Resurgence in Geometry and Physics

Brent Pym

Trinity Term 2016

Lecture 3

Abstract
We introduce the Borel transform as an operation that takes a germ
of a function on a Riemann surface, and produces a one-form on a vector
space. Its inverse is the Laplace transform. We then introduce the classes
of Gevrey series and Gevrey function, and use the Borel and Laplace
transforms to prove Watson’s theorem on the existence and uniqueness of
sums for Gevrey series.

Contents
1 Integral transforms 2
1.1 The Borel transform . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The Laplace transform . . . . . . . . . . . . . . . . . . . . . . . . 5

2 Formal transform and Gevrey functions 7

2.1 Formal Borel and Laplace transforms . . . . . . . . . . . . . . . . 7
2.2 Gevrey series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Gevrey functions . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Resummation of Gevrey series . . . . . . . . . . . . . . . . . . . . 10

At the end of the last lecture, we explained that every formal power series
can be realized as the asymptotic expansion of a function defined in an arbitrary
sectorial neighbourhood of a point. But because there are many functions whose
asymptotic expansions are zero, this procedure does not give a unique sum for
the asymptotic series. In this lecture, we will eventually single out a “nice” class
of series—the series of Gevrey class—for which better uniqueness results can be
obtained. But in order to motivate the introduction of this class of series, we
will first introduce another key set of players in our story: the Borel and Laplace
transform.

1
1 Integral transforms
An integral transform is a procedure that relates functions or tensors defined on
different spaces X and T. In our case, X and T will both be Riemann surfaces,
and we will construct a correspondence

'
holomorphic functions
g on X holomorphic one-forms on T

that will be used to approach the problem of resumming divergent series. In

fact, it will be important for us later in the course that T has an additional
structure: it is a one-dimensional complex vector space.
The constructions as presented here rely on the choice of a coordinate x on
X centred at a point p ∈ X. Meanwhile, the space T, being a one-dimensional
vector space, is equipped with a global linear coordinate t. It seems to be
conceptually helpful to think of T as the tangent space to X at p, and the
coordinate t as the natural coordinate t = dx|p ∈ T∗ induced by x. We then
imagine that the correspondence allows us to study functions on X by “zooming
in” on X near p so drastically that X is replaced by its tangent space.
With T interpreted in this way, the constructions will be invariant under
changes of coordinate in which x is rescaled by a constant. But arbitrary changes
of coordinate will result in actually different integral transforms, so this geomet-
ric interpretation should not be taken too seriously. We remark, however, that
if X is simply connected and non-compact, there is a special class of coordinates
on X: the uniformizing ones. These are the coordinates that identify (X, p) with
either (C, 0) or (∆, 0), where ∆ ⊂ C is a disk centred at zero. Since the ratio
of any two uniformizing coordinates is a constant, we can make our construc-
tions canonical by requiring x to be uniformizing. This solution has the obvious
drawback that uniformizations of a given abstract Riemann surface are rather
difficult to find.
We therefore assume throughout these notes that the coordinates x and t
have been fixed, and use the interpretation of T as the tangent space only as an
intuitive guide.

1.1 The Borel transform

We begin by describing the procedure that takes a function on X to a one-form
on T, called the Borel transform. Like any integral transform, it is defined
by integrating f against a “kernel” that depends on both x and t. In this case,
the kernel is the two-form
et/x
dx ∧ dt = −d(et/x ) ∧ dt
x2
defined on the product X × T using the coordinate functions x and t.

2
 This form evidently has an essential singularity where x = 0, i.e. along the
locus {p}×T. But if we integrate it along a path in X that avoids the singularity,
we clearly obtain a holomorphic one-form defined on all of T.
More precisely, suppose that f ∈ OX,p is the germ of a holomorphic function
defined in a neighbourhood U of p. By shrinking U if necessary, we may assume
without loss of generality that U is simply connected, so that U\{p} is isomorphic
to a punctured disk. Let γ be a loop in U \ {p} that wraps once around p in
the positively oriented sense, i.e. a generator for the first homology of U \ {p},
as shown in Figure 1. Then the Borel transform of f with respect to the
coordinate x is the one-form defined by the formula
Z 
1 t/x dx
B(f ) = fe · dt ∈ Ω1 (T)
2πi γ x2

By Cauchy’s theorem, the result is independent of the chosen loop γ. It is

therefore apparent that the Borel transform only depends on the behaviour of
f in an arbitrarily small neighbourhood of p, i.e. on the germ f ∈ OX,p .
Exercise 1. Show that the Borel transform is unchanged if the coordinates x
and t undergo a simultaneous rescaling by a constant λ, as would happen if T
were the tangent space of X at p and t were the linear coordinate dx|p ∈ T∗ .
What happens if we apply a more complicated change of coordinates?

X
U

Figure 1: Integration contour for the Borel transform

The Borel transform may be rewritten in various different ways. For exam-
ple, integration by parts using the identity

et/x dx
f· = −f · d(et/x ) = −d(f et/x ) + et/x df
x2
gives the formula Z 
1
B(f ) = et/x df · dt,
2πi γ

3
which shows that B(f ) is unchanged if we add a constant to f . Meanwhile, the
residue theorem gives

f et/x dx
 
B(f ) = Resx=0 · dt.
x2

which facilitates the computation of the transform.

Proposition 1. Let f ∈ OX,p be the germ of a holomorphic function at p ∈ X.

Then the following statements hold:
1. The Borel transform of f is a holomorphic one-form defined on all of T,
with at most exponential growth at infinity. In other words,

B(f ) = g dt

where g is an entire function such that

|g| ≤ CeM |t|

for some constants C, M > 0.

2. If f has the Taylor expansion
∞
X
f = f (p) + ak xk+1
k=0

then
∞
X ak tk
B(f ) = dt
k!
k=0

3. The Borel transform defines an injective C-linear map

B : mX,p → Ω1 (T)

where mX,p ⊂ OX,p denotes the ideal of functions that vanish at p.

4. The following identity holds for all f ∈ OX,p :

B(x2 ∂x f ) = tB(f )

Sketch of proof. The first statement follows easily from the fact that the in-
tegrand f et/x x−2 dx depends analytically on t, and evidently has a uniform
exponential bound on any circle of fixed radius |x| =  > 0.

4
 For the second statement, we observe that by linearity, it is enough to show
k
that B(xk+1 ) = tk! dt. For this, we use the residue formula and calculate
 
k+1 t/x dx
B(x k+1
) = Resx=0 x e · dt
x2
1 t2 1 t3
   
t k−1
= Resx=0 1 + + + + ··· x dx · dt
x 2 x2 3! x3
tk −1
   
k−1 k−2
= Resx=0 x + tx + · · · + x + · · · dx · dt
k!
tk dt
=
k!
as required.
The third statement is an immediate consequence of the second: clearly the
only information that is lost about the Taylor series of f is the constant term
f (p), and this ambiguity can be removed by requiring f to vanish at p. Since the
germ of an analytic function is determined by its Taylor series, the statement
about injectivity follows.
The fourth statement is obtained easily using integration by parts.

1.2 The Laplace transform

The Borel transform has given us an injective map

B : mX,p → Ω1 (T)

which takes a germ of a function that vanishes at p and produces an entire

one-form on T with exponential growth at infinity. Let us denote the set of such
one-forms by Ω1exp (T).
We claim that, in fact, we have an isomorphism
∼
B : mX,p −
→ Ω1exp (T).

To see that this is true, we will explicitly construct the inverse map, which is
given by the Laplace transform

L : Ω1exp (T) → mX,p ,

The construction again uses our coordinates x and t: it is an integral transform

whose kernel is given by the function

e−t/x

defined on X×T. But now some more care is required in defining the integration
cycle; we will be integrating along an infinite contour in T, and we must ensure
that the contour lies in a region where e−t/x decays.

5
 T
X
p

q0 γq
q
γq 0

Figure 2: Integration contours for the Laplace transform

To this end, we observe that, given a point q ∈ X, we can always find a ray
γq from 0 to ∞ in T such that t/x(q) takes on positive real values along γq .
Here x(q) ∈ C is the fixed complex value assigned to q by the coordinate x. If
we use our intuitive picture of T as the tangent space of X at p, then γq can be
thought of as a tangent ray at p that points towards q as shown in Figure 2. In
any case, with this choice of ray γq , the kernel function e−t/x(q) decays rapidly
as t → ∞ along γq .
Now suppose that ω = g dt ∈ Ω1exp (T) is a form on T that satisfies an
exponential bound
|g| ≤ CeM |t| ,
for some C, M > 0. Then the integral
Z Z
e−t/x(q) ω = e−t/x(q) g(t) dt
γq γq

will be guaranteed to converge as long as |x(q)| < M −1 .

In this way, we obtain the Laplace transform L (ω) ∈ OX,p , defined by
the formula Z
(L (ω))(q) = e−t/x(q) ω.
γq

This integral makes sense as long as q is sufficiently close to p. With a bit more
work it is easy to see that the value of the integral depends holomorphically on
q, and therefore defines a germ of a holomorphic function at p, as required.
Exercise 2. Show that the integral is independent of the ray γq , as long as this
path goes to infinity in a direction such that e−t/x(q) decays. This can be done
directly, by the usual trick of estimating the integral along an arc of large radius.
Or, for a more geometric approach, consider the manifold with boundary

T̂ = T ∪ S∞ T,

obtained by forming the real-oriented blowup of the Riemann sphere T ∪ {∞}

at infinity. Show that for fixed q, the one-form e−t/x(q) ω vanishes on a sector

6
A ⊂ S∞ T = ∂ T̂ in the boundary circle, and apply Stokes’ theorem. Interpret
the integration as a pairing on appropriate relative homology groups.
Exercise 3. Use the path independence of the integral and “differentiation under
the integral sign” to show that L (ω) is holomorphic.
Theorem 1. Given a coordinate x on X centred at p, the corresponding Borel
and Laplace transforms give mutually inverse isomorphisms
B
(
mX,p Ω1exp (T)
h
L

between germs of holomorphic functions on X that vanish at p, and entire forms

on T with at most exponential growth at infinity.
Proof. The proof is obtained by the usual trick of interchanging the order of
integration. We leave it as an exercise to the reader.

2 Formal transform and Gevrey functions

2.1 Formal Borel and Laplace transforms
In light of Proposition 1, which gave an explicit formula for the integral trans-
form in terms of Taylor expansions, we may define the formal Borel and
Laplace transforms B, b Lc by the formulae
∞ ∞
!
X X ak tk
Bb ak x k+1
= dt
k!
k=0 k=0

and
∞ ∞
!
X X
L
c bk tk dt = k!bk xk+1 .
k=0 k=0
using our chosen coordinates x and t. They give mutually inverse isomorphisms

B
)
b

m b 1 (T)
Ω
g
b X,p

L
c

where we denote by
b X,p ∼
m = xC[[x]] ⊂ C[[x]] = O
bX,p

the formal power series with zero constant coefficient, and

b 1 (T) ∼
Ω = C[[t]] dt
the one-forms on T with formal power series coefficients.

7
2.2 Gevrey series
Notice that, because the Borel transform divides the kth coefficient by k!, it is
perfectly possible for the Borel transform to have a nonzero radius of conver-
gence, even if the original series diverges.

Definition 1. A formal power series has Gevrey class 1 if its Borel transform
has positive radius of convergence. We denote by GbX,p ⊂ O
bX,p = C[[x]] the series
of Gevrey class one.
Convergent power series have coefficients that grow at most exponentially,
so the following lemma is immediate:

Lemma 1. The formal power series

∞
X
ak xk+1
k=0

has Gevrey class 1 if and only if there exists a constant M > 0 such that

|ak | < M k k!

for all k ∈ Z≥0 .

As the name suggests, there are other Gevrey classes, which correspond to
series with different bounds on their coefficients. We will not make much use
of these other classes here, so we will typically drop the terminology “class 1”
and simply say that f is a Gevrey series. As the following exercises show,
the Gevrey series define a natural subalgebra GbX,p ⊂ O bX,p of the ring of formal
functions at p, defined independently of any coordinate on X.
Exercise 4. Show that the product of two Gevrey series is Gevrey.
Exercise 5. Suppose that x = x(u) where u is another coordinate on X at p.
Using Lemma 1 above, together with Faà di Bruno’s formula for the composition
of power series, show that f (x) ∈ C[[x]] is Gevrey if and only if f (x(u)) ∈ C[[u]]
is Gevrey.
The importance of Gevrey series is that they are the ones which we can
hope to resum by the method of Borel summation, as follows. Given a Gevrey
series f ∈ GbX,p we consider its formal Borel transform B(f
b )∈Ω b 1 (T), which by
definition converges to define a holomorphic form ω on T in a neighbourhood of
zero. We then attempt to define a function of x by extending ω holomorphically
to all of T, and taking its Laplace transform.
We will spend most of the rest of the course dealing with the fact that ω
will not actually extend holomorphically to all of T. In good cases, it will have
only a discrete collection of branch cuts and poles; these are the “resurgent”
cases. But for the rest of this lecture, we will see that some interesting results
can be obtained even when the form does not extend beyond its original disk of
convergence.

8
2.3 Gevrey functions
Given that we have singled out the Gevrey series as particularly worthy of our
attention, it is natural to try to realize these series as asymptotic expansions of
some restricted class of functions.
Suppose that f ∈ O(U)
e admits an asymptotic expansion
∞
X
f∼ a k xk
k=0

in a sectorial neighbourhood U of a point p. Recall from the previous lecture

that the coefficients of the expansion in a coordinate x may be computed by a
Taylor-type formula
f (k) (x)
ak = lim
x→0 k!
and the limit is uniform in any subsectorial neighbourhood U0 ⊂ U whose closure
is contained in U. In fact, this statement can be upgraded so that the error term
on U0 is given by
 
N
 X  1
ak xk  ≤
 N +1 
f − sup f (y) .
 
  N ! y∈U0 ,|y|≤|x|
k=0

This estimate is proven in a fashion similar to the usual Taylor remainder the-
orem.
Definition 2. Let U be a sectorial neighbourhood of a point p ∈ X. A function
f ∈ O(U) is Gevrey if for every subsectorial neighbourhood U0 ⊂ U whose
closure lies in U, there exists a constant M > 0 such that
 (k) 
f 
sup   < M k k!
U0 k! 
for all k ∈ Z≥0 .
This condition ensures that every Gevrey function admits an asymptotic
expansion that is a Gevrey series. We therefore have a map

G(U) → GbX,p

defined by taking asymptotic expansions. Since we recall that asymptotic ex-

pansions are far from unique, we expect that the kernel of this map will be rather
large. For example, we may expect it to contain any function that decays like
β
e−1/|x| with β > 0, but the following lemma limits the possibilities:
Lemma 2. The kernel of the expansion map

G(U) → GbX,p

is the space of functions that decay exponentially in U.

9
Sketch of proof. See [1, p. 137] for details. Using the Gevrey condition, the
fact that f is asymptotic to zero, and the Taylor remainder formula for the
asymptotic expansion, we obtain in the coordinate x near p a bound of the form

|f | ≤ C(M |x|)n n!

on U0 . With a bit of work starting from Stirling’s approximation

√  n n
n! ∼ 2πn ,
e
the equivalence between this bound and an exponential bound of the form

|f | ≤ C 0 exp(−M 0 /|x|)

can be established.
β
Corollary 1. The function e1/x is not Gevrey for any β < 1.

2.4 Resummation of Gevrey series

We may now combine our considerations to obtain the following key result on
the resummation of Gevrey series, known as Watson’s theorem:
Theorem 2 (Watson [2]). Let p ∈ X be a point in a Riemann surface, and let
A be a sector at p. Consider the map

G(A) → GbX,p ,

which takes a germ of a Gevrey function defined in a sectorial neighbourhood of

A, and produces its asymptotic expansion. Then
1. If |A| > π this map is injective.
2. If |A| < π this map is surjective.
Proof. (Taken from [1, p. 138–139].) For the first statement, we note that any
function in the kernel of the expansion map must decay exponentially in A by
Lemma 2. But since |A| > π, the Fragmen–Lindelöf principle from Lecture 2
ensures that such a function is identically zero.
For the second statement, we apply Borel summation. Suppose that f ∈ GbX,p
is a Gevrey series and choose a coordinate x centred at p. We denote by U ⊂ X
a small sectorial neighbourhood with opening A, and assume without loss of
generality that U is contained in the domain of the coordinate x.
Subtracting a constant if necessary, we may assume that f has the form
∞
X
f= ak xk+1 .
k=0

Let ω = B(f
b ) be the formal Borel transform of the series with respect to the
coordinate x. Since f is Gevrey, ω defines a holomorphic one-form in some disk

10
V ⊂ T centred at zero. In general, it may be impossible to extend this form
beyond V, so we cannot apply the Laplace transform directly to ω. But for
our purposes, it is sufficient to consider a sort of truncated Laplace transform,
in which we integrate along a finite ray, rather than an infinite one. Indeed,
consider the function Z a
F (x) = e−t/x ω
0

where a ∈ V is chosen so that Re(a/x) > 0 as x → 0 in U; such an a can be

chosen precisely because |A| < π. The function F is evidently holomorphic in
U; we claim that F is, in fact, asymptotic to the original series f .
To this end, we write ω = g(t) dt where
∞
X ak tk
g= .
k!
k=0

Integrating by parts N + 1 times gives the equation

N
X N
X
F (x) = ak xk+1 − e−a/x ak xk+1 + xN +1 ψN (x)
k=0 k=0

where ψN is a bounded function. Since the function e−a/x decays exponentially

in U, this equation gives the desired bound
 N −1

 X 
F − ak xk+1  < C|x|N +1
 
 
k=0

uniformly on subsectors. With a bit more work one can check that the function
is Gevrey.

References
[1] B. Y. Sternin and V. E. Shatalov, Borel-Laplace transform and asymptotic
theory, CRC Press, Boca Raton, FL, 1996. Introduction to resurgent analy-
sis.
[2] G. N. Watson, A Theory of Asymptotic Series, Philosophical Transactions
of the Royal Society of London A: Mathematical, Physical and Engineering
Sciences 211 (1912), no. 471-483, 279–313.

11