Guided tour through resurgence theory.

Jean Ecalle
(Paris-Saclay, January 2022)

1 The starting point.

2 Alien derivations and alien calculus.
3 Notion of resurgent functions.
4 Critical variables and acceleration transforms.
5 The Great Divide: cohesive/loose.
6 Equational resurgence and the Bridge equation.
7 Co-equational resurgence.
8 Object synthesis.
9 Causes of divergence and sources of divergence.
10 Transseries and analyzable germs.
11 Pseudo-variables and display.
12 Resurgence as the impetus behind mould calculus.
13 Resurgence in mathematical physics.
14 Some loose thoughts by way of conclusion.
.

1 The starting point.

The initial nudge that got the whole thing rolling1 came serendipitously, almost
in text-book fashion. Back in the early 1970s I was, just for the heck of it,2 look-
ing at the continuous iterates f ˝t of identity-tangent, local analytic mappings
f , with the fixed point taken to 8 for technical convenience:
ÿ
f pzq “ z ` 1 ` an z ´n pz „ 8q (1)
ÿ
f ˝t pzq “ z ` t ` an ptq z ´n pt P C, an ptq polynomial in tq (2)

I knew that others had already pondered the question and established such basic
facts as the generic divergence of (2); the impossibility for entire functions to
admit a full set of analytic iterates (I.N.Baker); also this interesting, vaguely
1 As far as I was concerned, that is. I am telling here my side of the story, and in no way

implying the absence of parallel approaches to asymptotics.

2 I was, in effect, wondering what strange beasts the fractional iterates of familiar mappings

such as x ÞÑ x ` x2 or x ÞÑ x exppxq might be.

1
‘Tauberian’ dichotomy: for any given f , the set Wf of all iteration powers t
that keep f ˝t analytic is either C or p1 Z for some p P N (E.Jabotinsky).
1
To get beyond that, given the Gevrey-type behaviour lim sup n1 |an ptq| n ą 0
clearly displayed by the coefficients of divergent iterates, it was tempting to
consider the Borel transform of f ˝0 :
ÿ ζ n´1
fp˝t pζq “ δ1 ` t δ ` an ptq pδ “ dirac at 0q (3)
pn´1q!

From there it was but a small step to show that fp˝t pσq was not only convergent at
0, but analytically continuable to the entire universal covering of C ´ 2πiZ, with
at most exponential growth along each non-vertical axis. This automatically
allowed Laplace integration on the two real half-axes, and yielded two distinct
˝t ˝t
germs, f` pzq and f´ pzq, respectively defined on tC1 ă ˘<zu Y tC2 ă |=z|u,
and commuting there with f . This made them bona fide iterates of f , albeit only
˝t
‘sectorial’ ones. But on closer examination, it also became clear that, for f` pzq
˝t
and f´ pzq to relate the way they should on their common domains of definition
tC2 ă |=z|u, the singularities of fppζq at the points ω :“ 2πim pm P Z˚ q had to
be of a very specific type, namely closely related to each other and to fppζq itself,
viewed as a germ at the origin. To illuminate these elusive relations, it became
imperative to have linear operators ∆ p ω capable of measuring singularities at ω
or rather, due to multivaluedness, over ω. Moreover, in view of the non-linear
nature of the problem at hand, these ∆ p ω had to act as derivations:

p ω pϕ
∆ p1 ˚ ϕ p ω pϕ
p2 q ” ∆ p1 q ˚ ϕ
p2 ` ϕ p ω pϕ
p1 ˚ ∆ p2 q (4)

relative to the natural product in the Borel plane. That product, of course, is
the convolution ˚, which is first defined locally at the origin:
żζ
pϕ
p1 ˚ ϕ
p2 qpζq :“ ϕ p1 pζ1 q ˚ ϕ
p2 pζ ´ζ1 q dζ1 (5)
0

and then extended by analytic continuation in the large.

As for the functions that like fp˝t exhibit ‘self-replicating singularities’, let
us call them resurgent, pending the more precise definitions of §3.

2 Alien derivations and alien calculus.

Alien derivations.
The sought-after derivations ∆ p ω have to be of the form:
ÿ r p 1 ,..., r q p 1 ,..., r q
pω ϕ
∆ p pζq :“ δ ω1 ,..., ωr ϕ
p ω1 ,..., ωr pζ ` ωq ωr :“ ωq (6)
 “˘
2πi
i

with the ωi ’s denoting the singular points successively encountered when moving
from 0 to ω; with the i ’s indicating the mode of circumvention, right or left;

2
 1 ,..., r
and with ϕ pp ω1 ,..., ωr q standing for the corresponding branch of ϕ.p We first take
ζ on the interval r0, ωs but close to 0, then extend the definition in the large by
analytic continuation.
For ∆p ω to be an actual derivation, the scalar weights must verify certain
algebraic relations, and if we want them to depend only on the signs i (and not
on the points ωi ), these weights are unambiguously determined: we then get
the so-called standard system of alien derivations t∆ p ω u, with indices ω running
through C‚ “ C´t0u rather than C´t0u to account for possible ramifications
Č
at the origin.
This bounty of free3 alien derivations4 arising ex nihilo from a one variable
context immediately opens exhilarating vistas, conjuring up as it does an alien
calculus of extreme richness, and naturally endowed with two faces: differential
and integral. We shall in due course encounter applications galore, but here let
us give right away, as a first appetizer, three examples that stand on their own.

Resurgence monomials.
To deserve that name, resurgence monomials W x‚ should be defined for ω-strings
‚ of any length, and behave as simply as possible under convolution and alien
differentiation. Concretely, that means:5

x ω1 ˚ W
x ω2 qpζq ”
ÿ
pW x ω pζq
W p@ ω 1 , ω 2 q (7)
ωPshapω 1 ,ω 2 q

x ω1 ,...,ωr
∆ ω0 W x ω2 ,...,ωr
” δωω01 W pδ “ Kronecker symbol q (8)

Desingularisators.
Assuming convergence, the relation:
ÿÿ
E W pϕq
p :“ ϕ p´1qr W
x ω1 ,...,ωr ˚ p∆ω . . . ∆ω ϕq (9)
x
p` 1 r
p
1ďr ωi

defines a desingularisator E W , i.e. a convolution-respecting projector of the

x

algebra of resurgent function into that of resurgence constants:

EW EW EW (10)
x x x
”
W W W
E pϕ
p1 ˚ ϕ E pϕ1 q ˚ E pϕ2 q (11)
x x x
p2 q ”
W
∆ω0 E pϕq 0 (12)
x
p ” @ω0
3 The ∆ω are mutually independent.
4 Together with another system of operators ∇ω also acting as derivations, but relative to
ordinary multiplication and with a much more restricted domain of definition, the ∆ω are the
only natural instance of an infinite Lie algebra acting in complex Analysis.
5 In (7) ω 1 and ω 2 are two ω-strings, and ω runs through all their shuffle products.

3
Alien Taylor expansions.
Much like the classical Taylor formula, which expresses a function f pzq in terms
of simple universtal monomials z n {n! and f -dependent constants f pnq p0q, the
following formula:
ÿÿ ` x
p ” E W pϕq E W p∆ωr . . . ∆ω1 ϕq
˘
ϕ
x
p ` p ˚W x ω1 ,...,ωr (13)
1ďr ωi

expresses6 the resurgent function ϕ

p in terms of elementary resurgence monomials
x‚ and resurgence constants E W
W p∆ωr . . . ∆
x p p ω ϕq.
1
p

3 Algebra of resurgent functions.

Three models: formal, convolutive, geometric.
Resurgent ‘functions’ live simultaneously in three models:
ř
• in the formal model, typically as formal power series ϕpzq
r “ sÒ as z ´s ,
with Gevey-type bounds on the coefficients.
• in the convolutive model, as analytic germs ϕpζq
p at the origin 0‚ of C‚ , with
the property of endless analytic continuation7 on C‚ and exponentially
bounded growth at infinity.

• in the geometric (or sectorial) model,8 as analytic germs ϕθ pzq defined in

sectorial neighbourhoods of infinity: argpeiθ zq ă π{2 ` Const.
Pictorially, we get this triangle:

ϕpzq
r .... ϕθ pzq z-plane pmultiplicationq
Fig 1 B Œ Õ Lθ
ϕpζq
p ζ-plane pconvolutionq

The first arrow denotes the formal, term-wise Borel transform B

#
z ´σ ÞÑ ζ σ´1 {Γpσq pσ R ´Nq
B : (14)
z n ÞÑ δ pnq pn P N , δ “ Diracq

The second arrow denotes the Laplace transform L or, for distinctiveness, Lθ :
ż eiθ 8
Lθ : ϕpζq
p ÞÑ ϕθ pzq “ p e´z ζ dζ
ϕpζq parg ζ ” θq (15)
0
6 formally and, under suitable conditions, actually.
7 under avoidance of a discrete configuration of singular points (their projections on C,
however, may well be dense – somewhere or everywhere).
8 or, strictly speaking, models, since they depend on a polarisation angle θ.

4
The magic of resurgence equations.
The signature, name-giving property of resurgent functions – namely, the self-
replication, echo-like, of ϕpζq’s
p singularities – is too protean, too elusive a feature
to be made part of the formal definition. Yet it is something that we should al-
ways keep at the back of our minds. In any case, the alien derivations ∆ p ω , along
with their pull-backs ∆ω in the multiplicative models (formal or geometric) are
the pliant tools that, in each particular instance, allow an accurate description
of the phenomenon, by means of so-called resurgence equations :
Rω pϕ, ∆ω ϕq “ 0 pmultiplicative modelsq (16)
R
pω pϕ,
p ∆p ω ϕq
p “0 pconvolutive model q (17)
The form (16), being the simpler one, is often given preference in statements,
although it is the form (17) that makes concrete, tangible analytic sense.
Nonetheless – and we are touching here on the magic of resurgence – we
can often work directly at the level of (16), with a minimum of Analysis or
sometimes none at all. Indeed, suppose we are dealing with some differential
or functional equation9 Rpϕq “ 0, linear or not, but with divergent, formal
power series solutions. If these are resurgent (there exist simple criteria for
deciding that) and if the singularity locus in the Borel plane is a certain point
set Ω (again, there exist simple methods for determining Ω), we can find, for
ω P Ω and by purely formal means, the equation Rω pϕ, ∆ω ϕq “ 0 verified by
∆ω ϕ, just as we would form the equation R1 pϕ, ϕ1 q “ 0 verified by ϕ1 . Now,
since Rω is automatically linear homogeneous in ∆ω ϕ, we can calculate the
general solution, usually of the form ∆ω ϕ “ Aω ϕω , with Aω scalar and ϕω
some normalized power series. Here, the substantive factor ϕω results from a
strictly formal calculation. It is only to calculate the scalar Aω (for the actual
∆ω ϕ, that scalar is a well-defined, usually transcendental number – essentially
a Stokes constant) that some Analysis may be called for, though not always:
even Aω can sometimes be had ‘on the cheap’. Needless to say, the procedure
can be repeated to calculate all multiple alien derivatives ∆ωn ...∆ω1 ϕ.

The minor/major duality.

To the minor ϕpζq,
p which is unambiguously defined, it is often useful to associate
a major ϕpζq,
q defined up to regular germs at 0‚ and relating to the minor
according to:
1 ` ˘
ϕpζq
p “´ q πi ζq ´ ϕpe
ϕpe q ´πi ζq pζ near 0‚ q (18)
2πi
We can then remove all terms δ pnq from ϕpζq
p and replace them by terms n! ζ ´n´1
in ϕpζq.
q With the jarring diracs thus whisked away, the quantity of missing
information in the minor clearly equals the quantity of redundant information
in a major. This is but the first manifestation of a minor/major duality that
pervades the whole theory. See in particular the formulae in §4-§5.
9 or, to be technicaly accurate, an equation germ at a given point (here: 8).

5
Borel singularities: a hindrance and a mine of information.
Being responsible for the divergence of ϕpzq,
r the singularities of ϕpζq
p may seem
a nuisance, something that impedes the direct passage from formal solutions to
actual ones. But they are also, and above all, a valuable source of imformation,
because their leading terms (their residues, in the case of simple poles) carry
crucial ‘invariants’ pertaining to ϕpzq.
r These invariants manifest as Stokes con-
stants in the geometric model, where they govern the correspondence between
neighbouring sectorial germs ϕθ pzq.

Well-behaved convolution-preserving averages.

Physicists, who are mostly interested in Laplace integration along R` (to pre-
serve realness), often dismiss as ‘non-resummable’ functions ϕpζq
p with singular-
ities on R` . In actual fact, the presence of singular points ωi on the integration
axis is no obstacle, provided we apply Laplace to a suitable average µϕ p of ϕpζq:
p
B µ L
ϕpzq
r ÝÑ ϕpζq
p ÝÑ µϕpζq
p ÝÑ ϕpzq (19)
1 ,..., r 1 ,..., r
:“ i “˘ µp ω1 ,..., ωr q ϕ
pp ω1 ,..., ωr q pζq pif ωr ă ζ ă ωr`1 q and suitable
ř
Here µ ϕpζq
p
means that the average must respect realness10 as well as convolution:

µpϕ
p1 ˚ ϕ
p2 q ” pµϕ
p1 q ˚ pµϕ
p2 q pfirst ˚ local , second ˚ global q (20)

pω q
This imposes simple algebraic conditions on its weights µ . In presence of
infinitely many ωi , an additionnal, subtler condition (non-algebraic in nature)
must be added, to ensure exponential bounds on the growth of µ ϕpζq. p Such
averages will be referred to as well-behaved.

4 Critical variables and acceleration transforms.

Ascending the ladder of critical times.
Set zi :“ z σi p0 ă σ1 ă ... ă σr q. Next, take r resurgent functions ψri pzi q,
each resummable according to the scheme of Fig 1 but relative to its own vari-
able zi . Lastly, let ϕpzq
r be some inextricable, non-linear superposition of the
various ψi pzi q. Ordinary equations Rpϕq “ 0, differential or functional, often
r
enough present us with superpositions of this type. Clearly, ϕpzq
r ought to be
resummable. Just as clearly, each of the zi ’s ought to play its part in the pro-
cess, although the scheme of Fig 1 isn’t directly applicable to any of the series
ϕ
ri pzi q ” ϕpzq.
r So how are we to proceed? The short answer is this:

• We should order our ‘critical variables’ zi , also called ‘critical times’, from
slow- to fast-flowing: z1 ăă z2 ăă ... ăă zr .
10 µϕpζq
p must be real on the whole axis arg ζ “ 0 if ϕpζq
p itself is real there, for ζ small.

6
 • We should take the Borel transform ϕ p1 pζ1 q of ϕr1 pz1 q, since it is the only
one that converges at 0‚ . The ramified germ ϕ p1 pζ1 q duly retains the prop-
erty of endless continuation, but loses exponential boundedness at infinity.
This prohibits Laplace integration. Instead, we should apply, as in (21)
infra, a so-called acceleration transform, whose integral kernel has exactly
the right rate of decrease at infinity to accommodate the superexponential
growth rate of ϕ p1 pζ1 q.

• This results in a new ramified germ at 0‚ , denoted ϕ p2 pζ2 q. Though unob-

tainable as a direct Borel transform, ϕ p2 pζ2 q is morally the Borel counter-
part of ϕ
r2 pz2 q. It still has endless analytic continuation, but once again
with superexponential growth at infinity.

• We can then continue the acceleration process to successively construct

all germs ϕ
pi pζi q up to ϕ
pr pζr q.

• Only at the last stage do we get a function ϕpr pζr q with (at most) expo-
nential growth at infinity. This allows Laplace integration and produces
at last a geometric germ ϕr pzr q ” ϕp zq.
• Since each of the steps we went through is an algebra isomorphism11 , our
hard-won germ ϕpzq is automatically a proper solution of whatever equa-
tion Rpϕq “ 0, linear or non-linear, the initial formal series ϕpzq
r happened
to be a solution.

• But since each Borel plane ζi is saddled with its own singular locus Ωi ,
the end result ϕpzq is also highly polarized: it depends on the choice of r
integration axes argpζi q “ θi .

The accelero-summation scheme.

The process I have just sketched (I call it accelero-summation) can be repre-
sented pictorially as follows:12

ϕ
r1 pz1 q Ð ϕpzq
r ϕpzq Ð ϕr pzr q
Fig 2 ÓB LÒ
ϕ
p1 pζ1 q Ñ ϕ
p2 pζ2 q Ñ ¨¨¨ Ñ ϕ
pr´1 pζr´1 q Ñ ϕ
pr pζr q
C1 , 2 Cr´1 , r
11 First from multiplicative to convolutive, then from convolutive to convolutive over and

over again, and eventually from convolutive to multiplicative.

12 For simplicity, I left the majors ϕ
qi pζi q out of the picture. Actually, as shown in §5 infra,
by selecting suitably slow times zi´ „ zi in each critical time class rzi s, one can ensure the
smoothness of the minors ϕ pi´ and all their alien derivatives ∆ p ω .ϕ
p ω . . . . .∆ `
r 1 pi´ (ωk P R )
and, by the same token, render the corresponding majors redundant.

7
Acceleration/deceleration kernels.
Time now to construct the acceleration operators and their inverses, and to
describe their action on both minors and majors.
A single pair CF , C F of integral kernels does duty for the four combinations
of minor and major, acceleration and decelerations, but with a characteristic
diagonal ‘flip’:

acceleration deceleration
« ff
´ ¯
Fig 3 minor CF CF z1 ă z2 , z1 “ F pz2 q
major CF CF

These kernels depend on the germ F that expresses the slower ‘time’ z1 in terms
of the faster z2 .
ż c`i8
1
CF pζ2 , ζ1 q :“ ez2 ζ2 ´z1 ζ1 dz2 with z1 ” F pz2 q (21)
2πi c´i8
ż `8
C F pζ2 , ζ1 q :“ e´z2 ζ2 `z1 ζ1 dz2 with z1 ” F pz2 q and 1 ă u (22)
`u

Acceleration from ζ1 to ζ2 obeys the formula:

ż `8
ϕ
p2 pζ2 q “ CF pζ2 , ζ1 q ϕ
p1 pζ1 q dζ1 (23)
`0
ż c`i8
1
ϕ
q2 pζ2 q “ C F pζ2 , ζ1 q ϕ
q1 pζ1 q dζ1 (24)
2πi c´i8

Deceleration from ζ2 to ζ1 goes like this:

ż 02
1
ζ1 ϕ
p1 pζ1 q “ p2 pζ2 q C F pζ2 , ζ1 q dζ2 pu ą 0q
ζ2 ϕ (25)
2πi 01
ż `v
ζ1 ϕ
q1 pζ1 q “ ζ2 ϕ
q2 pζ2 q CF pζ2 , ζ1 q dζ2 (26)
`0

Here again, we notice a double flip between finite/infinite, path/loop inte-

grals. Integration in (23) is along an infinite path, in (26) along a finite one.
Integration in (24) is along an infinite loop that encircles 0 anticlockwise, in
(25) along a finite loop from 0 to 0 that encircles ζ1 ą 0 anticlockwise.
The basic, really indispensable transform is minor acceleration (23), and
the crucial point to note here is that the lower kernel CF pζ2 , ζ1 q, through some
minor miracle of pre-established harmony, has exactly the right faster-than-
exponential rate of decrease (as ζ1 Ñ `8) to make the acceleration integral
(23) convergent for small enough values of ζ2 ą 0. It therefore defines a germ
ϕ
p2 pζ2 q which then must, and can, be continued in the large.
One last remark: a ‘time’ zi is declared ‘critical’ if passage through the
corresponding Borel plane ζi is mandatory. But there is some latitude in the
choice of these critical times zi . They are actually defined up to an equivalence

8
lim zi {zi1 “ c ą 0, so that one should properly speak of critical time classes rzi s.
The next section is devoted to the corresponding transformations ζi Ñ ζi1 on
the Borel side, known as pseudo-accelerations and pseudo-deceleration.

5 The Great Divide: cohesive/loose.

The present section may be skipped at first, but is highly recommended in
a second reading. It is there ostensibly to guide us is the choice of proper
representatives zi in the various critical time classes rzi s. But it also sheds
a sharp and unexpected light on what is arguably the central watershed in
Analysis: the Great Divide between those smooth functions that are ‘of one
piece’ (analytic or quasi-analytic) and the functions not smooth enough to enjoy
that property (the ‘loose’ functions). First, though, we must define cohesiveness.

Notion of cohesive function.

We get the class COHES of cohesive functions by extending the classical Denjoy
classes αDEN to all transfinite orders α ă ω ω and then going to the limit: 13
α
` ˘´n
DEN :“ tf ; |f pnq ptq| ă c0,f pc1,f qn log1α`1 pnq u (27)
COHES :“ Yαăωω α DEN (28)
α
Like each DEN, the limit COHES is stable under `, ˆ, ˝, B and most other
operations. Crucially, it is also quasi-analytic : two cohesive functions defined
on a real interval J coincide as soon they coincide on a subinterval I Ă J.

Pseudoaccelerations/pseudodecelerations.
Here, the change is between two equivalent ‘times’, denoted for distinction by
z1´ and z1 with z1 “ z1´ `F pz1´ q and 1 ă F pxq ă x as above.14 The new trans-
forms serve a totally different purpose, but their integral kernels Cid`F , C id`F
are closely related to the old ones:
Cid`F pζ1´ , ζ1 q “ CF pζ1´ ´ ζ1 , ζ1 q (29)

C id`F pζ1´ , ζ1 q “ C F pζ1´ ´ ζ1 , ζ1 q (30)

In keeping with the more elementary character of the new transforms, all inte-
gration paths/loops now become finite.
Pseudodeceleration from ζ1 to ζ1´ obeys the formulae:
ż ζ1
ϕ
p1´ pζ1´ q “ Cid`F pζ1´ , ζ1 q ϕp1 pζ1 q dζ1 (31)
`0
ż v2
1
ϕq1´ pζ1´ q “ C id`F pζ1´ , ζ1 q ϕ
q1 pζ1 q dζ1 (32)
2πi v1
13 Despite the latitude in the analytic incarnation of the transfinite iterates logα`1 , each class
α DEN is unambiguously defined: the indeterminacy in logα`1 is absorbed by the constant
c1,f in (27).
14 The case when z
1´ and z1 are too close, i.e. when F pxq “ op1q, is uninteresting.

9
Pseudoacceleration from ζ1´ to ζ1 goes like this:
ż 02
1
ζ1 ϕ
p1 pζ1 q “ ζ1 ϕp1 pζ1 q C id`F pζ1´ , ζ1 q dζ1´ (33)
2πi 01 ´ ´ ´
żv
ζ1 ϕ
q1 pζ1 q “ ζ1´ ϕ
q1´ pζ1´ q Cid`F pζ1´ , ζ1 q dζ1´ (34)
ζ0

The most useful transforms are, paradoxically, the accelerations and pseudo-
decelerations. Despite going ‘in opposite directions’, both share a common reg-
ularising effect, albeit of crucially different force. To adequately describe that
common effect together with the discrepancy in regularising potency, we must
distinguish three sub-classes for each :
strong accelerations log z1 { log z2 Ñ 0 e.g. z1 “ log z2
medium accelerations log z1 { log z2 Ñ α Ps0, 1r e.g. z1 “ pz2 qα
weak accelerations log z1 { log z2 Ñ 1 e.g z1 “ logz2z2
z1
strong pseudodeceler . log z1 { logpz1´ ´z1 q Ñ 1 e.g. z1 “ z1´ ` log z´1
´
medium pseudodeceler . log z1 { logpz1´ ´z1 q Ñ α e.g. z1 “ z1´ `pz1´ qα
weak pseudodeceler . log z1 { logpz1´ ´z1 q Ñ 0 e.g. z1 “ z1´ `log z1´
Whatever the nature of the accelerand ϕp1 (provided it has the proper accelerable
growth at infinity), the corresponding accelerate ϕ
p2 is automatically guaranteed
a minimum of quasi-analytic smoothness – the weaker the accelaration, the less
the smoothness.
• Strong accelerates are always analytic in a spiralling neighbourhood of 0‚
with infinite aperture.
• Medium accelerates are always analytic in a neighbourhood of 0‚ with at
least finite aperture.
• Weak accelerates are always cohesive in a real right-neighbourhood s0, ...r
of 0‚ , but may lack an extension to the complex domain.
With pseudo-decelerations, the picture is the same, but on the other side of
the Great Divide — on the side of looseness: whatever the nature of the
pseudo-decelerand ϕ p1 , one can always, by suitably strenghtening the pseudo-
deceleration, ensure in the pseudo-decelerate ϕ
p1´ any given degree of smooth-
ness, short of cohesive.
Another difference is this : accelerations completely upset the singularity
landscape (they remove the old singular points and may create new ones)
whereas pseudo-decelerations keep all singular points ω in place and merely
smoothen the singularities there.

Cohesive continuation.
Any cohesive function given on an interval s0, ζr (viewed for the circumstance as
part of the axis argζ “ 0 in some Borel plane) can be constructively continued

10
to its maximal interval of cohesiveness s0, ζ˚ r by a suitably weak deceleration
followed by the reverse weak acceleration.

Cohesive singularities and their circumvention.

In some contexts like the Dulac problem15 , accelero-summation may produce
strictly cohesive germs on argζ “ 0 in some Borel planes, with any number of
cohesive singularities there. To proceed with accelero-summation, the germs in
question have to be cohesively continued (multivaluedly so) up to `8, which
means bypassing all intervening singularities to the right or to the left, while
never leaving the real axis! This sounds an impossibility, but is not.

6 Equational resurgence and the Bridge equation.

After describing, in its main outlines, the apparatus of resurgence theory, we
can now proceed to the applications. But let us first get some terminology
straight. By equational resurgence we mean resurgence that affects formal ex-
pansions in the variable of some equation (of whatever nature: differential or
functional). Coequational resurgence, by contrast, affects expansions in some
‘inert’ parameter. Both types, of course, rely on resurgence equations for their
description. Quite often, these resurgence equations are of a very specific type
– they unexpectedly connect alien and ordinary derivations, these denizens of
two distinct worlds – and are therefore called Bridge equations. So let us keep
these diverse notions cleanly apart.

Shape of the Bridge equation.

• In the present context of equational resurgence, the Bridge equation is
always of the form :

∆ω Y pz, uq “ Aω Y pz, uq p @ω P Ωq (35)

• Y pz, uq is the full (i.e. parameter-saturated) formal solution of, say, some
singular functional equation RpY q “ 0. To unclutter the notations, we
drop the tildas indicative of formalness. Here u :“ pu1 , .., ud q denotes a
maximal system of independent parameters, and z denotes the ‘critical
time’ of the problem or, if there be several, any one of them.
• ∆ω is an alien derivation taken in invariant form, i.e. with an exponential
factor that ensures commutation with Bz :

∆ω :“ e´ω z ∆ω ùñ r∆ω , Bz s “ 0 (36)

The index ω runs through a discrete set Ω Ă C‚ deducible from R.

15 See at the end of §10.

11
 • Aω is an ordinary, first order differential operator in z and in the pa-
rameters u1 , ..., ud . We may refer to it as the Stokes operator, because
its coefficients are indeed Stokes constants. They are also ‘holomorphic’
and ‘analytic’ invariants, since they depend holomorphically on the Taylor
coefficients of the equation R and remain unchanged when R undergoes
an analytic change affecting both the unknown and the variable.
• The outward shape of the differential operators Aω is quite simply the
most general of all possible shapes that makes formal sense. This means
that, when identifying the terms in front of any given monomial un on
both sides of (36), the exponential factors e´ωz should match.

• tAω ; ω P Ωu is a complete system of holomorphic-analytic invariants. In

the multicritical case, one should of course take all operators Aω relative
to all critical times zi , with ω running through the corresponding Ωi .
• The entire divergence of Y pz, uq is concentrated in the power series of z ´1 .
Once those have been resummed, there is no divergence left.

Scope and derivation of the Bridge equation.

The scope of the Bridge equation is breathtaking. It covers practically all singu-
lar analytic equations16 of all main types: differential, difference, mixed, general
functional. It also covers the so-called ‘local objects’, chiefly the singular vec-
tor fields X on Cν at 0, and the local self-mappings f of Cν with 0 as fixed
points. The main source of divergence for local objects is resonance, broadly
understood.17 That divergence manifests when we attempt to bring the ‘local
objects’ to some normal form by means of a change of coordinates, or when we
solve the equations of motion:

Bz xi pz, uq “ xi px1 pz, uq, . . . , xν pz, uqq p1 ď i ď νq (37)

fi pz`1, uq “ fi px1 pz, uq, . . . , xν pz, uqq p1 ď i ď νq (38)

As it happens, such divergence born of resonance is always amenable to resurgent

Analysis and fully describable by the Bridge equation, even in cases that are
completely out of bounds for alternative methods (geometric or others). A bulky
book and numerous papers have been devoted to the subject, and there can be
no question here of summarizing them. Let us just emphasize two points.
The first point is the relative ease of derivation of the Bridge equation, at
least in the monocritical case. Provided we judiciously decompose our local
object Ob as Ob 0`Ob 1 with a main part Ob 0 and a complement Ob 1 treated as
a perturbation, the analysis in the Borel plane becomes pretty straightforward:
it amounts to repeatedly dividing by expressions like pζ ´ ωq or pe´ω ζ ´ 1q, and
to calculating scores of convolution products that all have a strong regularizing
16 Properly speaking, local equations or germs of equation, with analytical germs as coeffi-

cients and analytical dependence on the unknown.

17 See §9. It includes the vanishing of one or several ‘multipliers’.

12
effect. As a result, convergence outside the singularity locus Ω is not overly
difficult to establish.
The second point to make is the unexpected unity that ř resurgence tends to
impart upon divergent objects. Let for instance Y pz, uq “ n un Yn pzq be the
formal solution of a singular differential equation RpY q “ 0, expressed in terms
of its critical time z and expanded as a power series of the parameters u. In
the generic case, i.e. if none of the Aω vanishes, the Bridge equation allows one
to recover (constructively so!) all components Yn from any one of them. This
would be clearly impossible in case of convergence, i.e. when all Aω vanish. A
rather apt simile may help bring the phenomenon into perspective. Let P pxq
be a polynomial with integer coefficients. If P is totally irreducible on Q, then
each root xi contains in spe all the others, but if P is fully reducible, these roots
become strangers to one another.

7 Co-equational resurgence.
The WKB or semi-classical approach.
My involvement with coequational resurgence dates back to my personal en-
counter with Yasutaka Sibuya and Andre Voros in the early 80s, and their work
on the one-dimensional, time-independent Schr:odinger equation with polyno-
mial potential:
#
2 x2 W pqq “ q ν `α1 q v´1 `...`αν
Bq ψpq, xq “ W pqqψpq, xq with 2
(39)
4 x “ hb ; α “ ´E penergyq

Sibuya had studied the q-dependence of the solutions in the spirit of Stir-
ling analysis, while Voros had tackled the x-dependence along the WKB ap-
proach and formulated remarkable conjectures regarding what he called a ‘boot-
strap’ phenomenon in the conjugate Borel plane – essentially a resurgence phe-
nomenon. On my part, I interpreted
şq a Sibuya’s result as a case of resurgence in the
‘critical variable’ z “ zpqq “ 0 W pq 1 q dq 1 ; proved the missing link in Voros’
conjectures; and observed an intriguing interplay between the two patterns of
resurgence, in z and in x, that seemed to point to a much wider phenomenon
— a duality of sorts between what I came to call equational and coequational
resurgence.

The second and third Bridge equations.

Coequational resurgence is distinctly trickier than the equational sort. For
one thing, it relies in the Borel plane on an intricate, ‘weighted’ convolution
product which is harder to handle than plain convolution, lacks its regularising
quality, and tends to generate singular points in far greater numbers. The main
difference, though, is the appearance of two new Bridge equations where there
was only one, and of universal ‘tesselation coefficients’.
Let x be the new multiplicative, resurgence-carrying variable, and ξ the
conjugate Borel variable. Our resurgent function ψpxq still verifies a Bridge

13
equation of familiar form:18
∆ω ψpx, uq “ Bω pxq ψpx, uq pω P Ω2 q (40)
except that now x reappears in the differential operators Bω pxq, so that a third
Bridge equation is needed to describe their resurgent behaviour:
∆ω Bω0 pxq “ hω,ω0 pBω1 , Bω2 , ...q pω P Ω3 q (41)
The exact shape of the right-hand side in (41) varies from case to case, but
essentially comprises two things:
(i) multiple Lie brackets of the Bωi ’s.
u1 ,..., u1
(ii) universal scalars tes p v1 ,..., v1 q – the so-called tesselation coefficients – that
depend only on two strings u anf v of complex numbers.

Tesselation coefficients.
Let us just mention two of their many fascinating properties:
(i) although they are piece-wise constant in each ui and vi , the only way of
expressing them without breaking their natural symmetries is as finite sums of
r!! hyperlogarithms, with r!! :“ 1.3.5...p2r´1q.
(ii) they are the most basic object exhibiting the important double symmetry
technically known as bialternality: see §12 towards the end.

Notion of autark function.

Many entire functions naturally occurring in Analysis, notably at the interface
of equational and coequational resurgence,19 display remarkable ‘finiteness’ and
‘closure’ properties: their behaviour at 8 depends on the sector, but in each
sector it can be modelled by divergent-resurgent asymptotic expansions, with
a closed system of alien derivatives. Let us call them autark functions, a name
suitably evocative of self-closure and self-sufficiency.
Autark functions have a quality of finiteness about them, and a predictabil-
ity of behaviour, that sets them apart from the wilder transcendental functions.
In fact, the dichotomy autark/non-autark is arguably no less basic than the di-
chotomy algebraic/transcendent. The paradigmatic example of an autark func-
1
tion is Γpzq . The paradigmatic example of a non-autark function is Ξpsq, the
entire fonction classically attached to Riemann’s zeta function. This of course
is due to the erratic behaviour of Ξpsq in the vertical strip |<psq| ă 21 , which
completely defies formalisation.

8 Object synthesis.
In the Bridge equation (35) of §6, we start from a singular equation RpY q “ 0
and derive its complete system tAω u of holomorphic invariants. But what about
18 Aswith the earlier Bridge equation, we drop the tildas indicative of formalness.
19 e.g.
the Stokes constants attached to singular ODE’s, when viewed as functions of one of
the ODE’s coefficients.

14
the reverse problem: starting from an admissible collection20 of differential op-
erators tAω u, search for an equation RpY q “ 0 admitting that collection as its
system of invariants? This is a problem of ‘object synthesis’, and solving it
is clearly a matter for ‘alien integration’. On the formal side, it reduces to a
rather mechanical exercise, over which there is no need to detain ourselves. It
is simply a question of iterating the Bridge equation by taking advantage of
the commutation r∆ω1 , Aω2 s “ 0 and of expanding Y pz, uq into highly multiple
series whose general term is of the form U ω1 ,...,ωr pzq Aω1 ...Aωr for some system
tU ‚ pzqu of resurgence monomials. The crux, of course, is to get these expansions
to converge in the space of resurgent functions.
Here, run-of-the-mill monomials won’t do. What is called for is a very spe-
cial type, the so-called paralogarithmic or spherical monomials. They are best
defined in the geometric model by means of the integrals:
ż 8 řj ωj pz´yj q`c2 řj ω̄j pz´1 ´yj´1 q
e dy1 ...dyr
Uω
c
1 ,..,ωr
pzq :“ SPA (42)
0 py r ´ yr´1 q . . . py 2 ´ y1 q py 1 ´ zq

The acronym SPA stands for ‘standard path averaging’ (a way of specifying how
the integration variables yi bypass each other). The crucial ingredient, however,
is the real parameter c which, if taken large enough21 , enforces convergence in
the ‘synthesis expansions’.
Another peculiarity of these monomials, which is almost obvious on their
definition and justifies the label spherical, is their broadly similar behaviour at
the antipodes z “ 8 and z “ 0 of the Riemann sphere. It is indeed a strange
feature of ‘spherical object synthesis’ that, while the avowed aim is to produce a
local object at z “ 8, it automatically creates an ‘antipodal shadow’ at z “ 0.

9 Causes of divergence and sources of resurgence.

Let us briefly discuss two important causes of divergence, which commonly go
by the name of small denominators. The divergence they call into being is non-
resurgent in nature, but may interfere with resurgence proper; and depending
on the situation, it may yield to resummation, or resist it absolutely.

The Liouvillian small denominators and the compensation technique.

An irrational number is Liouvillian if it is approximated by rational numbers
abnormally fast. Liouvillian small denominators typically occur with local vec-
tor fields or mappings whose multipliers (the eigenvalues of their linear part)
combine to produce Liouvillian numbers. As a rule of thumb, ‘Liouvillian diver-
gence’ is resummable whenever there stands a definite geometric object behind
the divergent series. In the opposite case, it is intrinsically unsurmountable.
20 admissible here means that the operators A
ω should have the proper shape, and that
their norms }Aω } should be properly bounded when ω increases.
21 Quite often, it is enough to take c ą 0.

15
 Consider for example a local analytic vector field on R2 , of the form:
ÿ λ1
X“ pλi xi ` . . . q Bxi with negative Liouvillian (43)
1ďiď2
λ2
ř
X is formally conjugate to its linear part Xlin “ 1ďiď2 λi xi Bxi , but the cor-
responding change of coordinates, having no geometric reality at its back, is
fated to remain formal. By contrast, fix two small numbers a1 , a2 ą 0 and
consider the well-defined geometric correspondence x1 Ø x2 such that pa1 , x2 q
and px1 , a2 q lie on the same ‘hyperbolic’ branch of X. The formal power series
that express x1 in terms or x2 (or x2 in terms of x1 ) are again clear cases of
Liouvillian divergence, but underwritten this time by geometric warranty, and
therefore resummable. How so?
The answer is that all abnormally large terms in these expansions can be ag-
gregated into clusters that guarantee mutual sign compensation. These clusters,
the so-called compensators, are of the form:22
ÿ ź
z ´tσ0 ,σ1 ,...,σr u :“ z ´σi pσi ´ σj q´1 pz P C‚ , σi P R` q (44)
0ďiďr j­“i

No matter how close the σi get to each other, the bound holds:
ˇ ˇ
ˇ ´tσ0 ,σ1 ,...,σr u ˇ 1 ˇˇ 1 ˇˇr ˇˇ 1 ˇˇσ˚
ˇz ˇ ă ˇ log ˇ ˇ ˇ pσ˚ :“ inf σi q (45)
r! z z
Now, the beauty is that this second type of Liouvillian divergence can amica-
bly coexist with resurgent divergence, and naturally fits into the general scheme
of accelero-summation, but with three essential nuances:
• Instead of the narrow critical time classes rzi s associated with resurgence
and corresponding to the equivalence relation lim zi1 {zi “ c ą 0, Liouvil-
lian divergence gives rise to much wider critical classes rrzi ss, corresponding
to the looser equivalence lim logzi1 {logzi “ c ą 0.
• On its own, Liouvillian divergence produces no singularities in the Borel
plane, and begets only analytic (as opposed to holomorphic) invariants.23
• In the geometric model, i.e. after resummation, Liouvillian divergence
leads to spiral-like neighbourhoods of 8‚ instead of sectorial ones.

The small denominators in Celestial Mechanics.

The small denominators associated with Celestial Mechanics and, more gener-
ally, with Hamiltonian systems, are of a quite distinct nature, and a cause for
much confusion.
The so-called Lindstedt-Poincare series – formally quasi-periodic – that de-
scribe the motion in the Many Body problem are, depending on the initial
22 For consistency, we revert to the large variable z „ 8.
23 Regarding the difference between holomorphic and analytic invariants, see §6.

16
conditions, sometimes convergent, sometimes divergent. In either case, there
is nothing to ‘compensate’. The confusion arises from the fact that even in
the convergent case, the coefficients cω of the L-P series that emerge from the
calculations involve finite sums of the form:
$
ÿ aω,k &r “ rpωq , K “ Kpωq
’
cω,r “ with λ1 , . . . , λν fixed (46)
ω ω . . . ωk,r
1ďkďK k,1 k,2
’

ωk,i “ ni , λ with ni P Zν
%

which may carry prohibitively small denominators, and therefore call for some
‘compensation’ mechanism. However, the finite expansion (46) for cω,r is by
no means unique, and if we conduct the calculations (inductively on r) with
deftness, by observing a neat set of rules, we can keep these abnormally small
denominators at bay.
More interesting is the case of Hamiltonian vector fields that, on top of the
intrinsic
ř resonance λi ` λi`ν “ 0, exhibit some extrinsic resonance, say λ1 “ 0
or ni λi “ 0. Attached to this extrinsic resonance, we have resurgence, carried
by a critical variable z and described as usual by the Bridge equation (35), but
with two eye-catching peculiarities:
• Each differential operator Aω now derives from a potential Aω – an ‘alien
potential’, so to speak.
• The shortest cut for calculating the potentials Aω involves writing the
original Hamiltonian H as H2 ` H, with a quadradic part H2 and a ‘per-
turbation’ H, and subjecting H to a remarkable involution H ÞÑ K :
1 1 1
K “ ´H´ tz, HuP ´ tz, tz, HuP uP ´ tz, tz, tz, HuP uP uP . . . (47)
2! 3! 4!
where t., .uP denotes the Poisson bracket, expressed in any map that iso-
lates the critical variable z.

Sources of resurgence.
By no means do equational and coequational exhaust the types of resurgence that
nature — physical nature or mathematical nature — may force on us. To keep
with mathematics for the moment,24 there is the wealth of power series whose
Taylor coefficients an have a pre-assigned form. Given sufficient regularity in
the make up of these an , resurgence is as good as guaranteed. To name but one
example, here is the case of coefficients with a so-called sum-product syntax:
ÿ ÿ ź k
ϕpζq
p :“ an ζ n with an :“ Fp q pF meromorphicq (48)
0ămăn 0ăkăm
n

The corresponding series ϕpζq

p arise naturally in Knot theory, and lead to resur-
gence equations entirely sui generis.
24 About physics, see §13 below.

17
10 Transseries and analyzable germs.
Transseries.
Very roughly, the algebra T r of transseries can be thought of as the natural
closure of Rrxs under t`, ˆ, B, ˝u and the inverse operations, with x living in
the real neighbourhood of `8. T r obviously contains E :“ exp, L :“ log and
the finite iterates En , Ln . After expulsion of all formal infinitesimals from the
exponentials, and of all terms other than x from Ln , each transseries Trpxq
decomposes into a sum of irreducible transmonomials, ordered from larger to
smaller. We crucially impose well-orderedness (each subseries of each T pxq
should have a first element), plus bounded logarithmic depth, plus, depending
on the context, various simplifying assumptions.

Analyzable germs.
r possesses a subsalgebra T
T r cv of directly convergent transseries (for x large
cv
enough) but T is radically unstable under integration, since even the simplest
r
transmonomials (think of Lα pxq or xα {Epxq for α R Z) have primitives that
are divergent (and resurgent). If we want both summability and stability, the
proper framework is Tr as , defined as the algebra of all transseries Trpxq that may
be resummed by accelero-synthesis. The corresponding sums T pxq, analytic or
cohesive on a real neighbourhood of `8, are dubbed analyzable germs.

Accelero-synthesis.
Accelero-synthesis is closely patterned on accelero-summation (see Fig 2 in §4),
but with four significant differences:
• In the early stages of the process, it is only the subexponential parts of
the transseries Tri pxi q ” Trpxq and of their individual transmonomials that
incarnate as analytic or cohesive germs in the Borel planes ξi . The other
parts provisionally retain their status as formal transseries.
• Since realness has to be preserved, integration must always take place
on the real axis arg ξi “ 0 of each critical Borel plane. When that axis
contains singular points, one must therefore resort to a well-behaved con-
volution average µi .

• Whenever two consecutive critical times xi´1 ăă xi are ‘close’25 , the germ
Tpi pξi q is liable to be non-analytic and merely cohesive. In the presence
of singularities on the real axis arg ξi “ 0, we therefore face the chal-
lenge, in order to calculate the average µi Tpi pξi q, of having to bypass these
singularities, to the right and to the left, without leaving the real axis!
Fortunately, that seeming impossibility can be overcome through some
delicate ‘cohesive acrobatics’.
25 e.g. when lim log xi { log xi´1 “ 1.

18
 • The final sum T pxq is always cohesive26 though not necessarily analytic.27

Decelero-analysis.
Being of the nature of construction, accelero-synthesis is a slow, arduous, step-
wise process, with mandatory passage through each critical Borel plane – or
Borel line, as the case may be. The reverse process, decelero-analysis, being
of the nature of destruction, is faster and easier, though not instantaneous: it
too may necessitate the passage through some critical Borel planes, and bypass
others. Nor need we always resort to the deceleration integrals: there exist
faster alternatives.

The Dulac problem and the return map.

The impetus behind the introduction of transseries and analyzable germs came
from the so-called Dulac problem. Given a polynomial vector field on R2 , prov-
ing the finiteness of isolated cycles reduces to proving the existence of only
finitely many isolated fixed points for the return map T pxq attached to any
given polycycle C. Once T pxq has been formalized to a transseries Trpxq, the
property becomes self-evident.

11 Pseudo-variables and display.

The notion of pseudovariable is dual to that of alien derivation ∆ω of the bold-
face, invariant sort: see (36). Pseudovarialbles carry as upper indices sequences
ω :“ pω1 , ..., ωr q of arbitrary length r. Multiplication for them reduces to se-
quence shuffling, while differentiation (ordinary or alien) and post-composition
obey the predictable rules:
1 2 ÿ ` ˘
Zω . Zω “ Zω ω P shufflepω 1 , ω 2 q (49)
∆ω0 Z ω1 ,...,ωr “ δωω01 Z ω2 ,...,ωr pδ “ Kronecker symbol q (50)
ω1 ,...,ωr
Bz Z “ 0 (51)
ω ω
Z ˝g “ Z if gpzq “ z ` opzq (52)

The display in the monocritical case.

The display Dpl is best thought of as some sort of ‘alien Taylor expansion’,
though of a radically different sort than the one mentioned in (13) of §2.
ÿÿ
Dpl ϕr :“ ϕr` Zω1 ,...,ωr ∆ωr . . . ∆ω1 ϕ
r (53)
r ωj

26 Think of sums like

ř
n 1{Es px ` nq ps ě 2q that always converge to cohesive sums, on the
strength of what we might call a ‘transserial Abel’s Lemma’.
27 Whereas with accelero-summation, the sum ϕ pzq is always analytic in some sectorial
θ
neighbourhood of 8.

19
The expansion (53) being formal in nature, the question of convergence does not
arise. The display has a double character ´ both local, via its z-dependence, and
global, via its Z-dependence. It encodes (displays, as it were) in ultra-compact
and user-friendly form, a huge amount of information about the function ϕpζq,
p
describing as it does the behaviour of ϕpζq
p at each ω and on each of its various
Riemann sheets.
What is more, thanks to the rules (50)-(52), any relation R between resur-
gent functions automatically extends to their displays :

t Rpϕ
r1 , . . . , ϕ
rs q ” 0 u ùñ t RpDpl ϕ
r1 , . . . , Dpl ϕ
rs q ” 0 u (54)

which can be extremely helpful for establishing transcendence or independence

results.

The display in the multicritical case.

The display is still defined in the multicritical case, but with two-tiered indices
$ “ p z$i q to specify the critical time:28
ÿ ÿ
Dpl.ϕ
r “ ϕ
r` Z$1 ,...,$r ∆$r . . . ∆$1 ϕ
r (55)
1ďr $1 ,...,$r

Any operation on the displays, like applying some alien derivation ∆$0 or going
from one multipolarised sum29
ÿ ÿ
pDpl.ϕqτ “ Sτ ` Z$1 ,...,$r p∆$r . . . ∆$1 ϕqτ (56)
1ďr $1 ,...,$r

to another multipolarised sum pDpl.ϕqτ 1 , reduces to purely formal manipula-

tions on the pseudovariables – munipulations that involve universal constants
depending on the pair pτ , τ 1 q, but strictly independent of ϕ. In that sense, the
display can be said to contain “everything there is to know” about the object
ϕ. In fact, it is only at the level of displays that the correspondence formal Ø
geometric attains perfection.

12 Resurgence as the impetus behind mould calculus.

Mould symmetries and mould operations.
Alien calculus presents us at every step with resurgence monomials U ω1 ,...,ωr pzq
or scalar monics U ω1 ,...,ωr that have their own symmetries (four main types,
scores of secondary ones) and undergo numerous operations that either pre-
serve or exchange these symmetries. The systematization of these operations
naturally led to mould calculus ( moulds are simply objects M ω1 ,...,ωr indexed
28 In the general transserial context, it is convenient to represent $ by the leading trans-

monomial of ω xi , with xi expressed as a transseries in x.

29 Any multipolarised sum depends on the choice, in each critical Borel plane ζ , of an
i
integration axis arg ζi “ θi plus, if need be, a well-behaved average µi (see at the end of §3).

20
by scalar sequences of any length), which later found many applications outside
its native context, in such fields as differential geometry, Lie or pre-Lie algebra,
etc. While at one level moulds with all their wherewithal may be dismissed as
just a glorified system of notations, the fact is that they often allow us to make
fully explicit what would otherwise remain implicit, and to go beyond mere ‘ex-
istence theorems’ (that all too often are sterile dead-ends) by illuminating the
innards of the object whose bare ‘existence’ has been proved.

Bimoulds, double symmetries, and the flexion structure.

u1 ,..., ur
The bimoulds M p v1 ,.., vr q are a rather different proposition. They crystallized,
together with their two-tier indexation, out of the intricate combinatorics that
underpins coequational resurgence, and inherited therefrom a plethora of struc-
ture. The fact is that they can undergo an incredibly rich array of operations,
unary and binary, resulting in what is known as flexion polyalgebra. But the
most salient feature is perhaps the existence:

• of a central involution swap:

u1 ,..., ur vr ,..., v2 ´v3 , v1 ´v2
pu q
pswap.Mqp v1 ,..., vr q
:“ M 1 `...`ur ,..., u1 `u2 , u1
(57)

• of bimoulds M ‚ possessed of a double symmetry, e.g. bialternality, mean-

ing that both M ‚ and its ‘swappee’ are simultaneously alternal:
#
ÿ ÿ @ w1 , w2
Mw “ swap.Mw “ 0 (58)
1 2 1 2 sha for shuffle
wPshapw ,w q wPshapw ,w q

• of binary operations that preserve these double symmetries.

This makes flexion algebra an ideal framework for unpicking arithmetical di-
morphy, a phenomenon in no way confined to the ring of multizetas, but preem-
inently manifest there, due the two canonical encodings of multizetas and the
two multiplication tables that go with them.

13 Resurgence in mathematical physics.

The ‘Michael Berry principle’.
The well-known quantum physicist Michael Berry30 is fond of saying that when
a new physical theory departs from its classical model by the introduction of a
small constant (like h for quantum mechanics), we should expect expansions in
power series of that small constant to diverge: their divergence merely reflects
the non-trivial nature of the transition from classical to non-classical. Withal,
30 Sir Michael, to give him his full due. His PhD advisor, incidentally, was Robert B.Dingle,

one of the pioneers of asymptotics. Other emblematic figures of that select club were Leonhard
Euler, James Stirling, and (somewhat controversially) Henri Poincare.

21
these expansions ought somehow to be resummable, this time to reflect the
physical relevance of the transition.
The Planck constant is an obvious case in point. We saw in §7 examples of
this with the so-called WKB method, A.Voros’ semi-classical treatment of the
harmonic oscillator, and the whole sprawling field of co-equational resurgence.
There exist other candidates than h for serving as ‘variable’ in formal expan-
sions, such as coupling constants. The trouble here is that only the very first
terms are accessible to experimental verification, and that anything beyond the
dominant effects (read: the closest singularities in the Borel plane) seems to lie
hopelessly beyond its reach.

Front-line physics.
In somewhat different directions, I hear of resummation methods being brought
to bear on such questions as tunneling effects in quantum mechanics, renor-
malons in quantum field theory, even some toy models in string theory, by
authors such as Mithat Unsal, Ricardo Schiappa, Ines Aniceto, and others. The
line of research I am most familiar with is Ovidiu Costin’s and Gerald Dunne’s.
I know of their attempts to squeeze the maximum of information out of scant
data, and I am greatly impressed by the odd-defying numerical accuracy their
methods often achieve. In the nature of things, however, the overall resurgence
picture seems to elude the current approaches. Missing, too, is the algebraic
side (crucial, from my viewpoint) of resurgence theory. I am told, however, that
efforts are afoot to fill this lacuna, and one certainly cannot rule out a concep-
tual breakthrough that would turn things around for good, and allow theory to
outpace experiment.

Asymptotics beyond all orders.

Going by this name is an asymptotic doctrine, pioneered by Michael Berry, that
has attracted wide attention over the last three decades and spawned a flood
of publications. The idea, roughly, is to indefinitely iterate the classical – and
numerically highly effective – trick known as least term truncation to achieve
any order of accuracy. The idea is seductive at first sight, but only at first sight.
In fact, a recent paper, coauthored by Ovidiu Costin and sir Michael himself,
has taken a closer look at the cost-effectiveness of the method, and laid bare the
somewhat pyrrhic nature of the gains it claims to achieve. Another drawback, no
less serious in my view, is that in the successive ‘models’ on which the method
relies, the resurgence picture (beginning with the resurgence equations) gets
completely distorted.

14 Some loose thoughts by way of conclusion.

Mathematicians, while in hot pursuit of their prize, understandably won’t let
others see into their cards. But even after the event, with all work done and
safely under their belt, many remain unwilling to open up about their initial

22
motivations, or to explain candidly what exactly it is that they cherish in their
brainchildren. Such reticence seems misplaced, for what merit can there be in
hiding the vital part of the creative process under the bushel, and what profit
in keeping the allure of a mathematical structure under burqa? Anyway, the
reticent are welcome to their reticence, but here I feel free to deliver myself of
a few thoughts about resurgence.
First to come to mind is the theory’s breadth of application along with its
unifying power – two traits best in evidence in the Bridge equation, which covers
huge ground and corrals seemingly disparate phenomena into a single framework
of compelling simplicity.
Next comes what we might call the analytical bent of the theory, which
culminates in the twin notions of transseries and analyzable germ. It is here or
nowhere that the dream of full formalization comes true – the dream of reducing
opaque, seemingly intractable geometric entities (and the operations on them)
to transparent formal objects (themselves subject to transparent operations).
Third (or should I say first ?) comes the wonder of a precise, sharp-contoured
structure – alien calculus – spontaneously arising from what would seem to be
the most unpromising and amorphous of contexts: divergence. Here lies, at
least to my subjective feeling, the core attraction of resurgence. It is indeed
often the case that mathematicians harbour secret preferences, centered on quite
concrete notions. Some, for instance, are in thrall to the identity d2 “ 0 and the
cohomological marvels that flow from it. I for one confess to an innate liking
for derivations, i.e. for operators that obey the Leibniz rule. I distinctly recall
how, aged 17, one fine day during the summer recess I found myself sitting
on the bank of a mountain stream, wondering: “Might there not exist a class
of smooth functions on which there operate deep derivations?” By which I
meant derivations that would somehow involve, all at once, the infinite string
of ordinary derivatives at a given point. When a few years later, as a result
of pursuing a seemingly unconnected line of investigation, it dawned on me
that such critters actually existed, in superabundant quantity and with no taint
of pathology about them, there was an immediate click of recognition, which
comforted me in my sense of being on the right track, and kept me hooked to
the subject for the next fifteen years.
Then there is – rarely acknowledged, but nonetheless essential – the symbolic
charge, or if you prefer the aura of associations, which nearly all key mathemat-
ical notions carry with them, not just in the eye of the beholder, but also, I
venture, in an almost objective sense. This is, however, a point that people are
apt to misconstrue, so let us clarify it on a striking example: analytic functions.
These functions31 have the fantastic property of being ‘of one piece’ – if you
know a little chunk, you know the whole thing. A truly magic quality, that in-
fuses them with life, and turns them into natural similes for this eternal theme:
the Whole in each of its Parts; the Macrocosm in the Microcosm; etc. We can
find echoes of this everywhere – in Oriental or Hermetic Philosophy; in Biology
(the full genome is encoded in each cell of a living organism); and, at an almost
31 Together with the cohesive functions which we encountered in §5.

23
literal level, in Newtonian physics: if space were truly analytic, then by knowing
the gravitational potential in a cubic inch of space to infinite accuracy, we could
in theory infer the position of all massive particles in the world, and to that
extent “know everything”. Taken literally, this is nonsense, of course, and we
should carefully avoid mistaking symbols for explanatory mechanisms. But this
in no way detracts from their evocative power or their vivifying potency for the
soul, not least the creative scientific soul.
So let me conclude by pointing to two such aspects in resurgence theory,
both highly loaded, and both arresting.
There is this hierarchy of ‘emergent’ levels of organization in the universe
– microphysics, chemistry, biology, history – each with its own laws and its
own patterns of ‘causality’, the higher the looser. Resurgence to me is strongly
evocative of these multi-track causalities (and others we are free to imagine).
Indeed, here we have in the Borel plane, riding piggyback on a local causality
(the step-by-step analytical continuation from one Weierstrass element to the
next), a long-range causality that ‘miraculously’ transports the situation at
the origin to distant singular points.32 I am not suggesting, heaven forbid, that
physical space-time might serve as a medium for such non-local transportation,33
or even that resurgence might model some real-world mechanism. I am just
saying that it provides a simile – nothing more, nothing less – for the peaceable
coexistence of quite distinct levels of causation.
Then we have the grand scheme of accelero-synthesis and decelero-analysis,
as sketched in §10, with its majestic double movement between a state of max-
imal dispersal (– the formal transseries Trpxq, with its disparate collection of
isolated coefficients hanging forlornly on a sprawling tree-like structure –) and
a state of maximal compactness-cum-cohesion (–the infrangible geometric germ
T pxq, of one piece on account of its analyticity or quasi-analyticity–). That dou-
ble movement – one of slow, arduous ascent; the other of sudden, precipitous
fall – to me carries a compelling, almost self-evident symbolic charge.

32 Meromorphic periodic functions also have that quality, but to a lesser degree, because

here the singularities merely repeat. And if we quotient the underlying space by the period,
the repetition disappears altogether. Not so with genuine resurgence: here we have creative
innovation on top of repetition, and no artifice of definition or change of framework can whisk
that aspect away.
33 For one thing, the consensus at the moment, or should we say the general suspicion, seems

to be that physical space-time is an emergent rather than a primary reality. In any case it is
certainly not, at the finest resolution, a ‘real analytic manifold’ in the mathematical sense.

24