if, conversely, each subset of G with reversed order type ! is of *nite
type, assuming G is *nitely generated to avoid trivial counterexamples. While this is
clearly the case for cyclic G, it is false in general: let G := xZ E(Zx) ⊆ G1 (generated
by x and E(x)), and note that the subset {xm E(−nx) : m6n2 } of G 61 has reversed
order type !, but is not of *nite type.
7.10. For any ordered multiplicative abelian group G we let k((G))ft consist of all series
f ∈ k((G)) such that Supp f is of *nite type, that is Supp f ⊆ g[g1 ; : : : ; gk ] for suitable
g; g1 ; : : : ; gk ∈ G with g1 ; : : : ; gk ¡1. It is easy to see that k((G))ft is a sub*eld of k((G)).
The sub*eld k((t))E; ft of k((t))E consisting of the E-series with support of hereditarily

4nite type is obtained as a union n Knft where Knft := k((Gnft ))ft , and (Gnft ))n∈N is a
certain increasing sequence of ordered groups, with each Gnft an ordered subgroup
of Gn .
We set G0ft := G0 = xk . For inductive reasons we also let G−1 ft
:= {1} ⊆ G0ft . Assume
ft
that for a certain n we have de*ned Gn−1 and Gnft as ordered subgroups of Gn−1 and
ft ft
Gn , respectively, and that Gn−1 = Gn−1 ∩ Gn . (This is clearly the case for n = 0.) Thus
ft
Gn−1 is a convex subgroup of Gnft , and Gnft = Gn ∩ Knft .
ft
Then we de*ne Gn+1 as follows. Let
Aftn := {f ∈ Knft : Supp f ¿ Gn−1

ft
} = An ∩ Knft ;
an additive subgroup of Knft with Gnft ∩ E(Anft ) = {1}. Now put Gn+1 ft
:= Gnft E(Anft ), ordered
as a subgroup of Gn+1 . It is easy to see that the inductive hypothesis above remains
true when n is replaced by n + 1. This *nishes the construction of the sequences
(Gnft ) and (Knft ). It is easy to see that E(Knft ) ⊆ Kn+1
ft
for all n, and thus k((t))E; ft is an
E
ordered exponential sub*eld of k((t)) . Note also that with G E; ft := Gnft , the group
of E-monomials of hereditarily 4nite type, we have
k((t))E;ft = k((G E;ft ))ft ⊆ k((G E;ft )):
7.11. Next we close o+ under logarithms. One veri*es easily that /(Gnft ) ⊆ Gn+1 ft
,
ft ft ft ft E; ft E; ft
/(An ) ⊆ An+1 and /(Kn ) ⊆ Kn+1 . As in 2.6 this implies /(k((t)) ) ⊆ k((t)) , and
each positive element of /(k((t))E; ft ) has its logarithm in k((t))E; ft . Because of these
two facts we can now close o+ under logarithms as in 2.7. Put
E;ft
Lfti := 1−1
i k((t)) :
Then Lfti is an ordered exponential sub*eld of Li , and the logarithm of each positive
element of Lfti lies in Lfti+1 . Finally, put

k((t))LE;ft := Lfti ;
i
the *eld of LE-series with support of hereditarily 4nite type, an ordered logarithmic-

exponential sub*eld of k((t))LE . Let G LE; ft := 1−1
i G
E; ft
be the group of LE-monomials
of hereditarily 4nite type. Then we have
k((t))LE;ft = k((G LE;ft ))ft ⊆ k((G LE;ft )):
Note that / maps k((t))LE; ft onto itself. It is also clear from the construction that
k((t))LE; ft is closed under subseries.
We can now state a strong version of the fact that the support of each series in
k((t))LE; F is of *nite type. (See [6, 5.13] for a much weaker version.)
7.12. Proposition. Let Fm = k[[X1 ; : : : ; Xm ]] for all m. Then
k((t))LE;F = k((t))LE;ft :
Thus for each f ∈ k((t))LE; F there are LE-monomials g; g1 ; : : : ; gk of hereditarily 4nite

type with g1 ; : : : ; gk ¡1 such that Supp f ⊆ g[g1 ; : : : ; gk ].
Proof. We *rst show that each *eld Knft = k((Gnft ))ft is weakly F-closed. Let 1 ; : : : ; m
be in*nitesimals of Knft and let F ∈ Fm . Then the sets Supp i are subsets of (Gnft )¡1 of
*nite type, hence their union U is a subset of (Gnft )¡1 of *nite type. Since Supp F(1 ; : : : ;
m ) ⊆ [U ] it follows that Supp F(1 ; : : : ; m ) is a subset of Gnft of *nite type, that is,
F(1 ; : : : ; m ) ∈ Knft .
Therefore k((t))E; ft is weakly F-closed, hence the *elds Lfti are weakly F-closed,
and thus k((t))LE; ft is weakly F-closed. This proves the inclusion
k((t))LE;F ⊆ k((t))LE;ft :
For the reverse inclusion we show *rst by induction on n that Knft ⊆ k((t))LE; F .
To see why this is true for n = 0, *rst note that all powers xr = E(r log x) with r ∈ k

r
belong to k((t))LE; F . Now consider a series f = ar x in K0ft = k((xk ))ft . Then
Supp f ⊆ xr0 [xr1 ; : : : ; xrm ] with r1 ; : : : ; rm ¡0. Thus each r with ar = 0 can be writ-
ten as r = r0 + ir; 1 r1 + · · · + ir; m rm for suitable ir; 1 ; : : : ; ir; m ∈ N. Put F(X1 ; : : : ; Xm ) :=
i i
ar X1r; 1 · · · Xmr; m ∈ k [[X1 ; : : : ; Xm ]], where the sum is over all r ∈ k such that ar = 0.
Then f = xr0 F(xr1 ; : : : ; xrm ) ∈ k((t))LE; F .
Assume inductively that Knft ⊆ k((t))LE; F . Then Gn+1 ft
= Gnft E(Aftn ) ⊆ k((t))LE; F . Con-

ft ft ft
sider a series f = a3 3 in Kn+1 = k((Gn+1 )) . Thus Supp f ⊆ 30 [31 ; : : : ; 3m ] with all
ft i i
3i ∈ Gn+1 and 31 ; : : : ; 3m ¡1. Write each 3 with a3 = 0 as 3 = 30 313;1 · · · 3m3; m for suitable
i3;1 ; : : : ; i3; m ∈ N. Put

i
F(X1 ; : : : ; Xm ) := a3 X13;1 · · · Xmi;3;m ∈ k[[X1 ; : : : ; Xm ]];
ft
where the sum is over all 3 ∈Gn+1 such that a3 = 0. Then f = 30 F(31 ; : : : ; 3m ) ∈
LE; F
k((t)) . This inductive argument shows that k((t))E; ft ⊆ k((t))LE; F . Composing on
the right with log x and its iterates gives Lfti ⊆ k((t))LE; F for all i, and thus k((t))E; ft ⊆
k((t))LE; F , as required.
The proof of the next result depends heavily on Lemma 7:8:
7.13. Proposition. For all n we have
Kn ∩ k((t))LE;ft = Knft and Ln ∩ k((t))LE;ft = Lftn :
ft
Proof. We *rst show that Kn ∩ Kn+1 = Knft . The inclusion ⊇ is clear. Let f ∈Kn ∩ Kn+1
ft
.
ft ft
Then Supp f is of *nite type in Gn+1 and is also contained in the subgroup Gn ∩ Gn+1
= Gnft , hence is of *nite type in Gnft by Lemma 7:8(3). Thus f ∈Knft . This proves the
ft
equality Kn ∩ Kn+1 = Knft , from which one easily obtains
Kn ∩ k((t))E;ft = Knft :
To go from here to the *rst identity stated in the proposition we will next show by
induction on n that
/p (Gn ) ∩ Gn+p
ft
= /p (Gnft ) and /p (Kn ) ∩ Kn+p
ft
= /p (Knft ) for all p ¿ 0:
For n = 0 the *rst identity is immediate since G0 = G0ft . For the second identity we
consider an element f ∈K0 with /p (f) ∈ Kpft , and we have to show that then f ∈K0ft .

Write f = ar x r in K0 where the sum is over the r ∈ k (all ar ∈k). Then /p (f) =
ar /p (x r ) ∈ Kpft implies that {/p (x r ) : ar = 0} is of *nite type in Gpft , hence of *nite

type in its subgroup /p (G0 ) by Lemma 7:8(3). Therefore Supp (f) = {x r : ar = 0} is
of *nite type in G0 . It follows that f ∈K0ft .
Let n¿0 and p¿0, and assume these identities hold for lower values of n. The two
inclusions ⊇ are clear from 7.10 and 7.11. Let g ∈ Gn such that /p (g) ∈ Gn+p ft
. For the
ft
*rst identity we have to show that then g ∈ Gn . Write g = 3E(a) with 3 ∈ Gn−1 and
a ∈ An−1 . Then
/p (g) = /p (3)E(/p a) ∈ Gn+p

ft ft
= G(n−1)+p E(Aft(n−1)+p )
⊆ Gn+p = G(n−1)+p E(A(n−1)+p ):
ft
The uniqueness properties of the product decompositions of Gn+p and Gn+p displayed
p ft p ft ft
here imply that / (3) ∈ G(n−1)+p and / a ∈ A(n−1)+p ⊆ K(n−1)+p . By the inductive as-
ft ft
sumption this gives 3 ∈ Gn−1 and a ∈ Kn−1 ∩ An−1 = Aftn−1 , and thus g ∈ Gnft . For the
second identity we consider an element f ∈Kn with /p (f) ∈ Kn+p ft
, and we have to
show that then f ∈Knft . Write f= cg g in k((Gn )) where the sum is over the g ∈ Gn

(all cg ∈ k). Then /p (f) = cg /p (g) ∈ Kn+p

ft
implies that all /p (g) with cg = 0 be-
ft ft
long to Gn+p ∩ Kn+p = Gn+p , and thus by the *rst equality all g with cg = 0 belong to
Gnft , that is Supp (f) ⊆ Gnft . Also, {/p (g) : cg = 0} is of *nite type in Gn+p
ft
, hence of *-
p ft
nite type in the subgroup / (Gn ) by Lemma 7:8(3). Therefore Supp(f) = {g : cg = 0}
is of *nite type in Gnft . It follows that f ∈Knft . This *nishes the proof of the two
identities. The second of these identities implies (for p¿0) that
/p (Kn ) ∩ k((t))LE;ft = /p (Kn ) ∩ (Kn+p ∩ k((t))LE;ft ) = /p (Kn ) ∩ Kn+p

ft
= /p (Knft ):
Applying /−p this gives Kn ∩ Lftp = Knft , and by letting here p go to in*nity we obtain
the *rst identity of the proposition. By letting n go to in*nity in this *rst identity we
obtain
k((t))E ∩ k((t))LE;ft = k((t))E;ft ;
which is the special case n = 0 of the second identity of the proposition. Applying /−n
to both sides of this special case we obtain the second identity in general.
Integration and compositional inversion in k((t))LE; ft
7.14. It follows immediately from Propositions 7:3, 7:12 and 7:13 that the derivation
on k((t))LE maps each Knft and each Lftn into itself. But this is only a small step in
the direction of showing that k((t))LE; ft is closed under integration and compositional
inverses. In trying to adapt the proofs in Sections 5 and 6 we run into problems, and to
get around these we have to redo a number of earlier items “with *nite type bounds on
supports”. That is, when a bound of *nite type on the support of a series is given and
an operation is applied to the series, the result should also have support contained in a
set that is of *nite type and entirely determined by the original bound. The following
considerations will make this general idea more concrete. We start with redoing the
material in 1.3 on small operators, by imposing everywhere bounds of *nite type.
Let D be a k-linear subspace of k((G)) such that for each family (ai )i∈I of elements

in D for which ai exists we have ai ∈D. Put D ft := D ∩ k((G)) ft . Let us say that
an operator P : D → D is small of 4nite type if there exists R ⊆ G ¡1 of *nite type in G
such that Supp P(s) ⊆ R · Supp s for all s ∈ D. It follows that then P is a small operator,
and that for all n and all s ∈ D:
Supp Pn (s) ⊆ [R]Supp s:

Thus for each sequence (cn ) of coe8cients in k the operator cn P n : D → D maps

ft

n
D into itself, and satis*es Supp ( cn P (s)) ⊆ [R]Supp s for all s ∈ D. In particular,
the perturbation I − P of the identity maps Dft bijectively onto itself, since its inverse
n
P maps Dft into itself.
Taking reciprocals preserves *nite type bounds on supports as follows, where G is

as usual an ordered abelian multiplicative group:
7.15. Lemma. Let S ⊆ G be of 4nite type, a ∈ S. Then there is a set S(a) ⊆ G of 4nite
type such that if f ∈k((G)) and Supp f ⊆ S, Lmf = a, then Supp f−1 ⊆ S(a).
Proof. By Lemma 7:8(2) we have S 6a ⊆ a[a1 ; : : : ; ak ] for certain a1 ; : : : ; ak ¡1 in G.

Then one easily checks that if f ∈k((G)) and Supp f ⊆ S, Lmf = a, then Supp f−1 ⊆
a−1 [a1 ; : : : ; ak ].
7.16. To integrate and take compositional inverses in k((t))LE; ft we need to know that
taking derivatives preserves *nite type bounds on supports in a rather strong way. For
f ∈K0 we clearly have Supp f (p) ⊆ x−p Supp f for all p ∈ N. For f ∈Knft with n¿0
some further preparation is necessary.
One complication in keeping track of supports is illustrated by the series
∞

f := (1 − xE(−x))−1 = xn E(−nx) ∈ K1ft :
n=0
Then Supp f = {x n E(−nx) : n ∈ N} ⊆ xk E(Aft0 ) is of *nite type, but Supp f is not con-
tained in any set of the form S0 S1 where S0 ⊆ xk and S1 ⊆ E(Aft0 ) are of *nite type.
It is for this reason that we introduce the notion of nice subgroup of Gnft , which allows
us to state and prove the important Lemma 7:20.
7.17. Recall that the weight of a monomial X1e(1) · · · Xje( j) (where e(1); : : : ; e(j) ∈ N)
is by de*nition the natural number e(1) + 2e(2) + · · · + je( j), and that a polynomial
in X1 ; : : : ; Xj over a *eld is said to have weight d (where d ∈ N) if all monomials
occurring in it have weight d. This is used below as follows: for f; g ∈ k((t))LE and
p ∈ N the pth derivative of fE(g) is given by

(p) (i) (j)
(fEg) = f Fi; j (g ; : : : ; g ) E(g);
i+j=p
where each Fi; j (X1 ; : : : ; Xj ) ∈ Q[X1 ; : : : ; Xj ] has weight j and does not depend on f
and g. (This is easily shown by induction on p.)
7.18. We now single out certain subgroups of Gnft as being nice. For n = 0 we declare
the nice subgroups of G0ft = xk to be exactly its *nitely generated subgroups that con-
tain x. Assume the class of nice subgroups of Gnft has been de*ned as a subclass of
the class of all subgroups of Gnft . Then we say that a subgroup H of Gn+1
ft
= Gnft E(Aftn )
ft
is nice if H = H · E(C) for some nice subgroup H of Gn and some *nitely generated
(additive) subgroup C of Aftn such that Supp a ⊆ H for all a ∈ C. Note that then H
and C are uniquely determined by H , since H = H ∩ Gnft and E(C) = H ∩ E(Aftn ).
7.19. Lemma. Any 4nite subset of Gnft , and thus any subset of 4nite type in this
group; is contained in some nice subgroup of Gnft .
ft
Proof. By induction on n. The case n = 0 is trivial. Let f1 ; : : : ; fk ∈ Gn+1 . We have
ft
to *nd a nice subgroup of Gn+1 containing f1 ; : : : ; fk . Write fi = gi E(ai ) with gi ∈ Gnft
and ai ∈ Aftn , for i = 1; : : : ; k. Take elements h0 ; h1 ; : : : ; hl ∈ Gnft with h1 61; : : : ; hl 61 such
that Supp ai ⊆ h0 [h1 ; : : : ; hl ] for i = 1; : : : ; k. Inductively we may take a nice subgroup
H of Gnft such that {g1 ; : : : ; gk ; h0 ; h1 ; : : : ; hl } ⊆ H . Let C be the additive subgroup of
Aftn generated by a1 ; : : : ; ak . Then H := H E(C) is a nice subgroup of Gn+1 ft
containing
f1 ; : : : ; fk .
ft
7.20. Lemma. Let H be a nice subgroup of Gn+1 ; with H = H E(C) as in 7.18 above.

Then there are h0 ; h1 ; : : : ; hk ∈ H with h1 61; : : : ; hk 61 such that for all f ∈Kn+1
with Supp f ⊆ H we have Suppf (p) ⊆ hp0 [h1 ; : : : ; hk ] Supp f for all p, and thus
Supp f (p) ⊆H for all p.
Proof. Let f ∈Kn+1 and Supp f ⊆ H . Write f= a fa E(a) where the sum is over all
a ∈ C, and all fa belong to Kn and satisfy Supp fa ⊆ H . Taking the pth derivative of
f gives

(p)
f = fa(i) Fi; j (a ; : : : ; a(j) ) E(a); (1)
a i+j=p
where the polynomials Fi; j ∈ Q[X1 ; : : : ; Xj ] are as in 7.17. The properties of H and
C imply that there are g0 ; g1 ; : : : ; gr ∈ H with g0 ¿1 and g1 61; : : : ; gr 61 such that
Supp a ⊆ g0 [g1 ; : : : ; gr ] for all a ∈ C. We now proceed by induction on n. Consider *rst
the case n = 0. For every monomial X1e(1) · · · Xje( j) occurring in Fi; j with i + j = p we
have e(1) + 2e(2) + · · · + je( j) = j, and thus, for all a ∈ C:
Supp((a )e(1) · · · (a( j) )e( j) ) ⊆ x−e(1) g0e(1) x−2e(2) g0e(2) · · · x−me(m) g0e(m) [g1 ; : : : ; gr ]
⊆ x−j g0e [g1 ; : : : ; gr ]; p ¿ e := e(1) + e(2) + · · · + e(m)
⊆ x−j g0p [g0−1 ; g1 ; : : : ; gr ]: (2)
Also Supp fa(i) ⊆ x−i Supp fa for all a ∈ C. In combination with (1) and (2) this gives
Supp f (p) ⊆ x−p g0p [g0−1 ; g1 ; : : : ; gr ] Supp f, and thus the lemma holds with h0 := x−1 g0
and {h1 ; : : : ; hk } := {g0−1 ; g1 ; : : : ; gr }. Suppose next that n¿0, and that the result holds
for lower values of n. Thus there are elements D0 ; D1 ; : : : ; Ds ∈ H ∩ Gn−1 ft
with D0 ¿1

and D1 61; : : : ; Ds 61 such that for all g ∈ Kn with Supp g ⊆ H and for all m
Supp g(m) ⊆ D0m [D1 ; : : : ; Ds ] Supp g:

In particular for all a ∈ C and m¿0:
Supp a(m) ⊆ D0m [D1 ; : : : ; Ds ] Supp a
⊆ D0m g0 [g1 ; : : : ; gr ; D1 ; : : : ; Ds ]:
A computation as in (2) above then shows that for each monomial X1e(1) · · · Xje( j)
occurring in Fi; j with i + j = p we have (for a ∈ C)
Supp ((a )e(1) · · · (a( j) )e( j) ) ⊆ D0j g0e [D1 ; : : : ; Ds ; g1 ; : : : ; gr ]
⊆ D0j g0p [g0−1 ; g1 ; : : : ; gr ; D1 ; : : : ; Ds ]: (3)
Also Supp fa ⊆ H for a ∈ C, and thus for i; j with i + j = p
Supp fa(i) ⊆ D0i [D1 ; : : : ; Ds ] Supp fa :
In combination with (1) and (3) this shows that the lemma holds with h0 := D0 g0 and
{h1 ; : : : ; hk } := {g0−1 ; g1 ; : : : ; gr ; D1 ; : : : ; Ds }.
We can now easily handle compositional inversion in k((t))LE; ft :
7.21. Proposition. Let g ∈ k((t))LE; ft , g¿k. Then there is f ∈k((t))LE; ft such that
f ◦ g = x.
Proof. This amounts to proving the “*nite type” versions of Lemmas 6:23, 6:24 and
Corollary 6:25, followed by the same argument as in 6.26. These “*nite type” versions
are as follows:
(1) Let ∈ Knft and Lm ¡x. Then there exists , ∈ K0ft with Lm ,¡x such that
(x + ,) ◦ (x + ) = x + ∗ with Lm ∗ ¡G0 .
(2) Let 0¡m6n, and let ∈ Knft with Lm ¡Gm−1 . Then there exists , ∈ Kmft with
Lm ,¡Gm−1 such that (x + ,) ◦ (x + ) = x + ∗ with Lm ∗ ¡Gm . In particular,
(x + ,) ◦ (x + ) = x if m = n.
(3) For each in*nitesimal ∈ Knft there is an in*nitesimal , ∈ Knft such that
(x + ,) ◦ (x + ) = x.
To obtain (1) we imitate the proof of Lemma 6:23, and *rst reduce to the case that
∈ K0ft . Then the operator P : D → D de*ned in that proof is small of *nite type.
Thus , as de*ned at the end of that proof belongs to Dft ⊆ K0ft . To establish (2)
we imitate the proof of Lemma 6:24, and *rst reduce to the case that ∈ Kmft , = 0,
and consider the operator P : D → D de*ned there. The key step is now to take a
nice subgroup H of Gmft such that Supp ⊆ H , and to consider the restriction of P
to D(H ) := {, ∈ D: Supp , ⊆ H }. By Lemma 7:20 there exist h0 ; h1 ; : : : ; hk ∈ H with
h1 61; : : : ; hk 61 such that Supp ,(p) ⊆ hp0 [h1 ; : : : ; hk ]Supp , ⊆ H for all , ∈ D(H ) and
all p. Write = Lm()a, with Supp a ⊆ H 61 . Thus for all , ∈ D(H ) and all p¿0
Supp ,(p) p ⊆ (h0 Lm())p [h1 ; : : : ; hk ][Supp a]Supp ,

⊆ h0 Lm()[(h0 Lm(); h1 ; : : : ; hk ][Supp a]Supp ,
⊆ H:
Since h0 Lm()¡Gm−1 it follows that P (D(H )) ⊆ D(H ), and that the restriction
P |D(H ) : D(H ) → D(H ) is small of *nite type. Hence , := (I + P )−1 (−) belongs
to D(H )ft , and therefore , ∈ Kmft as desired. The proof of (3) is just like that of
Corollary 6:25, appealing to (1) and (2) instead of Lemmas 6:23 and 6:24.
Integrating in k((t))LE; ft requires much more care in keeping track of supports.
7.22. Proposition. Each element of k((t))LE; ft has an integral in k((t))LE; ft .
Proof. We imitate Section 5, imposing suitable *nite type bounds on supports. Given
some n we assume inductively that for each nice subgroup H of Gnft and each set
ft
S ⊆ H of *nite type there is a set S ⊆ H of *nite type such that each g ∈ Kn with
ft
residue 0 and Supp g ⊆ S hasan integral f ∈ Kn with Supp f ⊆ S. (For n = 0 this
ft
is clearly the case by taking S := xS.) Now let H be a nice subgroup of Gn+1 and
ft
let S ⊆ H be of *nite type. Write H = H E(C) as in 7.18, and let g ∈ Kn+1 be such

that Supp g ⊆ S. Write g = a ga E(a) where a ranges over C and all ga ∈ Knft . As in

5:3 and Lemma 5:4 we *nd an integral f = a fa E(a) in Kn+1 of g. Here f0 ∈ Kn is

an integral of g0 , and for each a = 0 the series fa ∈ Kn is the solution of the equation
y − 3(a)y = h(a) obtained by applying a certain operator which depends on a to h(a)
(where 3(a) := (−a )−1 ∈ Kn and h(a) := (a )−1 ga ∈ Kn ). In particular fa = 0 whenever
ga = 0. Since Supp g0 ⊆ S ∩ H , and S∩ H is of *nite type, the inductive assumption
implies that f0 ∈ Knft with Supp f0 ⊆ (S ∩ H ) ⊆ H . For a = 0 we have the operator

i
Pa on Kn de*ned by Pa (y) = 3(a)y , so that fa = Pa (h(a)). We shall restrict these

operators to D := {y ∈ Kn : Supp y ⊆ H }, and prove that these restricted operators are
small of *nite type “uniformly in a”. Since Supp (3(a)) ⊆ H for a = 0 we clearly have
Pa (D) ⊆ D and h(a) ∈ D for a = 0. Next we show:
(*) There is U ⊆ H of *nite type (depending only on S and H ), with U ¡x if n = 0
and U ¡Gn−1 if n¿0, such that for all a = 0 we have Supp 3(a) ⊆ U .
To see this, let C = Za1 +· · ·+Zak . Then Lm(a ) takes at most k values as a ranges
over C\{0}. (This follows from a purely valuation-theoretic fact: if b1 ; : : : ; bN ∈ k((t))LE \
{0} and Lm(bi ) = Lm(bj ) for all i = j, then b1 ; : : : ; bN are linearly independent over
k
k.) Now apply Lemmas 3:5, 7:15, and the fact that Supp a ⊆ i=1 Supp ai ⊆ H for all
a ∈ C.
With U as in (*) we may assume that U = u0 [u1 ; : : : ; ur ] with u1 ; : : : ; ur ∈ H and
u1 6 1; : : : ; ur 6 1. Note that then u0 = max U , and thus u0 ¡x if n = 0 and u0 ¡Gn−1
for n¿0. We now *rst consider the case n = 0. Then Supp Pa (y) ⊆ u0 x−1 [u1 ; : : : ; ur ]
Supp y for non-zero a and y ∈ D. Since u0 x−1 ¡1 it follows that Pa |D is a small
operator of *nite type, and

∞

Supp fa = Supp Pia (h(a)) ⊆ [u0 x−1 ; u1 ; : : : ; ur ]Supp h(a)
i=0
⊆ u0 [u0 x−1 ; u1 ; : : : ; ur ]Supp ga :

Hence, with S := (S ∩ H ) the integral f = fa E(a) of g = ga E(a) satis*es
Supp f ⊆ S ∪ u0 [u0 x−1 ; u1 ; : : : ; ur ]Supp g

⊆ S ∪ u0 [u0 x−1 ; u1 ; : : : ; ur ]S
⊆ H:
This takes care of the case n = 0. Next, assume that n¿0. By Lemma 7:20 there are
v0 ; v1 ; : : : ; vs ∈ H ∩ Gn−1 with v1 61; : : : ; vs 61 such that Supp y ⊆ v0 [v1 ; : : : ; vs ]Supp y
for all y ∈ D. Hence
Supp Pa (y) ⊆ u0 v0 [u1 ; : : : ; ur ; v1 ; : : : ; vs ]Supp y for a = 0; and y ∈ D:
Since u0 ¡Gn−1 and v0 ∈ Gn−1 we have u0 v0 ¡1. Thus the considerations in 7.14 show
that if a = 0, then Pa |D is a small operator of *nite type, and
∞

Supp fa = Supp Pia (h(a)) ⊆ [u0 v0 ; u1 ; : : : ; ur ; v1 ; : : : ; vs ]Supp h(a)
i=0
⊆ u0 [u0 v0 ; u1 ; : : : ; ur ; v1 ; : : : ; vs ]Supp ga :

Hence, with S := (S ∩ H ) the integral f = fa E(a) of g = ga E(a) satis*es
Suppf ⊆ S ∪ u0 [u0 v0 ; u1 ; : : : ; ur ; v1 ; : : : ; vs ]Supp g

⊆ S ∪ u0 [u0 v0 ; u1 ; : : : ; ur ; v1 ; : : : ; vs ]S
⊆ H:
This *nishes the inductive proof of the “*nite type versions” of 5:3 and Lemma 5:4.
The rest of the argument imitates the proof of Theorem 5:6.
7.23. Final Remark. In a late stage of our work we became aware of J. van der
I
Hoeven’s treatise “Asymptotique automatique” (Ecole Polytechnique, Paris, 1997). Our
paper (and parts of [6]) clearly have much in common with Chapters 1 and 2 of this
remarkable Th@ese, although notations and terminology are di+erent. (For example, it
seems that our “sets of *nite type” are essentially his “grid-based sets”, and that our
“LE-series with support of hereditarily *nite type” are his “grid-based transseries”.
With hindsight, the “grid” terminology may be preferable.) Van der Hoeven’s thesis
also seems to realize (in Chapters 3–5) a large part of our good intentions in 5.10 as
to solvability of algebraic ODE’s in terms of LE-series.
References
[1] N.G. de Bruijn, Asymptotic Methods in Analysis, North-Holland, Amsterdam, 1970.

[2] B. Dahn, The limiting behaviour of exponential terms, Fund. Math. 124 (1984) 169–186.
[3] B. Dahn, P. GLoring, Notes on exponential logarithmic terms, Fund. Math 127 (1986) 45–50.
[4] L. van den Dries, Convergent series expansions of a new type for exponential functions, Math. Nachr.
142 (1989) 7–18.
[5] L. van den Dries, A. Macintyre, D. Marker, The elementary theory of restricted analytic *elds with
exponentiation, Ann. Math. 140 (1994) 183–205.
[6] L. van den Dries, A. Macintyre, D. Marker, Logarithmic-exponential power series, J. London Math.
Soc. (2) 56 (1997) 417–434.
[7] M. Eastham, Theory of Ordinary Di+erential Equations, Van Nostrand Reinhold, London, 1970.
[8] I
J. Ecalle, Introduction aux fonctions analysables et preuve constructive de la conjecture de Dulac,
Hermann, Paris, 1992.
[9] J. Esterle, Solution d’un problYeme d’Erdős, Gillman et Henriksen et application aY l’Ietude des
homomorphismes de C(K), Acta Math. (Hungarica) 30 (1977) 113–127.
[10] L. Fuchs, Partially Ordered Algebraic Structures, Pergamon Press, New York, 1963.
[11] G.H. Hardy, Properties of logarithmico-exponential functions, Proc. London Math. Soc. 10 (1912)
54–90.
[12] L. Harrington, D. Marker, S. Shelah, Borel Orderings, Trans. Amer. Math. Soc. 310 (1988) 293–302.
[13] L. Harrington, S. Shelah, Counting equivalence classes for co-H-Souslin relations, in: D. van Dalen,
T.J. Smiley (Eds.), Logic Colloquium ‘80, North-Holland, Amsterdam, 1982.
[14] Il’yashenko, Finiteness theorems for limit cycles, Translations of Mathematical Monographs, Vol. 94,
AMS, Providence, RI, 1991.
[15] F.-V. Kuhlmann, S. Kuhlmann, Explicit construction of exponential-logarithmic power series, preprint.
[16] F.-V. Kuhlmann, S. Kuhlmann, S. Shelah, Exponentiation in power series *elds, Proc. Amer. Math.
Soc. 125 (1997) 3177–3183.
[17] D. Marker, C. Miller, Levelled o-minimal structures, Rev. Mat. Univ. Complut. (Madrid) 10 (1997)
241–249.
[18] M.H. Mourgues, J.P. Ressayre, Every real closed *eld has an integral part, J. Symbolic Logic 58 (1993)
641–647.
[19] B.H. Neumann, On ordered division rings, Trans. Amer. Math. Soc. 66 (1949) 202–252.
[20] P. Ribenboim, ThIeorie des valuations, Les Presses de L’UniversitIe de MontrIeal (1964).
[21] M. Rosenlicht, Di+erential valuations, Paci*c J. Math. 86 (1980) 301–319.
[22] M. Rosenlicht, Hardy Fields, J. Math. Anal. Appl. 93 (1983) 297–311.
[23] M. Rosenlicht, The rank of a Hardy Field, Trans. Amer. Math. Soc. 280 (1983) 659–671.
[24] M. Rosenlicht, Growth properties of functions in Hardy *elds, Trans. Amer. Math. Soc. 299 (1987)
261–272.
[25] M. Rosenlicht, Asymptotic solutions of Y = F(x)Y , J. Math. Anal. Appl. 189 (1995) 640–650.
[26] A. Tarski, A Decision Method for Elementary Algebra and Geometry, 2nd Edition revised, Berkeley
and Los Angeles, Rand Corporation, 1951.
[27] A.J. Wilkie, Some model completeness results for expansions of the ordered *eld of real numbers by
Pfa8an functions and the exponential function, J. Amer. Math. Soc. 9 (1996) 1051–1094.

1 s2.0 S0168007201000355 Main PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S0168007201000355 Main PDF

Uploaded by

Copyright:

Available Formats

Annals of Pure and Applied Logic 111 (2001) 61–113

f(x) = ak xk + ak−1 xk−1 + · · · + a1 x + a0 + a−1 x−1 + a−2 x−2 + · · ·

= ea0 (1 + b1 x−1 + b2 x−2 + · · ·)

for certain reals b1 ; b2 ; : : : .

(2) Formal integration:

(3) Formal composition:

f(g(x)) = x log x + log(x log x)

(4) Compositional inversion:

1. Fields of generalized series

with coe8cients cg ∈ k, such that Supp s := {g ∈ G: cg = 0} is reverse well ordered

k((S)) := {s ∈ k((G)) : Supp s ⊆ S};

which is a k-linear subspace of k((G)). Note that

k((G)) = k((G ¿1 )) ⊕ k ⊕ k((G ¡1 )) (direct sum of k-linear subspaces)

and that k((G 61 )) = k ⊕ k((G ¡1 )) = {s ∈ k((G)) : Lm(s) 6 1} is a valuation ring of

Supp s = {t : ∈ supp s}:

sum i∈I ai exists if the following two conditions are satis*ed:

de*ned element ai ∈ k((G)). (Note that each element a = cg g of k((G)) is indeed

elements of k((G)) such that ai , bj and i; j ci; j exist. Then

(1) for each d ∈ k((G)), dai exists, and dai = d ai ,

(2) if I = J , then (ai + bi ) exists, and (ai + bi ) = ai + bi ;

(3) i; j ai bj exists, and i; j ai bj = ( ai ) · ( bj );

(4) if % : I → J is a bijection and ai = b%(i) for all i, then ai = bj ,

(5) j ci; j exists for each i ∈ I , and i( j ci; j ) exists, and

power series F = !∈Nn a! Y1!1 · · · Yn!n ∈ k[[Y1 ; : : : ; Yn ]] and in*nitesimals 1 ; : : : ; n ∈ m

of elements in D for which ai exists we have ai ∈ D. (Example: D = {s ∈ k((G)) :

Supp Pn (s) ⊆ Rn · Supp s

bijection from D onto itself with inverse Pn .

We shall use small operators in Section 5 to integrate logarithmic-exponential series,

of k((G1 ))-algebras given by

Note that in k((G)) we actually have

k((G)) = A ⊕ B (direct sum of k((G1 ))-linear subspaces);

where A = {s ∈ k((G)) : Supp s¿G1 } and B = {s ∈ k((G)) : Supp s 6 g1 for some

by setting A := {s ∈ k((xk )) : Supp s¿1}, B := {s ∈ k((xk )) : Supp s 6 1}, and de*ning

2. The #eld of logarithmic-exponential series

From now on in this paper we 4x a logarithmic-exponential ordered 4eld k.

topological *eld k((G E )).

2.2. Lemma. (i) k((Gn )) = An ⊕ An−1 ⊕ · · · ⊕ A0 ⊕ k ⊕ mn (a direct sum of k-linear

2.3. Lemma. Let f ∈ Kn ; f¿0. Then

Proof. By Lemma 2.2(iv) we have Lm f = xr E(a) where r ∈ k and a ∈ An−1 ⊕ · · · ⊕ A0 .

Conversely, suppose f = E(g), with g ∈ Kl , l ¿ n. Write g = + + s + , with

Hence Lm f = E(+) = xr E(a), and thus r = 0 by Lemma 2.2(iv).

and K0 = R((t R )) (in [6]) with -( ar xr ) = ar t . Inductively, for f = fa E(a)

2.6. The substitution map /. To extend k((t))E to a logarithmic-exponential ordered

It is easy to see that / also preserves in*nite sums: if f = fi in k((t))E , then

/(f) = /(fi ) in k((t))E .

2.7. The construction of k((t))LE . We construct an increasing chain of exponential or-

E0 = L0 := identity map on k((t))LE ;

denote it by / as well. Thus we have an automorphism /k for each (positive or

2.8. k((t))LE as a #eld of generalized series. It will be convenient to construe Ln as a

of ordered *elds. There is no loss of generality in taking 1−1 n to be the restriction

G E = G E;0 ⊂ G E;1 ⊂ G E;2 ⊂ · · ·

then /k (f) = /k (fi ) in k((t))LE for all integers k.

v : k((t))LE → log(G LE ) ∪ {∞}

is de*ned by v(f) := − log(Lm(f)) for f = 0. The value group log(G LE ) is here

2.10. A countability property of logarithmic-exponential series. Here we show that

3. The derivation on k((t))LE

3.1. We start by de*ning inductively a derivation f → f on k((t))E , to be thought of

For f = ar x in K0 we set f = rar xr−1 . Then the map f → f : K0 → K0

f → f : Kn → Kn . We then extend this map to Kn+1 as follows: for f = a∈An fa E(a)

3.2. Lemma. Suppose fi exists in Kn . Then fi also exists in Kn and ( fi ) =

The (inductive) proof is entirely routine and left to the reader.

3.3. Corollary. Let F ∈ k[[X1 ; : : : ; Xm ]] and let 1 ; : : : ; m be in4nitesimals of k((t))E .

Proof. Write F = !∈Nm a! X ! , and suppose all i ∈ Kn . Then F() = a! ! in Kn ,

power series F = !∈Nn a! Y1!1 · · · Yn!n ∈ k[[Y1 ; : : : ; Yn ]] and in*nitesimals 1 ; : : : ; n ∈ m

3.1. We start by de*ning inductively a derivation f → f on k((t))E , to be thought of

For f = ar x in K0 we set f = rar xr−1 . Then the map f → f : K0 → K0

f → f : Kn → Kn . We then extend this map to Kn+1 as follows: for f = a∈An fa E(a)

3.3. Corollary. Let F ∈ k[[X1 ; : : : ; Xm ]] and let 1 ; : : : ; m be in4nitesimals of k((t))E .

Proof. Write F = !∈Nm a! X ! , and suppose all i ∈ Kn . Then F() = a! ! in Kn ,

Suppose f ∈ Kn . By Lemma 2.2(i) we have f = a+r+ where a ∈ An ⊕· · ·⊕A0 ; r ∈ k

By Corollary 3:3 we have ( i =i!) = i =i!. Thus,

Let D denote the derivation f → f on k((t))E . To motivate how we extend D to

has a solution fa ∈ Kn ; then f := ga =0 fa E(a) ∈ Kn+1 is an integral of g. For a = 0

(f; g) → f ◦ g : k((t))LE × k((t))LE

Write each + ∈ A as + = a + r + with a ∈ AM; N ⊕ AM; N −1 ⊕ · · · ⊕ AM; 0 ; r ∈ k and

(4)m If m = 0 and f ∈ K0 ; ∈ LM; n with Lm()¡Lm(g), then

If m¿0 and f ∈ Km ; ∈ LM; n ; ||¡k((Gm−1 (g)))¿0 , then