Fun with ECH capacities 2

This is a continuation of the previous post about computing some examples of ECH capacities.  My goals in this post are (1) to recap some of the previous discussion using better notation, (2) to explain an effective way to compute the ECH capacities of triangles, and (3) to give an update on the status of the conjectures in the previous post.

(1) Let me briefly recall what we are doing.  For every domain \Omega in the first quadrant of the plane we have a subset X_\Omega of {\mathbb C}^2, namely the inverse image of \Omega under the map sending (z_1,z_2)\mapsto\pi(|z_1|^2,|z_2|^2).  We also have a sequence of numbers

0 = c_0(\Omega) \le c_1(\Omega) \le c_2(\Omega) \le \cdots \le \infty

which are the ECH capacities of X_\Omega.  We would like to calculate c_k(\Omega), at least for some examples of interest.

Let’s recall what we know about the numbers c_k(\Omega).  To do so it will help to introduce a bit of notation.  Let {\mathcal S} denote the set of nondecreasing sequences A=(A_k)_{k=0,1,\ldots} where each A_k is a real number (or \infty) and A_0=0.  The ECH capacities of X_\Omega give such a sequence which we denote by c(\Omega).  Given two sequences A,B\in{\mathcal S}, define a new sequence A+B\in{\mathcal S} by

(A+B)_k = \max_{i+j=k}(A_i+B_j).

Observe that the operation + is associative and commutative.  Also, write A\le B if A_k\le B_k for all k.  Write A < B if A\le B and \lim_{k\to\infty}(B_k-A_k)=\infty.  (For this purpose we define \infty-\infty=\infty.  But that is not important for now because the examples of interest have finite ECH capacities.)  If A < B, define a new sequence B-A\in{\mathcal S} by

(B-A)_k = \min_{l\ge 0} (B_{k+l}-A_l).

Note that the operation - is not inverse to +.  In particular, it is easy to verify that (B-A) + A \le B \le (B+A)-A, but the reverse inequalities do not in general hold.

Now here are some properties of c(\Omega):

  • If \Omega' and \Omega are in the interior of the first quadrant, and “affine equivalent”, meaning that one can be obtained from the other by the action of SL_2{\mathbb Z} and translation, then c(\Omega)=c(\Omega').
  • If \Omega\subset \Omega', or more generally if X_\Omega symplectically embeds into X_{\Omega'}, then c(\Omega) \le c(\Omega').
  • If \Omega=\Omega_1\cup\Omega_2 where \Omega_1 and \Omega_2 intersect only along their boundaries, then c(\Omega_1)+c(\Omega_2)\le c(\Omega), and c(\Omega_1) \le c(\Omega)-c(\Omega_2).
  • If \Omega is convex and intersects the axes in at most two points, then c(\Omega) = m(\Omega), where m_k(\Omega) is defined to be the minimum, over convex paths \Lambda in the plane that enclose at least k+1 lattice points, of the length of \Lambda, as measured using the dual of a norm whose unit ball is a translate of \Omega.

Here the first property follows from the symplectomorphism invariance of ECH capacities.  The second property follows from the monotonicity axiom for ECH capacities.  In the third property, the first inequality follows from the monotonicity and disjoint union axioms for ECH capacities, and the second inequality is a consequence of the first.  These inequalities are not sharp in general, but they are sharp in some lucky cases.  The fourth property follows from Theorem 1.1 in my paper “Quantitative ECH”, as discussed in the previous post.  The assumption on intersecting the axes in at most two points can be dropped for rectangles and right triangles with legs on the axes, and quite possibly more generally.  In any case we always know using monotonicity that c_k(\Omega) \ge m_k(\Omega).

(2) The formula for c(\Omega) when \Omega is convex, while combinatorial, is rather difficult to compute with.  I discussed some ways to simplify it in the previous post when \Omega is a polygon, and now I would like to explain a way to simplify it even more when \Omega is a triangle.  First recall that if \Omega is affine equivalent to a right triangle with legs on the axes, then we can compute c(\Omega) very easily:  If the vertices are (0,0),(a,0),(0,b), then we simply write down all nonnegative integer linear combinations of a and b in nondecreasing order. Second, note that a general triangle can be approximated by a rational triangle, i.e. a triangle with vertices in {\mathbb Q}^2.  Third, observe that any rational triangle is affine equivalent to a triangle with all vertices on the axes. The ECH capacities of the latter kind of triangle can be effectively computed using the following proposition, which is a generalization of Lemma 2.6(ii) in Dusa McDuff’s paper “The Hofer conjecture on embedding symplectic ellipsoids”:

PROPOSITION. Suppose \Omega_1 is affine equivalent to a triangle with vertices (a,0),(b,0),(0,c) where a<b. Let \Omega_2 denote the triangle with vertices (0,0),(a,0),(0,c), and let \Omega_{12} denote the triangle with vertices (0,0),(b,0),(0,c). Then

c(\Omega_1) = c(\Omega_{12}) - c(\Omega_2).

PROOF:  We know from the general properties above that c(\Omega_1)\le c(\Omega_{12})-c(\Omega_2), so we just need to show that c(\Omega_1)\ge c(\Omega_{12})-c(\Omega_2).  For this purpose it is enough to show that m(\Omega_1)\ge m(\Omega_{12})-m(\Omega_2), since we know that c(\Omega_1)\ge m(\Omega_1) while c(\Omega_{12})=m(\Omega_{12}) and c(\Omega_2)=m(\Omega_2).

By an approximation argument, we may assume that a/c and b/c are irrational.

Recall from the previous post that if \Omega is any triangle, then m_k(\Omega) is the minimum of 2A(\Omega)\lambda over triangles \Lambda that enclose at least k+1 lattice points and are obtained from \Omega by scaling by \lambda and translating.  Also A(\Omega) denotes the area of \Omega.  (In the previous post I also rotated \Omega by 90 degress, as that was natural in the context of the discussion there, but since this does not change the answer I will not do it here.)  Let us call a triangle “minimal” if it achieves this minimum.  Note that if a triangle is minimal, then every edge must contain a lattice point;  otherwise we could move an edge towards an opposite vertex so as to decrease \lambda.  If \Omega is affine equivalent to a right triangle with legs on the axes, then the converse is also true, i.e. \Lambda is minimal if and only if every edge of \Lambda encloses a lattice point.

Now fix k and let \Delta_1 be a minimal triangle for computing m_k(\Omega_1).  Since \Delta_1 is minimal, it follows as above that the bottom, horizontal edge of \Delta_1 must have integer y coordinate. So by translating we may assume that the bottom edge of \Delta_1 is on the x axis, and by further translating we can assume that the top vertex of \Delta_1 has x coordinate in the interval (-1,0].  Now let \Delta_2 denote the triangle formed by the left edge of \Delta_1 together with the axes, and let l+1 denote the number of lattice points enclosed by \Delta_2.  Let \Delta_{12} denote the triangle formed by the right edge of \Delta_1 and the axes.  Note that each of the nonhorizontal edges of \Delta_1 contains exactly one lattice point (because \Delta_1 is minimal and a/c and b/c are irrational).  Consequently, \Delta_{12} encloses k+l+1 lattice points,  \Delta_2 is minimal for computing m_l(\Omega_2), and \Delta_{12} is minimal for computing m_{k+l}(\Omega_{12}).  Now write the bottom vertices of \Delta_1 as (a',0),(b',0) with a'<b'.  Then it follows from the above that

m_k(\Omega_1) = (b'-a')c,

m_l(\Omega_2) = a'c,

m_{k+1}(\Omega_{12}) = b'c.

Thus m_k(\Omega_1) = m_{k+l}(\Omega_{12})-m_l(\Omega_2) \ge (m(\Omega_{12})-m(\Omega_2))_k as desired.


(3) Enough for now about computing c(\Omega) when \Omega is convex (although there are still some interesting mysteries which I plan to discuss later).  If \Omega is a nonconvex polygon, can we compute c(\Omega) combinatorially using the above properties?  Or do we have to go back to the definition of ECH capacities and count holomorphic curves?  I don’t know.

In the previous post I considered the example where \Omega is a quadrilateral with vertices (0,0),(A,0),(1-a,a),(0,1) where a<1<A.  One can get a lower bound on c_k(\Omega) by cutting \Omega into two triangles (each affine equivalent to a right triangle with legs on the axes).  And one can get an upper bound on c_k(\Omega) by adding a triangle (whose ECH capacities are determined by the above proposition) to get a right triangle with legs on the axes.  I noticed that in some examples I worked out, the lower and upper bounds are equal.

Since then, I tried a couple more examples.  I found one case where the upper and lower bounds are not equal.  This case is where a=2/5, A=3, k=5.  In this case the lower bound on c_5 is 14/5, while the upper bound on c_5 is 3.  However, in all other examples that I computed (which admittedly were not very many), the upper and lower bounds were equal. So I don’t know what’s going on here.

Anyway, this does not disprove Conjecture 2 in the previous post, although it does cast some doubt on it. Recall that Conjecture 2 in the previous post was introduced because it implies Conjecture 1 in the previous post.  However Conjecture 2 is much more than is needed to prove Conjecture 1.  To prove Conjecture 1, one does not need c(\Omega) to equal the lower bound; one just needs good enough upper bounds on c(\Omega).

Anyway, I think a lot more calculation and/or intelligent thought is needed to sort all this out.  Meanwhile, if anyone is still reading this and has questions about how the calculations work, please post a comment or email me.  I am happy to clarify.  Also, since these are just informal notes, I may well have made mistakes.

This entry was posted in ECH, Open questions. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s