Weak KAM theory in higher-dimensional

holonomic measure flows

Rodolfo Rı́os-Zertuche
arXiv:1901.08181v1 [math.DS] 24 Jan 2019

Abstract
We construct a weak kam theory for higher-dimensional holonomic
measures. We define their slices and curves of those slices. We find a
weak kam solution in that context, and we show that in many cases
it corresponds to an exact form that satisfies a version of the Hamilton-
Jacobi equation. Along the way, we give a characterization of minimizable
Lagrangians, as well as some abstract weak kam machinery.

1 Introduction
In this paper, we construct a weak kam theory (see for example [2] for an
introduction) for Lagrangian action functionals of the following kind: Given
a smooth manifold M of dimension d, we let T n M be the Whitney sum of
1 6 n < d copies of the tangent bundle, and let L : T n M → R be a smooth
function. Then, for an embedding ϕ : U ⊂ Rn → M , the action of L is defined
to be Z
∂ϕ ∂ϕ
L(ϕ(x), ∂x 1
(x), . . . , ∂x n
(x)) dx1 . . . dxn .
U
An important example of such a functional is the surface area, which corresponds
to setting q
L = det(g(vi , vj ))ni,j=1
for a Riemannian metric g. These actions can in greater generality be evaluated
at all compactly-supported, Radon measures on T n M , which constitute the
weak* closure of the measures induced by integration of embeddings like ϕ, and
which we term holonomic measures.
In order to construct the weak kam theory, we propose a way to think about
the flow of holonomic measures in terms of their time-slices, which gives rise to
the concept of curves of slices. We are able to find weak kam solutions on the
space of slices and to relate them to exact forms on M .
The contributions of the paper are as follows:
1. A concept of slices of a holonomic measure that allows the slice to be
simultaneously an object of dimension n and an object of dimension n − 1.
This is achieved using a cotangential component. See Definition 2.
2. A concept of curves in the space of slices that, without considerably re-
stricting the space of minimizers, is adequate for most optimization prob-
lems in the literature, and clarifies the idea of flow in the space of slices.
See Definition 3.

1
 3. A weak kam theorem for the space of slices, Theorem 6, which produces
a Lipschitz function on the space of slices with the usual weak kam prop-
erties.
4. A characterization of all Lagrangians whose actions are minimizable by
holonomic measures, Theorem 8.
5. A way to connect the function produced by the weak kam theorem with
the existence of an exact form on the manifold that satisfies a sort of
Hamilton-Jacobi equation. See Corollary 11 and Remark 9.
6. Some very general weak kam machinery, Theorem 6, abstracted from the
work of Maderna–Fathi [3], suitable for σ-compact, metric spaces.
The organization of the paper is as follows. Section 2 presents the concepts
of slices and their curves, as well as the weak kam theorem. Section 3 gives
the characterization of minimizable Lagrangians and the connection of weak
kam solutions to exact forms. Finally, Section 4 treats the abstract weak kam
machinery.

Acknowledgements. I am deeply grateful for the advice of Albert Fathi,

whose question unleashed this line of research. I am deeply indebted to Patrick
Bernard, Valentine Roos, and Stefan Suhr for helpful discussions.
I am also very grateful to the École Normale Superieure de Paris and the
Université de Paris – Dauphine for their hospitality and support during the
development of this research.

2 Slices and curves of slices

In this section we give a definition of slices of holonomic measures, and of curves
of such slices. We also give Example 5 as a way to clarify where these definitions
come from. The main result of this section is Theorem 6, which gives the weak
kam theory in this context.
Let M be a smooth, connected manifold of dimension d, and let n > 0 be
an integer. We let T n M be the Whitney sum of n copies of the tangent bundle
T M , so that the fibre Txn M = Tx M ⊕ Tx M ⊕ · · · ⊕ Tx M is a Euclidean vector
space of dimension nd.

Definition 1. Let c be a normal current on M of dimension n − 1. A positive,

compactly-supported, Radon measure µ on T n M has boundary c if, for all
ω ∈ Ωn−1 (M ),
h∂Tµ , ωi = hTµ , dωi = hc, ωi.
We will denote by H n (c) the set of measures on T n M with boundary c, and
we will refer to these measures as being holonomic with boundary c.

Let X = T n M ⊕ T ∗ M , and denote by π1 and π2 the canonical projections

X ❋
② ❋❋
π1 ②② ❋❋π2
②②② ❋❋
|②
② ❋"
T nM T ∗M .

2
We will denote a point in X by (x, v1 , v2 , . . . , vn , t) for x ∈ M , vi ∈ Tx M ,
t ∈ Tx∗ M .
Definition 2. We define the set S of slices to comprise those compactly-
supported Radon measures ν on X such that their induced (n − 1)-dimensional
currents Tν ,
Z
hTν , ωi = (ω ∧ t)x (v1 , v2 , . . . , vn ) dν(x, v1 , . . . , vn , t), Ωn−1 (M ),
X

have null boundary ∂Tν = 0.

Note that the measure π1∗ ν on T n M induces a current of dimension n on
M . As should become clear soon, the t factor is there to perform the transition
between the (n − 1)-dimensional slice and the n-dimensional current it is a slice
of.
Let Sn be the group of permutations on n elements. It acts son T n M
by permuting the vectors. For example, on T 3 N , the cycle (1, 2, 3) acts on
(x, v1 , v2 , v3 ) by
(1, 2, 3)(x, v1 , v2 , v3 ) = (x, v2 , v3 , v1 ).
We can thus form the bundle T n M/Sn in which the order of the vectors does
not matter. Let π̃ be the projection X → (T n M/Sn ) ⊕ T ∗ M .
Definition 3. For T > 0, a curve of slices (on [0, T ]) is a family γ = (νt )t∈[0,T ] ,
such that
C1. νt ∈ S for all t ∈ [0, T ],
C2. t 7→ π̃∗ νt is continuous in the weak* topology, and
C3. for t ∈ [0, T ], the current Tγ|[0,t] defined for ω ∈ Ωn (M ) by
Z tZ
hTγ|[0,t] , ωi = ωx (v1 , . . . , vn ) d(π1∗ νs )(x, v1 , . . . , vn ) ds
0 T nM
Z tZ
= ωx (v1 , . . . , vn ) dνs (x, v1 , . . . , vn , t) ds,
0 T nM

has boundary
∂Tγ|[0,t] = Tνt − Tν0 .
We say that the curve γ starts at ν0 and ends at νT . The mass M(γ) of the
curve γ is X
M(γ) = sup sup |hTγ|[ti ,ti+1 ] , ωi|,
kωk61 0=t0 <t1 <···<tk =T i
where the first supremum is taken over all ω ∈ Ωn (M ) with comass at most 1,
and the second supremum is taken over all finite partitions of [0, T ].
The flat distance between two slices ν1 and ν2 in S is defined to be
dist(ν1 , ν2 ) = inf M(γ)
γ

where the infimum is taken over all curves γ starting at ν1 and ending at ν2 .
To a curve of slices γ = (νt )t∈[0,T ] , we assign its associated holonomic mea-
sure µγ , to be defined by
Z T
µγ = π1∗ νt dt.
0

3
Remark 4. Item C2 is appropriate when the minimizers are expected to be of
class C 1 ; however, it is not used in the proof of Theorem 6, and the theory can
be developed replacing it with a reasonable alternative, like the continuity of
t 7→ Tνt with respect to the flat distance.
Example 5. Let S 1 = R/Z be the circle. Assume that d = dim M > 2, let
ϕ : [0, 1] × S 1 → M be an immersion of the cylinder, and let t : [0, T ] × S 1 → R
be a function that will have the role of “time,” and which we assume to be a C 1
function with nowhere-vanishing differential dt and such that for t ∈ [0, T ], the
set t−1 (t) is homeomorphic to a circle R/Z. In particular, t must be constant
on the boundary circles {0} × S 1 and {1} × S 1 , and it must be equal to 0 on
one of them and to T on the other one.
Fix t ∈ [0, T ]. Parameterize t−1 (t) the circle isometrically using a function
ψ : [0, b) → [0, 1] × S 1
with b equal to the arclength of the circle t−1 (t), so that t◦ψ(s) = t for s ∈ [0, b).
With coordinates x1 and x2 in [0, 1] × S 1 , we define νt by
Z Z b
∂ϕ ∂ϕ
f dνt = f (ψ(s), (ψ(s)), (ψ(s)), t(s))ds,
X 0 ∂x1 ∂x2
where f ∈ C 0 (X ), ∂ϕ/∂xi (ψ(s)) are vectors in Tψ(s) M , and t(s) is an element
∗
of Tψ(s) M that we choose with the property that
 
  dt ∂
∂ϕ ψ(s) ∂xi
t(s) (ψ(s)) = , i = 1, 2, s ∈ [0, b).
∂xi kdtψ(s) k
Unless d = 2, this does not completely determine t, so we just pick any choice;
what matters is that, restricted to the subspace generated by the vectors ∂ϕ/∂xi ,
t is a pushforward of dt. Note that, with this definition, we are basically
transferring the information about the directions of ∂/∂x1 , ∂/∂x2 , and dt (which
all live on the cylinder) to M using ϕ.
With this setup, we have that for all ω ∈ Ω1 (M ), which is a unit vector,
Z Z b  
∂ϕ ◦ ψ
ω= ωϕ◦ψ(s) ds (1)
ϕ(t−1 (t)) 0 ∂s
Z b  
∂ϕ ∂ϕ
= (ω ∧ t)ϕ◦ψ(s) (ψ(s)), (ψ(s)) ds
∂x1 ∂x2
Z0
= ω ∧ t dνt = hTνt , ωi. (2)
T 2M

The significance of (1) is that it is the integral of ω along the image of t−1 (t)
under ϕ, in other words, it is the 1-dimensional current that would naturally
represent a slice of the immersed cylinder ϕ([0, 1] × S 1 ). With the help of the
operation ∧t, we see that it is exactly equal to Tνt . We thus have, for t ∈ [0, T ],
Z tZ
h∂Tγ|[0,t], ωi = hTγ , dωi = dω d(π1∗ νs ) ds =
0 T 2M
Z Z Z
dω = ω− ω = hTνt − Tν0 , ωi.
ϕ(t−1 ([0,t])) ϕ(t−1 (t)) ϕ(t−1 (0))

4
In other words, γ satisfies item C3 in Definition 3. Since it is clear that it also
satisfies C1 and C2, we conclude that γ is indeed a curve.
Theorem 6 (Weak kam for slices). Let L : T n M → R be a C ∞ function. For
m > 0, let the function h : S × S → R be given by
Z TZ
hm (ν1 , ν2 ) = inf L(x, v1 , . . . , vn ) d(π1∗ νs )(x, v1 , . . . , vn )
γ 0 T nM

where the infimum is taken over all T > 0 and all curves γ = (νt )t∈[0,T ] starting
at ν1 and ending at ν2 with mass M(γ) = m. Assume that h satisfies
i. (Lipschitzity) there is some P > 0 such that

hdist(ν1 ,ν2 ) (ν1 , ν2 ) 6 P dist(ν1 , ν2 )

for ν1 , ν2 ∈ S with dist(ν1 , ν2 ) < +∞,

ii. (Superlinearity) for every K0 > 0 there is K1 > 0 such that for all ν1 and
ν2 in S with dist(ν1 , ν2 ) < +∞ and all m > 0, we have

K0 dist(ν1 , ν2 ) − K1 m 6 hm (ν1 , ν2 ).

Then there are a Lipschitz function u : S → R and a number c0 ∈ R such that,

for all m > 0,
u(ν) = inf u(ν̃) + hm (ν̃, ν) + c0 m.
ν̃∈S

Proof. This follows immediately from Theorem 13 in Section 4.

Remark 7. Just as an exact form dω ∈ Ωn (M ) induces a continuous function
ν 7→ hTν , ωi on the space of slices S , one may ask whether u ∈ Lip(S ) descends
to one or several exact forms on M , at least when restricted to certain subsets
of S . This question is addressed in Corollary 11 in Section 3.

3 Characterization of minimizable Lagrangians

In this section we define holonomic measures and we characterize the Lagrangian
actions minimizable by them in Theorem 8. We then connect weak kam solu-
tions with exact forms in Corollary 11.
We denote by C ∞ (T n M ) the space of infinitely-differentiable functions on
the bundle T n M , and we let E ′ (T n M ) denote the space of compactly-supported
distributions on T n M , which is dual to C ∞ (T n M ). The set E ′ (T n M ) contains,
in particular, all compactly-supported, Radon measures µ on T n M . Each such
measure µ defines an n dimensional current Tµ by integration, considering forms
ω ∈ Ωn (M ) as a function on T n M :
Z
hTµ , ωi = ω dµ, ω ∈ Ωn (M ).
T nM

Let L : T n M → R be a Borel-measurable function. The action AL of L is

defined by Z
AL (µ) = L dµ, µ ∈ H n (c).
T nM

5
The measure µ ∈ H n (c) minimizes the action AL if

AL (µ) 6 AL (ν) for all ν ∈ H n (c).

The following result can be obtained in essentially the same way as [5, The-
orem 3]; we include a full proof for completeness.
Let E be a complete, sequential, locally-convex topological vector space of
Borel measurable functions on T n M that contains C ∞ (T n M ) as a subspace.
Assume that every element of H n (M ) induces a continuous linear functional
on E, i.e., H n (M ) ⊆ E ∗ , and that the topology of C ∞ (T n M ) is finer than the
one this subspace inherits from E, or, in other words, that every open set in the
inherited topology is an open set in the topology induced by the seminorms
X  
|f |K,k = sup ∂ I f (x) ,
x∈K
|I|6k

where k > 0, K ⊂ T n M is compact,P and the sum is taken over all multi-indices
I = (i1 , . . . , in ) with |I| = i
r r 6 k. This assumption implies that every
continuous linear functional ϑ ∈ E ∗ defines a compactly-supported distribu-
tion when restricted to T n M . For example, E can be the space C k (T n M ),
k ∈ [0, +∞], with the topology of uniform convergence on compact sets of the
derivatives of order 6 k.
The following is a multi-dimensional version of [5, Theorem 3].
Theorem 8. Let c be a normal current on M that is a boundary, so that
H n (c) 6= ∅.
If L is an element of E such that its action functional AL reaches its min-
imum within H n (c) at some measure µ, then there exist differential forms
ω1 , ω2 , . . . in Ωn−1 (M ), and nonnegative functions g1 , g2 , . . . in E such that
Z
lim hc, ωi i = 0, lim gi dµ = 0,
i→+∞ i→+∞ T nM

and Z
1
L= L dµ + lim (dωi + gi ),
µ(T n M ) T nM i→+∞

where the limit is taken in E.

Remark 9. In many cases, like in the situation of Corollary 11 below, using
the version of the Arzelà–Ascoli theorem given in Lemma 12, one can extract a
Lipschitz limit ω of the forms ωi . It then satisfies

dω − L 6 c0 ,
R
for c0 = − L dµ/µ(T n M ), with equality µ-almost everywhere. Here one can
think of H(dω) = dω − L as the Hamiltonian, so that we are looking at a
sort of Hamilton-Jacobi equation. In this sense, we can say that ω is a critical
subsolution of the Hamilton-Jacobi equation; cf. [2].
In order to prove the theorem, we need

6
Lemma 10. In the setting of Theorem 8, let
Z
Q = {L ∈ E : L dµ > 0 for all µ ∈ H n (c)},
T nM
R = {L ∈ E : L > dω for some ω ∈ Ωn−1 (M ) with hc, ωi > 0}.

Then we have Q = R in E.
Proof of Lemma 10. For a set A ⊂ E, we will denote by A′ ⊆ E ∗ the dual
cone to A, that is, the set of all continuous linear functionals θ ∈ E ∗ such that
θ(a) > 0 for all a ∈ A. It is a consequence of the Hahn-Banach Separation
Theorem that, if two cones A and B satisfy A′ = B ′ , then A = B.
We will prove that Q′ = R>0 H n (M ) = R′ , which implies that their closures
coincide, R = Q = Q, thus proving the lemma.
Let q ∈ Q′ and r ∈ R′ . We know that the restrictions of q and r to C ∞ (T n M )
are compactly-supported distributions. Now, the set of all nonnegative C ∞
functions on T n M is obviously a subset of Q, and it is also a subset of R
because the null function is itself an exact differential form of order n on M (it
is the differential of itself). Thus by [4, §6.22] both q and r must be positive,
compactly-supported, Radon measures on T n M . R
Let us prove that q has boundary c. Take ω ∈ Ωn−1 (M ). Then dω dµ =
hc, ωi for µ ∈ H n (c), so dω ∈ Q if hc, ωi > 0. Thus h∂q, ωi = hq, dωi > 0 for all
ω ∈ Ωn−1 (M ) with hc, ωi > 0, which implies that ∂q = c by the coincidence of
the dual cones.
The measure r also has boundary c because hc, ωi > 0.
Proof of Theorem 8. Consider the function
Z
1
L0 = L − L dµ.
µ(T n M ) T nM

It belongs to the set Q in the statement of Lemma 10, and by the lemma it also
belongs to R. The sequentiality of E implies that the topological closure equals
the sequential closure, so there exists a sequence of functions Li = gi + dωi ∈ E,
i = 1, 2, . . . , converging to L0 and such that gi ∈ E, ωi ∈ Ωn−1 (M ), gi > 0, and
hc, ωi i > 0. Since we also have that
Z Z Z
0= L0 dµ = lim Li dµ = lim gi dµ,
T nM i T nM i T nM

where the last equality is true because µ is closed, and since gi > 0, we have that
the limits of the integrals of gi and dωi vanish, which proves the theorem.
Corollary 11. Let L : T n M → R be a C 0 function satisfying the hypotheses
of Theorem 6. Assume that there is a family {γi }i∈I of curves of slices each of
which minimizes the action AL within the class of curves that share its beginning
and ending slices, and such that the support of the curves in the family covers
almost all of M , that is, such that γi = (νsi )s∈[0,Ti ] for i ∈ I, then the complement
of the set [ [
πM (supp νsi ) ⊆ M
i∈I s∈[0,Ti ]

has Lebesgue measure zero on M .

7
 Moreover, assume that the curves γi minimize AL simultaneously, in the
sense that every convex combination of the associated holonomic measures µγi
minimizes AL with respect to all positive, compactly-supported, Radon measures
with the same boundary.
Then the function u in the conclusion of Theorem 6 corresponds, up to a
constant c0 ∈ R, to a Lipschitz form ω ∈ Ωn−1 (M ), and for all i ∈ I, t ∈ [0, Ti ]
we have
hTνti , ωi = u(νti ) + c0 .
Proof. Form a convex combination of all the the associated holonomic measures
µγi using a probability measure supported throughout I. Apply Theorem 8
to obtain forms {ωj }∞
j=1 ⊂ Ω
n−1
as in the statement of that result, and then
apply also Lemma 12 to obtain a subsequence of the forms ωj converging to a
Lipschitz form.
Lemma 12 (Arzelà–Ascoli for sections of a vector bundle). Let α1 , α2 , . . . be
a sequence of smooth, uniformly bounded as sections of a vector bundle F → M
on the compact manifold —M , and assume that their derivatives dα1 , dα2 , . . .
are also uniformly bounded. Then there is a subsequence {ij }∞ j=1 ⊂ N such that
the sequence αi1 , αi2 , . . . converges uniformly to a Lipschitz section α of F .
Proof. Take a finite set of sections β1 , β2 , . . . , βN of F such that, for each x ∈ M ,
the set {β1 (x), . . . , βN (x)} is a basis for the fiber of F at x. Express each αi as
PN
αi (x) = j=1 φji (x)βN , for some smooth functions φji . Note that the uniform
boundedness of dαi implies the uniform boundedness of the derivatives dφji .
Apply the classical Arzelà–Ascoli result for Lipschitz functions to the sequences
{φji }∞
i=1 , j = 1, . . . , N , successively so as to obtain a subsequence for which all
N sequences converge simultaneously P to Lipschitz functions φj , which gives the
statement of the lemma with α(x) = N j
j=1 φ (x)βj .

4 Weak KAM machinery for noncompact met-

ric spaces
The purpose of this section is to present and prove Theorem 13, which is an
abstraction of some of the material in [3].
Theorem 13. Let X be a metric space that can be covered by countably-many
compact sets (i.e, it is a σ-compact space). Assume that for each m > 0 there
is a function hm : X × X → R satisfying the following hypotheses:
1. hm+m′ (x, y) = inf z∈X hm (x, z) + hm′ (z, y),
2. there is some P > 0 such that

hdist(x,y) (x, y) 6 P dist(x, y)

for all x, y ∈ X, and

3. for every K0 > 0 there is K1 > 0 such that for all x, y ∈ X, m > 0, we
have
K0 dist(x, y) − K1 m 6 hm (x, y).

8
For m > 0, define maps ϕm that operate on functions on X by

ϕm u(x) = inf u(y) + hm (y, x).

y∈X

Then there are a Lipschitz function u : X → R and a number c0 ∈ R such that,

for all m > 0,
u = ϕm u + c0 m.
To prove Theorem 13, we need an auxiliary lemma and a definition. Let
u ∈ C 0 (X). For c ∈ R, we say that u is c-dominated if, for every m > 0 and
every γ ∈ Γm ,
u(y) − u(x) 6 hm (x, y) + cm.
We denote H(c) the set of c-dominated functions.
Lemma 14. Under the assumptions of Lemma 13, we have:
i. ϕm ◦ ϕm′ = ϕm+m′ .
ii. The functions u ∈ H(c) are Lipschitz with constant c + P .
iii. The set H(c) is nonempty for c large enough.
iv. ϕm (H(c)) ⊆ H(c).
Proof of Lemma 14. To prove the first assertion, we observe that

ϕm+m′ (x) = inf u(y) + hm+m′ (y, x)

y∈X
 
= inf u(y) + inf hm′ (y, z) + hm (z, x)
y∈X z∈X

= inf inf [u(y) + hm′ (y, z) + hm (z, x)]

y∈X z∈X

= inf inf [u(y) + hm′ (y, z) + hm (z, x)]

z∈X y∈X
 
= inf inf [u(y) + hm′ (y, z)] + hm (z, x)
z∈X y∈X

= inf ϕm′ u(z) + hm (z, x) = ϕm [ϕm′ u].

z∈X

The second assertion follows immediately from Hypothesis 2 in Theorem 13.

For the third assertion note that from Hypothesis 3, if u is Lipschitz with
Lipschitz constant k0 , then there is some k1 > 0 such that

u(y) − u(x) 6 k0 dist(x, y) 6 hm (x, y) + k1 m.

So u ∈ H(k1 ).
For the fourth assertion, observe first that u ∈ H(c) if, and only if,

u(x) 6 u(y) + hm (x, y) + cm

which is true if, and only if, (taking the infimum on the right hand side)

u(x) 6 ϕm u(x) + cm. (3)

9
Moreover, since by the definition of ϕm we clearly have that if u 6 v then
ϕm u 6 ϕm v, we see that if applying ϕm to both sides of (3) and using that
ϕm (u + k) = ϕm u + k for k ∈ R, we get
ϕm u(x) 6 ϕm ϕm u(x) + cm,
whence ϕm u ∈ H(c) again.
Proof of Theorem 13. The proof is essentially the same as that in [3, Section 4].
We denote by C b 0 (X) the quotient of the vector space C 0 (X) by its subspace
of constant functions. If qb: C 0 (X) → C b0 (X) is the quotient map, then since
ϕm (u + k) = k + ϕm (u) for k ∈ R, the semigroup ϕm (cf. item (i) in Lemma
14) induces a semigroup on C b0 (X) that we denote ϕ bm .
b 0
The topology on C (M ) is the quotient of the compact open topology (i.e.,
the topology of uniform convergence on compact sets). With this topology, the
space Cb 0 (X) becomes a locally convex topological vector space.
Denote by H(c)b the image qb(H(c)) in C b 0 (X). The subset H(c)
b b 0 (X)
of C
is convex and compact. The convexity of H(c) b follows from that of H(c). To
b 0
prove that H(c) is compact, we introduce Cx (X) the set of continuous functions
X → R vanishing at some fixed x ∈ X. The map qb induces a homeomorphism
from Cx0 (X) onto C b0 (X). Since H(c) is stable under addition of constants, its
b
image H(c) is also the image under qb of the intersection Hx (c) = H(c) ∩ Cx0 (X).
The subset Hx (c) is closed in C 0 (M ) for the compact-open topology. Moreover,
it consists of functions which all vanish at x and are (c+P )-Lipschitzian (because
of item (ii) in Lemma 14). It follows from the Arzelà-Ascoli theorem (and a
diagonal argument) that Hx (c) is a compact set, hence its image H(c) b by qb is
also compact. The restriction of qb to Hx (c) induces a homeomorphism onto
b
H(c).
As a first consequence we conclude that if
c(h) = inf{c ∈ R|H(c) 6= 0}
T
b
then c>c(h) H(c) 6= ∅ as the intersection of a decreasing family of compact
hT that H(c(h))
nonempty subsets (cf. item (iii) in Lemma 14). It follows i is also
nonempty because it contains the nonempty subset qb−1 b
H(c) .
c>c(h)
It is obvious that ϕ q (u)) = qb[ϕm u − ϕm u(x)] for u ∈ Hx (c). Since the
bm (b
map
[0, +∞) × Hx (c) → Hx (c)
(m, u) 7→ ϕm (u) − ϕm (u)(x)

is continuous, we conclude that ϕ bt induces a continuous semigroup of H(c) b into

itself (cf. item (iv) in Lemma 14). Since this last subset is a nonempty convex
compact subset of a locally convex topological vector space C b 0 (X), we can apply
the Schauder-Tychonoff theorem [1, pages 414–415] to conclude that ϕ bt has a
fixed point u b
b in H(c) if H(c) 6= ∅, that is, for every c > c(h).
If we let u ∈ H(c) be such that qbu = u b, then the fact that u b is a fixed point
for ϕbm means that ϕt u = u + a(m). Using that ϕt is a semigroup, we get that
a(m) = a(1)m. The equality u = ϕt u − a(1)m shows that for all x, y ∈ X
u(y) − u(x) 6 hm (x, y) + a(1)m.

10
Hence −a(1) > c(h). Since u ∈ H(c(h)), we must have u 6 ϕm u + c(h)m,
which gives ϕm u − a(1)m 6 ϕm u + c(h)m for all m > 0, and −a(1) 6 c(h). We
conclude that −a(1) = c(h).

References
[1] James Dugundji. Topology. Allyn and Bacon, Inc., Boston, Mass.-London-
Sydney, 1978. Reprinting of the 1966 original, Allyn and Bacon Series in
Advanced Mathematics.
[2] Albert Fathi. Weak KAM theorem in lagrangian dynamics. Preliminary
Version Number 10, June 2008.
[3] Albert Fathi and Ezequiel Maderna. Weak KAM theorem on non compact
manifolds. NoDEA Nonlinear Differential Equations Appl., 14(1-2):1–27,
2007.
[4] Elliott H. Lieb and Michael Loss. Analysis, volume 14 of Graduate Studies
in Mathematics. American Mathematical Society, Providence, RI, second
edition, 2001.
[5] Rodolfo Rios-Zertuche. Characterization of minimizable Lagrangian action
functionals and a dual Mather theorem. Preprint. arXiv:arXiv:1810.03433
[math.AP].

11