You are on page 1of 32

ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS

ELY KERMAN AND YUANPU LIANG

Abstract. In this paper we settle three basic questions concerning the Gutt-Hutchings
capacities from [9] which are conjecturally equal to the Ekeland-Hofer capacities from [7, 8].
Our primary result settles a version of the recognition question from [4], in the negative.
We prove that the Gutt-Hutchings capacities together with the volume, do not constitute
arXiv:2109.01792v1 [math.SG] 4 Sep 2021

a complete set of symplectic invariants for star-shaped domains with smooth boundary.
In particular, we construct a smooth family of such domains, all of which have the same
Gutt-Hutchings capacities and volume, but no two of which are symplectomorphic.
We also establish two independence properties of the Gutt-Hutchings capacities. We
prove that, even for star-shaped domains with smooth boundaries, these capacities are
independent from the volume by constructing a family of star-shaped domains with smooth
boundaries, whose Gutt-Hutchings capacities all agree but whose volumes all differ. We also
prove that the capacities are mutually independent by constructing, for any j ∈ N, a family
of star-shaped domains, with smooth boundary and the same volume, whose capacities are
all equal but the j th .
The constructions underlying these results are not exotic. They are convex and concave
toric domains as defined in [9], where the authors also establish beautiful formulae for their
capacities. A key to the progress made here is a significant simplification of these formulae
under an additional symmetry assumption. This simplification allows us to identify new
blind spots of the Gutt-Hutchings capacities which are used to construct the desired exam-
ples. This simplification also yields explicit computations of the Gutt-Hutchings capacities
in many new examples.

1. Introduction
In the papers [7] and [8], Ekeland and Hofer introduce the formal notion of a symplectic
capacity and construct a sequence of capacities, {cEH 2n
k }k∈N , for subsets of (R , ω). These rich
symplectic invariants are defined in terms of the closed orbits of autonomous Hamiltonian
flows. They are difficult to compute and their values are still only known completely for
model subsets like symplectic ellipsoids and polydisks. In [9], Gutt and Hutchings use S 1 -
equivariant Floer theory to construct another sequence of symplectic capacities, {ck }k∈N ,
for star-shaped domains in (R2n , ω).1) These are also defined in terms of closed orbits of
Hamiltonian flows and are conjectured to be equal to the Ekeland-Hofer capacities. In
[9], the authors also derive combinatorial formulae for their capacities for both convex and
concave toric domains. Here we establish a significant simplification of these formulae in
the presence of an additional symmetry. This simplification reveals new blind spots of the
capacities which allow us to settle several basic questions concerning their properties.

Date: September 7, 2021.


The first named author is supported by a grant from the Simons Foundation.
1)Adopting the convention of [9], a domain here will refer to the closure of an open subset of a Euclidean
space.
1
2 ELY KERMAN AND YUANPU LIANG

1.1. Results. Motivation for the questions addressed in this paper can already be found in
the simplest nontrivial computations of the Gutt-Hutchings capacities. Fix a number a > 1.
The kth capacity of the symplectic ellipsoid
2
 
2 2 π|z2 |
E(1, a) = (z1 , z2 ) ∈ C | π|z1 | + ≤1
a
is the kth element of the sequence obtained by ordering the set {N ∪ aN} in nonincreasing
order with repetitions. In meta-code,
(1) ck (E(1, a)) = (Sort[{N ∪ aN}]) [[k]].
The kth capacity of the polydisk
P (1, a) = (z1 , z2 ) ∈ C2 | π|z1 |2 ≤ 1, π|z2 |2 ≤ a .


is
(2) ck (P (1, a)) = k.
Formulas (1) and (2) are established in Section 2 of [9], and agree with the corresponding
formulas for the Ekeland-Hofer capacities from Section III of [8].

How do the ck develop blind spots at corners?


It follows from (1) that the collection of capacities {ck (E(1, a))} sees the defining parameter
a. In contrast, it follows from (2) that for P (1, a) the parameter a is invisible to the capacities.
It hides from them in the corners of the boundary of P (1, a). In Section 2, we investigate the
formation of these well-known blind spots of capacities at corners by analyzing the capacities
of the domains
p
π|z2 |2
  
2 2 p
(3) Ep (1, a) = (z1 , z2 ) ∈ C | (π|z1 | ) + ≤1 .
a
As p goes from 1 to ∞, these domains connect the ellipsoid E(1, p) to the polydisk P (1, a).
By refining the formulas from [9], we give an explicit description of the process by which
the capacities ck (Ep (1, a)) loose sight of a along the way. In short, for a fixed k this blind
spot develops in an instant. Above a specific value of p the capacity ck (Ep (1, a)) ceases to
depend on a (see Lemma 2.2). On the other hand, for each p ∈ [1, ∞) there are infinitely
many ck (Ep (1, a)) which depend nontrivially on a (see Lemma 2.3). In other words, the set
of capacities {ck (Ep (1, a))} only looses sight of a at infinity.

Can the ck detect the volume of domains with smooth boundaries?


It is clear from formula (2) that the Gutt-Hutchings capacities do not see volume in general
and, even worse, can fail to detect symplectic factors in symplectic product manifolds.2) One
can preclude product domains by restricting to the case of domains with smooth boundary.
In this setting, the analysis of the capacities ck (Ep (1, a)) in Section 2 (in particular Lemma
2.3) suggests that it might be possible for the set of capacities to detect volume (which is
determined by a in that setting). We show that this is a false hope.
Theorem 1.1. There is a smooth family Vδ of toric star-shaped domains in R4 with smooth
boundary such that δ 7→ ck (Vδ ) is constant for all k and volume(Vδ ) = volume(V0 ) + δ.
2)For c1 this also occurs, over a limited range, in locally trivial symplectic bundles, [11], [13].
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 3

Are the ck mutually independent?


Within the symplectic category of ellipsoids or polydisks there are no variations which
vary one capacity, or even finitely many capacities, and leave the others fixed. It is not
immediately clear whether this interdependence is a general feature of the Gutt-Hutchings
capacities or an exception corresponding to the simplicity of the closed characteristics on the
boundaries of ellipsoids and polydisks. For a generic star-shaped domain U with a smooth
boundary, ∂U , there are infinitely many geometrically distinct closed characteristics on ∂U
and they are all nondegenerate. In this generic setting, each ck (U ) is equal to a multiple
of the symplectic action of one of the closed characteristics, [9]. It might be that the ck (U )
always correspond to multiples of the actions of only finitely many characteristics, and that
any variation of U that alters the actions of one of these characteristics must then alter
infinitely many capacities. Our second result shows that this is not the case.
Theorem 1.2. For every j ∈ N, there is a smooth family Vδj of toric star-shaped domains
in R4 with smooth boundary such that δ 7→ ck (Vδj ) is constant for all k 6= j, and cj (Vδj ) =
cj (V0j ) + δ.

If U ⊂ R2n is a star-shaped domain with smooth boundary, do the capacities ck (U )


together with the volume of U determine it up to symplectomorphism?
This is a version of the Recognition Question from [4] (see Question 2 in §3.6 of [4]).
Without the assumption that the boundary is smooth, the answer is known to be negative.
For example, it follows from [9] that ck (P (1, 2, 6)) = ck (P (1, 3, 4)) = k for all k ∈ N. These
polydisks also have the same volume. However, the rigidity of closed symplectic polydisks,
established by L. Bates in [2] 3), implies that they are not symplectomorphic.
Here we prove that the answer is still negative even under the additional assumption of
smooth boundaries. Our main theorem is the following.
Theorem 1.3. There is a smooth family Vδ of toric star-shaped domains in R4 with smooth
boundary all of which have the same Gutt-Hutching capacities and volume, but no two of
which are symplectomorphic.
These domains are distinguished using the ECH capacities constructed by Hutchings in
[10], using the formulas for them established in [5].
1.2. Methods and Ideas. The examples underlying the proofs of Theorems 1.1 , 1.2 and
1.3 are all either convex or concave toric domains as defined by Gutt and Hutchings in
[9]. The key tool developed here is a significant simplification of the combinatorial formulas
from [9] for the capacities of such regions, which holds in the presence of addition symmetry.
As described below, these simplifications reveal new blind spots of the capacities which are
exploited to construct the relevant examples. They also reveal some new representation
patterns for the capacities, and allow for new explicit computations.
Symmetry and collapse. We first recall the formulas for the capacities of convex and
concave toric domains from [9]. Let µ: Cn = R2n → Rn≥0 be the standard moment map
defined by
µ(z1 , . . . , zn ) = (π|z1 |2 , . . . , π|zn |2 ).
3)It is still not known whether this rigidity holds for open polydisks.
4 ELY KERMAN AND YUANPU LIANG

To a domain Ω in Rn≥0 we associate the toric domain XΩ = µ−1 (Ω) in R2n . Following, [9],
the domain XΩ is said to be convex if
Ω̂ = {(x1 , . . . , xn ) ∈ Rn | (|x1 |, . . . , |xn |) ∈ Ω}
is compact and convex. The domain XΩ is said to be concave if Ω is compact and its
complement, in Rn≥0 , is convex.
Theorem 1.4 ([9], Theorem 1.6 and Theorem 1.14). If XΩ ⊂ R2n is convex, then for any k
in N we have
( )
n
X
ck (XΩ ) = min k(v1 , . . . vn )kΩ v1 , . . . , vn ∈ {0} ∪ N, vi = k


i=1

where
k(v1 , . . . vn )kΩ = max {hv, wi | w ∈ Ω} .
If XΩ ⊂ R2n is concave, then for any k in N we have
( )
n
X
ck (XΩ ) = max [(v1 , . . . vn )]Ω v1 , . . . , vn ∈ N, vi = k + n − 1


i=1

where
[(v1 , . . . vn )]Ω = min {hv, wi | w ∈ Ω} .
Remark 1.5. Given a domain Ω ⊂ Rn≥0 with topological boundary ∂Ω, let ∂¯+ Ω be the
closure of {w ∈ ∂Ω | w ∈ Rn>0 }. It follows from the definitions above that in the formulas for
¯
k(v1 , . . . vn )kΩ with max hv, wi | w ∈ ∂+ Ω and [(v1 , . . . vn )]Ω with
ck (XΩ ) one can replace
¯
min hv, wi | w ∈ ∂+ Ω .
Remark 1.6. In principle, to compute ck (XΩ ) with the formulae of Theorem
 1.4, one must
analyze k+n−1
n−1
optimization problems on Ω in the convex case, and k+n−2
n−1
such problems
4)
in the concave case.
We say that Ω ⊂ Rn≥0 is symmetric if
(x1 , . . . , xn ) ∈ Ω =⇒ (xσ(1) , . . . , xσ(n) ) ∈ Ω
for any permutation σ ∈ Sn . The following result asserts that for symmetric convex domains
Ω the formula for ck (XΩ ) from [9] collapses to a single optimization problem on ∂¯+ Ω.
Theorem 1.7. If Ω ⊂ Rn≥0 is symmetric and XΩ is convex, then
ck (XΩ ) = max hV (k, n), wi | w ∈ ∂¯+ Ω .


where  
 k     
k k
 
k 
V (k, n) =  ,..., , ,..., .

 n n n n 
| {z }
(k mod n)−times

A similar collapse occurs for symmetric concave domains.


4)These are the number of weak compositions of k, and the number of compositions of k+n−1, respectively.
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 5

Theorem 1.8. If Ω ⊂ Rn≥0 is symmetric and XΩ is concave, then


ck (XΩ ) = min hV̌ (k, n), wi | w ∈ ∂¯+ Ω


where
 
 k + n − 1  
k+n−1
 
k+n−1
  
k+n−1 
V̌ (k, n) =  ,..., , ,..., .
 
 n n n n 
| {z }
(k+n−1 mod n)−times

New blind spots. The following simple case demonstrates how Propositions 1.7 and 1.8
can be used to reveal new and useful blind spots of the Gutt-Hutching’s capacities. Let
f : [0, 1] → R≥0 be a continuous strictly decreasing function such that f (0) = 1 and f (1) = 0.
Let Ωf be the domain in R2≥0 that is bounded by the axes and the graph of f and set
Xf = µ−1 (Ωf ). Suppose further that f is smooth on [0, 1) and satisfies f 0 (0) = 0, f 00 < 0
and f −1 = f . Then Xf is a (strictly) convex toric domain in R4 which is symmetric. In this
case, Theorem 1.7 implies that
ck (XΩ ) = max {hV (k, 2), (x, f (x))i | x ∈ [0, 1]} .
A simple argument, see Proposition 4.2 and Corollary 5.1 below, implies that for even values
of k this reduces to
ck (Xf ) = kx(f ),
where x(f ) is the unique fixed point of f . For odd values of k it reduces to
k−1 k+1
ck (Xf ) = xk + f (xk )
2 2
where xk is the unique solution of
k−1
(4) f0 = − .
k+1
The important point here is that the capacities ck (Xf ) are all determined by the
1−jet of f at x(f ) and at the sequence of points xk , for k odd, that converges
monotonically to x(f ). This is illustrated in Example 1.9 and Figure 1, below.
The blind spots referred to in the title of this paper are epitomized by the values of f
away from the points of the sequence xk → x(f ) and, by symmetry, its mirror sequence
f (xk ) → x(f ). By altering f away from all these points, in a manner which preserves
its crucial properties, one can change the volume while keeping the capacities fixed. By
altering f near one of the points xj and its mirror image f (xj ), in a manner which preserves
the function’s crucial properties, one can vary the jth capacity while leaving the others
unchanged. One can also alter f away from all these points, in a manner which preserves its
crucial properties, and deform the region, while keeping its volume fixed, with the goal of
changing its symplectomorphism class. These are the simple ideas that underly the proofs
of Theorems 1.1, 1.2, and 1.3, respectively. 5)

5)These ideas were motivated by the examples that appear in the authors’ previous work, [12].
6 ELY KERMAN AND YUANPU LIANG

Example 1.9. For the function f2 (x) = 1 − x2 , the capacities ck (Xf2 ) are determined by
the values of f2 on the sequence of points
k−1
xk = √ , for odd k
2k 2 + 2
and its limit x(f2 ) = √12 . They are given by the formula
k
 √2
 , if k is even,
ck (Xf2 ) = √
 k√2 +1 , if k is odd.

2

Figure 1. A view of the infinite symplectic rib cage of Xf2 .

1.0

0.8

0.6

0.4

0.2

0.2 0.4 0.6 0.8 1.0

The red points correspond to the sequence (x2n−1 , f2 (x2n−1 )) and determine all the odd
index capacities of Xf2 . They converge to the purple point (x(f2 ), x(f2 )) which determines
all the even index capacities of Xf2 .

Infinite symplectic rib cages. The domains Xf above yield the first examples of star-
shaped domains whose capacities are represented by infinitely many geometrically distinct
closed characteristics. In the spirit of Arnold’s evocative analogy from [1], we refer to the
collection of these characteristics as an infinite symplectic rib cage.
The carriers of the odd index capacities are distinct ribs, that is distinct closed character-
istics. A representative for c1 (Xf ) is a closed characteristic with image µ−1 ((0, 1)). For each
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 7

odd k = 2n − 1 > 1 the capacity c2n−1 (Xf ) is represented by (any of) the closed character-
istics on the Lagrangian torus µ−1 ((xk , f (xk ))) ⊂ ∂Xf . These characteristics represent the
class (n − 1, n) ∈ H1 (S 1 × S 1 , Z) ' Z2 with respect to the standard basis.
For different even values, k = 2n, the capacity carriers may not be geometrically distinct.
In particular, the capacity c2n (Xf ) is represented by (any of) the closed characteristics on the
Lagrangian torus µ−1 ((x(f ), x(f ))) ⊂ ∂Xf in class (n, n). So, for example, representatives
of c4 (Xf ) cover representatives of c2 (Xf ). It is a curious fact that this covering phenomenon
corresponds to the monotone Lagrangian torus µ−1 ((x(f ), x(f ))).
One also observes infinite symplectic rib cages for concave toric domains of the form Xh
where h is convex (h00 ≥ 0), see Proposition 4.3. In that case, all the even index capacities
are represented by distinct closed characteristics.
Question 1.10. Is there a star-shaped domain U of R2n such that all the capacities ck (U )
are represented by distinct closed characteristics?

New capacity computations. Propositions 1.7 and 1.8 yield formulas for the capacities
of many interesting domains. Several of these computations are presented in Section 4. Here
we mention just the following.
Example 1.11 (The Lagrangian Bidisk). Let ΩL be the concave region in R2≥0 bounded by
the axes and the curve
        
t t t t
α(t) = 2 sin − t cos , 2 sin + (2π − t) cos , t ∈ [0, 2π].
2 2 2 2
In [15], Ramos proves that the interior of the corresponding toric domain XΩL is symplecto-
morphic to the interior of the Lagrangian bidisk
PL = (x1 + iy1 , x2 + iy2 ) ∈ C2 | x21 + x22 ≤ 1, y12 + y22 ≤ 1 .


Theorem 1.8 implies that for an odd k we have


1
ck (XΩL ) = min {h(k + 1, k + 1), α(t)i}
t∈[0,2π] 2

= 2k + 2.
For even k we have
1
ck (XΩL ) = min {h(k + 2, k), α(t)i}
t∈[0,2π] 2
  
π k+2
= (4k + 2) sin
2 k+1
Invoking Ramos’s symplectomorphism from [15], we then get the following simple formula
for the Gutt-Hutchings capacities of the Lagrangian Bidisk:

2k + 2,
 for odd k
(5) ck (PL ) =
(4k + 2) sin π k+2  , for even k.

2 k+1

Remark 1.12. The first three Ekeland-Hofer capacities of PL were computed only recently
by Baracco, Fassina and Pinton in [3].
8 ELY KERMAN AND YUANPU LIANG

Rigidity of symplectic embeddings to and from balls. The following is a simple


application of Propositions 1.7 and 1.8.
Corollary 1.13. If XΩ ⊂ R2n is a symmetric convex toric domain, then
(6) c1 (XΩ ) = max{δ | (0, . . . , 0, δ) ∈ Ω},
and for every ` ∈ N we have
(7) c`n (XΩ ) = `n max{δ | (δ, . . . , δ) ∈ Ω}.
If XΩ ⊂ R2n is a symmetric concave toric domain, then for every ` ∈ N we have
(8) cn(`−1)+1 (XΩ ) = `n max{δ | (δ, . . . , δ) ∈ Ω}.
Proof. For the convex case, Theorem 1.7 implies that for k = 1 we have
c1 (XΩ ) = max{wn | (w1 , . . . wn ) ∈ ∂¯+ Ω}
which is equivalent to (6). For k = `n, Theorem 1.8 implies
c`n (XΩ ) = max{`(w1 + · · · + wn ) | (w1 , . . . wn ) ∈ ∂¯+ Ω}
By continuity of the capacities we may assume that ∂¯+ Ω is smooth. It then follows from
the symmetry assumption and elementary constrained optimization theory, that max{`(w1 +
· · · + wn ) | (w1 , . . . wn ) ∈ ∂¯+ Ω} is realized at the intersection of ∂¯+ Ω with the super diagonal.
Equation (7), follows from this.
For the concave case, Theorem 1.8 implies that for k = n(` − 1) + 1
cn(`−1)+1 (XΩ ) = max{`(w1 + · · · + wn ) | (w1 , . . . wn ) ∈ ∂¯+ Ω}.
A similar argument to that above yields (8).

Corollary 1.13 immediately implies a rigidity result. For a symmetric convex toric domain
XΩ ⊂ R2n , the largest standard ball contained in XΩ has radius equal to
p
max{δ | (0, . . . , 0, δ) ∈ Ω}/π
and the largest standard ball containing XΩ has radius equal to
p
n max{δ | (δ, . . . , δ) ∈ Ω}/π.
Applying Corollary 1.13, it follows that one can not do better with nonstandard balls. In
particular, we have the following.
Theorem 1.14. If XΩ ⊂ R2n is a symmetric convex toric domain, then
(9) sup{a | ∃ a symplectic embedding B 2n (a) → XΩ } = max{δ | (0, . . . , 0, δ) ∈ Ω}
and
(10) inf{A | ∃ a symplectic embedding XΩ → B 2n (A)} = n max{δ | (δ, . . . , δ) ∈ Ω}.
Equation (9) also holds if XΩ is concave. For concave toric domains, equation (10) is
known to be false in general (see, for example, [14]).
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 9

1.3. Organization. Section 2 contains an analysis of how the capacities ck (Ep (1, a)) loose
sight of a as p goes to infinity. The simplified capacity formulas of Theorem 1.7 and Theorem
1.8 are established in Section 3. These formulas are then used, in Section 4, to derive explicit
formulas for the Gutt-Hutchings capacities in several new examples. The proofs of Theorems
1.1,1.2 and 1.3 are presented in Section 5. The last section contains a discussion of some
questions motivated by these results and their proofs.

2. The development of blind spots at corners


For a fixed a > 1, consider the family of domains
p
π|z2 |2
  
2 2 p
(11) Ep (1, a) = (z1 , z2 ) ∈ C | (π|z1 | ) + ≤1 .
a
As p varies from 1 to ∞ these domains connect the ellipse E(1, a) to the polydisk P (1, a).
In this section we analyze the process by which the capacities ck (Ep ((1, a)) loose sight of a
along the way.
Each Ep (1, a) is a convex toric domain of the form µ−1 (Ωfp ) where Ωfp is the region in
R2≥0 bounded by the axes and the graph of the smooth strictly concave function
1
fp (x) = a(1 − xp ) p .
Since a > 1, this domain is not symmetric. However, the formula from Theorem 1.4 can still
be significantly simplified in this setting.
Working a little more generally, let V be the set of continuous functions f : [0, 1] → R≥0
with the following properties:
(1) f (0) ≥ 1.
(2) f (1) = 0.
(3) f is smooth on [0, 1).
(4) f 0 (0) = 0.
(5) limx→1− f 0 (x) = −∞.
(6) f 00 < 0 on (0, 1).
Each f ∈ V has a unique fixed point in (0, 1) which we denote by x(f ). For k ≥ 2 and
j = 0, . . . k − 2 set  
k −j − 1 −j
Ij = , .
k−j−1 k−j
k
Let Ik−1 = (−∞, −k + 1] . We then define Jk = Jk (f ) to be the integer in [0, k − 1] that is
determined uniquely by the condition
f 0 (x(f )) ∈ IJkk .
Proposition 2.1. For f ∈ V, let Ωf be the region in R2≥0 bounded by the axes and the graph
of f . Let Xf = µ−1 (Ωf ). Then c1 (Xf ) = 1 and for k ≥ 2
(12) ck (Xf ) = min{k, F (k, f )}
where

x1 + (k − 1)f (x1 )
 if Jk = 0,
F (k, f ) = min {Jk xJk + (k − Jk )f (xJk ), (Jk + 1)xJk +1 + (k − Jk − 1)f (xJk +1 )} if 1 ≤ Jk < k − 1

(k − 1)x
k−1 + f (xk−1 ) if Jk = k − 1,
10 ELY KERMAN AND YUANPU LIANG

−`
and x` = x` (f ) is the unique solution of f 0 (x) = k−`
.
Proof. Since the compact convex domain Ωf is defined by the graph of f , the formula for
ck (Xf ) can be simplified to
 
ck (Xf ) = min max {`x + (k − `)f (x)} .
`∈{0,1,...,k} x∈[0,1]

The assertion about c1 (Xf ) follows immediately. Assuming now that k ≥ 2, consider the
partition
{0, 1, . . . , k} = {0, k} ∪ {1, . . . , k − 1} .
| {z } | {z }
Ak Bk
We then have  
min max {`x + (k − `)f (x)} = min{kf (0), k} = k.
`∈Ak x∈[0,1]

The condition that f ∈ V satifsies f 00 < 0 on (0, 1), implies that for ` in Bk the function
x 7→ `x + (k − `)f (x)
−`
has a unique critical point at the point x` defined uniquely by the equation f 0 (x) = k−`
.
Moreover, x` is a global maximum. Hence, for ` ∈ Bk we have
max {`x + (k − `)f (x)} = `x` + (k − `)f (x` ).
x∈[0,1]

Now consider
c(`) = `x` + (k − `)f (x` )
as a function of a real variable ` ∈ [1, k − 1] where x` is now the smooth function defined
−`
implicitly by the equation f 0 (x` ) = k−` . We then have
(13) c0 (`) = x` − f (x` ).
Hence, c(`) takes its minimum value at the value of ` corresponding to the fixed point x(f ).
Viewing ` as a discrete variable in {1, 2, . . . , k − 1}, the discussion above implies that if
Jk is contained in [1, k − 1), then
 
min max {`x + (k − `)f (x)} = min {Jk xJk + (k − Jk )f (xJk ), (Jk + 1)xJk +1 + (k − Jk − 1)f (xJk +1 )} .
`∈Bk x∈[0,1]

On the other hand, if Jk = 0, then (13) implies that c(`) is increasing (as a function on the
discrete set {1, 2, . . . , k − 1}) and so
 
min max {`x + (k − `)f (x)} = x1 + (k − 1)f (x1 ).
`∈Bk x∈[0,1]

Similarly, if Jk ≥ k − 1, then (13) implies that c(`) is decreasing and


 
min max {`x + (k − `)f (x)} = (k − 1)xk−1 + f (xk−1 ).
`∈Bk x∈[0,1]

Putting these observations together we arrive at the formula of Proposition 2.1. 


1
We now apply Proposition 2.1 to the function fp (x) = a(1 − xp ) defining Ep (1, a). We
p

first prove that for each k the capacity ck (Ep (1, a)) looses sight of a for all sufficiently large
p.
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 11

Lemma 2.2. For each k ∈ N there is a p(k) ∈ [1, ∞) such that ck (Ep (1, a)) = k for all
p ≥ p(k).
Proof. A simple computation yields
fp0 (x(fp )) = −ap .
For all ` ∈ [1, k − 1] we also have
 1
 p−1
`
(k−`)a
x` (p) =  p 
1
  p−1 p
`
1+ (k−`)a

and 1
p !−
  p−1 p
`
fp (x` (p)) = a 1 + .
(k − `)a
Consider a fixed k ≥ 2. For p near 1, the capacity ck (Ep (1, a)) is determined by which
interval, say Ijk1 , contains −a. If j1 = k − 1, then −ap remains in Ik−1 k
as p increases.
p k k k
Otherwise, −a moves from Ij1 to Ij1 +1 and continues in this way until it enters Ik−1 for
good. In either case, for p sufficiently large we have
F (k, fp ) = (k − 1)xk−1 + fp (xk−1 )
 p p
 p−1
p
= a p−1 + (k − 1) p−1 .
Lemma 2.2, now follows from Proposition 2.1 (see (12)) and the fact that
lim F (k, fp ) = k − 1 + a > k.
p→∞


Figure 2 illustrates the formula from Proposition 2.1 and the behavior described in Lemma
2.2 when k = 123 and a is equal to Euler’s number, e.

Figure 2. The graph of p 7→ c123 (Ep (1, e)).

120

115

110

105

100

95

2 4 6 8 10

Next we prove that a is visible to the collection of capacities, {ck (Ep (1, a))}k∈N for all
p < ∞.
12 ELY KERMAN AND YUANPU LIANG

Lemma 2.3. For each p ∈ (1, ∞) there is a k(p) ∈ N such that ck (Ep (1, a)) depends on a
for all k ≥ k(p).
Proof. Suppose that
(14) ap < k − 1.
Then −ap lies in some some Ijk for j ≤ k − 2 and
 p p
 p−1
p
F (k, fp ) = (a(k − m)) p−1 +m p−1

for some m ≥ 1. It follows from the proof of Proposition 2.1 that


F (k, fp ) ≤ (k − 1)xk−1 + fp (xk−1 )
 p p
 p−1
p
= a p−1 + (k − 1) p−1

Hence, one has F (k, fp ) < k if


 p p
p−1
ap < k p−1 − (k − 1) p−1 .
The right hand side goes to infinity as k does. With this, Proposition 2.1 implies that if k
is large enough for inequalities (14) and (2) to both hold, then
 p p
 p−1
p
ck (Ep (1, a)) = (a(k − m)) p−1 +m p−1

for some 1 ≤ m ≤ k − 1.


3. Symmetry and collapse


Here we present the proofs of Theorem 1.7 and Theorem 1.8. Recall that Ω ⊂ Rn≥0 is
symmetric if
(x1 , . . . , xn ) ∈ Ω =⇒ (xσ(1) , . . . , xσ(n) ) ∈ Ω
for any permutation σ ∈ Sn .

3.1. Proof of Theorem 1.7. We must show that if Ω ⊂ Rn≥0 is symmetric and XΩ is convex,
then
ck (XΩ ) = kV (k, n)kΩ
where  
       
 k k k k 
V (k, n) = 
 n , . . . , , , . . . , .
n n n 
| {z }
k mod n
n
We say that an n-tuple y = (y1 , . . . yn ) ∈ R is ordered if
y1 ≤ y2 ≤ . . . ≤ yn .
Since Ω is symmetric, we have
(15) k(v1 , . . . vn )kΩ = k(vσ(1) , . . . vσ(n) )kΩ
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 13

for all σ ∈ Sn . This allows us to rewrite the expression for ck (XΩ ), as


n o
~
ck (XΩ ) = min kvkΩ v ∈ C(k, n)

where ( )
Xn
~ n) =
C(k, v ∈ Zn≥0

vi = k, v is ordered .

i=1

~ n) there exists an ordered w = (w1 , . . . , wn ) ∈ Ω such that


Lemma 3.1. For every v ∈ C(k,
kvkΩ = hv, wi.
Proof. Choose w̃ = (w̃1 , . . . , w̃n ) in Ω such that kvkΩ = hv, w̃i. If w̃ is ordered we are done.
Otherwise, w̃i > w̃j for some 1 ≤ i < j ≤ n . Let τ ∈ Sn be the transposition τ = (i j), and
set w̃τ = (w̃τ (1) , . . . , w̃τ (n) ). Since Ω is symmetric w̃τ is in Ω and
hv, w̃τ i − hv, w̃i = (vi w̃j + vj w̃i ) − (vi w̃i + vj w̃j )
(16) = (vi − vj )(w̃j − w̃i )
≥ 0.
As hv, w̃i is already realizing the maximum, kvkΩ , this inequality must be an equality. Hence
kvkΩ = hv, w̃τ i. Proceeding in this manner we can continue to order w̃ and obtain the desired
w ∈ Ω. 
~ n), set
For v ∈ C(k,
∆(v) = vn (#{vi | vi = vn }) − v1 (#{vi | vi = v1 }).
Consider the transfer map T: C(k, ~ n) → C(k, ~ n) defined by

 . . . v1 , v1 + 1 . . . , vn − 1, vn , . . . vn )
( v|1 , {z if vn > v1 + 1,
} | {z }
v = (v1 , . . . v1 , . . . , vn , . . . vn ) 7→ (t−1)−times (T −1)−times
| {z } | {z } 
v
t−times T −times otherwise.
Note that the vector V (k, n) in the statement of Theorem 1.7 is the unique fixed point of T
and that ∆(T(v)) ≤ ∆(v) with equality if and only if v = V (k, n). Hence,
(17) T j (v) = V (k, n) for all sufficiently large j.
~ n) we have kT(v)kΩ ≤ kvkΩ .
Lemma 3.2. For all v ∈ C(k,
Proof. By Lemma 3.1 there is an ordered w ∈ Ω such that kT(v)kΩ = hT(v), wi. We then
have
kvkΩ −kT(v)kΩ ≥ hv, wi − hT(v), wi
(18) = (v1 wt + vn wn−T ) − ((v1 + 1)wt + (vn − 1)wn−T )
= wn−T − wt
which is nonnegative since w is ordered. 
~ n), Lemma 3.2 implies that
Since ck (XΩ ) = kvkΩ for some v in C(k,
ck (XΩ ) = kT j (v)kΩ
for any j ∈ N. It then follows from (17) that ck (XΩ ) = kV (k, n)kΩ . This completes the proof
of Theorem 1.7.
14 ELY KERMAN AND YUANPU LIANG

3.2. Proof of Theorem 1.8. For a symmetric Ω ⊂ Rn≥0 such that XΩ is concave, we must
show that
ck (XΩ ) = [V̌ (k, n)]Ω
where
 
 k + n − 1  
k+n−1
 
k+n−1
  
k+n−1 
V̌ (k, n) =  ,..., , ,..., .
 
 n n n n 
| {z }
k+n−1 mod n

As the proof is very similar to that of Theorem 1.7 it is presented in less detail. A vector
y = (y1 , . . . yn ) ∈ Rn is said to be ordered backwards if
y1 ≥ y2 ≥ . . . ≥ yn .
For symmetric concave domain XΩ we then have
n o
~ n) ,
ck (XΩ ) = max [v]Ω v ∈ C(k,

~ n) = { v ∈ Nn | Pn vi = k + n − 1, v is ordered backwards }.
where C(k, i=1

~ n) there exists an ordered w = (w1 , . . . , wn ) ∈ Ω such that


Lemma 3.3. For every v ∈ C(k,
[v]Ω = hv, wi.
Proof. Choose w̃ = (w̃1 , . . . , w̃n ) in Ω such that [v]Ω = hv, w̃i. If w̃ is ordered we are done.
Otherwise, w̃i > w̃j for some 1 ≤ i < j ≤ n . Let τ ∈ Sn be the transposition τ = (i j), and
set w̃τ = (w̃τ (1) , . . . , w̃τ (n) ). We then have
hv, w̃τ i − hv, w̃i = (vi − vj )(w̃j − w̃i )
(19)
≤0
Since hv, w̃i is already realizing the minimum, [v]Ω , this inequality must be an equality. Hence
kvkΩ = hv, w̃τ i and we can proceed in this manner to obtain the desired ordered w ∈ Ω 
Now define the backwards transfer map B: C(k, ~ n) → C(k, ~ n) by

 . . . v1 , v1 − 1 . . . , vn + 1, vn , . . . vn )
( v|1 , {z if v1 > vn + 1,
} | {z }
v = (v1 , . . . v1 , . . . , vn , . . . vn ) 7→ (t−1)−times (T −1)−times
| {z } | {z } 
v
t−times T −times otherwise.
We then have Bj (v) = V̌ (k, n) for all sufficiently large j.
~ n) we have [B(v)]Ω ≥ [v]Ω .
Lemma 3.4. For all v ∈ C(k,
Proof. By Lemma 3.3 there is an ordered w ∈ Ω such that [B(v)]Ω = hB(v), wi. Hence,
[v]Ω − [B(v)]Ω ≤ hv, wi − hB(v), wi
(20) = (v1 wt + vn wn−T ) − ((v1 − 1)wt + (vn + 1)wn−T )
= wt − wn−T
which is nonpositive since w is ordered. 
Finally, given ck (XΩ ) = [v]Ω , Lemma 3.4 implies that, for all sufficiently large j,
cK (XΩ ) = [Bj (v)]Ω = [V̌ (k, n)]Ω .
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 15

4. New capacity computations


In this section we use the formulas of Propositions 1.7 and 1.8 to derive explicit formulas
for the Gutt-Hutchings capacities in several new examples.
4.1. Graphs for n = 2. We start with simple but illuminating case when the domain Ω is
defined by the graph of a nice function of one variable. Let g: [0, λ] → R≥0 be a piecewise-
smooth nonincreasing function such that g(0) > 0 and g(λ) = 0. Denote by Ωg the domain
in R2≥0 that is bounded by the axes and the graph of g and set Xg = µ−1 (Ωg ).
By the Regular Value theorem we have the following.
Lemma 4.1. If g is smooth and g 0 is finite and bounded away from zero, then the boundary
of Xg is smooth.
4.1.1. Convex graph domains. For λ > 0, let V(λ)
b be the set of continuous functions f : [0, λ] →
R≥0 with the following properties:
(f1) f is smooth on [0, λ),
(f2) f (0) = λ,
(f3) f 0 (0) ≤ 0,
(f4) f 00 < 0 on [0, λ),
(f5) f −1 = f .
Properties (f1)-(f4) imply that the toric domain Xf is convex. Property (f5) implies that it
is also symmetric. In this setting, Theorem 1.7 yields the following.
Proposition 4.2. Suppose f is in V(λ).
b For even values of k,
ck (Xf ) = kx(f )
where x(f ) is the unique fixed point of f . For odd values of k,

k−1 k+1 k−1 0
 2 xk + 2 f (xk ), if − k+1 < f (0)

ck (Xf ) =
 k+1 λ,

otherwise,
2
where xk is defined uniquely by the equation
k−1
(21) f 0 (xk ) = − .
k+1
Proof. For even values of k, we have
 
k k
V (k, 2) = , .
2 2
Theorem 1.7 then yields
 
k k
ck (Xf ) = max x + f (x)
x∈[0,λ] 2 2
k
= max {λ, 2x(f )}
2
= kx(f ).
Here we have used (f1)-(f5) to infer that x(f ) is the unique solution of f 0 (x) = −1, and that
x(f ) > λ/2.
16 ELY KERMAN AND YUANPU LIANG

For odd values of k we have


     
k k k−1 k+1
V (k, 2) = , = ,
2 2 2 2
and Theorem 1.7 yields
 
k−1 k+1
ck (Xf ) = max x+ f (x) .
x∈[0,λ] 2 2
Set
k−1 k+1
f˜k (x) = x+ f (x)
2 2
It is straightforward to check that if
k−1
f 0 (0) > −
k+1
then we have fk (0) > 0 and, by (f5), fk (λ) < 0. Together with (f4), this implies that f˜k
˜ ˜
takes its maximum value at its unique critical point xk ∈ (0, λ) defined by
k−1
f 0 (xk ) = − .
k+1
If, on the other hand,
k−1
f 0 (0) ≤ − ,
k+1
then f˜k is strictly decreasing and its maximum value is f˜k (0) = k+1
2
λ. 

4.1.2. Concave graph domains. For λ > 0, let C(λ)


b be the set of continuous functions
h: [0, λ] → R≥0 with the following properties:
(h1) h is smooth on (0, λ],
(h2) h(λ) = 0,
(h3) h0 (λ) ≤ 0,
(h4) h00 > 0 on (0, λ],
(h5) h−1 = h.
The toric domain Xh = µ−1 (Ωh ) is concave and symmetric. As before, each h ∈ C(λ)
b has a
unique fixed point x(h) ∈ (0, 1) and
h0 (x(h)) = −1.
Theorem 1.8 implies the following.
Proposition 4.3. Suppose h is in C(λ).
b For odd values of k,
ck (Xh ) = (k + 1)x(h)
where x(h) is the unique fixed point of h. For even values of k,

k+2 k k 0
 2 x̌k + 2 h(x̌k ), if − k+2 < h (λ)

ck (Xh ) =
 k λ,

otherwise,
2
where x̌k is the number that is defined uniquely by the condition
k+2
(22) h0 (x̌k ) = − .
k
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 17

Proof. For odd values of k, we have


 
k+1 k+1
V̌ (k, 2) = , .
2 2
In this case, Theorem 1.8 together with conditions (h1)-(h5), yields
 
k+1 k+1
ck (Xh ) = min x+ h(x)
x∈[0,λ] 2 2
k+1
= min {λ, 2x(h)}
2
= (k + 1)x(h).
For even values of k we have
     
k+1 k+1 k+2 k
V̌ (k, 2) = , = ,
2 2 2 2
and Theorem 1.8 yields
 
k+2 k
ck (Xh ) = min x + h(x) .
x∈[0,λ] 2 2
Arguing as in the proof of Proposition 4.2 one can show that if
k+2
h0 (0) ≥ − ,
k
or equivalently if h0 (λ) ≤ − k+2
k
, then the function
k+2 k
x 7→ x + h(x)
2 2
has no critical points in (0, λ) and a minimum value of k2 λ. Otherwise, it has a unique critical
point x̌k ∈ (0, λ) defined by
k+2
h0 (x̌k ) = −
k
at which it takes its minimum value. 
Example 4.4. Motivated by the work of Ostrover and Ramos in [14] we apply Proposition
4.2 and Proposition 4.3 to the `p -sum of two Lagrangian disks, for 1 ≤ p ≤ ∞. These
domains are defined as follows
n p p
o
PL (p) = (x1 + iy1 , x2 + iy2 ) ∈ C2 | (x21 + x22 ) 2 + (y12 + y22 ) 2 ≤ 1 .

Proposition 4.5. For 1 ≤ p ≤ 2, we have


 2Γ(1+ 1 )2

 k Γ(1+ p2 ) for even k,
p



 k+1 2π 1


2 41/p
for odd k < q ,
ck (PL (p)) = 2
p
−1


 1
kgp (−vk ) + (k + 1)πvk for odd k ≥ q ,



2
−1


p
18 ELY KERMAN AND YUANPU LIANG

where
√1
( 12 + −v p )1/p
r
v2
Z
4
(23) gp (v) = 2 √1 (1 − rp )2/p − dr.
( 21 − 4
−v p )1/p r2

and vk is defined by
k+1
(24) gp0 (−vk ) = −π .
k
For 2 < p < ∞, we have
2Γ(1+ p1 )2



 (k + 1) Γ(1+ p2 )
for odd k,

 q

 2
p

k 2π


 1/p
24
for even k < q ,
ck (PL (p)) = 1 − p2

 q

 2
p


(k + 2)πv̌k + (k + 1)gp (−v̌k ) for even k ≥ q ,



1− 2


p

where gp is as above and v̌k is defined by


k
(25) gp0 (−v̌k ) = −π .
k+1
Proof. The crucial starting point is the following result from [14].
Theorem 4.6 (Theorem 5, [14]). The interior of PL (p) is symplectomorphic to the interior
of the toric domain Xp = µ−1 (Ωp ) ⊂ R4 where Ωp ⊂ R2≥0 is the region bounded by the axes
and the curve

1/p
(gp (−v), −2πv + gp (−v)) , for v ∈ [−(1/4) , 0]

αp (v) =
, for v ∈ [0, (1/4)1/p ]

(2πv + g (v), g (v))
p p

where gp (v) is defined as in (23).


In the process of computing the first two ECH capacities of Xp , Ostrover and Ramos also
prove the following.
Lemma 4.7 (Lemma 18, [14]). For all for p ≥ 1:
2Γ(1+ p1 )2
(i) gp (0) = Γ(1+ p2 )
and gp (1/41/p ) = 0.

(ii) gp is strictly decreasing.


r
2
(iii) lim gp0 (v)
= −π and lim =− π. gp0 (v)
v→0 v→1/4 1/p p
Moreover, for 1 < p < 2 the function gp is strictly concave and for p > 2 the function gp is
strictly convex.
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 19

Case 1: 1 ≤ p ≤ 2. In this range, the toric region Xp is convex and symmetric. Let fp be
the function whose graph corresponds to the image of αp . Then
2Γ(1 + p1 )2
x(fp ) = gp (0) = .
Γ(1 + p2 )
k−1
As well, − k+1 ≤ fp0 (0) if and only if k ≥ q1 .
2
If this holds, then we have
p
−1

xk (fp ) = gp (−vk )
where vk ∈ [0, (1/4)1/p ] is determined uniquely by the equation
k+1
(26) gp0 (−vk ) = −π
.
k
With this, the first capacity formula in Proposition 4.5 follows immediately from Proposition
4.2.
Case 2: p > 2. In this range, the toric region Xp is convcave and symmetric. Let hp be the
function whose graph corresponds to the image of αp . As before,
2Γ(1 + p1 )2
x(hp ) = gp (0) = .
Γ(1 + p2 )
q
2

Now, we have k
− k+2 ≤ h0p ( 42π
1/p ) if and only if k ≥ q .
p
In this case
1− p2

x̌k (hp ) = 2πv̌k + gp (v̌k )


where v̌k ∈ [−(1/4)1/p , 0] is determined uniquely by the equation
k
(27) gp0 (−v̌k ) = −π .
k+1
The second capacity formula in Proposition 4.5 now follows from Proposition 4.3.

4.2. Symplectic `p -sums (Graphs for n > 2). One can also consider toric domains Xg
with projections Ωg ⊂ Rn , for n > 2, that are determined by the graph of a function
g: Λ ⊂ Rn−1 → R≥0 . The corresponding capacity formulas may involve many cases. A
simple but interesting set of examples of this type are the symplectic `p -sums
Bpn = (z1 , . . . zn ) ∈ Cn | π p/2 (|z1 |p + · · · + |zn |p ) ≤ 1 ,


for p > 0. Note that Bpn = µ−1 (Ωnp ) where Ωnp ⊂ Rn≥0 is the region bounded by the coordinate
hyperplanes and the graph of the function
p/2 p/2 2
fpn (x1 , . . . , xn−1 ) = (1 − x1 − · · · − xn−1 ) p
with domain n o
p/2 p/2
Λn−1
p/2
n−1
= (x1 , . . . xn ) ∈ R≥0 | x1 + · · · + xn−1 ≤ 1 .
Each toric domain Bpn is symmetric and B2n is the standard closed unit ball.
20 ELY KERMAN AND YUANPU LIANG

Proposition 4.8. For p > 2, we have


 p
  p−2 p  p−2
k    k  p−2 p
n
(28) ck (Bp ) = (n − (k mod n)) + k mod n .
n n
For 0 < p < 2, we have
p
 0  p−2 p  p−2

k    k 0  p−2 p
n 0 0
(29) ck (Bp ) = (k mod n) + n − (k mod n) ,
n n
where k 0 = k + n − 1.
Remark 4.9. For n = 2 and p > 2, equation (28) simplifies to

k
2 , for even k


2p

(30) ck (Bp2 ) =
  p  p  p−2
 k+1 p−2 + k−1 p−2 p , for odd k.


2 2

As p → ∞, Bp2 converges to the polydisk P (1, 1) in the Hausdorff topology and equation
(30) yields
lim ck (Bp2 ) = k
p→∞
which agrees with the formula for ck (P (1, 1)) from [9]. Similarly, as p → 2+ the domains Bp2
converge to the ball E(1, 1) and (30) yields

k

2 , for even k,
2
lim ck (Bp ) =
p→1+  k+1 , for odd k.

2
which agrees with the formula for ck (E(1, 1)) from [9].
For the case n = 2 and p < 2, equation (29) simplifies to

k+1
2 , for odd k


 2p

(31) ck (Bp2 ) =
   p  p  p−2
 k p−2 + k+2 p−2 p , for even k.


2 2

As p → 2− the domains Bp2 converge to the ball E(1, 1) and (31) yields

k+1

 2 , for odd k,
2
lim ck (Bp ) =
p→1+ k

2
, for even k.
which again agrees with the formula for ck (E(1, 1)) from [9].

Proof. Case 1: p > 2. In this case Bpn is (strictly) convex and so by Theorem 1.7
( n−1 )
X p/2 p/2 2
ck (Xpn ) = max V (k, n)i xi + V (k, n)n (1 − x1 − · · · − xn−1 ) p | (x1 , . . . , xn−1 ) ∈ Λn−1
p/2 .
i=1

(For the moment it is useful to forget that we know the components V (k, n)i of V (k, n).)
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 21

The function
n−1
X p/2 p/2 2
F (x1 , . . . , xn−1 ) = V (k, n)i xi + V (k, n)n (1 − x1 − · · · − xn−1 ) p
i=1

attains it maximum value at its unique critical point which we now solve for. The equation
∂F
∂xi
= 0 is equivalent to
  p
p/2 V (k, n)i p−2  p/2 p/2

xi = 1 − x1 − · · · − xn−1 .
V (k, n)n
  p
p/2 V (k,n)i p−2
Setting yi = xi and αi = V (k,n)n we can rewrite this as
αi y1 + · · · + (αi + 1)yi + · · · + αi yn−1 = αi .
Hence the critical point of F corresponds to the unique solution of the linear system
 
~ · (~1)T ~y = α
1n−1 + α ~

where (~1)T = (1, 1, . . . , 1). Using the Sherman-Morrison formula we arrive at the expression
!
1
~y = α
~.
1 + n−1
P
i=1 α i

Hence
n−1   p ! −2
p
X V (k, n)i p−2 −2 2
xj = 1+ (V (k, n)n ) p−2 (V (k, n)j ) p−2
i=1
V (k, n)n
n
! −2
p
X p 2
= (V (k, n)i ) p−2 (V (k, n)j ) p−2 .
i=1

Evaluating F at this critical point we get


 p
  p−2 p  p−2
k    k  p−2 p
n
(32) ck (Bp ) = (n − (k mod n)) + k mod n ,
n n
as desired.
Case 2: 0 < p < 2. Here, Bpn is concave, and
( n−1 )
X p/2 p/2 2
ck (Xpn ) = min V̌ (k, n)i xi + V̌ (k, n)n (1 − x1 − · · · − xn−1 ) p | (x1 , . . . , xn−1 ) ∈ Λn−1
p/2 .
i=1

Arguing as above, it follows that the function


n−1
X p/2 p/2 2
V̌ (k, n)i xi + V̌ (k, n)n (1 − x1 − · · · − xn−1 ) p
i=1

has a unique global minimum at


n
! −2
p
X p
 p−2 2
 p−2
(33) xj = V̌ (k, n)i V̌ (k, n)j .
i=1
22 ELY KERMAN AND YUANPU LIANG

Hence
p
 0  p−2 p  p−2

k    k 0  p−2 p
(34) ck (Bpn ) = (k 0 mod n) + n − (k 0 mod n) ,
n n
where k 0 = k + n − 1.

4.3. Symmetric toric polytopes. Let P be a convex polytope in Rn≥0 with verticies
{pj }j=1,...,N . Note that P is symmetric if every permutation matrix of Rn maps the set
of vertices of P onto itself.
Proposition 4.10. If P is symmetric and the toric domain µ−1 (P ) is convex then
(35) ck (µ−1 (P )) = max{hV (k, n), pj i}.
j

Proof. This follows from Theorem 1.7 and standard linear programming. 
Remark 4.11. The simplicity of this formula for the Gutt-Hutchings (conjecturally Ekeland-
Hofer) capacities of this class of star-shaped domains in any even dimension is quite note-
worthy. Especially since any symmetric convex toric domain can be C 0 -approximated by
such a domain. The formula is used to provide models in Section 6.
Example 4.12. For r ∈ [1/2, 1], let Ωr be the symmetric convex polytope in R2≥0 with
vertices
{(0, 0), (1, 0), (0, 1), (r, r)}.
For even values of k, (35) implies that
ck (µ−1 (Ωr )) = kr.
For odd values of k, (35) implies that
 
−1 k+1
ck (µ (Ωr )) = max , kr .
2
k+1
1 k+1

This piecewise linear function of r is constant and equal to 2
for r ∈ ,
2 2k
, and then
grows linearly as kr.

5. The proofs of Theorems 1.1, 1.2, and 1.3


5.1. Preliminaries. Here we will use the following useful simplifications of Propositions 4.2
and 4.3.
Corollary 5.1. Let f be a function in V(λ)b such that f 0 (0) ∈ (−1/2, 0). Then Xf is a
(strictly) convex toric domain in R4 with a smooth boundary. For even values of k we have
(36) ck (Xf ) = kx(f ),
where x(f ) is the unique fixed point of f . The first capacity of Xf is equal to λ, and for odd
k > 1 we have
k−1 k+1
(37) ck (Xf ) = xk + f (xk )
2 2
where xk is the unique solution of
k−1
(38) f0 = − .
k+1
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 23

The sequence of points xk increases monotonically and converges to x(f ).


Corollary 5.2. Let h be a function in C(λ)
b such that h0 (λ) ∈ (−1/2, 0). Then Xh is a
4
strictly concave toric domain in R with a smooth boundary. For odd values of k,
(39) ck (Xh ) = (k + 1)x(h)
where x(h) is the unique fixed point of h. For even values of k we have
k+2 k
ck (Xh ) = x̌k + h(x̌k )
2 2
where x̌k is the unique solution of
k+2
(40) h0 = − .
k
The sequence of points x̌k increases monotonically and converges to x(h).
5.1.1. Symmetric perturbations. Let g be a function in either V(λ)
b or C(λ).
b Let β: [0, λ] → R
be a smooth function with support [a, b] contained in (0, x(g)). Then for all sufficiently small
 > 0 the restriction of the function g +β to [0, x(g)] can be extended to a unique function in
the same set V(λ)
b or C(λ)
b as g. This (symmetric) extension of g is of the form g +(β + β̃) for
a unique function β̃: [0, λ] → R with support [g(b), g(a)] ⊂ (x(g), λ). A simple computation,
using integration by parts, implies that
Z λ Z λ
(41) β= β̃.
0 0

5.2. Proof of Theorem 1.1. Let f be a function in V(1) b as in Corollary 5.1. For an odd
integer j > 1 choose a smooth bump function ηj : [0, 1] → R≥0 such that
(i) the support of ηj is contained in (xj , xj+2 ).
R1
(ii) 0 ηj = 21 .
For all δ ≥ 0 small enough the function f + δ(ηj + η̃j ) is in V(1)
b and we set Vδ = Xf +δ(ηj +η̃j ) .
Condition (i) and Corollary 5.1 imply that ck (Vδ ) = ck (V0 ) for all k ∈ N. Condition (ii) and
(41) imply that
volume(Vδ ) = volume(V0 ) + δ.
5.3. Proof of Theorem 1.2. For even values of j the desired family of domains Vδj will
be comprised of strictly concave toric domains. For odd values of j it will be comprised of
strictly convex toric domains.
5.3.1. Even Indices. Let h be a function in C(1),
b as in Corollary 5.2, with h0 (1) ∈ (−1/2, 0).
For j ∈ 2N, let βj : [0, 1] → R be a smooth function with the following three properties:
(i) βj = 2j in a neighborhood of x̌j ,
(ii) βj (x) is compactly supported in (x̌j−2 , x̌j+2 ),
R1
(iii) 0 βj = 0.
For all sufficiently small δ ≥ 0 the function h + δ(βj + β̃j ) is in C(1).
b Restricting ourselves
to such values of δ we set Vδj = Xh+δ(βj +β̃j ) . Properties (i) and (ii) imply that the solution of
k+2
(h + δ(βj + β̃j ))0 = −
k
24 ELY KERMAN AND YUANPU LIANG

is the same for all for all even k and all δ > 0. By Corollary 5.2 and (i), we then have
j+2 j 
cj (Vδj ) = x̌j + h(x̌j ) + δ(βj (x̌j ) + β̃j (x̌j ))
2 2
j
= cj (V0 ) + δ.
On the other hand, (ii) implies that ck (Vδj ) = ck (V0j ) for all even k 6= j. Similarly, (ii) implies
that x(h + δ(βj + β̃j )) = x(h) for all sufficiently small δ > 0, and so ck (Vδj ) = ck (V0j ) for all
odd k as well. Finally, it follows from (iii) and (41) that
volume(Vδj ) = volume(V0j ).

5.3.2. Odd Indices. Here the argument is entirely similar except for a few details in the case
j = 1. We describe this case and leave the others to the reader. Let f be a function in V(1)
b
as in Corollary 5.1. Let β1 : [0, 1] → R be a smooth function such that
(i) β1 = 1 in a neighborhood of 0.
R 11(x) = 0 for all x ≥ x3 /2.
(ii) β
(iii) 0 β1 = 0.
For all sufficiently small δ > 0, the restriction of the function f + δβ1 to [0, x(f )] can be
extended to a unique function in V(1 b + δ) which we again denote by f + δ(β1 + β̃1 ). Set
1
Vδ = Xf +δ(β1 +β̃1 ) .
By (i), we then have c1 (Vδ1 ) = 1 + δ = c1 (V01 ) + δ. It follows from (ii) that x(Vδ1 ) = x(f )
and so ck (Vδ1 )) = ck (V01 )) for all even k. As well, for odd values of k > 1, property (ii)
implies that the solution of
k−1
(42) (f + (β1 + β̃1 ))0 = − .
k+1
is identical to that of (38). Hence, ck (Vδ1 )) = ck (V01 )) for these values of k as well. Finally,
it follow from (iii) and (41) that
volume(Vδ1 )) = volume(V01 )).

5.4. Proof of Theorem 1.3.


Lemma 5.3. Let h be a function in C(λ)
b with h0 (λ) ∈ (−1/2, 0). Let ρ: [0, 1] → R≥0 be a
nonconstant smooth function such that

R 1 support of ρ does not contain x(h) or x̌k for and even k ∈ N.


(i) the
(ii) 0 ρ = 0.
Then, for all sufficiently small δ > 0 the function h + δ(ρ + ρ̃) is in C(λ),
b we have
ck (Xh+δ(ρ+ρ̃) ) = ck (Xh )
for all k ∈ N, and
volume(Xh+δ(ρ+ρ̃) ) = volume(Xh ).
Proof. The assertion about the capacities follows from (i) and Corollary 5.2. The assertion
about the volumes follows from (ii) and (41). 
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 25

We will define explicit functions h and ρ and prove that the symplectomorphism type of
Xh+δ(ρ+ρ̃) varies with δ for all sufficiently small δ. To do this we will use the Embedded
Contact Homology capacities, {cECH k }k∈N , defined by Hutchings in [10] and the formula
for the ECH capacities of concave toric domains established by Choi, Cristofaro-Gardiner,
Frenkel, Hutchings and Ramos in [5]. In particular, we will prove, in Lemma 5.8 that
cECH
9 (Xh+δ(ρ+ρ̃) ) = cECH
9 (Xh ) + δ.
Remark 5.4. Given the results of [6], a similar argument should work for any choices of h and
ρ, as above. In particular, it should be possible to show that cECH
k (Xh+δ(ρ+ρ̃) ) = cECH
k (Xh )+δ
for some k ∈ N.
Remark 5.5. It is well known that the ECH capacities have their own blind spots cor-
responding to singular boundaries. In particular, E(1, 2) and P (1, 1) have the same ECH
capacities (and volumes). It is not known whether star-shaped regions in R4 with smooth
boundaries and the same ECH capacities must be symplectomorphic.

ECH capacities of concave toric domains. We recall here the formula from [5] for the
ECH capacities of a concave toric domain XΩ ⊂ R4 .
Theorem 5.6 ([5]). If XΩ is a concave toric domain in R4 and the ordered weight expansion
of Ω is
w(Ω)
~ = {w1 , w2 , w3 . . . },
then
( k )
k
X X d2i + di
(43) cECH (XΩ ) = max di w i ≤ k, di ∈ {0} ∪ N .

k
i=1

i=1
2

To make use of this we must also recall the definition of the the ordered weight expansion
w(Ω)
~ of a concave domain Ω. We begin with the basic procedure of concave subdivision
which derives three smaller (possibly empty) concave domains from Ω. We denote these by
T (Ω), Ω1 and Ω2 .
Let T (c) be the triangle in R2≥0 with vertices
{(0, 0), (c, 0), (0, c)} .
Setting τ (Ω) = max{c | T (c) ⊂ Ω}, we define
T (Ω) = T (τ (Ω)).
Now Ω r T (Ω) as the disjoint union of two, possibly empty, subsets whose closures we denote
by Ω̃1 and Ω̃2 , If either of these sets is nonempty, then the labels are chosen so that Ω̃1 is
disjoint from the vertical axis and Ω̃2 is disjoint from the horizontal axis. If both these sets
are nonempty their intersection is the point (τ (Ω), τ (Ω)). If Ω̃1 is nonempty then it has a
unique obtuse angle and we define Ω1 to be the concave domain obtained by translating Ω̃1
by (−τ (Ω), 0), and applying the transformation
 
1 1
.
0 1
26 ELY KERMAN AND YUANPU LIANG

Similarly, if Ω̃2 is nonempty, then it has a unique obtuse angle and we define Ω2 be the
concave domain obtained by translating Ω̃2 by (0, −τ (Ω)), and applying the transformation
 
1 0
.
1 1
If Ω̃j = ∅, we set Ωj = ∅.
Applying concave subdivision to Ωj we get concave domains T (Ωj ), Ωj1 , and Ωj2 . Con-
tinuing in this manner we get the collection of convex domains Ωj1 ...jd where ji ∈ {1, 2} and
d ∈ N. The weight expansion of Ω is the, possibly finite, multiset
w(Ω) = {τ (Ω)} ∪ {τ (Ωj1 ...jd ) | ji ∈ {1, 2}, d ∈ N, Ωj1 ...jd 6= ∅}.
Ordering this multiset with repetitions, we get the ordered weight expansion of Ω
w(Ω)
~ = {w1 , w2 , w3 . . . }
with
w1 ≥ w2 ≥ . . . .
Lemma 5.7. Suppose that h is in C(λ)
b and that h0 (0) < −4. For the concave domain Ω
bounded by the axes and the graph of h, we have
τ (Ω) = y0 + h(y0 )
τ (Ω2 ) = τ (Ω1 ) = 2y2 + h(y2 ) − τ (Ω)
τ (Ω22 ) = τ (Ω11 ) = 3y22 + h(y22 ) − τ (Ω) − τ (Ω2 )
τ (Ω21 ) = τ (Ω12 ) = 3y21 + 2h(y21 ) − 2τ (Ω) − τ (Ω2 )
τ (Ω222 ) = τ (Ω111 ) = 4y222 + h(y222 ) − τ (Ω) − τ (Ω2 ) − τ (Ω22 )
τ (Ω221 ) = τ (Ω112 ) = 5y221 + 2h(y221 ) − 2τ (Ω) − 2τ (Ω2 ) − τ (Ω22 )
τ (Ω212 ) = τ (Ω121 ) = 5y212 + 3h(y212 ) − 3τ (Ω) − 2τ (Ω2 ) − τ (Ω21 )
τ (Ω211 ) = τ (Ω122 ) = 4y211 + 3h(y211 ) − 3τ (Ω) − τ (Ω2 ) − τ (Ω21 )
where the y values are determined uniquely by the conditions
h0 (y0 ) = −1
h0 (y2 ) = −2
h0 (y22 ) = −3
3
h0 (y21 ) = −
2
h0 (y222 ) = −4
5
h0 (y221 ) = −
2
5
h0 (y212 ) = −
3
4
h0 (y211 ) = −
3
Proof. The leftmost set of equalities all follow from the symmetry of h. The condition
h0 (0) < −4 implies that all the points y1 , . . . , y211 exist. The rest of the formulas follow
easily from the concave subdivision process described above. 
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 27

Let ρ: [0, λ] → R≥0 be a smooth function such that


(i) the support of ρ is contained in an arbitrarily small interval around y22 .
(ii) ρ
R 1= 1 in a smaller neighborhood of y22 .
(iii) 0 ρ = 0.
Choose δ0 > 0 small enough so that the function
hδ = h + δ(ρ + ρ̃)
is in C(λ)
b for all δ ≤ δ0 . We will refer to process of moving from h to hδ as the δ-shift. Let
Ω be the region bounded by the axes and the graph of hδ and set Xhδ = µ−1 (Ωδ ). Note that
δ

y0 = x(h) and that y2 = x̌2 . It then follows from Lemma 5.3 that
ck (Xhδ ) = ck (Xh ) for all k ∈ N and δ ∈ [0, δ0 ],
and
volume(Xhδ ) = volume(Xh ) for all δ ∈ [0, δ0 ].
By construction, we have the following formulas relating τ -values before and after the
δ-shift:
τ (Ωδ ) = τ (Ω)
τ (Ωδ2 ) = τ (Ωδ1 ) = τ (Ω2 )
τ (Ωδ22 ) = τ (Ωδ11 ) = τ (Ω22 ) + δ
τ (Ωδ21 ) = τ (Ωδ12 ) = τ (Ω21 )
(44)
τ (Ωδ222 ) = τ (Ωδ111 ) = τ (Ω222 ) − δ
τ (Ωδ221 ) = τ (Ωδ112 ) = τ (Ω221 ) − δ
τ (Ωδ212 ) = τ (Ωδ121 ) = τ (Ω212 )
τ (Ωδ211 ) = τ (Ωδ122 ) = τ (Ω211 )
To say anything about the ordered weight expansion of Ωδ we now need to consider an
explicit function h. We will define h by parameterizing its graph. We start with the curve
from Example 1.11,
        
t t t t
α(t) = 2 sin − t cos , 2 sin + (2π − t) cos , t ∈ [0, 2π].
2 2 2 2
For a fixed, suitably small  > 0, we then set
γ(t) = α(t) − (, ) = (γ1 (t), γ2 (t))
where the domain of γ is now [ξ, 2π − ξ] for the number ξ > 0 defined uniquely by the
condition    
ξ ξ
2 sin − ξ cos = .
2 2
Let h be the function defined by the image of γ. Then h is in C(2
b − ). By Lemma 4.1,
Xh has a smooth boundary thanks to the shift by . A simple computation yields
γ20 (t) 2π − t
(45) h0 (γ1 (t)) = 0
=−
γ1 (t) t
28 ELY KERMAN AND YUANPU LIANG

and hence
2π − ξ
h0 (0) = − .
ξ
This is finite and less than −4 for all sufficiently small  > 0.
Let Ω be the domain defined by h. It then follows from (45) and Lemma 5.7 that:
τ (Ω) = 4 − 2

τ (Ω2 ) = τ (Ω1 ) = 3 3 − 4 −  ≈ 1.19615 − 
√ √
τ (Ω22 ) = τ (Ω11 ) = 4 2 − 3 3 −  ≈ .46070 − 

 

τ (Ω21 ) = τ (Ω12 ) = 10 sin − 3 3 − 4 ≈ .31441
5
π  √
τ (Ω222 ) = τ (Ω111 ) = 10 sin − 4 2 −  ≈ .22010 − 
5 
2π √ √
τ (Ω221 ) = τ (Ω112 ) = 14 sin − 3 3 − 4 2 ≈ .09263
7

   
3π 3π
τ (Ω212 ) = τ (Ω121 ) = 16 sin − 3 3 − 10 sin +  ≈ .07535 + 
8 5
   
3π 3π
τ (Ω211 ) = τ (Ω122 ) = 14 sin − 4 − 10 sin +  ≈ 0.13843 + 
7 5
For all sufficiently small  > 0 we then have
w(Ω)
~ = {w1 , w2 , . . . }
= {τ (Ω), τ (Ω2 ), τ (Ω1 ), τ (Ω22 ), τ (Ω11 ), τ (Ω21 ), τ (Ω12 ), . . . }.
Similarly, for all sufficiently small  > 0 and δ > 0 we have
~ δ ) = {w1δ , w2δ , . . . }
w(Ω
= {τ (Ωδ ), τ (Ωδ2 ), τ (Ωδ1 ), τ (Ωδ22 ), τ (Ωδ11 ), τ (Ωδ21 ), τ (Ωδ12 ), . . . }
= {τ (Ω), τ (Ω2 ), τ (Ω1 ), τ (Ω22 ) + δ, τ (Ω11 ) + δ, τ (Ω21 ), τ (Ω12 ), . . . }
To complete the proof of Theorem 1.3 it suffices to prove the following.
Lemma 5.8. For all sufficiently small  > 0 and δ > 0 we have
(46) cECH
9 (Xh+δ(ρ+ρ̃) ) = cECH
9 (Xh ) + δ.
Proof. By equation (43) we have
( k )
k 2
X X di + di
cECH (Xh ) = max di w i ≤ 9, di ∈ {0} ∪ N .

9

i=1
2 i=1
The two largest terms in the set
( k )
k
X X d2i + di
di wi ≤ 9, di ∈ {0} ∪ N

i=1
2
i=1
are 3w1 + 2w2 and 3w1 + w2 + w3 + w4 . For us, w2 = w3 and w4 > 0 and so
cECH
9 (Xh ) = 3w1 + w2 + w3 + w4
= 3τ (Ω) + τ (Ω2 ) + τ (Ω1 ) + τ (Ω22 ).
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 29

An identical argument then yields


cECH
9 (Xh+δ(ρ+ρ̃) ) = 3τ (Ωδ ) + τ (Ωδ2 ) + τ (Ωδ1 ) + τ (Ωδ22 )
= 3τ (Ω) + τ (Ω2 ) + τ (Ω1 ) + τ (Ω22 ) + δ
= cECH
9 (Xh ) + δ,
as required. 

6. Further Questions
Here we discuss some unresolved questions motivated by Theorems 1.1, 1.2 and 1.3 and
their proofs, and describe some relevant examples.
6.1. On varying capacities one at a time. The proof of Theorem 1.2 requires the con-
sideration of both convex and concave toric domains to allow for the independent variation
of both even and odd index capacities.
Question 6.1. Does there exist a star-shaped domain U ⊂ R2n , such that for any integer j,
there is star-shaped domain Vj ⊂ R2n with
ck (Vj ) = ck (U ) for all k 6= j
and
cj (Vj ) 6= cj (U )?
It is also not clear from the proof of Theorem 1.2 whether a fixed capacity can be indepen-
dently varied through the full range of its possible values while keeping the other capacities
fixed.
Question 6.2. Does there exist a star-shaped domain U ⊂ R2n with cj−1 (U ) < cj+1 (U ) for
some j ≥ 2, such that for each a ∈ (cj−1 (U ), cj+1 (U )) there is star-shaped domain Va with
ck (Va ) = ck (U ) for all k 6= j
and
cj (Va ) = a?
6.2. Isocapacity variations of volume. Let U is a star-shaped domain in R2n . Define the
isocapacity volume ratio of U to be
volume(V )
IVR(U ) = sup
volume(W )
where the supremum is taken over all star-shaped domains V, W ⊂ R2n such that ck (V ) =
ck (W ) = ck (U ) for all k ∈ N.
Example 6.3. For any star-shaped domain U ⊂ R2 we have IVR(U ) = 1
Example 6.4. For any symplectic polydisk P ⊂ R2n with n > 1 we have IVR(P ) = ∞.
With these simple cases in mind we restrict our attention again to the case of star-shaped
domains, with smooth boundaries, in R2n for n > 1. The proof of Theorem 1.1 provides a
mechanism, in the strictly convex toric setting, for varying volumes while keeping capacities
fixed. This mechanism fails in the case of symplectic ellipsoids, since the usable gaps between
capacity carriers vanish as one relaxes strict convexity. This observation motivates the
following.
30 ELY KERMAN AND YUANPU LIANG

Question 6.5. Is the isocapacity volume ratio of any symplectic ellipsoid equal to 1?
The general question concerning isocapacity volume ratios is the following.
Question 6.6. Is the isocapacity volume ratio of every star-shaped domain U in R2n with
smooth boundary finite? If so, is there some universal upper bound, dependent on the di-
mension, for the isocapacity volume ratio of all star-shaped domains in R2n with smooth
boundary?
Example 6.7 (Convex graphs for n = 2.). Let f be a function in V(λ) b such that f 0 (0) ∈
(−1/2, 0). Here, we use Corollary 5.1 to obtain a lower bound for IVR(Xf ). It is not clear
whether this bound can be improved.
We first construct a function f : [0, λ] → R whose graph lies below that of f , such that
the domain Xf has the same capacities as Xf . Define f to be the piecewise linear function
obtained by first connecting (xk−1 , f (xk−1 )) to (xk , f (xk ))) for all odd k ∈ N, where x0 = λ,
and then extending this to [x(f ), λ] as a symmetric function. Note that for any  > 0 there
is a function f in V(λ)
b such that kf − f kC 0 < ,
f (xk ) = f (xk ),
and
k−1
f0 (xk ) = −
k+1
for all odd k ∈ N. It follows from this, and the continuity of the ck , that
ck (Xf ) = ck (Xf )
for all k ∈ N.
Next we construct a function f¯: [0, λ] → R whose graph lies above that of f such that
the domain Xf¯ has the same capacities as Xf . For odd k, let Lk be the tangent line to the
graph of f at (xk , f (xk )). Let L0 be the line through (0, λ) with slope f 0 (0). Denote the
intersection point of Lk−1 and Lk by pk . Let f¯: [0, λ] → R be the piecewise linear function
obtained by first connecting pk−1 to pk for all odd k ∈ N, and then extending this to [x(f ), λ]
as a symmetric function. Arguing as above, we have
ck (Xf¯) = ck (Xf )
for all k ∈ N.
By construction, we have f ≤ f ≤ f¯ and hence Xf ⊂ Xf ⊂ Xf¯. From this it follows that
volume(Xf¯)
(47) IVR(Xf ) ≥ .
volume(Xf )
Since both f¯ and f are piecewise linear, the volumes on the right are easily computable.

For example, for the function f2 (x) = 1 − x2 from Example 1.9, it follows from (47) that
IVR(Xf2 ) is at least
√ P hp 2 p  i
5−2+ ∞
p
k + (k + 1) 2 (k − 1) 2 + k2 + (k + 1) 2 + (k + 2)2 − 2(k 2 + (k + 1)2 )
k=1
− 12
P∞ 4
k=1 (4k + 1)
or, approximately, 1.0335.
Question 6.8. Is inequality (47) really an equality?
ON SYMPLECTIC CAPACITIES AND THEIR BLIND SPOTS 31

Example 6.9 (Toric polytopes in R4 ). Here we use the capacity formula for symmetric toric
polytopes from Proposition 4.10, to find simple lower bounds for the isocapacity volume
ratios of the family of simple star-shaped domains in R4 from Example 4.12. For r ∈ [2/3, 1)
let Ωr be the convex hull of {(0, 0), (1, 0), (0, 1), (r, r)}. For a, b > 0 let Ωrab be the convex
hull of
{(0, 0), (1, 0), (0, 1), (r, r), (a, b), (b, a)}.
Set Xr = µ−1 (Ωr ) and Xab r
= µ−1 (Ωrab ).
r
It is straight forward to check that the symmetric toric domain Xab is convex for all (a, b)
in the region
Ir = {(a, b) | 0 ≤ b ≤ a ≤ 1, a + b ≤ 2r}.
r
Lemma 6.10. The domains Xr and Xab have the same capacities for all (a, b) in the region
Jr = Ir ∩ {(a, b) | 2a + b ≤ 3r}.
Proof. It follows from Proposition 4.10, and our choice of r ≥ 2/3, that for (a, b) in Ir we
have
r
(48) ck (Xab ) = ck (Xr )
for all k ∈ N if and only if
(n + 1)a + nb ≤ (2n + 1)r
for all n ∈ N. Given the condition a + b ≤ 2r from Ir , it is sufficient to require the first of
these inequalities,
2a + b ≤ 3r.

r
The volume of Xab is equal to 2π(r(a − b) + b)). This function of (a, b) takes its maximum
value on Jr , 2π(6r − 3r2 − 2), at the boundary point (1, 3r − 2). Hence,
2
(49) IVR(Xr ) ≥ 3(2 − r) − .
r

The maximum lower
q bound for the isocapacity volume ratios attained in this way is 6−2 6≈
2
1.10102 at r = 3
.

Remark 6.11. For r ∈ [1/2, 2/3] the lower bounds for IVR(Xr ) are smaller and more
difficult to derive since the conditions for the isocapacity property are more complicated.
This is the reason that these allowable values of r are disregarded in the discussion above.
Remark 6.12. Adding more than two vertices to the original polytope Ωr does not improve
the lower bound in (49). It would be interesting to know if inequality (49) is really an
equality. If this were true then the function r 7→ IVR(Xr ) would approach 1 as r → 1− ,
whereas IVR(X1 ) = IVR(P (1, 1)) = ∞.
Remark 6.13. The authors have not found domains, including symmetric toric polytopes

in higher dimensions, which have finite isocapacity volume ratios larger than 6 − 2 6.
32 ELY KERMAN AND YUANPU LIANG

References
[1] V. I. Arnol’d, First steps in symplectic topology, Russian Math. Surveys 41 (1986), 1–21.
[2] L. Bates, A Symplectic Rigidity Theorem, Manuscripta Mathematica 66 (1990),109–112.
[3] L. Baracco, M. Fassina, S. Pinton, On the Ekeland-Hofer symplectic capacities of the real bidisk, Pacific
J. Math. 305 (2020), 423-446.
[4] K. Cieliebak, H. Hofer, J. Latschev, and F. Schlenk, Quantitative symplectic geometry, Dynamics,
Ergodic Theory and Geometry, MSRI 54 (2007), 1–44.
[5] K. Choi, D. Cristofaro-Gardiner, D. Frenkel, M. Hutchings and V. G. B. Ramos, Symplectic embeddings
into four-dimensional concave toric domains, J . Topol., 7 (2014), 1054–1076.
[6] D. Cristofaro-Gardiner, M. Hutchings, and V.G.B.Ramos, The asymptotics of ECH capacities. Invent.
math. 199 (2015), 187–214.
[7] I. Ekeland and H. Hofer, Symplectic topology and Hamiltonian dynamics, Math. Z., 200 (1989), 355–
378.
[8] I. Ekeland and H. Hofer, Symplectic topology and Hamiltonian dynamics II, Math. Z., 203 (1990),
553–567.
[9] J. Gutt, M. Hutchings. Algebraic & Geometric Topology 18 (6), (2018), 3537–3600.
[10] M. Hutchings, Quantitative embedded contact homology, J. Diff. Geom. 88 (2011), 231–266.
[11] E. Kerman, Squeezing in Floer theory and refined Hofer-Zehnder capacities of sets near symplectic
submanifolds, Geometry &Topology, 9 (2005), 1775–1834.
[12] E. Kerman and Y. Luang, Higher index symplectic capacities do not satisfy the symplectic Brunn-
Minkowski inequality, to appear in Israel Journal of Mathematics.
[13] Lu, Guangcun, Finiteness of the Hofer-Zehnder capacity of neighborhoods of symplectic submanifolds,
International Mathematics Research Notices, (2006), 10.1155/IMRN/2006/76520.
[14] Y. Ostrover, V.G.B. Ramos, Symplectic embeddings of the `p –sum of two disks, preprint
arXiv:1911.06062.
[15] V.G.B. Ramos, Symplectic embeddings and the Lagrangian bidisk, Duke Math. J., 166 (2017), 1703–
1738.

Department of Mathematics, University of Illinois at Urbana-Champaign, 1409 West


Green Street, Urbana, IL 61801, USA.

You might also like