Handle Addition for doubly-periodic Scherk Surfaces
HHANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES
MATTHIAS WEBER AND MICHAEL WOLF
Abstract.
We prove the existence of a family of embedded doubly periodic minimalsurfaces of (quotient) genus g with orthogonal ends that generalizes the classical doublyperiodic surface of Scherk and the genus-one Scherk surface of Karcher. The proof of thefamily of immersed surfaces is by induction on genus, while the proof of embeddedness isby the conjugate Plateau method. Introduction
In this note we prove the existence of a sequence { S g } of embedded doubly-periodicminimal surfaces, beginning with the classical Scherk surface, indexed by the number g ofhandles in a fundamental domain. Formally, we prove Theorem 1.1.
There exists a family { S g } of embedded minimal surfaces, invariant undera rank two group Λ g generated by horizontal orthogonal translations. The quotient of eachsurface S g by Λ g has genus g and four vertical ends arranged into two orthogonal pairs. Our interest in these surfaces has a number of sources. First, of course, is that theseare a new family of embedded doubly periodic minimal surfaces with high topologicalcomplexity but relatively small symmetry group for their quotient genus. Next, unlike thesurfaces produced through desingularization of degenerate configurations (see [24], [25] forexample), these surfaces are not created as members of a degenerating family or are evenknown to be close to a degenerate surface. More concretely, there is now an abundance ofembedded doubly periodic minimal surfaces with parallel ends due to [3], while in the caseof non-parallel ends, the Scherk and Karcher-Scherk surfaces were the only examples.Third, one can imagine these surfaces as the initial point for a sheared family of (quo-tient) genus g embedded surfaces that would limit to a translation-invariant (quotient)genus g helicoid: such a program has recently been implemented for case of genus one byBaginsky-Batista [1] and Douglas [5].Our final reason is that there is a novelty to our argument in this paper in that wecombine Weierstrass representation techniques for creating immersed minimal surfaces ofarbitrary genus with conjugate Plateau methods for producing embedded surfaces. Theresult is then embedded surfaces of arbitrary (quotient) genus. Mathematics Subject Classification.
Primary 53A10 (30F60).The first author was partially supported by NSF grant DMS-0139476.The second author was partially supported by NSF grants DMS-9971563 and DMS-0139887. a r X i v : . [ m a t h . DG ] J u l MATTHIAS WEBER AND M. WOLF
Intuitively, our method to create the family of immersed surfaces — afterwards provenembedded — is to add a handle within a fundamental domain, and then flow within amoduli space of such surfaces to a minimal representative. We developed the method ofproof in [28] and [29] of using the theory of flat structures to add handles to the classicalEnneper’s surface and the semi-classical Costa surface; here we observe that the methodeasily extends to the case of the doubly-periodic Scherk surface — indeed, we will computethat the relevant flat structures for Scherk’s surface with handles are close cousins to therelevant flat structures for Enneper’s surface with handles. (This is a small surprise as thetwo surfaces are not usually regarded as having similar geometries.)Finally, we look at a fundamental domain on the surface for the automorphism groupof the surface and analyze its conjugate surface. As this turns out to be a graph, Krust’stheorem implies that our original fundamental domain is embedded.
Figure 1.1.
Scherk’s surface with four additional handlesOur paper is organized as follows: in the second section, we recall the background infor-mation about the Weierstrass representation, conjugate surfaces, and Teichm¨uller Theory,which we will need to construct our family of surfaces. In the third section, we outlineour method and begin the construction by computing triples of relevant flat structurescorresponding to candidate for the Weierstrass representation for the g -handled Scherksurfaces. In the fourth section, we define a finite dimensional moduli space M g of suchtriples and define a non-negative height function H : M g → R on that moduli space; awell-defined g -handled Scherk surface S g will correspond to a zero of that height function.Also in section 4, we prove that this height function is proper on M g .In section 5, we show that the only critical points of H on a certain locus Y g ⊂ M g arcat H − { } ∩ Y , proving the existence of the desired surfaces. We define this locus Y g ⊂ M g as an extension of a desingularization of the ( g − S g − , viewedas an element of M g − ⊂ ∂ M g , itself a stratum of the boundary ∂ M g of the closure M g of M g .In section 6, we show that the resulting surfaces { S g } are all embedded. ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 3 Background and Notation
History of doubly-periodic Minimal Surfaces.
In 1835, Scherk [21] discovereda 1-parameter family of properly embedded doubly-periodic minimal surfaces S ( θ ) ineuclidean space. These surfaces are invariant under a lattice Γ = Γ θ of horizontal euclideantranslations of the plane which induce orientation-preserving isometries of the surface S ( θ ).If we identify the xy -plane with C , this lattice is spanned by vectors 1 , e iθ .In the upper half space, S ( θ ) is asymptotic to a family of equally spaced half planes.The same holds in the lower half space for a different family of half planes. The anglebetween these two families is the parameter θ ∈ (0 , π/ S ( θ ) / Γ θ isconformally equivalent to a sphere punctured at ± , ± e iθ .Lazard-Holly and Meeks [15] have shown that all embedded genus 0 doubly-periodicsurfaces belong to this family.Since then, many more properly embedded doubly-periodic minimal surfaces in euclideanspace have been found:Karcher [11] and Meeks-Rosenberg [16] constructed a 3-dimensional family of genus-oneexamples where the bottom and top planar ends are parallel. Some of these surfaces canbe visualized as a fence of Scherk towers.P´erez, Rodriguez and Traizet [20] have shown that any doubly-periodic minimal surfaceof genus one with parallel ends belongs to this family.The first attempts to add further handles to these surfaces failed, and similarly it seemedto be impossible to add just one handle to Scherk’s doubly-periodic surface between every pair of planar ends.However, Wei [30] added another handle to Karcher’s examples (where all ends areparallel) by adding the handle between every second pair of ends. This family has beengeneralized by Rossman, Thayer and Wohlgemuth [26] to include more ends. Recently,Connor and Weber [3] adapted Traizet’s regeneration method to construct many examplesof arbitrary genus and arbitrarily many ends.Soon after Wei’s example, Karcher found an orthogonally-ended doubly-periodic Scherk-type surface with handle by also adding the handle only between every second pair of ends,see figure 2.2.Baginski and Ramos-Batista [1] as well as Douglas [5] have shown that the Karcherexample can be deformed to a 1-parameter family by changing the angle between the ends.On the theoretical side, Meeks and Rosenberg [17] have shown the following: Theorem 2.1.
A complete embedded minimal surface in E / Γ has only finitely many ends.In particular, it has finite topology if and only if it has finite genus. Theorem 2.2.
A complete embedded minimal surface in E / Γ has finite total curvatureif and only if it has finite topology. In this case, the surface can be given by holomorphicWeierstrass data on a compact Riemann surface with finitely many punctures which extendmeromorphically to these punctures. Weierstrass Representation.
Let S be a minimal surface in space with metric ds ,and denote the underlying Riemann surface by R . The stereographic projection of the MATTHIAS WEBER AND M. WOLF
Gauss map defines a meromorphic function G on R , and the complex extension of thethird coordinate differential dx defines a holomorphic 1-form dh on R , called the heightdifferential. The data ( R , G, dh ) comprise the Weierstrass data of the minimal surface. Via ω = 12 ( G − − G ) dhω = i G − + G ) dhω = dh one can reconstruct the surface as z (cid:55)→ Re (cid:90) z · ( ω , ω , ω )Vice versa, this Weierstrass representation can be used on any set of Weierstrass data todefine a minimal surface in space. Care has to be taken that the metric becomes complete.This procedure works locally, but the surface is only well-defined globally if the periodsRe (cid:90) γ
12 ( G − − G ) dh, i G − + G ) dh, dh )vanish for every cycle γ ⊂ R . The problem of finding compatible meromorphic data ( G, dh )which satisfies the above conditions on the periods of ω i is known as ‘the period problemfor the Weierstrass representation’.These period conditions are equivalent to(2.1) Re (cid:90) γ dh = 0and(2.2) (cid:90) γ Gdh = (cid:90) γ G − dh. For surfaces that are intended to be periodic, one can either define Weierstrass data onperiodic surfaces, or more commonly, one can insist that equations (2.1) and (2.2) holdfor only some of the cycles, with the rest of the homology having periods that generatesome discrete subgroup of Euclidean translations. Our setting will be of the latter type,with periods that either vanish or are in a rank-two abelian group of orthogonal horizontaltranslations.2.3.
Flat Structures.
The forms ω i lead to singular flat structures on the underlyingRiemann surfaces, defined via the line elements ds ω i = | ω i | . These singular metrics areflat away from the support of the divisor of ω i ; on elements p of that divisor, the metricshave cone points with angles equal to 2 π ( ord ω i ( p ) + 1). More importantly, the periodsof the forms are given by the Euclidean geometry of the developed image of the metric ds ω i — a period of a cycle γ is the (complex) distance C between consecutive images of adistinguished point in γ . We reverse this procedure in Section 3: we use putative developed ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 5 images of the one-forms
Gdh , G − dh , and dh to solve formally the period problem for someformal Weierstrass data. For more details about the properties of flat structures associatedto meromorphic 1-forms in connection with minimal surfaces, see [27].2.4. The Conjugate Plateau Construction and Krust’s Theorem.
The materialhere will be needed in Section 6 where we will prove the embeddedness of our surfaces.General references for the cited theorems of this subsection are [19] and [4].Given a minimal immersion F : z (cid:55)→ Re (cid:90) z ω , then the immersions F t : z (cid:55)→ Re (cid:90) z e it ω define the associate family of minimal surfaces. Among them, the conjugate surface F ∗ = F π/ is of special importance because symmetry properties of F correspond to symmetryproperties of F ∗ as follows: Theorem 2.3.
If a minimal surface patch is bounded by a straight line, the conjugate patchis bounded by a planar symmetry curve, and vice versa. Angles at corresponding verticesare the same.If (cid:96) and (cid:96) are a pair of intersecting straight lines on the conjugate patch correspondingto the intersection of a pair of (planar) symmetry curves lying on planes P and P , thenthe lines (cid:96) and (cid:96) span a plane orthogonal to the line common to P and P .Proof. The first paragraph is well-known: see [12] for example. The second paragraph iselementary, for if P is a plane of reflective symmetry, then the normal to the surface mustlie in the plane. At the intersection of two such planes, the normal must lie in both planes,hence in the line L of intersection of the two planes. But the Gauss map is preservedby the conjugacy correspondence, hence both of the corresponding straight lines (cid:96) and (cid:96) are orthogonal to L . Thus the plane spanned by (cid:96) and (cid:96) is normal to L , the line ofintersection of P and P . (cid:3) The best-known example of a conjugate pair are the catenoid and one full turn of thehelicoid.The second-best-known examples are the singly- and doubly-periodic Scherk surfaces.To get started with the conjugate Plateau construction, one can take a boundary contourbounded by straight lines and solves the Plateau problem using the classic result of Douglasand Rad´o (see [14] for a proof):
Theorem 2.4.
Let Γ be a Jordan curve in E bounding a finite-area disk. Then thereexists a continuous map ψ from the closed unit disk ¯ D into E such that (1) ψ maps S = ∂D monotonically onto Γ . (2) ψ is harmonic and almost conformal in D . (3) ψ ( ¯ D ) minimizes the area among all admissible maps. MATTHIAS WEBER AND M. WOLF
Here almost conformal allows a vanishing derivative and admissible maps on the diskare required to be in H , ( D, E ) so that their trace on ∂D can be represented by a weaklymonotonic, continuous mapping ∂D → Γ.For good boundary curves, one obtains the embeddedness and uniqueness of the Plateausolution for free by
Theorem 2.5. If Γ has a one-to-one parallel projection onto a planar convex curve, then Γ bounds at most one disk-type minimal surface which can be expressed as the graph of afunction f : E → E . The embeddedness of a Plateau solution sometimes implies the embeddedness of theconjugate surface. This observation is due to Krust (unpublished), see [12].
Theorem 2.6 (Krust) . If a minimal surface is a graph over a convex domain, then theconjugate piece is also a graph.
Teichm¨uller Theory.
For M a smooth surface, let Teich( M ) denote the Teichm¨ullerspace of all conformal structures on M under the equivalence relation given by pullbackby diffeomorphisms isotopic to the identity map id: M −→ M . Then it is well-known thatTeich( M ) is a smooth finite-dimensional manifold if M is a closed surface.There are two spaces of tensors on a Riemann surface R that are important for theTeichm¨uller theory. The first is the space QD( R ) of holomorphic quadratic differentials,i.e., tensors which have the local form Φ = ϕ ( z ) dz where ϕ ( z ) is holomorphic. Thesecond is the space of Beltrami differentials Belt( R ), i.e., tensors which have the local form µ = µ ( z ) d ¯ z/dz .The cotangent space T ∗ [ R ] (Teich( M )) is canonically isomorphic to QD( R ), and the tan-gent space is given by equivalence classes of (infinitesimal) Beltrami differentials, where µ is equivalent to µ if (cid:90) R Φ( µ − µ ) = 0 for every Φ ∈ QD( R ) . If f : C → C is a diffeomorphism, then the Beltrami differential associated to thepullback conformal structure is ν = f ¯ z f z d ¯ zdz . If f (cid:15) is a family of such diffeomorphisms with f = id , then the infinitesimal Beltrami differential is given by dd(cid:15) (cid:12)(cid:12) (cid:15) =0 ν f (cid:15) = (cid:18) dd(cid:15) (cid:12)(cid:12) (cid:15) =0 f (cid:15) (cid:19) ¯ z We will carry out an example of this computation in section 5.2A holomorphic quadratic differential comes with a picture that is a useful aid to one’sintuition about it. The picture is that of a pair of transverse measured foliations, whoseproperties we sketch briefly (see [6] for more details).A C k measured foliation on R with singularities z , . . . , z l of order k , . . . , k l (respec-tively) is given by an open covering { U i } of R − { z , . . . , z l } and open sets V , . . . , V l around z , . . . , z l (respectively) along with real valued C k functions v i defined on U i s.t.(1) | dv i | = | dv j | on U i ∩ U j ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 7 (2) | dv i | = | Im( z − z j ) k j /z dz | on U i ∩ V j Evidently, the kernels ker dv i define a C k − line field on R which integrates to give afoliation F on R − { z , . . . , z l } , with a k j + 2 pronged singularity at z j . Moreover, givenan arc A ⊂ R , we have a well-defined measure µ ( A ) given by µ ( A ) = (cid:90) A | dv | where | dv | is defined by | dv | U i = | dv i | . An important feature that we require of this measureis its “translation invariance”. That is, suppose A ⊂ R is an arc transverse to the foliation F , with ∂A a pair of points, one on the leaf l and one on the leaf l (cid:48) ; then, if we deform A to A via an isotopy through arcs A t that maintains the transversality of the image of A at every time, and also keeps the endpoints of the arcs A t fixed on the leaves l and l (cid:48) ,respectively, then we require that µ ( A ) = µ ( A ).Now a holomorphic quadratic differential Φ defines a measured foliation in the followingway. The zeros Φ − (0) of Φ are well-defined; away from these zeros, we can choose acanonical conformal coordinate ζ ( z ) = (cid:82) z √ Φ so that Φ = dζ . The local measuredfoliations ( { Re ζ = const } , | d Re ζ | ) then piece together to form a measured foliation knownas the vertical measured foliation of Φ, with the translation invariance of this measuredfoliation of Φ following from Cauchy’s theorem.Work of Hubbard and Masur [9] (see also alternate proofs in [13, 7, 34], following Jenkins[10] and Strebel [22], showed that given a measured foliation ( F , µ ) and a Riemann surface R , there is a unique holomorphic quadratic differential Φ µ on R so that the horizontalmeasured foliation of Φ µ is equivalent to ( F , µ ).2.6. Extremal length.
The extremal length ext R ([ γ ]) of a class of arcs Γ on a Riemannsurface R is defined to be the conformal invariantsup ρ (cid:96) ρ (Γ)Area( ρ )where ρ ranges over all conformal metrics on R with areas 0 < Area( ρ ) < ∞ and (cid:96) ρ (Γ)denotes the infimum of ρ -lengths of curves γ ∈ Γ. Here Γ may consist of all curves freelyhomotopic to a given curve, a union of free homotopy classes, a family of arcs with endpointsin a pair of given boundaries, or even a more general class. Kerckhoff [13] showed thatthis definition of extremal lengths of curves extended naturally to a definition of extremallengths of measured foliations.For a class Γ consisting of all curves freely homotopic to a single curve γ ⊂ M , (or moregenerally, a measured foliation ( F , µ )) we see that ext ( · ) (Γ) (or ext ( · ) ( µ )) can be construedas a real-valued function ext ( · ) (Γ): Teich( M ) −→ R . Gardiner [7] showed that ext ( · ) ( µ ) isdifferentiable and Gardiner and Masur [8] showed that ext ( · ) ( µ ) ∈ C (Teich( M )). In ourparticular applications, the extremal length functions on our moduli spaces will be realanalytic: this will be explained in Proposition 4.4.Moreover Gardiner computed that d ext ( · ) ( µ ) (cid:12)(cid:12) [ R ] = 2Φ µ MATTHIAS WEBER AND M. WOLF so that(2.3) (cid:16) d ext ( · ) ( µ ) (cid:12)(cid:12) [ R ] (cid:17) [ ν ] = 4 Re (cid:90) R Φ µ ν. A Brief Sketch of the Proof.
In this subsection, we sketch basic logic of theapproach and the ideas of the proofs, as a step-by-step recipe.
Step 1. Draw the Surface.
The first step in proving the existence of a minimalsurface is to work out a detailed proposal. This can either be done numerically, as in thework of (i) Thayer [23] for the Chen-Gackstatter surfaces we discussed in [28], (ii) Boix andWohlgemuth [2, 31, 32, 33] for the low genus surfaces we treated in [29] and (iii) Figures 2.1and 2.2 below for the present case; or it can be schematic, showing how various portionsof the surface might fit together, using plausible symmetry assumptions.
Step 2. Compute the Divisors for the Forms
Gdh and G − dh . From the modelthat we drew in Step 1, we can compute the divisors for the Weierstrass data, which wejust defined to be the Gauss map G and the ’height’ form dh . (Note here how important itis that the Weierstrass representation is given in terms of geometrically defined quantities— for us, this gives the passage between the extrinsic geometry of the minimal surface asdefined in Step 1 and the conformal geometry and Teichm¨uller theory of the later steps.)Thus we can also compute the divisors for the meromorphic forms Gdh and G − dh onthe Riemann surface (so far undetermined, but assumed to exist) underlying the minimalsurface. Of course the divisors for a form determine the form up to a constant, so the divisorinformation nearly determines the Weierstrass data for our surface. Here our schematicssuggest the appropriate divisor information, and this is confirmed by the numerics. Step 3. Compute the Flat Structures for the Forms
Gdh and G − dh requiredby the period conditions. A meromorphic form on a Riemann surface defines a flatsingular (conformal) metric on that surface: for example, from the form
Gdh on our puta-tive Riemann surface, we determine a line element ds Gdh = | Gdh | . This metric is locallyEuclidean away from the support of the divisor of the form and has a complete Euclideancone structure in a neighborhood of a zero or pole of the form. Thus we can develop theuniversal cover of the surface into the Euclidean plane.The flat structures for the forms Gdh and G − dh are not completely arbitrary: becausethe periods for the pair of forms must be conjugate (formula 2.2), the flat structures mustdevelop into domains which have a particular Euclidean geometric relationship to oneanother. This relationship is crucial to our approach, so we will dwell on it somewhat. Ifthe map dev : Ω −→ E is the map which develops the flat structure of a form, say α , ona domain Ω into E , then the map dev pulls back the canonical form dz on C ∼ = E tothe form α on ω . Thus the periods of α on the Riemann surface are given by integrals of dz along the developed image of paths in C , i.e. by differences of the complex numbersrepresenting endpoints of those paths in C .We construe all of this as requiring that the flat structures develop into domains thatare “conjugate”: if we collect all of the differences in positions of parallel sides for thedeveloped image of the form Gdh into a large complex-valued n -tuple V Gdh , and we collectall of the differences in positions of corresponding parallel sides for the developed image of
ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 9 the form G − dh into a large complex-valued n-tuple V G − dh , then these two complex-valuedvectors V Gdh and V G − dh should be conjugate. Thus, we translate the “period problem”into a statement about the Euclidean geometry of the developed flat structures. This isdone at the end of section 3.The period problem 2.1 for the form dh will be trivially solved for the surfaces we treathere. Step 4. Define the moduli space of pairs of conjugate flat domains.
Nowwe work backwards. We know the general form of the developed images (called Ω
Gdh and Ω G − dh , respectively) of flat structures associated to the forms Gdh and G − dh , but ingeneral, there are quite a few parameters of the flat structures left undetermined; this holdseven after we have assumed symmetries, determined the Weierstrass divisor data for themodels and used the period conditions 2.2 to restrict the relative Euclidean geometries ofthe pair Ω Gdh and Ω G − dh . Thus, there is a moduli space ∆ of possible candidates of pairsΩ Gdh and Ω G − dh : our period problem (condition 2.2) is now a conformal problem of findingsuch a pair which are conformally equivalent by a map which preserves the correspondingcone points. (Solving this problem means that there is a well-defined Riemann surfacewhich can be developed into E in two ways, so that the pair of pullbacks of the form dz give forms Gdh and G − dh with conjugate periods.)The condition of conjugacy of the domains Ω Gdh and Ω G − dh often dictates some restric-tions on the moduli space, and even a collection of geometrically defined coordinates. Wework these out in section 3. Step 5. Solve the Conformal Problem using Teichm¨uller theory.
At thisjuncture, our minimal surface problem has become a problem in finding a special pointin a product of moduli spaces of complex domains: we will have no further referencesto minimal surface theory. The plan is straightforward: we will define a height function H : ∆ −→ R with the properties:(1) (Reflexivity) The height H equals 0 only at a solution to the conformal problem(2) (Properness) The height H is proper on ∆. This ensures the existence of a criticalpoint.(3) (Non-degenerate Flow) If the height H at a pair (Ω Gdh , Ω G − dh ) does not vanish,then the height H is not critical at that pair (Ω Gdh , Ω G − dh ).This is clearly enough to solve the problem: we now sketch the proofs of these steps. Step 5a. Reflexivity.
We need conformal invariants of a domain that provide acomplete set of invariants for Reflexivity, have estimable asymptotics for Properness, andcomputable first derivatives (in moduli space) for the Non-degenerate Flow property. Oneobvious choice is a set of functions of extremal lengths for a good choice of curve systems,say Γ = { γ , . . . , γ g } on the domains. These are defined for our examples in section 4.1. Wethen define a height function H which vanishes only when there is agreement between allof the extremal lengths ext Ω Gdh ( γ i ) = ext Ω G − dh ( γ i ) and which blows up when ext Ω Gdh ( γ i )and ext Ω G − dh ( γ i ) either decay or blow up at different rates. See for example Definition4.2 and Lemma 4.11. Step 5b Properness.
Our height function will measure differences in the extremallengths ext Ω Gdh ( γ i ) and ext Ω G − dh ( γ i ). A geometric degeneration of the flat structure ofeither Ω Gdh or Ω G − dh will force one of the extremal lengths ext • ( γ i ) to tend to zero orinfinity, while the other extremal length stays finite and bounded away from zero. Thisis a straightforward situation where it will be obvious that the height function will blowup. A more subtle case arises when a geometric degeneration of the flat structure forces both of the extremal lengths ext Ω Gdh ( γ i ) and ext Ω G − dh ( γ i ) to simultaneously decay (or ex-plode). In that case, we begin by observing that there is a natural map between the vector < ext Ω Gdh ( γ i ) > and the vector < ext Ω G − dh ( γ i ) > . This pair of vectors is reminiscent ofpairs of solutions to a hypergeometric equation, and we show, by a monodromy argumentanalogous to that used in the study of those equations, that it is not possible for corre-sponding components of that vector to vanish or blow up at identical rates. In particular,we show that the logarithmic terms in the asymptotic expansion of the extremal lengthsnear zero have a different sign, and this sign difference forces a difference in the rates ofdecay that is detected by the height function, forcing it to blow up in this case. The mon-odromy argument is given in section 4.3, and the properness discussion consumes section4.2. Step 5c. Non-degenerate Flow.
The domains Ω
Gdh and Ω G − dh have a remark-able property: if ext Ω Gdh ( γ i ) > ext Ω G − dh ( γ i ), then when we deform Ω Gdh so as to de-crease ext Ω Gdh ( γ i ), the conjugacy condition forces us to deform Ω G − dh so as to increaseext Ω G − dh ( γ i ). We can thus always deform Ω Gdh and Ω G − dh so as to reduce one term ofthe height function H . We develop this step in Section 5. Step 5d. Regeneration.
In the process described in the previous step, an issue arises:we might be able to reduce one term of the height function via a deformation, but thismight affect the other terms, so as to not provide an overall decrease in height. We thusseek a locus Y in our moduli space where the height function has but a single non-vanishingterm, and all the other terms vanish to at least second order. If we can find such a locus Y , we can flow along that locus to a solution. To begin our search for such a locus, weobserve which flat domains arise as limits of our domains Ω Gdh and Ω G − dh : commonly,the degenerate domains are the flat domains for a similar minimal surface problem, maybeof slightly lower genus or fewer ends.We find our desired locus by considering the boundary of the (closure) of the modulispace ∆: this boundary has strata of moduli spaces ∆ (cid:48) for minimal surface problems oflower complexity. By induction, there are solutions of those problems represented on sucha boundary strata ∆ (cid:48) (with all of the corresponding extremal lengths in agreement), andwe prove that there is a nearby locus Y inside the larger moduli space ∆ which has theanalogues of those same extremal lengths in agreement. As a corollary of that condition,the height function on Y has the desired simple properties.2.8. The Geometry of Orthodisks.
In this section we introduce the notion of or-thodisks.Consider the upper half plane H and n ≥ t i on the real line.The point t ∞ = ∞ will also be a distinguished point. We will refer to the upper half ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 11 plane together with these data as a conformal polygon and to the distinguished points as vertices . Two conformal polygons are conformally equivalent if there is a biholomorphicmap between the disks carrying vertices to vertices, and fixing ∞ .Let a i be some odd integers such that(2.4) a ∞ = − − (cid:88) i a i By a Schwarz-Christoffel map we mean the map(2.5) F : z (cid:55)→ (cid:90) zi ( t − t ) a / · . . . · ( t − t n ) a n / dt A point t i with a i > − finite , otherwise infinite . By Equation 2.4, there is atleast one finite vertex. Definition 2.7.
Let a i be odd integers. The pull-back of the flat metric on C by F definesa complete flat metric with boundary on H ∪ R without the infinite vertices. We call sucha metric an orthodisk . The a i are called the vertex data of the orthodisk. The edges ofan orthodisk are the boundary segments between vertices; they come in a natural order.Consecutive edges meet orthogonally at the finite vertices. Every other edge is parallelunder the parallelism induced by the flat metric of the orthodisk. Oriented distancesbetween parallel edges are called periods . We will discuss the relationship of these periodsto the periods arising in the minimal surface context in Section 3.The periods can have 4 different signs: +1 , − , + i, − i .The interplay between these signs is crucial to our monodromy argument, especiallyLemma 4.11. Remark 2.8.
The integer a i corresponds to an angle ( a i +2) π/ − θ ) lies at infinity and isthe intersection of a pair of lines which also intersect at a finite point, where they make a positive angle of + θ .In all the drawings of the orthodisks to follow, we mean the domain to be to the left ofthe boundary, where we orient the boundary by the order of the points t i .2.9. Scherk’s and Karcher’s Doubly-Periodic Surfaces.
The singly- and doubly-periodic Scherk surfaces are conjugate spherical minimal surfaces whose Weierstrass datalead to no computational difficulties and whose orthodisk description illustrates the basicconcepts in an ideal way.We discuss first the Weierstrass representation of the doubly-periodic Scherk surfaces S ( θ ) (see figure 2.1). G ( z ) = z and dh = z − idzz + z − − φ on the Riemann sphere punctured at the four points q = ± e ± iφ . Figure 2.1.
Scherk’s surfaceThe residues of dh at the punctures have the real values ±
14 sin 2 φ and hence we haveno vertical periods. At the punctures (which correspond to the ends) the Gauss map ishorizontal and takes the values ± e ± iφ so that the angle between two ends is 2 φ . Becauseof this, the horizontal surface periods around a puncture q are given as complex numbersby Re (cid:73) q i (cid:18) G + 1 G (cid:19) dh + i Re (cid:73) q (cid:18) G − G (cid:19) dh == Re (cid:16) πi (cid:16) G ( q ) + G ( q ) (cid:17) Res q dh (cid:17) − i Re (cid:16) πi (cid:16) G ( q ) − G ( q ) (cid:17) Res q dh (cid:17) = − πG ( q ) Res q dh. These numbers span a horizontal lattice in C so that the surface is indeed doubly-periodic.Indeed we can now regard the result as being defined over the even squares of a shearedcheckerboard with vertices given by the period lattice.Our principal interest in this paper will be with handle addition for the orthogonal Scherksurface S = S ( π/ φ = π/ shear these surfaces, as in the work of Ramos-Batista-Baginsky[1] or Douglas [5], so that the periods span a non-orthogonal horizontal lattice, or to addhandles to a sheared surface S ( θ ).The construction of S is due to Hermann Karcher [11] who found a way to ‘add ahandle’ to the classical Scherk surface. Here we mean that he found a doubly-periodicminimal surface whose fundamental domain is equivariantly isotopic to a doubly-periodicsurface formed by adding a handle to the Scherk surface above.We now reprove the result of Karcher from the perspective of orthodisks.The quotient surface of S by its horizontal period lattice is a square torus with fourpunctures corresponding to the ends. We will construct this surface using the Weierstrass ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 13
Figure 2.2.
Scherk’s surface with an additional handledata given by figure 2.3 below. We begin with the figure on the far right. The pointswith labels 1 to 4 are 2-division points on the torus and correspond to the points withvertical normal on S . The points E , E and their symmetric counterparts (not labelled)correspond to the ends. The point E is placed on the straight segment between 1 and 2,and its position is a free parameter that will be used to solve the period problem. Theother poles are then determined by the reflective symmetries of the square. G dh labels ∞ ∞ ∞∞ ∞ ∞
E1E2
Figure 2.3.
Divisors for the doubly-periodic Scherk surface with handleWe obtain the domains | Gdh | and | G dh | by developing just the shaded square the regionsgiven by figure 2.3.As usual, the domains lie in separate planes and the half-strips extend to infinity.Observe that these domains are arranged to be symmetric with respect to the y = − x diagonal. Remark 2.9.
Four copies of the (say) | Gdh | domain fit together to form a region as inFigure 2.5 with orthogonal half-strips, where a square from the center has removed. Thissquare (with opposite edges identified) corresponds precisely to the added handle.The period condition requires Gdh and G dh to have the same residues at E and E (sothat the half-infinite strips need to have the same width) and the remaining periods need E1E2 E2E11 2 3 4 1 2 3 4 |G dh| |1/G dh|
Figure 2.4. | Gdh | and | G dh | orthodisks of a quarter piece Figure 2.5. | Gdh | and | G dh | orthodisks of a fundamental domainto be complex conjugate so that the 23 and 34 edges need to have the same length in bothdomains.This is a one-dimensional period problem, and as is often the case, one can solve it viaan intermediate value argument. There are two versions of this argument, one approachingthe problem from the perspective of the period integrals on a fixed Riemann surface, andthe other from the perspective of the conformal moduli of the pair of orthodisks. We beginwith the version based on the behavior of the period integrals in the limit cases. We keepthe discussion as close as possible to the orthodisk description by using Schwarz-Christoffelmaps from the upper half plane to parametrize the domains for Gdh and G : f Gdh : z (cid:55)→ (cid:90) z √ r √− x √− r √ x ( − r + x ) dxf /Gdh : z (cid:55)→ (cid:90) z √− r √ x √ r √− x ( − r + x ) dx ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 15
Here the point − < − r < < r < < ∞ on the real axis are mapped to the labels4 , E , , E , ,
3, respectively. The normalization is choosen so thatRes r Gdh = 12 r = Res r G dh
The parameter r determines the relative position of E and will now be determined. Forthis we compute the lengths of the edge 23 as functions of r using hypergeometric functionsas follows: A Gdh = (cid:90) ∞ G dh = √ π √ r √ − r Γ(5 / / F (cid:18) , , , r (cid:19) A /Gdh = (cid:90) ∞ Gdh =2 √ π √ − r √ r Γ(3 / / F (cid:18) , , , r (cid:19) The period condition requires A Gdh ( r ) = A /Gdh ( r ) . To determine r , notice the ‘boundary conditions’ A Gdh (0) = ∞ A /Gdh (0) = 0 A Gdh (1) = π A /Gdh (1) = ∞ so that the intermediate value theorem implies the existence of a solution.Alternatively, we can give an intermediate value theorem argument based on extremallength. This is more in keeping with our theme of converting the period problem forminimal surfaces into a conformal problem for orthodisks.In terms of the orthodisks, the family of possible pairs { Ω Gdh , Ω G − dh } of orthodisks canbe normalized so that each half-strip end of either Ω Gdh or Ω G − dh has width one. Thenthe family of pairs { Ω Gdh , Ω G − dh } is parametrized by the distance, say d , between thepoints 1 and 3 in the domain Ω Gdh : there is degeneration in the domain Ω
Gdh as d −→ other domain Ω G − dh as d −→ √ F of curves connecting the side E E
4. Let usexamine the extremal length of this family under the pair of limits. In the first case, as d −→
0, the two edges E E Gdh , while the domainΩ G − dh is converging to a non-degenerate domain. Thus the extremal length of F in Ω Gdh is becoming infinite, while the corresponding extremal length in Ω G − dh is remaining finiteand positive. The upshot is that near this limit point, ext Ω Gdh ( F ) > ext Ω G − dh ( F ).Near the other endpoint, where d is nearly √ Gdh , the opposite inequality holds.This claim is a bit more subtle, as since the pair of segments 23 and 34 are converging toa single point, we see both extremal lengths ext Ω Gdh ( F ) and ext Ω G − dh ( F ) are tending to zero. Yet it is quite easy to compute that the rates of vanishing are quite different, yieldingext Ω Gdh ( F ) < ext Ω G − dh ( F ) near this endpoint. To see this let Ω Gdh ( (cid:15) ) and Ω G − dh ( (cid:15) )denote the domains Ω Gdh (and Ω G − dh , respectively) with the lengths of sides 23 or 34being (cid:15) . We are interested in the Schwarz-Christoffel maps F Gdh,(cid:15) : H → Ω Gdh ( (cid:15) ) and F /Gdh,(cid:15) : H → Ω G − dh ( (cid:15) ), and in particular at the preimages x ( (cid:15) ), x ( (cid:15) ), x ( (cid:15) ) (and y ( (cid:15) ), y ( (cid:15) ), y ( (cid:15) ), resp. ) under F Gdh,(cid:15) (and F /Gdh ( (cid:15) ), resp.) of the vertices marked 2, 3 and 4.It is straighforward to see that up to a factor that is bounded away from both 0 and ∞ as (cid:15) →
0, these positions are given by the positions of the corresponding pre-images of thesimplified (and symmetric) maps F Gdh,(cid:15) ( z ) = (cid:90) z t − ( t − x ) − / ( t − x ) / ( t − x ) − / dt and F /Gdh,(cid:15) = (cid:90) z t − ( t − y ) / ( t − y ) − / ( t − y ) / dt where { x i } and { y i } are bounded away from zero and we have suppressed the dependenceon (cid:15) in the expressions for the vertices. (We can ignore the factor because we can, forexample, normalize the positions of the points so the the distance x − x is the only freeparameter. Once that is done, the factor is determined by an integral of a path beginningin the interval [ x , E ] to a point in the interval [ x , E ]; in this situation, both the lengthof the path and the integrand are bounded away from both zero and infinity, proving theassertion.) Thus we may compute the asymptotics by setting (cid:15) = (cid:90) x x t (cid:115) t − x ( t − x )( t − x ) dt (cid:16) ( x − x ) / (cid:90) (cid:114) s ( s + 1)( s − ds using that s = t − x x − x , that we have bounded X , x , x away from zero, and that we haveassumed the symmetry x − x = x − x .Thus x ( (cid:15) ) − x ( (cid:15) ) = 0( (cid:15) ).A similar formal substitution into the integral expression for F /Gdh ( (cid:15) ), again using thesymmetry of the domain, yields that (cid:15) = (cid:90) y y t (cid:115) ( t − y )( t − y )( t − y ) dt (cid:16) ( y − y ) / (cid:90) (cid:114) ( s − s + 1) s ds so ( y − y ) = 0( (cid:15) / ).As the extremal length ext Ω Gdh ( F ) and ext Ω G − dh ( F ) of F are given by the extremallengths in H for a family of arcs between intervals that surround x , x , and x (or y , ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 17 y , and y ) in H , and those extremal lengths are monotone increasing in the length ofthe excluded interval x x (or y y , we see from the displayed formulae above and that (cid:15) / > (cid:15) for (cid:15) small implies that(2.6) ext Ω Gdh ( F ) < ext Ω G − dh ( F )for (cid:15) small, as desired. 3. Orthodisks for the Scherk family
In this section, we begin our proof of the existence of the surfaces S g , the doubly-periodic Scherk surfaces with g handles. We begin by deciding on the form of the relevantorthodisks; our plan is to adduce these orthodisks from the orthodisks for the classicalScherk surface S and the Karcher surface S . It will then turn out that these orthodisksare quite similar to the orthodisks we used in [28] to prove the existence of the surfaces E g of genus g with one Enneper-like end.In this section we will introduce pairs of orthodisks and outline the existence proof forthe S g surfaces, using the E g surfaces as the model case.The existence proof consists of several steps. The first is to set up a space ∆ = ∆ g of geometric coordinates such that each point in this space gives rise to a pair of conjugateorthodisks as described in section 2.Given such a pair, one canonically obtains a pair of marked Riemann surfaces withmeromorphic 1-forms having complex conjugate periods. If the surfaces were conformallyequivalent, these two 1-forms would serve as the 1-forms Gdh and G dh in the Weierstrassrepresentation.After that, it remains to find a point in the geometric coordinate space so that thetwo surfaces are indeed conformal. To achieve this, a nonnegative height function H isconstructed on the coordinate space with the following properties:(1) H is proper;(2) H = 0 implies that the two surfaces are conformal;(3) Given a surface S g − , there is a smooth locus Y which lies properly in the S g coordinate space ∆ g whose closure contains S g − ∈ ∂ ∆ g . On that locus Y ⊂ ∆ g , if d H = 0, then actually H = 0.The height should be considered as some adapted measurement of the conformal distancebetween the two surfaces. Hence it is natural to construct such a function using conformalinvariants. We have chosen to build an expression using the extremal lengths of suitablecycles.The first condition on the height poses a severe restriction on the choice of the geometriccoordinate system: The extremal length of a cycle becomes zero or infinite only if thesurface develops a node near that cycle. Hence we must at least ensure that when reachingthe boundary ∂ ∆ g of the geometric coordinate domain ∆ g , at least one of the two surfacesdegenerates conformally .This condition is called completeness of the geometric coordinate domain ∆ g . Fortunately, we can use the definition of the geometric coordinates for E g to derivecomplete geometric coordinates for S g .We recall the geometric coordinates that we used in [28] to prove the existence of theEnneper-ended surfaces E g . There, both domains Ω Gdh and Ω G − dh were bounded bystaircase-like objects we referred to as ’zigzags’: in particular, the boundary of a domainwas a properly embedded arc, which alternated between ( g + 1) purely vertical segmentsand ( g + 1) purely horizontal segments and was symmetric across a diagonal line. Any suchboundary is determined up to translation by the lengths of its initial g finite-length sides,and up to homothety by any subset of those of size g −
1. Thus, the geometric coordinateswe used for such a domain Ω
Gdh or Ω G − dh were the lengths of the first g − Remark 3.1.
We were fortunate in [28], as we will be in the present case, to be able torestrict our attention to orthodisks which embed in the plane. For orthodisk systems thatbranch over the plane (see [29]) or are not planar (see [5]), the description of the geometriccoordinates can be quite subtle.Recall that the orthodisks for the Chen-Gackstatter surfaces of higher genus were ob-tained by taking the negative y -axis and the positive x -axis and replacing the subarc from(0 , − a ) to ( a,
0) by a monotone arc consisting of horizontal and vertical segments whichwere symmetric with respect to the diagonal y = − x . The two regions separated by this‘zigzag’ constituted a pair of orthodisks. The geometric coordinates were given by the edgelengths of the finite segments above the diagonal y = − x .For our new surfaces, we continue the above construction as follows. Denote the vertexof the new subarc that meets the diagonal by ( c, − c ). Choose b > c . We then intersect theupper left region with the half planes { x > − b } and { y < b } . Similarly, we intersect thelower right region with the half planes { x < b } and { y > − b } . This procedure defines twodomains which we denote by Ω Gdh and Ω G − dh . We use the convention that Ω Gdh is thedomain where the vertex ( c, − c ) makes a 3 π/ b of the half-infinite vertical and horizontal strips. Theorem 3.2.
This coordinate system for S g is complete.Proof. Certainly if one of the finite edges degenerates, the conformal structure also leavesall compact sets in its moduli space. Next, if the geometric coordinate b − c tends to 0,the two vertices on the diagonal y = − x coalesce, so that the extremal length of the arcconnecting P E to E P g tends to ∞ , and so the surface has also degenerated. (cid:3) Why should such an orthodisk system correspond to a doubly-periodic minimal surfaceof genus g ? Here we are both generalizing the intuition given by Karcher’s surface, or al-ternatively relying on numerical simulation (see Figure 1.1). Either way, we can conjecturethe divisor data for a fundamental (and planar) piece of the surface S g , and use this todefine the orthodisk of the surface, hence the developed image of a fundamental piece.To formalize the discussion, we introduce: ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 19
Definition 3.3.
A pair of orthodisks is called reflexive if there is a vertex- and label-preserving holomorphic map between them.Then we have:
Theorem 3.4.
Given a reflexive pair of orthodisks of genus g , there is a doubly-periodicminimal surface of genus g in T × E with two orthogonal top and two bottom ends.Proof. We first construct the underlying Riemann surface by taking the Ω
Gdh orthodisk,doubling it along the boundary, and then taking a double branched cover of that, branchedat the vertices. This gives us a Riemann surface X g of genus g .That Riemann surface carries a natural cone metric induced by the flat metric of Ω Gdh .As all identifications are done by parallel translations, this cone metric has trivial holonomyand hence the exterior derivative of its developing map defines a 1-form which we call
Gdh .This 1-form is well-defined, up to multiplication by a complex number.By the reflexitivity condition, the very same Riemann surface X g carries another conemetric, being induced from the Ω G − dh orthodisk and the canonical identification of theΩ Gdh and Ω G − dh orthodisks by a vertex-preserving conformal diffeomorphism. This secondorthodisk defines a second 1-form, denoted by G dh , also well-defined only up to scaling.The free scaling parameters are now fixed (up to an arbitrary real scale factor whichonly affects the size of the surface in E ) so that the developed Ω Gdh and Ω G − dh are trulycomplex conjugate if we use the same base point and base direction for the two developingmaps.This way we have defined the Weierstrass data G and dh on a Riemann surface X g .We show next that the resulting minimal surface has the desired geometric properties.The cone points on X g come only from the orthodisk vertices: the finite vertices P j , beingbranch points, lift to a single cone point (also denoted P j ). The other finite cone points V + and V − give also only one cone point on the surface, denoted by V . The half stripslead to four cone points E i . From the cone angles we can easily deduce the divisors of theinduced 1-forms as ( Gdh ) = P P · · · P g E − E − E − E − ( 1 G dh ) = P P · · · P g − V E − E − E − E − ( dh ) = P · · · P g V E − E − E − E − These data guarantee that the surface is properly immersed, without singularities, andcomplete. The points P i and V correspond to the points with vertical normal at theattached handles, while the E i correspond to the four ends.As dh has only simple zeroes and poles, its periods will all have the same phase, andusing a local coordinate it is easy to see that the periods must all be real.For the cycles in Ω Gdh and Ω G − dh corresponding to finite edges, the conjugacy conditionensures that all of the periods are purely real. The cycles around the ends are similarlyconjugate by the construction of the orthodisks. The symmetry of the domain ensures thatthe ends are orthogonal. (cid:3) Existence Proof: The Height Function
Definition and Reflexivity of the Height Function.
For a cycle c connectingpairs of edges denote by ext Ω Gdh ( c ) and ext Ω G − dh ( c ) the extremal lengths of the cycle inthe Gdh and G dh orthodisks, respectively. Recall that this makes sense as we have a nat-ural topological identification of these domains (up to homotopy) mapping correspondingvertices onto each other.The height function on the space of geometric coordinates will be a sum over severalsummands of the following type: Definition 4.1.
Let c be a cycle. Define H ( c ) = (cid:12)(cid:12)(cid:12) e / ext Ω Gdh ( c ) − e / ext Ω G − dh ( c ) (cid:12)(cid:12)(cid:12) + (cid:12)(cid:12)(cid:12) e ext Ω Gdh ( c ) − e ext Ω G − dh ( c ) (cid:12)(cid:12)(cid:12) The rather complicated shape of this expression is required to prove the propernessof the height function: Because there are sequences of points in the space of geometriccoordinates which converge to the boundary so that both orthodisks degenerate for thesame cycles, the above expression must be very sensitive to different rates with which thishappens.Due to the Monodromy Theorem 4.6, it is sometimes possible to detect such rate differ-ences in the growth of exp c ) for degenerating cycles with ext( c ) → S g as { P , . . . , P g } , the endvertices as E and E , and the finite vertices on the outside boundary components of Ω Gdh and Ω G − dh as V + and V − , respectively. Note that in the combined Figure 4.1, the vertices P i proceed in a different order for the domain Ω G − dh than they do for the domain Ω Gdh .At this point, there is a difficulty in keeping the notation consistent; a consistent choiceof orientation of the Gauss map G results in the two regions switching labels as we increasethe genus by one; we will circumvent that notational issue by requiring the Gauss map G to have the orientation for odd genus opposite to that which is has for even genus – thus,the angle at P g in Ω Gdh will always be 3 π/
2, independently of g . See Figure 4.1.Now let’s introduce the cycles formally.Let c i denote the cycle in a domain which encircles the segment P i − P i ; here i rangesfrom 1 to g −
1, and from g + 2 to 2 g . In addition, let δ connect the segment E P to thesegment P g E . This last segment δ is loosely analogous in its design and purpose to thearc we used in the second proof of the existence of the Karcher surface S .We group these cycles in pairs symmetric with respect to the y = − x diagonal and alsorequire that the cycles are symmetric themselves:To this end, set γ i = c i + c g +1 − i , i = 1 , . . . , g − . ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 21
P0 P1 P2gPgV- E2 E2E1 E1 P0P1P2g Pg V+|G dh||1/G dh|
Figure 4.1.
Geometric CoordinatesThese cycles will detect degeneracies on the boundary with many finite vertices, while δ detects degeneration of the pair of boundaries in Ω Gdh .We next use these cycles to define a proper height function on the moduli space ∆ g ofpairs of orthodisks. Note that dim ∆ g = g , so we are using g cycles. Definition 4.2.
The height for the E g surface is defined as H = g − (cid:88) i =1 H ( γ i ) + H ( δ ) Lemma 4.3. If H = 0 , the two orthodisks are reflexive, i.e. there is a vertex preservingconformal map between them.Proof. Map the Ω
Gdh orthodisk conformally to the upper half plane H so that P g is mappedto 0, and V + to ∞ . As the domain Ω Gdh is symmetric about a diagonal line connecting P g with V + , our mapping is equivariant with respect to that symmetry and the reflection in H about the imaginary axis — in particular, E is taken to 1, while E is taken to −
1. Thevertices P j , E k , V ± are mapped to points ˜ P j , ˜ E k , ˜ V ± ∈ R and the cycles γ j are carried tocycles in the upper half plane which are symmetric with respect to reflection in the y − axis.Now, note that if the height H vanishes, then so do each of the terms H ( γ i ) and H ( δ ).Thus the corresponding extremal lengths ext Ω Gdh (Γ) and ext Ω G − dh (Γ) agree on the curvesΓ = δ, γ , . . . , γ g − . It is thus enough to show that that set { ext Ω Gdh ( γ ) , . . . ext Ω Gdh ( γ g − ) , ext Ω Gdh ( δ ) } of extremal lengths determines the conformal structure of Ω Gdh , or equivalently in thiscase of a planar domain Ω
Gdh , the positions of the distinguished points { ˜ P j , ˜ e k , ˜ V ± } on theboundary of the image H . Now, ˜ P = − ˜ P g , and as ext Ω Gdh ( δ ) is monotone in the position of ˜ P = − ˜ P g (havingfixed ˜ e = 1 and ˜ e = − Ω Gdh ( δ ) determines the position of ˜ P and˜ P g = − ˜ P . Next we regard ˜ P as a variable, with the positions of ˜ P , . . . , ˜ P g − dependingon ˜ P . The point is that any choice of ˜ P , together with the datum ext Ω Gdh ( γ ) uniquelydetermines a corresponding position of ˜ P ; moreover, as our choice of ˜ P tends to − P also tends to −
1, and as our choice of ˜ P tends to 0,the correspondingly determined ˜ P pushes towards 0. Thus since we know that there isat least one choice of points { ˜ P j , ˜ E k , ˜ V ± } on the boundary of the image H for which theextremal lengths will agree for corresponding curve systems, we see there is a range ofpossible values in ( − ,
0) for the position of ˜ P , each uniquely determining a position of ˜ P in ( ˜ P , P and ˜ P , the extremal length ext Ω Gdh ( γ )uniquely determines a value for ˜ P in ( ˜ P , P j − and ˜ P j and the datum ext Ω Gdh ( γ j ) to determine ˜ P j +1 . In the end, we have, for eachchoice of ˜ P , a sequence of uniquely determined positions ˜ P , . . . , ˜ P g − , with the positionsof all the determined points depending monotonically on the choice of ˜ P . Of course thepositions of ˜ P g − , ˜ P g − , ˜ P g − and ˜ P g = 0 determine the value ext Ω Gdh ( γ g − ), which is partof the data. By the monotonicity of the dependence of the choice of positions ˜ P , . . . , ˜ P g − on the choice of ˜ P , we see that the choice of ˜ P , and hence all of the values, is uniquelydetermined.Thus all of the distinguished points on the boundary of H are determined, hence so isthe conformal structure of Ω Gdh . (cid:3) As we clearly have that
H ≥
0, we see that our task in the next few sections is to findzeroes of H . This we accomplish, in some sense, by flowing down −∇H along a nice locus Y ⊂ ∆ g avoiding both critical points and a neighborhood of ∂ ∆ g .An essential property of the height is its analyticity: Proposition 4.4.
The height function is a real analytic function on ∆ g .Proof. The height is an analytic expression in extremal lengths of cycles connecting edges ofpolygons. That these are real analytic, follows by applying the Schwarz-Christoffel formulatwice: first to map the polygon conformally to the upper half plane, and second to mapthe upper half plane to a rectangle so that the edges the cycle connects become paralleledges of the rectangle. Then it follows that the modulus of the rectangle depends realanalytically on the geometric coordinates of the orthodisks. (cid:3)
The properness of the height function.Theorem 4.5.
The height function is proper on the space of geometric coordinates.
The proof is based on the following fundamental principle we have used for the identicalpurpose in [28] and [29].
Theorem 4.6.
Let c be a cycle as above. Consider a sequence of pairs of conjugateorthodisks Ω Gdh and Ω G − dh indexed by a parameter n such that either c encircles an edge ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 23 shrinking geometrically to zero and both ext Ω Gdh ( γ ) → and ext Ω G − dh ( γ ) → or c foots onan edge shrinking geometrically to zero and both ext Ω Gdh ( γ ) → ∞ and ext Ω G − dh ( γ ) → ∞ .Then H ( c ) → ∞ as n → ∞ . We postpone the proof of this theorem until after the proof of Theorem 4.5.
Proof of Theorem 4.5.
To show that the height functions from section 4.1 are proper,we need to prove that for any sequence of points in ∆ converging to some boundary point,at least one of the terms in the height function goes to infinity. The idea is as follows.By the completeness of the geometric coordinate system (Theorem 3.2), at least one ofthe two orthodisks degenerates conformally. We will now analyze those possible geometricdegenerations.Begin by observing that we may normalize the geometric coordinates such that theboundary of Ω
Gdh containing the vertices { P i } has fixed ‘total length’ 1 between P and P g , i.e. the sum of the Euclidean lengths of the finite length edges is 1. If the geometricdegeneration involves degeneration in this outer boundary component of ∂ Ω Gdh , then oneof the cycles γ j that either encircles or ends on an edge (or in the case where P g − , P g and P g +1 coalesce, a pair of edges) must shrink to zero. By the Monodromy Theorem 4.6, thecorresponding term of the height function goes to infinity, and we are done.Alternatively, if there is no geometric degeneration on the boundary component of Ω Gdh containing the vertices { P i } , then the degeneration must come from the vertex V + eitherlimiting on P g , or tending to infinity. In the first case, as in our discussion of the extremallength geometry behind Karcher’s surface, this then forces the extremal length ext Ω Gdh ( δ )to go to ∞ , while, in the dual orthodisk, no degeneration is occuring and ext Ω G − dh ( δ ) isconverging to a positive value. Naturally, this also sends the corresponding term H ( δ ) to ∞ .In the latter case of V + tending to infinity, and no other degeneration on ∂ Ω Gdh , itis convenient to adopt a different normalization: for this case, we set d ( P g , V + ) = 1.This forces all V , . . . , V g to coalesce simultaneously. Then the argument proceeds quiteanalogously to the argument we gave in section 3 for the existence of Karcher’s surface.In particular, the present case follows directly from that case, once we take into account awell-known background fact. Claim:
Let Ω ⊂ Ω (cid:48) , let Γ be a curve system for Ω and let Γ (cid:48) be a curve system for Ω (cid:48) .Suppose that Γ ⊂ Γ (cid:48) . Then ext Ω (Γ) ≥ ext Ω (cid:48) (Γ (cid:48) ). Proof of Claim:
Any candidate metric ρ (cid:48) for ext Ω (cid:48) (Γ (cid:48) ) restricts to a metric ρ forext Ω (Γ). The minimum length of elements of Γ in this restricted metric is at least as largeas the minimum length of Γ (cid:48) ⊃ Γ in the extended metric; moreover, the area of the metricrestricted to Ω is no larger than that of the ρ (cid:48) -area of Ω (cid:48) ⊃ Ω. Thus (cid:96) ρ (Γ)Area( ρ ) ≥ (cid:96) ρ (cid:48) (Γ (cid:48) )Area( ρ (cid:48) ) . The claim follows by comparing these ratios for an extremizing sequence ρ (cid:48) n for ext Ω (cid:48) (Γ (cid:48) ).Then observe that the orthodisk Ω Gdh for S g sits strictly outside the orthodisk Ω Gdh for S , where here we compare corresponding orthodisks whose first and last vertices ( P and P g ) agree, while P g for S is constructed using the existing geometric data. (See Figure4.2.) Thus the extremal length, say ext g Ω Gdh ( δ ), for the curve δ in the genus g version ofthe domain Ω Gdh , is less than the genus one version ext Gdh ( δ ) of the extremal length of δ for that domain, i.e. ext g Ω Gdh ( δ ) ≤ ext Gdh ( δ ). P0 P1 P2gPgV- E2 E2E1 E1 P0P1P2g Pg V+ |G dh||1/G dh| δ δ
Figure 4.2.
Orthodisk comparisonOn the other hand, the corresponding orthodisk Ω G − dh for S g sits strictly inside thecorresponding orthodisk Ω G − dh for S , using the standard correspondence of Ω Gdh andΩ G − dh orthodisks. Observe that for the V i close enough together, the vertex V of S willlie outside that of Ω G − dh of S g . Thus ext G − dh ( δ ) ≤ ext G − dh ( δ ).Thus because we have ext G − dh ( δ ) > ext Gdh ( δ ) for the case of S (see (2.6)), withboth quantities tending to zero (at different rates), the claim implies that we have theanalogous inequality ext g Ω G − dh ( δ ) >> ext g Ω Gdh ( δ ) holding for S g . Moreover, the claim (andthe notation) also implies that ext Ω Gdh ( δ ) tends to zero at a rate distinct from that ofext Ω G − dh ( δ ). Thus the height function H ( δ ) in such a case tends to infinity.There is one final case to consider, which is hidden a bit because of our usual choiceof conventions: it is only here that this normalizing of notation can be misleading. Theissue is that, in Figure 4.2 for instance, the angle at P g and the angle of V − are both π/ Gdh , and the angles are 3 π/ P g and V + in Ω G − dh . However, we of courseneed to consider degenerations when the corresponding angles do not agree, for examplewhen the angle at P g in Ω Gdh is 3 π/ V − (also) in Ω Gdh is π/
2. [In thatsituation, we will also be in the situation where the angle at P g in Ω G − dh is π/ V + in Ω G − dh is 3 π/ V − in Ω Gdh is π/ P g in Ω Gdh is 3 π/ ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 25 imagine ‘cutting a notch out of Ω G − dh ’ near P g : more precisely, replace a neighborhood of ∂ Ω Gdh near P g with three vertices P ∗ g − , P ∗ g , P ∗ g +1 and edges between them that alternate π/ π/ ∗ Gdh for a surface ofquotient genus g + 1, where the angle at P ∗ g is now π/
2, now equaling the angle at V − opposite P ∗ g . Of course, this notch-cutting also determines a conjugate domain Ω ∗ G − dh ,where the angle at the (new) central point P ∗ g is now 3 π/
2, also equaling the angle at V + opposite it. Thus, in considering the domains Ω ∗ Gdh and Ω ∗ G − dh , we have returned to thethird case we just finished considering. Fortunately, the comparisons between the extremallengths on Ω Gdh and Ω ∗ Gdh and those between Ω G − dh and Ω ∗ G − dh allow for us to concludethat ext Ω Gdh ( δ ) >> ext Ω G − dh ( δ ) as follows:ext Ω Gdh ( δ ) ≥ ext ∗ Ω Gdh ( δ ) by the claimed principle ≥ ext G − dh ( δ ) as in the third case >> ext Gdh ( δ ) ≥ ext ∗ Ω G − dh ( δ ) ≥ ext Ω G − dh ( δ ) . This treats the four possible cases, and the theorem is proven. (cid:3)
A monodromy argument.
In this section, we prove that the periods of orthodiskshave incompatible logarithmic singularities in suitable coordinates and apply this to provethe Monodromy Theorem 4.6. The main idea is that to study the dependence of extremallengths of the geometric coordinates, it is necessary to understand the asymptotic depen-dence of extremal lengths of the degenerating conformal polygons (which is classical andwell-known, see [18]), and the asymptotic dependence of the geometric coordinates of thedegenerating conformal polygons. This dependence is given by Schwarz-Christoffel mapswhich are well-studied in many special cases. Moreover, it is known that these maps possessasymptotic expansions in logarithmic terms. Instead of computing this expansion explic-itly for the two maps we need, we use a monodromy argument to show that the cruciallogarithmic terms have a different sign for the two expansions.Let ∆ g be a geometric coordinate domain of dimension g ≥
2, i.e. a simply connecteddomain equipped with defining geometric coordinates for a pair of orthodisks Ω
Gdh andΩ G − dh as usual.Suppose γ is a cycle in the underlying conformal polygon which joins two non-adjacentedges P P with Q Q . Denote by R the vertex before Q and by R the vertex after Q and observe that by assumption, R (cid:54) = P but that we can possibly have P = R .Introduce a second cycle β which connects R Q with Q R .The situation is illustrated in the figure below; we have replaced the labels of P i , V ± and E j that we use for vertices in ∂ Ω Gdh and ∂ Ω G − dh with generic labels of distinguishedpoints on the boundary of the region: these will represent in general the situations that wewould encounter in the orthodisk. Of course, we retain the convention of using the samelabel name for corresponding vertices in ∂ Ω Gdh and ∂ Ω G − dh . P P R Q Q R γ β Figure 4.3.
Monodromy argumentWe formulate the claim of Theorem 4.6 more precisely in the following two lemmas:
Lemma 4.7.
Suppose that for a sequence p n ∈ ∆ with p n → p ∈ ∂ ∆ we have that ext Ω Gdh ( p n ) ( γ ) → and ext Ω G − dh ( p n ) ( γ ) → . Suppose furthermore that γ is a cycleencircling an edge which degenerates geometrically to as n → ∞ . Then (cid:12)(cid:12)(cid:12) e / ext Ω Gdh ( pn ) ( γ ) − e / ext Ω G − dh ( pn ) ( γ ) (cid:12)(cid:12)(cid:12) → ∞ Lemma 4.8.
Suppose that for a sequence p n ∈ ∆ with p n → p ∈ ∂ ∆ we have that ext Ω Gdh ( p n ) ( γ ) → ∞ and ext Ω G − dh ( p n ) ( γ ) → ∞ . Suppose furthermore that γ is a cyclefooting on an edge which degenerates geometrically to as n → ∞ . Then (cid:12)(cid:12)(cid:12) e ext Ω Gdh ( pn ) ( γ ) − e ext Ω G − dh ( pn ) ( γ ) (cid:12)(cid:12)(cid:12) → ∞ Proof.
We first prove Lemma 4.7.Consider the conformal polygons corresponding to the pair of orthodisks. Normalize thepunctures by M¨obius transformations so that P = −∞ , P = 0 , Q = (cid:15), Q = 1for Ω Gdh and P = −∞ , P = 0 , Q = (cid:15) (cid:48) , Q = 1for Ω G − dh .If α is a curve in a domain Ω ⊂ C , then define Per α (Ω) = (cid:82) α dz . Here our focus is onperiods of the one-form dz as we are typically interested in domains Ω which are developedimages of pairs (Ω , ω ) of domains and one-forms on those domains, i.e. z ( p ) = (cid:82) pp ω . Bythe assumption of Lemma 4.7, we know that (cid:15), (cid:15) (cid:48) → n → ∞ .We now allow Q to move in the complex plane and apply the Real Analyticity Alter-native Lemma 4.11 below to the curve (cid:15) = (cid:15) e it : here we are regarding the position of Q as traveling along a small circle around the origin, i.e. its defined position (cid:15) ∈ R has beenextended to allow complex values. We will conclude from that lemma that either(4.1) | Per γ (Ω Gdh ) || Per β (Ω Gdh ) | + 1 π log (cid:15) =: F ( (cid:15) ) ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 27 is single-valued in (cid:15) and(4.2) | Per γ (Ω G − dh ) || Per β (Ω G − dh ) | − π log (cid:15) (cid:48) =: F ( (cid:15) (cid:48) ) = F ( (cid:15) (cid:48) ( (cid:15) ))is single-valued in (cid:15) (cid:48) or vice versa, with signs exchanged. Without loss of generality, wecan treat the first case.Now suppose that (cid:15) (cid:48) is real analytic (and hence single-valued) in (cid:15) and comparable to (cid:15) near (cid:15) = 0. Then using that Ω Gdh and Ω G − dh are conjugate implies that | Per γ (Ω Gdh ) || Per β (Ω Gdh ) | = | Per γ (Ω G − dh ) || Per β (Ω G − dh ) | . By subtracting the function F ( (cid:15) ) in 4.1 from the function F ( (cid:15) (cid:48) ) in 4.2 (both of which aresingle-valued in (cid:15) ) we get that log( (cid:15)(cid:15) (cid:48) ( (cid:15) ))is single-valued in (cid:15) near (cid:15) = 0 which contradicts that (cid:15), (cid:15) (cid:48) → Gdh ( p n ) we have ext( γ ) = O (log | (cid:15) | )(see [18]). We conclude that | e / ext Ω Gdh ( pn ) ( γ ) − e / ext Ω G − dh ( pn ) ( γ ) | = O (cid:18) (cid:15) − (cid:15) (cid:48) (cid:19) which goes to infinity, since we have shown that (cid:15) and (cid:15) (cid:48) tend to zero at different rates.This proves Lemma 4.7.The proof of Lemma 4.8 is very similar: For convenience, we normalize the points of thepunctured disks such that P = −∞ , P = 0 , Q = 1 , Q = 1 + (cid:15) for Ω Gdh and P = −∞ , P = 0 , Q = 1 , Q = 1 + (cid:15) (cid:48) for Ω G − dh .By the assumption of Lemma 4.8, we know that (cid:15), (cid:15) (cid:48) → n → ∞ . We now apply theReal Analyticity Alternative Lemma 4.11 below to the curve 1 + (cid:15) e it and conclude thatPer γ (Ω Gdh )Per β (Ω Gdh ) + 1 π log (cid:15) is single-valued in (cid:15) while Per γ (Ω G − dh )Per β (Ω G − dh ) − π log (cid:15) (cid:48) is single-valued in (cid:15) (cid:48) . The rest of the proof is identical to the proof of Lemma 4.7. (cid:3) To prove the needed Real Analyticity Alternative Lemma 4.11, we need asymptoticexpansions of the extremal length in terms of the geometric coordinates of the orthodisks.Though not much is known explicitly about extremal lengths in general, for the chosencycles we can reduce this problem to an asymptotic control of Schwarz-Christoffel integrals.Their monodromy properties allow us to distinguish their asymptotic behavior by the signof logarithmic terms.We introduce some notation: suppose we have an orthodisk such that the angles at thevertices alternate between π/ − π/ π . (We will also allow some angles tobe 0 modulo 2 π but they will not be relevant for this argument.) Consider the Schwarz-Christoffel map F : z (cid:55)→ (cid:90) z ( t − t ) a / · . . . · ( t − t n ) a n / dt from a conformal polygon with vertices at t i to this orthodisk: here the exponents a j alternate between − π/ − π/ , (mod 2 π ), respectively. Choose four distinct vertices t i , t i +1 , t j , t j +1 (not necessarilyconsecutive). Introduce a cycle γ in the upper half plane connecting edge ( t i , t i +1 ) withedge ( t j , t j +1 ) and denote by ¯ γ the closed cycle obtained from γ and its mirror image atthe real axis. Similarly, denote by β the cycle connecting ( t j − , t j ) with ( t j +1 , t j +2 ) and by¯ β the cycle together with its mirror image.Now consider the Schwarz-Christoffel period integrals F ( γ ) = 12 (cid:90) ¯ γ ( t − t ) a / · . . . · ( t − t n ) a n / dtF ( β ) = 12 (cid:90) ¯ β ( t − t ) a / · . . . · ( t − t n ) a n / dt as multivalued functions depending on the now complex parameters t i . Lemma 4.9.
Under analytic continuation of t j +1 around t j the periods change their valueslike F ( γ ) → F ( γ ) + 2 F ( β ) F ( β ) → F ( β ) Proof.
The path of analytic continuation of t j +1 around t j gives rise to an isotopy of C which moves t j +1 along this path. This isotopy drags β and γ to new cycles β (cid:48) and γ (cid:48) .Because the curve β is defined to surround t j and t j +1 , the analytic continuation merelyreturns β to β (cid:48) . Thus, because β (cid:48) equals β , their periods are also equal. On the other hand,the curve γ is not equal to γ (cid:48) : informally, γ (cid:48) is obtained as the Dehn twist of γ around ¯ β .Now, the period of γ (cid:48) is obtained by developing the flat structure of the doubled orthodiskalong γ (cid:48) . To compute this flat structure, observe the crucial fact that the angles at theorthodisk vertices are either π/ − π/
2, modulo 2 π . In either case, we see from thedeveloped flat structure that the period of γ (cid:48) equals the period of γ plus twice the periodof β . (cid:3) ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 29
Now denote by δ := t j +1 − t j and fix all t i other than t j +1 : we regard t j +1 as theindependent variable, here viewed as complex, since we are allowing it to travel around t j . Lemma 4.10 (Analyticity Lemma) . The function F ( γ ) − log δπi F ( β ) is single-valued andholomorphic in δ in a neighborhood of δ = 0 .Proof. By definition,the function is locally holomorphic in a punctured neighborhood of δ = 0. By Lemma 4.9 it extends to be single valued in a (full) neighborhood of δ = 0. (cid:3) We will now specialize this picture to the situation at hand — an orthodisk where γ represents one of the distinguished cycles γ i . Then F ( γ ) and F ( β ) are either real orimaginary, and are perpendicular. Thus Lemma 4.10 implies that | F ( γ ) | ± log δπ | F ( β ) | isreal analytic in δ with one choice of sign. The crucial observation is now that whateveralternative holds, the opposite alternative will hold for the conjugate orthodisk. Moreprecisely:Let F and F be the Schwarz-Christoffel maps associated to a pair of conjugate or-thodisks. These will be defined on different but consistently labeled punctured upper halfplanes. Let δ i refer to the complex parameter δ introduced above for the maps F i , respec-tively. Then Lemma 4.11 (Real Analyticity Alternative Lemma) . Either | F ( γ ) | − log δ π | F ( β ) | or | F ( γ ) | + log δ π | F ( β ) | is real analytic in δ for δ = 0 . In the first case, | F ( γ ) | + log δ π | F ( β ) | is real analytic in δ , while in the second case, | F ( γ ) | − log δπ | F ( β ) | is real-analytic in δ .Proof. We have already noted that either alternative holds in both cases. It remains to showthat it holds with opposite signs. For some special values δ , δ >
0, the two orthodisks areconjugate. For instance, we can assume that for these values, F ( γ ) = F ( γ ) >
0. Then F ( β ) and F ( β ) are both imaginary with opposite signs, and the claimed alternative holdsfor these values of δ , δ . By continuity, the alternative holds for all δ and δ . (cid:3) Remark 4.12.
A concrete way of understanding the phenomenon here is that the asymp-totic expansion of the period of a curve meeting a degenerating cycle β , where the edge for β has preimages b and b + (cid:15) , has a term of the form ± (cid:15) k log (cid:15) , where the sign relates to thegeometry of the orthodisk. 5. The Flow to a Solution
The last part of the proof of the Main Theorem requires us to prove the
Lemma 5.1 (Regeneration Lemma) . There is, for a given genus g , a certain (good) locus Y ⊂ ∆ g in the space ∆ g of geometric coordinates for with the following properties: • Y lies properly within the space of geometric coordinates; • if d H = 0 at a point on the locus Y , then actually H = 0 at that point. This locus will be defined by the requirement that all but one of the extremal lengths ofthe distinguished cycles of the
Gdh and G dh orthodisks are equal. Overall Strategy.
In this section we continue the proof of the existence of the sur-faces { S g } . In the previous sections, we defined an associated moduli space ∆ = ∆ g of pairsof conformal structures { Ω Gdh , Ω G − dh } equipped with geometric coordinates (cid:126) t = ( t i , ..., t g ).We defined a height function H on the moduli space ∆ and proved that it was a properfunction: as a result, there is a critical point for the height function in ∆, and our overallgoal in the next pair of sections is a proof that this critical point represents a reflexiveorthodisk system in ∆, and hence, by our fundamental translation of the period problemfor minimal surfaces into a conformal equivalence problem, a minimal surface of the form S g . Our goal in the present section is a description of the tangent space to the the modulispace ∆: we wish to display how infinitesimal changes in the geometric coordinates (cid:126) t affectthe height function. In particular, it would certainly be sufficient for our purposes to provethe statement Model 5.2. If (cid:126) t is not a reflexive orthodisk system, then there is an element V of thetangent space T (cid:126) t ∆ for which D V H (cid:54) = 0.This would then have the effect of proving that our critical point for the height functionis reflexive, concluding the existence parts of the proofs of the main theorem.We do not know how to prove or disprove this model statement in its full generality.On the other hand, it is not necessary for the proofs of the main theorems that we do so.Instead we will replace this statement by a pair of lemmas.
Lemma 5.3.
Let
Y ⊂ ∆ is a real one-dimensional subspace of ∆ which is defined by theequations H ( γ ) = H ( γ ) = ... = H ( γ g − ) = H ( δ ) = 0 . If (cid:126) t ∈ Y has positive height, i.e. H ( (cid:126) t ) > , then there is an element V of the tangent space T (cid:126) t Y for which D V H (cid:54) = 0 . Lemma 5.4.
There is an analytic subspace
Y ⊂ ∆ = ∆ g , for which Y = {H ( γ ) = H ( γ ) = ... = H ( γ g − ) = H ( δ ) = 0 } . Given these lemmas, the proof of the existence of a pair of conformal orthodisks { Ω Gdh , Ω G − dh } is straightforward. Proof.
Proof of Existence of Reflexive Orthodisks. Consider the locus Y guaranteed byLemma 5.4. By Theorem 4.5, the height function H is proper on Y , so the height function H | Y has a critical point (on Y ). By Lemma 5.3, this critical point represents a point of H = 0, i.e a reflexive orthodisk by Lemma 4.3. (cid:3) The proof of Lemma 5.3 occupies the current section while the proof of Lemma 5.4 isgiven in the following section.
Remarks on deformations of conjugate pairs of orthodisks.
Let us discuss informallythe proof of Lemma 5.3. Because angles of corresponding vertices in the Ω
Gdh ↔ Ω G − dh correspondence sum to 0 (mod 2 π ), the orthodisks fit together along corresponding edges,so conjugacy of orthodisks requires corresponding edges to move in different directions: ifthe edge E on Ω Gdh moves “out,” the corresponding edge E ∗ on Ω G − dh edge moves “in”,and vice versa (see the figures below). Thus we expect that if γ has an endpoint on E ,then one of the extremal lengths of γ decreases, while the other extremal length of γ on ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 31 the other orthodisk would increase: this will force the height H ( γ ) of γ to have a definitesign, as desired. This is the intuition behind Lemma 5.3; a rigorous argument requires usto actually compute derivatives of relevant extremal lengths using the formula 2.3. We dothis by displaying, fairly explicitly, the deformations of the orthodisks (in local coordinateson Ω Gdh / Ω G − dh ) as well as the differentials of extremal lengths, also in coordinates. Aftersome preliminary notational description in section 5.2, we do most of the computing insection 5.3. Also in section 5.3 is the key technical lemma, which relates the formalism offormula 2.3, together with the local coordinate descriptions of its terms, to the intuitionwe just described.5.2. Infinitesimal pushes.
We need to formalize the previous discussion. As always weare concerned with relating the Euclidean geometry of the orthodisks (which correspondsdirectly with the periods of the Weierstrass data) to the conformal data of the domainsΩ
Gdh and Ω G − dh . From the discussion above, it is clear that the allowable infinitesimalmotions in ∆, which are parametrized in terms of the Euclidean geometry of Ω Gdh andΩ G − dh , are given by infinitesimal changes in lengths of finite sides, with the changesbeing done simultaneously on Ω Gdh and Ω G − dh to preserve conjugacy. The link to theconformal geometry is the formula 2.3: a motion which infinitesimally transforms Ω Gdh ,say, will produce an infinitesimal change in the conformal structure. This tangent vector tothe moduli space of conformal structures is represented by a Beltrami differential. Later,formula 2.3 will be used, together with knowledge of the cotangent vectors d ext Ω Gdh ( · )and d ext Ω G − dh ( · ), to determine the derivatives of the relevant extremal lengths, hence thederivative of the height.To begin, we explicitly compute the effect of infinitesimal pushes of certain edges on theextremal lengths of relevant cycles. This is done by explicitly displaying the infinitesimaldeformation and then using this formula to compute the sign of the derivative of theextremal lengths, using formula 2.3. There will be two different cases to consider.Case A. Finite non-central edges of the type P i − P i for i < g .Case B. An edge (finite or infinite) and its symmetric side meet in a corner, for instance P g − P g .For each case there are two subcases, which we can describe as depending on whetherthe given sides are horizontal or vertical. The distinction is, surprisingly, a bit important,as together with the fact that we do our deformations in pairs, it provides for an importantcancelation of (possibly) singular terms in Lemma 5.5. We defer this point for later, whilehere we begin to calculate the relevant Beltrami differentials in the cases.While logically it is conceivable that each infinitesimal motion might require two differenttypes of cases, depending on whether the edge we are deforming on Ω Gdh corresponds onΩ G − dh to an edge of the same type or a different type, in fact this issue does not arise forthe particular case of the Scherk surfaces we are discussing in this paper. By contrast, itdoes arise for the generalized Costa surfaces we discussed in [29].Case A. Here the computations are quite analogous to those that we found in [28];they differ only in orientation of the boundary of the orthodisk. We include them for thecompleteness of the exposition. R3R5 R2 R6R4R1 R3*R1*R4* R6*R2*R5*a b
Figure 5.1.
Beltrami differential computation — case AWe first consider the case of a horizontal finite side; as in the figure above, we see that theneighborhood of the horizontal side of the orthodisk in the plane naturally divides into sixregions which we label R ,..., R . Our deformation f (cid:15) = f (cid:15),b,δ differs from the identity onlyin such a neighborhood, and in each of the six regions, the map is affine. In fact we have atwo-parameter family of these deformations, all of which have the same infinitesimal effect,with the parameters b and δ depending on the dimensions of the supporting neighborhood.(5.1) f (cid:15) ( x, y ) = (cid:0) x, (cid:15) + b − (cid:15)b y (cid:1) , {− a ≤ x ≤ a, ≤ y ≤ b } = R (cid:0) x, (cid:15) + b + (cid:15)b y (cid:1) , {− a ≤ x ≤ a, − b ≤ y ≤ } = R (cid:18) x, y + (cid:15) + b − (cid:15)b y − yδ ( x + δ + a ) (cid:19) , {− a − δ ≤ x ≤ − a, ≤ y ≤ b } = R (cid:18) x, y − (cid:15) + b − (cid:15)b y − yδ ( x − δ − a ) (cid:19) , { a ≤ x ≤ a + δ, ≤ y ≤ b } = R (cid:18) x, y + (cid:15) + b + (cid:15)b y − yδ ( x + δ + a ) (cid:19) , {− a − δ ≤ x ≤ − a, − b ≤ y ≤ } = R (cid:18) x, y − (cid:15) + b + (cid:15)b y − yδ ( x − δ − a ) (cid:19) , { a ≤ x ≤ a + δ, − b ≤ y ≤ } = R ( x, y ) otherwisewhere we have defined the regions R , . . . , R within the definition of f (cid:15) . Also note thathere the orthodisk contains the arc { ( − a, y ) | ≤ y ≤ b } ∪ { ( x, | − a ≤ x ≤ a } ∪ { ( a, y ) |− b ≤ y ≤ } . Let E denote the edge being pushed, defined above as [ − a, a ] × { } .Of course f (cid:15) differs from the identity only on a neighborhood of the edge E , so that f (cid:15) takes the symmetric orthodisk to an asymmetric orthodisk. We next modify f (cid:15) in aneighborhood of the reflected (across the y = − x line) segment E ∗ in an analogous waywith a map f ∗ (cid:15) so that f ∗ (cid:15) ◦ f (cid:15) will preserve the symmetry of the orthodisk.Our present conventions are that the edge E is horizontal; this forces E ∗ to be verticaland we now write down f ∗ (cid:15) for such a vertical segment; this is a straightforward extension ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 33 of the description of f (cid:15) for a horizontal side, but we present the definition of f ∗ (cid:15) anyway, aswe are crucially interested in the signs of the terms. So set(5.2) f ∗ (cid:15) = (cid:0) − (cid:15) + b − (cid:15)b x, y (cid:1) , {− b ≤ x ≤ , − a ≤ y ≤ a } = R ∗ (cid:0) − (cid:15) + b + (cid:15)b x, y (cid:1) , { ≤ x ≤ b, − a ≤ y ≤ a } = R ∗ (cid:18) x − − (cid:15) + b − (cid:15)b x − xδ ( y − δ − a ) , y (cid:19) , {− b ≤ x ≤ , a ≤ y ≤ a + δ } = R ∗ (cid:18) x + − (cid:15) + b − (cid:15)b x − xδ ( y + δ + a ) , y (cid:19) , {− b ≤ x ≤ , − a − δ ≤ y ≤ − a } = R ∗ (cid:18) x − − (cid:15) + b + (cid:15)b x − xδ ( y − δ − a ) , y (cid:19) , { ≤ x ≤ b, a ≤ y ≤ a + δ } = R ∗ (cid:18) x + − (cid:15) + b + (cid:15)b x − xδ ( y + δ + a ) , y (cid:19) , { ≤ x ≤ b, − a − δ ≤ y ≤ − a } = R ∗ ( x, y ) otherwiseNote that under the reflection across the line { y = − x } , the region R i gets taken to theregion R ∗ i .Let ν (cid:15) = ( f (cid:15) ) ¯ z ( f (cid:15) ) z denote the Beltrami differential of f (cid:15) , and set ˙ ν = dd(cid:15) (cid:12)(cid:12) (cid:15) =0 ν (cid:15) . Similarly,let ν ∗ (cid:15) denote the Beltrami differential of f ∗ (cid:15) , and set ˙ ν ∗ = dd(cid:15) (cid:12)(cid:12) (cid:15) =0 ν ∗ (cid:15) . Let ˙ µ = ˙ ν + ˙ ν ∗ . Now˙ µ is a Beltrami differential supported in a bounded domain in one of the domains Ω Gdh orΩ G − dh . We begin by observing that it is easy to compute that ˙ ν = [ dd(cid:15) (cid:12)(cid:12) (cid:15) =0 ( f (cid:15) )] ¯ z evaluatesnear E to(5.3)˙ ν = b , z ∈ R − b , z ∈ R b [ x + δ + a ] /δ + i (1 − y/b ) δ = bδ (¯ z + δ + a + ib ) , z ∈ R − b [ x − δ − a ] /δ − i (1 − y/b ) δ = bδ ( − ¯ z + δ + a − ib ) , z ∈ R − b [ x + δd + a ] /δ + i (1 + y/b ) δ = bδ ( − ¯ z − δ − a + ib ) , z ∈ R b [ x − δ − a ] /δ − i (1 + y/b ) δ = bδ (¯ z − δ − a − ib ) , z ∈ R z / ∈ supp( f (cid:15) − id)We further compute(5.4) ˙ ν ∗ = − b , R ∗ b , R ∗ bδ ( i ¯ z − δ − a + bi ) R ∗ bδ ( − i ¯ z − δ − a − bi ) R ∗ bδ ( − i ¯ z + δ + a + bi ) R ∗ bδ ( i ¯ z + δ + a − bi ) R ∗
64 MATTHIAS WEBER AND M. WOLF
Case B. We have separated this case out for purely expositional reasons. We can imaginethat the infinitesimal push that moves the pair of consecutive sides along the symmetryline { y = − x } is the result of a composition of a pair of pushes from Case A, i.e. ourdiffeomorphism F (cid:15) ; b,δ can be written F (cid:15) ; b,δ = f (cid:15) ◦ f ∗ (cid:15) , where the maps differ from theidentity in the union of the supports of ˙ ν b,δ and ˙ ν ∗ b,δ .It is an easy consequence of the chain rule applied to this formula for F (cid:15) ; b,δ that theinfinitesimal Beltrami differential for this deformation is the sum ˙ ν b,δ + ˙ ν ∗ b,δ of the infinites-imal Beltrami differentials ˙ ν b,δ and ˙ ν ∗ b,δ defined in formulae 5.3, 5.4 for Case A (even in aneighborhood of the vertex along the diagonal where the supports of the differentials ˙ ν b,δ and ˙ ν ∗ b,δ coincide).5.3. Derivatives of Extremal Lengths.
In this section, we combine the computationsof ˙ ν b,δ with formula 2.3 (and its background in section 2) and some easy observations onthe nature of the quadratic differentials Φ µ = d ext ( · ) ( µ ) (cid:12)(cid:12) · to compute the derivatives ofextremal lengths under our infinitesimal deformations of edge lengths.We begin by recalling some background from section 2. If we are given a curve γ , theextremal length of that curve on an orthodisk, say Ω Gdh , is a real-valued C function on themoduli space of that orthodisk. Its differential is then a holomorphic quadratic differentialΦ γ = d ext ( · ) ( γ ) (cid:12)(cid:12) Ω Gdh on that orthodisk; the horizontal foliation of Φ γ consists of curveswhich connect the same edges in Ω Gdh as γ , since Φ γ is obtained as the pullback of thequadratic differential dz from a rectangle where γ connects the opposite vertical sides.We compute the derivative of the extremal length function using formula 2.3, i.e. (cid:16) d ext · ( γ ) (cid:12)(cid:12) Ω Gdh (cid:17) [ ν ] = 4 Re (cid:90) Ω Gdh Φ γ ν It is here where we find that we can actually compute the sign of the derivative of theextremal lengths, hence the height function, but also encounter a subtle technical problem.The point is that we will discover that just the topology of the curve γ on Ω Gdh willdetermine the sign of the derivative on an edge E , so we will be able to evaluate the signof the integral above, if we shrink the support of the Beltrami differential ˙ ν b,δ to the edgeby sending b, δ to zero. (In particular, the sign of Φ γ depends precisely on whether thefoliation of Φ = Φ γ is parallel or perpendicular to E , and on whether E is horizontal orvertical.) We then need to know two things: 1) that this limit exists, and 2) that we mayknow its sign via examination of the sign of ˙ ν b,δ and Φ γ on the edge E . We phrase this as Lemma 5.5. (1) lim b → ,δ → Re (cid:82) Φ ˙ ν exists, is finite and non-zero. (2) The (horizontal) fo-liation of
Φ = Φ γ is either parallel or orthogonal to the segment which is lim b → ,δ → (supp ˙ ν ) ,and (3) The expression
Ψ ˙ ν has a constant sign on the that segment E , and the integral 2.3also has that (same) sign. Of course, in the statement of the lemma, the horizontal foliation of the holomorphicquadratic differential Φ = Φ γ has regular curves parallel to γ .This lemma provides the rigorous foundation for the intuition described in the finalparagraph of strategy section 5.1. ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 35
Proof of the Technical Lemma 5.5.
Proof.
Let S Gdh denote the double of Ω
Gdh across the boundary; the metric space S Gdh isa flat sphere with conical singularities, two of which are metric cylinders.The foliation of Φ, on say Ω
Gdh , lifts to a foliation on the punctured sphere, symmetricabout the reflection about the equator. This proves the second statement. The thirdstatement follows from the first (and from the above discussion of the topology of thevertical foliation of Φ γ ), once we prove that there is no infinitude of (cid:82) Φ ˙ ν as b, δ → R and R for the finite vertices. This finiteness will follow from the proof of the first statement.Thus, we are left to prove the first statement which requires us once again to consider thecases A and B.Case A: Suppose γ connects two non-central finite edges E (cid:48) and E (cid:48)(cid:48) on Ω Gdh . To un-derstand the singular behavior of Φ = Φ γ near a vertex of the orthodisk, say Ω Gdh , webegin by observing (by formula 2.3) that on a preimage on S Gdh of such a vertex, thelifted quadratic differential, say Ψ, has a simple pole. This is consistent with the natureof the foliation of Ψ, whose non-singular horizontal leaves are all freely homotopic to thelift of γ ; the fact itself follows from following the lift of the canonical quadratic differentialon a rectangle. Thus the singular leaves of Ψ are segments on the equator of the sphereconnecting lifts of endpoints of the edges E (cid:48) and E (cid:48)(cid:48) .Now let ω be a local uniformizing parameter near the preimage of the vertex on S Gdh and ζ a local uniformizing parameter near the vertex of Ω Gdh on C . There are two cases toconsider, depending on whether the angle in Ω Gdh at the vertex is 3 π/ π/
2. In the firstcase, the map from Ω
Gdh to a lift of Ω
Gdh in S Gdh is given in coordinates by ω = ( iζ ) / ,and in the second case by ω = ζ . Thus, in the first case we write Ψ = c dω ω so thatΦ = − / c ( iζ ) − / dζ , and in the second case we write Φ = 4 cdζ ; in both cases, theconstant c is real with sign determined by the direction of the foliation.With these expansions for Φ, we can compute lim b → ,δ → Re (cid:82) Φ ˙ ν .Clearly, as b + δ →
0, as | ˙ ν | = O (cid:0) max (cid:0) b , δ (cid:1)(cid:1) , we need only concern ourselves with thecontribution to the integrals of the singularity at the vertices of Ω Gdh with angle 3 π/ E is horizontal so thatΩ Gdh has a vertex angle of 3 π/ P . This means that Ω Gdh also has avertex angle of 3 π/ P ∗ , on E ∗ . It is convenient to rotate aneighborhood of E ∗ through an angle of − π/ ν is a reflection ofthe support of ˙ ν ∗ (see equation 5.1 through a vertical line. If the coordinates of supp ˙ ν and supp ˙ ν ∗ are z and z ∗ , respectively (with z ( P ) = z ∗ ( P ∗ ) = 0), then the maps which liftneighborhoods of P and P ∗ , respectively, to the sphere S Gdh are given by z (cid:55)→ ( iz ) / = ω and z ∗ (cid:55)→ ( z ∗ ) / = ω ∗ . Now the poles on S Gdh have coefficients c dω ω and − c dω ∗ ω ∗ , respectively, so we find thatwhen we pull back these poles from S Gdh to Ω
Gdh , we have Φ( z ) = − c dz /ω whileΦ( z ∗ ) = − c dz / ( ω ∗ ) in the coordinates z and z ∗ for supp ˙ ν and supp ˙ ν ∗ , respectively.But by tracing through the conformal maps z (cid:55)→ ω (cid:55)→ ω on supp ˙ ν and z ∗ (cid:55)→ ω ∗ (cid:55)→ ( ω ∗ ) , we see that if z ∗ is the reflection of z through a line, then1( ω ( z )) = 1 /ω ∗ ( z ∗ ) so that the coefficients Φ( z ) and Φ( z ∗ ) of Φ = Φ( z ) dz near P and of Φ( z ∗ ) dz ∗ near P ∗ satisfy Φ( z ) = Φ( z ∗ ), at least for the singular part of the coefficient.On the other hand, we can also compute a relationship between the Beltrami coeffi-cients ˙ ν ( z ) and ˙ ν ∗ ( z ∗ ) (in the obvious notation) after we observe that f ∗ (cid:15) ( z ∗ ) = − f (cid:15) ( z ).Differentiating, we find that ˙ ν ∗ ( z ∗ ) = ˙ f ∗ ( z ∗ ) z ∗ = − ˙ f ( z ) z ∗ = ( ˙ f ( z )) z = ˙ f ( z ) z = ˙ ν ( z ) . Combining our computations of Φ( z ∗ ) and ˙ ν ( z ∗ ) and using that the reflection z (cid:55)→ z ∗ reverses orientation, we find that (in the coordinates z ∗ = x ∗ + iy ∗ and z = x + iy ) forsmall neighborhoods N κ ( P ) and N κ ( P ∗ ) of P and P ∗ respectively,Re (cid:90) supp ˙ ν ∩ N κ ( P ) Φ( z ) ˙ ν ( z ) dxdy + Re (cid:90) supp ˙ ν ∗ ∩ N κ ( P ∗ ) Φ( z ∗ ) ˙ ν ( z ∗ ) dx ∗ dy ∗ = Re (cid:90) supp ˙ ν ∩ N κ ( P ) Φ( z ) ˙ ν ( z ) − Φ( z ∗ ) ˙ ν ( z ∗ ) dxdy = Re (cid:90) supp ˙ ν ∩ N κ Φ( z ) ˙ ν ( z ) − [Φ( z ) + O (1)] ˙ ν ( z ) dxdy = O ( b + δ )the last part following from the singular coefficients summing to a purely imaginary termwhile ˙ ν = O (cid:0) b + δ (cid:1) , and the neighborhood has area bδ . This concludes the proof of thelemma for this case.Case B. Here we need only consider the singularities resulting at the origin, as wetreated the other singularities in Case A. The lemma in this case follows from a pairof observations. First, because of the symmetry across the line through the vertex underdiscussion, the differential Ψ (the lift of Φ) on the sphere is holomorphic, and so the behaviorof Φ near the vertex is at least as regular as in the previous cases. Moreover, becausethe infinitesimal Beltrami differential in this case is the sum of infinitesimal Beltramidifferentials encountered in the previous cases A, the arguments there on the cancellationof the apparent singularities of the sum ˙ ν b,δ + ˙ ν ∗ b,δ continue to hold here for the singlesingularity. ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 37
This concludes the proof of Lemma 5.5. (cid:3)
Proof.
Conclusion of the proof of Lemma 5.3. Conjugacy of the domains Ω
Gdh and Ω G − dh allows that the there is a Euclidean motion which glues the domains Ω Gdh and Ω G − dh together through identifying the side P i P i +1 with P g − i − P g − i : this is evident from theconstruction and is illustrated in Figure 4.1. Thus, if we push an edge E ⊂ ∂ Ω Gdh intothe domain Ω
Gdh , we will change the Euclidean geometry of that domain in ways that willforce us to push the corresponding edge E ∗ ⊂ ∂ Ω G − dh out of the domain Ω G − dh .Now, given this geometry of the glued complex D = Ω Gdh ∪ Ω G − dh , we observe thatwe can reduce the term H ( γ g − ) of the height function H by an infinitesimal push on theedges meeting the boundary of γ g − . Moreover, because the rest of the terms of the heightfunction H vanish along the locus Y to second order in the deformation variable, we seethat any deformation of the orthodisk will not alter (infinitesimally) the contribution ofthese terms to H . Thus the only effect of an infinitesimal deformation of an orthodisksystem on Y to the height function H is to the term H ( γ g − ), which is non-zero to firstorder by Lemma 5.5. This concludes the proof of the lemma. (cid:3) Regeneration.
In the previous section we showed how we might reduce the heightfunction H at a critical point of a locus Y , where the locus Y was defined as the nulllocus of all but one of the heights H ( γ g − ). In this section, we prove Lemma 5.4, whichguarantees the existence of such a locus Y .Let us review the context for this argument. Basically, we will prove the existence of thegenus g Scherk surface, S g , by using the existence of the genus g − S g − ,to imply the existence of a locus Y ⊂ ∆ g — the Lemma 5.3 and the Properness Theorem4.5 then prove the existence of S g .Indeed, our proof of the main theorem is by induction: we make the Inductive Assumption A:
There exists a genus g − S g − .Thus, all of our surfaces are produced from only slightly less complicated surfaces; thisis the general principle of ’handle addition’ referred to in the title.For concreteness and ease of notation, we will prove the existence of S assuming theexistence of S . The general case follows with only more notation. Thus, our present goalis the proof of Theorem 5.6.
There is a reflexive orthodisk system for the configuration S .Proof. Let us use the given height H for S and consider how the height H for S relatesto it, near a solution for the genus 2 problem.Our notation is given in section 4.1 and is recorded in the diagrams below: for instance,the curve system δ connects the edges E P and P E .We are interested in how an orthodisk system might degenerate. One such degenerationis shown in the next figure, where the points P , P , and P have coalesced. The pointis that the degenerating family of (pairs of) Riemann surfaces in ∆ limits on (a pairof) surfaces with nodes. (We recall that a surface with nodes is a complex space whereevery point has a neighborhood complex isomorphic to either the disk {| z | < } or a pair P V- V+e e P P P P P P δ δ Figure 5.2.
Curve system used for regenerationof disks { ( z, w ) | zw = 0 } in C .) In the case of the surfaces corresponding to Ω Gdh andΩ G − dh , the components of the noded surface (i.e. the regular components of the nodedsurface in the complement of the nodes) are difficult to observe, as the flat structures onthe thrice-punctured sphere components are simply single points.An important issue in this section is that some of our curves cross the pinching locuson the surface, i.e. the curve on the surface which is being collapsed to form the node. Inparticular, in the diagram, the dotted curves γ are such curves, so their depiction in thedegenerated figure is, well, degenerate: the curves connect a point and an edge.Note that when we degenerate, we are left with the orthodisks for the surface of onelower genus, in this case that of S .Our basic approach is to work backwards from this understanding of degeneration — weaim to “regenerate” the locus Y in ∆ from the solution X ∈ ∆ ⊂ ∂ ¯∆ .We focus on the curves δ and γ , ignoring the degenerate curve γ .(In the general case for ∆ g , there are g − { δ, γ , ..., γ g − } ), andone degenerate curve γ g − .)We restate Lemma 5.4 in terms of the present (simpler) notation. Lemma 5.7. : There is a one-dimensional analytic closed locus
Y ⊂ ∆ g so that both ext Ω Gdh ( γ i ) = ext Ω G − dh ( γ i ) for i = 1 , . . . , g − and ext Ω Gdh ( δ ) = ext Ω G − dh ( δ ) on Y , and Y is proper in ∆ g . ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 39 P P P P Figure 5.3.
Regenerating orthodisks for the Scherk surfaces
Proof.
We again continue with the notation for g = 3.As putatively defined in the statement of the lemma, Y would be clearly closed, andwould have non-empty intersection with ∆ as ∆ contains the solution S to the genus 2problem.We parametrize ∆ near X as ∆ × [0 , (cid:15) ) and consider the mapΦ : ( X, t ) : ∆ × [0 , (cid:15) ) −→ R given by ( X, t ) (cid:55)→ (ext Ω Gdh ( δ ) − ext Ω G − dh ( δ ) , ext Ω Gdh ( γ ) − ext Ω G − dh ( γ )) . Here, the coordinate t refers to a specific choice of normalized geometric coordinate, i.e. t = Im( P P −→ P P ) = Re( P P −→ P P ), where the periods ( P P −→ P P ) and( P P −→ P P ) are measured on the domain Ω Gdh . In terms of these coordinates, we notethat whenever either t = 0, we are in a boundary stratum of ∆ . The locus { t > } ⊂ ∆ is a neighborhood in Int (∆ ) with X in its closure.Note that Φ( X ,
0) = 0 as X is reflexive.Now, to find the locus Y , we apply the implicit function theorem. The implicit functiontheorem says that if(i) the map Φ is differentiable, and(ii) the differential d Φ (cid:12)(cid:12) T X ∆ is an isomorphism onto R , then there exists a differentiable family Y ⊂ ∆ for which Φ (cid:12)(cid:12) Y ≡ ∈ (cid:100) M × (cid:100) M isdifferentiable (here (cid:100) M refers to a smooth cover of the relevant neighborhood of S ⊂ M ,where M is the Deligne-Mostow compactification of the moduli space of curves of genustwo) the theorem of Gardiner-Masur [GM] implies that Φ is differentiable, as we have beenvery careful to choose curves { δ, γ } which are non-degenerate in a neighborhood of ∆ near the genus two solution S , with both staying in a single regular component of thenoded surface.We are left to treat (ii), the invertibility of the differential d Φ (cid:12)(cid:12) T S ∆ . To show that d Φ (cid:12)(cid:12) T S ∆ is an isomorphism, we simply prove that it has no kernel. To see this, choosea tangent direction in T S ∆ interpreted as a perturbation of the geometric coordinatesfor S . To be concrete, we might fix the distance between the parallel semi-infinite sidesand vary the finite lengths (or periods) of the sides P i P i +1 . Now, up to replacing theinfinitesimal variation with its negative, one of the finite-length edges has moved into theinterior of Ω Gdh as in this case, with our normalization, the only edges free to move arethose finite edges. Connect each of those positively moving edges with a curve system from E P (and, symmetrically, P E = P g E ). The result is a large curve system (say Γ)consisting of classes of curves from (possibly) several free homotopy classes. In addition,let ν be the associated Beltrami differential to this variation, as in section 5.2.The flow computations in section 5 then say that this entire curve system Γ has extremallength in Ω Gdh which has decreased by an amount proportional to | ν | while the extremallength on Ω G − dh has increased by an amount proportional | ν | on Ω G − dh . Thus, in any setof differentiable coordinates for the Teichmuller space of the surfaces in ∆ , the differenceof coordinates for Ω Gdh and Ω G − dh is by an amount proportional to | ν | . Thus d Φ( ν ) ≥ c | ν | (for c >
0) which proves the assertion.To finish the proof of the lemma, we need to show that Y (cid:12)(cid:12) ∆ is an analytic submanifoldof T × T , where T is the Teichm¨uller space of genus three curves: this follows on theinterior of ∆ from the fact that Ohtsuka’s formulas for extremal length are analytic andthe map from ∆ to extremal lengths has non-vanishing derivative.This concludes the proof of Lemma 5.7 for the case g = 3 and hence also the proof ofTheorem 5.6. We have already noted that the argument is completely general, despite ourhaving presented it in the concrete case of S ; thus, by adding more notation, we haveproven Lemma 5.7 in full generality. Naturally, this also completes the proof of Lemma5.4. (cid:3) Embeddedness of the doubly-periodic Scherk surface with handles
In this section, we follow Karcher [11] and use the conjugate surface method from sec-tion 2.4 to prove
Proposition 6.1.
The doubly-periodic Scherk surface with handles ( S g ) is embedded. ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 41
We consider a quarter of the surface, as defined by the shaded lower left quarter squareof the fundamental domain of the torus, as in Figure 2.3.Of course, this is also the part of the surface used to define the orthodisks Ω
Gdh andΩ G − dh . This surface patch, say Σ g , is bounded by planar symmetry curves and one verticalend and is contained within the infinite box over a ’black’ checkerboard square. It will besufficient to show that this patch Σ g is embedded, as the rest of the surface S g in space isobtained by reflecting the image Σ g of this quarter surface across vertical planes. To showthat the image of the quarter surface is embedded, we prove that the conjugate surface isa graph over a convex domain; the result then follows by Krust’s Theorem in section 2.4.Thus we prove Lemma 6.2.
The conjugate surface for Σ g is a minimal graph over a convex domain. We apply the basic principles that straight lines and planar symmetry curves get inter-changed by the operation of conjugation of minimal surfaces, and angles get preserved bythis operation. Using these principles, we compute the conjugate surface for Σ g .We assert that the conjugate surface has the form depicted in the figure 6.1 and de-scribed below. The surface in the figure extends horizontally along the positive horizontalcoordinate directions to infinity. Figure 6.1.
Conjugate surface patch of the doubly-periodic Scherk surfacewith handlesThe surface patch will be bounded by two polygonal arcs in space. The first one cor-responds to the P . . . P g polygonal arc of the orthodisks. As the original surface is cutorthogonally by two symmetry planes along this arc, the corresponding arc of the conjugatesurface will consist of orthogonal line segments which stay in the same horizontal plane:to see this, begin by observing that the Gauss map has vertical normals at the pointscorresponding to the vertices of the orthodisk boundaries, hence also at the corners of thestraight line segments on the conjugate surfaces. Yet where two of these segments meet,the tangent plane is tangent to both of them, hence the normal is in the unique direction normal to both; as the Gauss map is vertical, both segments must be horizontal. We con-clude then that all of the straight segments must be horizontal, and hence this connectedcomponent of the boundary must lie in a horizontal plane.Simlarly the second connected arc, this time made up of a pair of infinite arcs, is alsohorizontal, thus parallel to the first boundary arc.We next claim that the two horizontal connected components of the boundary have (apair of) parallel infinite edges as ends which lie on the same (pair of) vertical planes. Tosee this, consider the (pair of) cycles around the two ends E and E : we have constructedthe Weierstrass data one-forms Gdh and G dh so that the coordinate one-forms ( G − G ) dh and i ( G + G ) dh have purely real periods, while the coordinate one-form dh has a purelyimaginary period. Thus for the conjugate surface, the coordinate one-forms ( G − G ) dh and i ( G + G ) dh have purely imaginary periods, while the coordinate one-form dh has a purelyreal period. As any representative of this cycle lifts to connect semi-infinite ends of theconnected components of the boundary, and any such lift must have endpoints differing bya period, we see that the semi-infinite ends of the connected components of the boundarydiffer (respectively) by a purely vertical translation.We now produce a minimal surface which spans this pair of boundary components.To do this, we first approximate the boundary by a compact boundary, then solve thecorresponding Plateau problem for compact boundary values, and then finally take a limit.More precisely, consider a boundary formed from the boundary arcs described above byintroducing vertical segments Γ b, and Γ b, connecting the two pairs of parallel semi-infiniteboundary edges at distance b from the (image of the) point P g . This boundary is nowcompact and projects injectively onto a rectangle boundary by using the projection in thedirection of the ’diagonal’ vector (1 , , , ,
0) by Rad´o’s theorem. We now look ata sequence of such Plateau solutions for increasing values of b → ∞ . Using solutionsfor smaller values of b as barriers for the solutions corresponding to the larger values of b , by the maximum principle we see that the solutions S ∗ g,b form an increasing sequenceof graphs. To show that this sequence actually converges, it suffices to show that theintersections A b of the family of surfaces with the vertical plane Π passing through P g and V ± lie in a single compact set. To see this, we consider a pair of (quarters of) Scherksurfaces (conjugate to S ), each of whose boundary components consist of an infinite L and a vertical translate of that L by the same amount as for a conjugate surface for S g .The first such conjugate Scherk passe through the point P g , while the second is displacedto pass through the points V ± . This pair of boundary arcs lie to the other side of theboundary arcs of S g in the respective horizontal planes. Thus, this pair of surfaces meetsthe plane Π in a pair of arcs, which, by the maximum principle lie to either side of the arcs A b on the strip of Π between our fixed pair of horizontal planes.Thus, by Harnack, the approximate solutions S ∗ g,b converge to a solution S ∗ g for theinfinite boundary problem; as the approximate solutions are all graphs and minimal, thelimit S ∗ g is also a graph (here convergence of the graphed functions u b in C implies their ANDLE ADDITION FOR DOUBLY-PERIODIC SCHERK SURFACES 43 convergence in C by standard elliptic theory, hence to a graph). Then, by Krust’s theorem,the conjugate patch to that limit graph is also a graph, and since that conjugate patch isa fundamental piece of our surface S g , we see that S g is embedded. (cid:3) References [1] Frank Baginski and Valerio Ramos-Batista. Solving period problems for minimal surfaces with thesupport function. preprint, 2008.[2] E. Boix and M. Wohlgemuth. Numerical results on embedded minimal surfaces of finite total curvature.In perpetual preparation.[3] P. Connor and M. Weber. Doubly periodic minimal surfaces with parallel ends. Preprint, 2009.[4] U. Dierkes, S. Hildebrandt, A. K¨uster, and O. Wohlrab.
Minimal Surfaces I . Grundlehren der mathe-matischen Wissenschaften 295. Springer-Verlag, 1992. MR1215267, Zbl 0777.53012.[5] C. Douglas.
Doubly Periodic Minimal Surfaces of Genus 1 . PhD thesis, Rice University, Houston, 2008.[6] A. Fathi, F. Laudenbach, and V. Poenaru.
Traveaux de Thurston sur les Surfaces . Soci´et´eMath´ematique de France. Ast´erisque.[7] F.P. Gardiner.
Teichmuller Theory and Quadratic Differentials . Wiley Interscience, New York, 1987.[8] F.P. Gardiner and H. Masur. Extremal length geometry of teichm¨uller space.
Complex Analysis andits Applications , 16:209–237, 1991.[9] J. Hubbard and H. Masur. Quadratic differentials and foliations.
Acta Math. , 142:221–274, 1979.[10] H. Jenkins. On the existence of certain general extremal metrics.
Ann. of Math. , 66:440–453, 1957.MR0190811 , Zbl 171.08301.[11] H. Karcher. Embedded minimal surfaces derived from Scherk’s examples.
Manuscripta Math. , 62:83–114, 1988. MR0958255, Zbl 658.53006.[12] H. Karcher. Construction of minimal surfaces.
Surveys in Geometry , pages 1–96, 1989. University ofTokyo, 1989, and Lecture Notes No. 12, SFB256, Bonn, 1989.[13] S. Kerckhoff. The asymptotic geometry of teichm¨uller space.
Topology , 19:23–41, 1980.[14] H. B. Lawson, Jr.
Lectures on Minimal Submanifolds . Publish or Perish Press, Berkeley, 1980.MR0576752, Zbl 0434.53006.[15] H. Lazard-Holly and W. H. Meeks III. The classification of embedded doubly–periodic minimal surfacesof genus zero.
Invent. Math. , 143:1–27, 2001. MR1802791, Zbl 992.18190.[16] W. H. Meeks III and H. Rosenberg. The geometry, topology, and existence of doubly periodic minimalsurfaces.
C. R. Acad. Sci. Paris , 306:605–609, 1988. Research Announcement.[17] W. H. Meeks III and H. Rosenberg. The global theory of doubly periodic minimal surfaces.
Invent.Math. , 97:351–379, 1989. MR1001845, Zbl 676.53068.[18] M. Ohtsuka.
Dirichlet Problem, Extremal Length and Prime Ends . Van Nostrand Reinhold, New York,1970.[19] R. Osserman.
A Survey of Minimal Surfaces . Dover Publications, New York, 2nd edition, 1986.MR0852409, Zbl 0209.52901.[20] J. P´erez, M. Rodr´ıguez, and M. Traizet. The classification of doubly periodic minimal tori with parallelends.
Journal of Diff. Geometry , 69(3):523–577, 2007. MR2170278, Zbl pre05004289.[21] H. F. Scherk. Bemerkungen ¨uber die kleinste Fl¨ache innerhalb gegebener Grenzen.
J. R. Angew. Math. ,13:185–208, 1835. ERAM 013.0481cj.[22] K. Strebel.
Quadratic Differentials . Springer, Berlin, 1984.[23] E. Thayer.
Complete Minimal Surfaces in Euclidean 3-Space . PhD thesis, University of Massachusettsat Amherst, 1994.[24] M. Traizet. An embedded minimal surface with no symmetries.
J. Differential Geometry , 60(1):103–153, 2002. MR1924593, Zbl 1054.53014.[25] M. Traizet. Adding handles to Riemann minimal examples.
J. Differential Geom. , 79:243–275, 2008.MR2420019, Zbl pre05294962. [26] M. Wohlgemuth W. Rossman, E. Thayer. Embedded, doubly periodic minimal surfaces.
Exp. Math. ,9(2):197–219, 2000.[27] M. Weber, D. Hoffman, and M. Wolf. An embedded genus-one helicoid.
Ann. of Math. , 169(2):347–448,2009. MR2480608.[28] M. Weber and M. Wolf. Minimal surfaces of least total curvature and moduli spaces of plane polygonalarcs.
Geom. Funct. Anal. , 8:1129–1170, 1998. MR1664793, Zbl 0954.53007.[29] M. Weber and M. Wolf. Teichm¨uller theory and handle addition for minimal surfaces.
Ann. of Math. ,156:713–795, 2002. MR1954234, Zbl 1028.53009.[30] F. Wei. Some existence and uniqueness theorems for doubly periodic minimal surfaces.
Invent. Math. ,109:113–136, 1992. MR1168368, Zbl 773.53005.[31] M. Wohlgemuth. Higher genus minimal surfaces by growing handles out of a catenoid.
ManuscriptaMath. , 70:397–428, 1991.[32] M. Wohlgemuth.
Vollst¨andige Minimalfl¨achen h¨oheren Geschlechts und endlicher Totalkr¨ummung . PhDthesis, University of Bonn, April 1993.[33] M. Wohlgemuth. Higher genus minimal surfaces of finite total curvature. Preprint, 1994.[34] Michael Wolf. On realizing measured foliations via quadratic differentials of harmonic maps to R -trees. J. Anal. Math. , 68:107–120, 1996.
Indiana University
E-mail address : [email protected] URL : Rice University
E-mail address : [email protected] URL ::