‘SYMPLECTIC GEOMETRY AND TOPOLOGY Vladimir. ARNOLD ne 9410 2402/94Symplectic geometry and topology V.LAmold February 21, 1994 1 What is Symplectic Geometry About. 1.1 From Hamilton Dynamics to Symplectic Geometry. Symplectic geometry is the product of a long evolution of such branches of mathematics, as the variational calculus, the theory of dynamical systems, especially of Hamilton systems of classical mechanics, the geometrical optics, the theory of wave propagation, the study of the short waves or quasiclassical asymptotics in quantum mechanics, the microlocal analysis of PDEs and the Lie theory of diffeomorphism groups and Poisson algebras. ‘The geometrization of mathematics and of physics, originating in the pioneering works of Poincaré, has led to the description of the time evolution of the state of a dynamical system in terms of a flow of the so-called phase fluid in the phase space, whose points represent different states of the system (say, the positions and the velocities of its particles). The phase flow consists of transformations g' of the phase space, de- pending on time and sending any initial state to the new state of the system after time t. The transformations forming the phase flow of a Hamilton dy- namical system are not arbitrary diffeomorphisms. For instance, they can not have attracting fixed points or other attractors, accordingly to the Li- ouville theorem, saying that the phase flow of a Hamilton system preserves the volume of any domain of the phase space. One can thus imagine that the phase fluid, filling the phase space, is incompressible. This fact implies many special features of the transforma- tions of the phase flow, a typical example being the Poincare return theorem and the numerous applications of the ergodic theory to dynamical systems. Statistical mechanics is based on this incompressibility of phase flows. The applications of the incompressibility property form the volume- Preserving geometry, the study of the geometrical properties of different 1objects on a manifold with a fixed volume element. A property is here called geometrical i invariant under the volume-preserving diffeomorphisms, “The symplectic geometry arises from the understanding of the fact that the transformations of the phase flows of the dynamical systems of classical mechanics and of variational calculus (and hence also of the optimal control theory) belong to a narrower class of diffeomorphisms of the phase space, than the incompressible ones. Namely, they preserve the so-called symplectic structure of the phase space - a closed nondegenerate differential two-form. This form can be integrated along two-dimensional surfaces in the phase space. The integral, which is called the Poincaré integral invariant, is preserved by the phase flows of Hamilton dynamical systems. The diffeomorphisms, preserving the symplectic structure - they are called symplectomorphisms - form a group and have peculiar geometrical and topological properties. For instance, they preserve the natural volume element of the phase space (the exterior power of the symplectic structure 2form) and hence can not have attractors, However this is not the only restriction - the preservation of the symplectic structure is responsible for many astonishing facts in mechanics, optics and other parts of mathematical physics. While the first examples of such phenomena were explicitly described by Poincaré (and might be traced to the works of Hamilton, Jacobi, Lie, Cayley and others), the systematical study of the geometry on symplectic manifolds is mostly due to the works of the second half of the 20th century. Applications to statistical physics (where the physicists stil freely permute the pieces of the phase space provided that they have equal volumes) are still to be found. ‘The word symplectic was invented by H. Weyl ([1}, 1939) to describe the finite-dimensional version of the symplectomorphism group - the group of the linear transformations, preserving a nondegenerate skew-symmetric bilinear form. C. L. Siegel ([2], 1943) had christened symplectic geometry rather the geometry of this linear group, than that of the group of symplectomorhisms, but now this name is used for the extended domain of nonlinear symplectic manifolds and mappings. The symplectic topology is even younger ({3], 1965; [a], 1986). Before I start a more systematical description of the subject, I shall discuss some very particular examples of results of symplectic geometry and topology.1.2 Symplectic Phenomena and Invariants. The conjecture that the symplectomorphisms may behave differently than the volume-preserving diffeomorphisms was first formulated in the following form. Example: the symplectic camel problem. Consider the“eye of the needle” (a hole in a vertical plane in the three-space). In volume-preserving geometry any camel, however large it is, can snake from one half-space to the other through the hole (Fig.1). (Fig.1 here) The symplectic camel problem asks whether it is possible for a symplectic camel in one half of the symplectic space to be continuously transformed into @ camel in the other half, passing through a small hole in the hyperplane separating the half-spaces. The transformation should be defined by time- dependent symplectomorphisms of the complement of the hyperplane with a hole. These diffeomorphisms should connect the identity mapping with the mapping sending the camel from one halfspace to the other. Any symplectic space is even-dimensional. Symplectic geometry in di- mension two coincides with the volume (area)-preserving geometry. The two-dimensional camel can percolate through any hole from the left half- plane to the right one. However in dimension four the situation is different: a symplectic camel have symplectic ribs that do not permit him to pass from one half-space to the other, if the hole is not large enough [5]. ‘These ribs are defined in terms of the periods of periodic solutions of some Hamilton differential equations associated to the camel. The impossibility of the crossing of the hole can be considered as an extension from linear to nonlinear oscillations of the Rayleigh - Fisher - Courant theorem on the behaviour of the eigenvalues under the imposing of constraints. Example: the symplectic packing problem. Any bounded domain of the plane, whose area is smaller than the area of the unit disc, can be sent into this disc by an area-preserving diffeomorphism. A similar result holds in the volume-preserving geometry in any dimension. The symplectic packing problem requiers to send by a symplectomor- phism into a given bounded domain of the symplectic space another bounded domain of smaller volume. It is impossible, for instance, to embedd by a symplectomorphysm the unit ball of the standard symplectic four-space into the product of two two- dimensional symplectic discs, one of which has a radius smaller than one(see (6]). The system of the disjoint images of equal balls (Fig.2) under a system of the symplectomorphisms of the symplectic four-space can fill no more than 1-1/N of the volume of the symplectic four-ball, and can fill almost 1-1/N of the volume, where WV is equal to 2,4, 00, 5, 25, 64, 289, oo for k= 2,3, (Fig.2 here) Moreover, for every positive integer p the symplectic 2n-ball can be filled with gaps of arbitrary small volume by the disjoint images of p" equal sym- plectic balls (V = co, see [7]). The universal obstacles in the symplectic embedding and packing prob- lems are called Gromov’s width and symplectic capacities (see [8], [9]). 1.3 Last Geometric Theorem of Poincaré. ‘The first theorem of symplectic topology has been discovered by Poincaré in his studies of periodic orbits in celestial mechanics. It is called the last geometric theorem of Poincaré, because he had announced it just before his death being unable to prove it. The proof has been given later by Birkhoff. Poincaré had formulated his theorem for the area-preserving mappings of an annulus to itself: such a mapping has at least two fized points, provided that it moves the two boundary circles in opposite directions. From the modern point of view this is a particular case of the general facts of the symplectic topology. Example: the symplectic fixed points theorem. Consider a torus of dimension two with its area element as a symplectic manifold. A sym- plectomorphism of the torus is called ezact, if it can be connected to the identity mapping by a continuous path in the group of symplectomorphisms, preserving the "center of masses” of the torus. The "preservation of the center of masses ” condition can be written in the terms of a coordinate system on the plane, covering the torus (Fig.3). (Fig.3 here) If the coordinates are (z, y)modl, the area clement is dzdy and the map- ping sends the point (z,y) to (x + f(z,y),y + 9(2,y)), then the center of masses preservation condition requires the vanishing of the average shift, that is of the integral of the vector (f,9) along the unit square.‘The theorem (see [10]) says that any ezact symplectomorphism of a torus has at least four fized points, and that at least three of them are geometrically different. Remark. The original statement of Poincaré on the annulus mappings follows, since one can construct a torus glueing together two equal annuli ‘The theorem says that an ezact diffeomorphism of the torus has at least as many fized points, as a function on the torus has critical points. In this form the theorem holds also on all surfaces and on many higher-dimensional manifolds, but is still neither proved nor disproved for arbitrary compact symplectic manifolds (the exact eymplectomorphisms forming the commu- tator subgroup of the connected component of the identity in the symplec- tomorphism group). Example. On a surface of genus g the number of fixed points counted with multiplicities is at least 2g +2, and at least 3 among them are geo- metrically different. For the torus J" these numbers are 2°" and 2n +1 respectively. For the complex projective space CP" both numbers are n+1 The proofs for different classes of symplectic manifolds are in [10], (6), (11), [12}, (44) e.t.c. ‘The Morse numbers and the critical points occur in this problem not by a coincidence. The whole theory can be viewed as an extension of Morse theory to generalized multivalued functions, called Legendrian manifolds. Infinitesimal exact symplectomorphisms are defined by the Hamilton vec- tor fields. The critical points of the Hamilton function are fixed points of these symplectomorphisms. The symplectic fixed point theorem extends ‘the Morse-theoretical minoration of the number of the fxied points of an infinitesimal exact symplectomorphism to the finite ones. Some other facts of the classical calculus may also be considered as the infinitesimal versions of theorems in symplectic geometry. 1.4 Toplogy of Caustics and Wavefronts. We start with the following result of Euclidean geometry. Example: the four vertices theorem. Consider a convex plane curve (Fig.4). A verter is a point of extremum of the curvature. The classical four vertices theorem says that the number of vertices is at least four (see (13]). (Fig.4 here) It seems that it is a theorem in Riemannian geometry, but it it is not, since the result does not hold for the geodesical curvature of curves on theplane with a Riemannian metric, even when it is closed to the standard Euclidean one. ‘The real meaning of this theorem can only be understood in terms of symplectic topology, where it appears together with quite a few theorems on the singularities of the caustic and wave fronts. The infinitesimal version of these theorems is a Sturm-type result on the oscillatory properties of the function /”” + f’ on the circle. According to a general theorem of Tabachnikov, any 2x-periodical function of this form has at least 4 zeroes on every period. This infinitesimal result have global counterparts which are general the- orems of symplectic and contact topology. In the next two examples the applications of these general theorems to the problems of the classical Rie- mannian geometry and of the calculus of variation are discussed. Example: focal points. A focal point of a point on a surface equipped with a Riemannian metric is an intersection point of a geodesic ray issued from the point with an infinitely close geodesic ray (also issued from this point). The set of the focal points of a given point of a surface form a curve,called the caustic of the initial point. The caustic of the North pole of a sphere consists of its South pole and North pole. This is a degenerate caustic. Perturbing the metric of the sphere (say, transforming it into an ellipsoid) one transforms the caustics of its points into small but complicated curves. Jacobi has proved that the caustic of any point on any convex surface has cusps (see Fig.5). (Fig.5 here) In fact the caustic consists of several branches. The first branch consists of the first focal points along each ray, the second of the next ones and so on. If the perturbation is small the first branch of the caustic of the North pole is close to the South pole, the second to the North pole and so on. The smalleness of the perturbations one needs depends of course of the order of the branch. ‘The four cusps theorem says that the caustic (say, of a generic point on a generic convex surface) has at least four cusps. This is classicaly proved for the first branch of the caustic. It holds also for any branch, provided that the perturbation is sufficiently small [14]. Some standard conjectures in symplectic topology imply that it should be true with no smallness restrictions, but this is not proved.Symplectic geometry of even-dimensional phase spaces has an odd- di- mensional twin, the contact geometry. This geometry of wave fronts and ‘wave propagation is the mathematical base of geometrical optics. It is highly nontrivial even in the simplest case of the wave propagation in the Euclidean space, even in the case of the plane, where the propagating fronts are the equidistant curves of the initial one. Example: the inside-out reversal of a front. Consider the equidis- tant curve of an ellipse (Fig.6). When the distance from the initial ellipse grows the equidistant curve first shrinks and later starts to grow. If one starts from a circle it shrinks to a point. For the ellipse the transition from the shrinking to the grows is decomposed into several steps. (Fig.6 here) At one of the steps the equidistant curve has four cusps. This is a very Particular case of a general theorem of symplectic and contact topology, accordingly to which any inside-out reversal process requiers at least four cusps on a moving front at some moment. This is proved ({15]) for fronts having no parallel coorienting normals (for instance, for the equidistant curves of convex plane curves). The result, probably holds for all fronts reversals with no cooriented selftangencies at any moment. The proof depends on the Tabachnikov theorem of Sturm type [13], describing the oscillation properties of linear combinations of eigenfunctions of linear differential equations on the circle. These combinations define the infinitesimal perturbations of the reversal of the circular front. Consider a periodic function, whose Fourier series contains only the har- monics of higher order, coskz and sinkz with k > n. The Tabachnikov theorem says that the number of zeroes of any such function on a period is at least equal to the number 2n of zeroes of its lowest harmonics. ‘The extension of the theorem on the cusps of the equidistant curves to the general theorems in symplectic and contact topology suggests generalizations of the Sturm theory to the multivalued and higher-dimensional cases in the same sence in which the symplectic fixed point theorem extends Morse theory.2 Symplectic Manifolds and Their Lagrangian Submanifolds. 2.1 Symplectic Structures. Symplectic manifolds are the natural generalizations of the phase spaces of classical mechanics. Definition. A symplectic manifolds a smooth manifold equiped with closed nondegenerate differential form, which is called the symplectic struc- ture. Example. A plane with the oriented area element is a symplectic man- ifold. Example. The cartesian product of two symplectic manifolds has a natural symplectic structure - the sum of (the pull-backs of) the two given structures, Example. The space R™ has a natural symplectic structure. Its value on a pair of vectors is equal to the sum of the oriented areas of the projections of these two vectors on the n coordinate plains (ps, 91). (Pas Ja)> Theorem of Darboux. Ail symplectic manifolds of any given dimen- sion are locally aymplectomorphic one to the other. Corollary (Darboux coordinates). Any symplectic structure is de- fined in a sufficiently small neighbourhood of any point of any symplectic manifold of dimension 2n by the form described in the preceeding example in suitable local coordinates. In other words all the difference between the symplectic manifolds is global, locally all the symplectic manifolds of the same dimension are alike (similarly to the Euclidean spaces but unlike the Riemannian manifolds). Moreover, any point of a connected symplectic manifold can be carried to any other by a symplectomorphism (diffeomorphism preserving the sym- plectic structure), and even by a flow of the symplectomorphisms. ‘The most important examples of symplectic manifolds are the phase spaces of the classical mechanical systems - the cotangent bundle spaces of smooth manifolds (which are called the base manifolds ot the configuration ‘space Definition. A cotangent vector on a manifold is a linear function on the space of tangent vectors at a point of the manifold. All the cotangent vectors (which in physics are usually called momenta vectors) at all the points of a base manifold B" form the cotangent bundle space T*B of dimension 2n.Manifolds which have serious reasons to be even-dimensional usually carry natural symplectic structures. Definition. The action form on the cotangent bundle space is the tau- tological differential one-form on the cotangent bundle space, namely it is the form, whose value on any vector tangent to the cotangent bundle space at a point p is equal to the value of the cotangent vector p on the projection of this tangent vector to the base manifold. Example. The action form on the cotangent bundle space of the vector- space with linear coordinates (4;,...,gn) has the form Pidgy +. + Prddn where p; are the natural coordinates of a cotangent vector. Definition. The natural symplectic structure of the cotangent bundle space is the differential 2-form e w=da. Other examples of symplectic manifolds are the Kaeblerian manifolds in complex geometry and the orbits of the coadjoint representation of any Lie group in the dual space of its Lie algebra. Example. The complex projective space carries a natural symplectic structure. Example. The manifold of all complex matrices with fixed and simple eigenvalues carries a natural symplectic structure. 2.2 Submanifolds of Symplectic Manifolds. The geometry of a symplectic manifold is refreshingly different from the usual Euclidean or volume-preserving geometry- some of its submanifolds are locally different from others submanifolds of the same dimension. Definition. A submanifold of a symplectic manifold is called isotropic if the symplectic structure's restriction to this submanifold vanishes. An isotropic submanifold of maximal possible dimension (which is n in a sym- plectic manifold of dimension 2n) is called a Lagrangian submanifold. Example. Any smooth curve on the plane is a Lagrangian submanifold. Example. Any smooth function f defines a Lagrangian section of the cotangent bundle space of its domain of definition, pad,associating to each point of the base manifold the differential of the function at this point. Example. ‘The fibers of the cotangent bundle are Lagrangian subman- ifolds. Locally all the Lagrangian submanifolds are symplectomorphic one to the other, Weinstein Theorem.(See [16]). Some neighbourhood of any La- grangian submanifold in any symplectic manifold is symplectomorphic to some neighbourhood of this Lagrangian submanifold in any other symplectic manifold, for instance in its own cotangent bundle space. Remark. In general there is no global symplectomorphysm - this is clear from the example of plane curves. Indeed, the area bounded by a closed curve is a symplectic invariant of the curve. Givental - Darboux Theorem. (See [17]). The restriction of the symplectic structure to a submanifold of a symplectic manifold defines this submanifold locally in a neigbourhood of any of its points up to a local sym- plectororphisms. In other words in symplectic geometry, in contrast with the Rieman- nian geometry, there exist no local exterior invariants of submanifolds: the interior geometry controls the exterior one. Example. Any two smooth hypersurfaces (submanifolds of codimen- sion one, that is defined locally by one nondegenerate equation each) in a symplectic manifold are locally symplectomorphic in the neighbourhoods of any two pointe, 2.3 Hamilton Vector Fields and Poisson Bracket: All points of a smooth hypersurface in a symplectic manifold are equal, but the hypersurface is (even locally) nonisotropic. At each point of the hyper- surface there exists a prefered (characteristic) tangent direction, intrinsically defined by the symplectic structure. There is only one such direction - it is the skeworthocomplementary direction of the tangent hyperplane of the hypersurface. Definition. The Hamilton vector field of a function on a symplectic manifold is a field, for which the value of the symplectic structure on a pair formed by a vector of the field and by any second tangent vector of the symplectic manifold at the same point is equal to the derivative of the function along the second vector. 10The function is then called the Hamilton function of the field. It is de- fined by the field up to a locally constant summand. The Hamilton vector field is the only vector field intrinsically associated to a function on a sym- plectic manifold (up to the multiplication of the field by a function locally constant on the level sets of the Hamilton function). Example. A hypersurface in a symplectic manifold is folliated (locally fibered) into the characteristics - lines tangent to the vectors of the Hamilton fields, defined by all fanction constant along the hypersurface. The characteristics are indepedent of the choice of the equation defin- ing the hypersurface. Hence the structure of the decomposition of the hy- persurface into characteristics (which may be very complicated as it is well known from celestial mechanics and Hamiltonian chaos theory) is a symplec- tic topology invariant of the hypersurface. It is also an important source of symplectic invariants of the domains, bounded by hypersurfaces in the sym- plectic space. ‘The Hamilton vector fields form a Lie algebra - a subalgebra of the algebra of all the vector fields with the usual operation - the Poisson bracket. ‘The Hamilton functions also form a Lie algebra, whose operation is also called the Poisson bracket. The Poisson bracket of two functions is the derivative of one of them along the Hamilton vector field of the other. The algebra of functions is a central extension of the algebra of fields (the center consisting of the locally constant functions). The flows of the Hamilton vector fields preserve the symplectic structure. If the flow preserves the symplectic structure the field is locally a Hamilton field, but globally the Hamilton function may be multivalued. The Lie algebra of Hamilton vector fields is a subalgebra of the algebra of symplectic vector fields preserving the symplectic stracture. Example. The translations of the torus preserve the area element and hence the constant (translations invariant) vector fields on the torus are symplectic, However they are not Hamilton vector fields. The flows of Hamilton vector fields on the torus are precisely the flows consisting of area-preserving and center of masses preserving. diffeomor- phisms. These fields form the commutator subalgebra of the Lie algebra of symplectic vector fields. The quotient space is the one-dimensional coho- mology space of the symplectic manifold. u2.4 Symplectic Algebraic and Analytic Geometry Return to the Lagrangian submanifolds of a symplectic manifold. According to a general priciple of A. Weinstein, in symplectic geometry every important object is a Lagrangian submanifold. Example. A symplectomorphism between two symplectic manifolds is defined by its graph, which is a submanifold of middle dimension in the carthesian product of the two manifolds. Equip the product with a sym- plectic structure, which is the difference of (the pull-backs of) the given two structures. The graph of a diffeomorphism is then a Lagrangian submanifold of the product if and only if the diffeomorphism is symplectic. Remark. Replacing here the graph by any Lagrangian submanifold of the product we obtain an interesting generalization of the notion of sym- plectomerphism: the symplectic correspondence. In many problems it is also ‘usefull to consider the Lagrangian varieties with singularities. The algebraic (or better analytic) geometry of symplectic manifolds and of their subvarieties should be based on the interaction of two algebra struc- tures in the space of functions: the ordinary commutative multiplication and the Poisson bracket Lie algebra structure. Many facts,known in the regular case of transversal submanifolds, should be extended to the general situation. One of the motivations for this extension is the remark (due to R. Mel- rose,who had based on it his works on diffraction singularities near gliding rays) that the geometry of submanifolds of a Riemannian manifold can be considered as the aymplectic geometry of pairs of submanifolds in a symplec- tie manifold. Example. Consider the case of a hypersurface in the usual Euclidean space (the cases of a curve in a plane and of a surface in the three-space being quite instructive). ‘The Riemannian metric can be viewed as the hypersurface p? = 1 of momenta of lenght one in the phase space (space of the cotangent bundle of the Euclidean space). Consider a hypersurface in the Euclidean space (think of the boundary of the obstacle in the diffraction problem), say F(g) = 0. This equation defines a second hypersurface in the phase space. The differential geometry of the boundary surface in the Euclidean space can be read from the symplectic geometry of this pair of hypersurfaces in the phase space (think of the asymptotical directions and on the parabolic points line on a surface in the usual three-space and try to understand their 12meaning in the symplectic geometry of the pair). Return to the general theory of Lagrangian submanifolds. 2.5 Lagrangian Singularities. Lagrangian singularities are the singularities of Lagrangian mappings.The simplest examples are the mappings of Lagrangian submanifolds of the phase space to the configuration space. The Gauss mapping of Euclidean differen- tial geometry also is a Lagrangian mapping (of a hypersurface to the unite sphere). Definition. A Lagrangian fibration is a fibration which fibers are La- grangian submanifolds. Example. The cotangent bundle is a Lagrangian fibration. Theorem. All the Lagrangian fibrations of the same dimension are locally symplectomorphic to each other (locally means here “in a neighbour- hood of any point of the symplectic manifold”) and hence are locally sym- plectomorphic to the cotangent bundle of the Euclidean space. Remark. The fibers of a Lagrangian fibration (or foliation) have natural local affine structures, intrinsically defined by the symplectic geometry of the fibration. This is crucial for the theory of integrable Hamilton systems and is very close to the Liouville theorem on integrable system, which claims that the smooth compact common level manifolds of commuting Hamilton functions are nested tori (see Fig.7 and (18]). (Fig.7 here) Suppose now that we have a Lagrangian submanifold in the space of a Lagrangian fibration. ‘Example. Consider a hypersurface in an Euclidean space, say an el- lipsoid. All the unit vectors of the lines normal to the hypersurface form a Lagrangian submanifold in the cotangent bundle space of the Euclidean space (we identify the tangent and cotangent vectors in an Euclidean space). One may think of particles emanated by the surface along its normals with fixed energy. Their states form our Lagrangian submanifold, Consider the positions of these particles. They are clustered along some hypersurface of the Euclidean space - the caustic. Definition. The caustic of a Lagrangian submanifold of a Lagrangian fibration space over some base manifold is the set of critical values of the projection of the Lagrangian submanifold to the base manifold along the fibers. The singularities of the restriction of the projection to the Lagrangian submanifold are called Lagrangian singularities. 13Example. The caustic of the Lagrangian submanifold constructed in the preceeding example is the focal set of the initial hypersurface. For instance, let the initial submanifold be an ellipse in the Euclidean plane. In this case the caustic is an astroid (Fig. 8). It has four cusps. (Fig.8 here) ‘The classical four vertices theorem garantees the existence of four cusps on the caustic defined by any generic convex curve in the Euclidean plane. The general results of symplectic geometry show that it is not by chance and is not a pecularity of the Euclidean geometry of the plane. We might have started from a curve in the plane with a different Rie- mannian metric. The Lagrangian variety is well defined, and its caustic still generically has four cusps. The Riemannian metric might be deformed to a Finslerean metric, there still will be four cusps. And there might be no metric at all. The classification of generic singularities of caustics is at present known in spaces of dimensions smaller then six. In these dimensions the classifica- tion is discreete: there is only a finite number of classes of singularities (up to the fibered symplectomorphisms, see {19]). Namely, the singularities are classified by the simple Lie algebras, As, Dy, Ex, where k is at most n +1 if the dimension of the base manifold is n. All singularities of generic caustics in the three-space are shown in Fig.9. (Fig. 9 here) iFrom dimension 6 on the classification of the Lagrangian singularities up to symplectomorphisms (or even of caustics singularities up to the diffeo- morphisms) is no longer discrete. The topological classification is discreete in any dimension. But it is unknown even in dimension six. 2.6 Unimodular Singularities, Strange Duality and Mirror Symmetry. ‘The nondiscrete classification of the Lagrangian singularities of higher di- mensions remains algebraically appealing at least at the beginning, when they depend on few parameters. The classification of unimodular singular- ities (whose classes depend om one parameter) consists of several series of ‘one- parameter families of singularities and of fourteen sporadic families. The 14 sporadic families of exceptional unimodular singularities corre- spond to the 14 exceptional triangles on the Lobachevski plane. These 14 14 |triangles have angles /p, t/q, x/r, where (p,q, r) is one of the 14 triples (239)(247)(336)(256)(345)(444)(238)(246)(335)(255)(344)(237)(245)(334) Among the 14 unimodular singularities there had been discovered long ago a strange duality [20]. This duality (permuting the so called Gabriefov numbers and Dolgachev numbers of the same singularity) is an involution on the set of the 14 triangles. Thad called this duality strange because it is hard (impossible 2) to guess which triangle is dual to which. Try to find the rule, knowing the answer: the dual pairs are (239) — (334), (247) — (335), (238) — (245), (256) — (344). Each of the remaining six triangles is dual to itself. The sum of all the six Gabrielov and Dolgachev numbers is equal to 24 for any of the 14 singular- ities, It has been recently discovered that this strange duality is the manifes- tation of the so-called mirror symmetry of three-folds studied by the physi- cists.It is perhaps the first manifestation of this symmetry, which permute the Hodge numbers of different-dimensional cohomologies and which itself is strange and rather poorly understood. ‘More general cases of mirror symmetries (for the hypersurfaces in the so-called toric varieties of arbitrary dimension, not only of dimension 3 of physicists - see (21]) are the manifestations of the Legendre duality of convex geometry (applied to the Newton polyhedra) - the duality between a function and its Legendre transformation, that we shall describe below. ‘The theory of mirror symmetry for the toric varieties themselves is closely related to the symplectic fixed point and Lagrangian intersections theories. The multiplication in the Floer type cohomology, studied in the framework of symplectic topology by Givental [45] , is called now quantum cohomology and even simply cohomology by the physicis 2.7. Simple Lagrangian Singularities and Simple Lie algebras. ‘The classification of Lagrangian singularities remains discrete in higher di- mensions in the neighbourhoods of some special simplest singularities, which are generically stable (but from dimension six on one can not eliminate by small deformations also some singularities with nondiscrete classification). 15These simplest singularities are called simple and are classified by the classical Dynkin diagrams Ay, Ds, Ee, Er, Es (Fig-10) of the simple Lie al- gebras, of quivers theory ete (classifying also' the regular polyhedra in the usual three-dimensional Euclidean space and the discrete subgroups of the Spin(3) group SU(2)). (Fig.10 here) ‘The complete lst of simple Lie algebras contains also some diagrams with double and triple lines, By,C1, Fi,Gz. They correspond to the boundary singularities of caustics (in the cases with double lines). Example. The caustic singularity F, occurs on the focal set of a surface with boundary in the usual Euclidean three-space. The focal set of a surface with boundary consists of three components. One is the focal set of the surface, the other of the boundary curve and the third (B, in Fig. 11) is formed by the normals to the surface at the points of the boundary curve. (Fig.11 here) ‘The F, singularity, shown in Fig.11, occurs on the canstic near the focal point of the surface at a point of the boundary curve where this curve is ‘tangent to the principal curvature direction of the surface [22]. 2.8 Simple Lie Algebras, Euclidean Reflections Groups and the Hypericosahedron. The list of the simple Lie algebras forms a part of a larger list of the Cox- ‘eter Euclidean reflection groups. An Euclidean reflection is an orthogonal transformation preserving pointwise fixed exactly a hyperplane (called ite mirror). The reflections in a finite set of mirrors may generate a finite or an infinite groups. Example. Two mirrors in the plane define a finite reflection group if and only if the angle between them is commensurable with 2r. This reflection group is then the symmetry group of a regular p-gone, denoted by J,(p). Remark. The Dynkin diagram provides a short description of a reflec- tion group in terms of the angles between the generating mirrors. Each vertex represents a normal vector of a mirror. Two vertices are connected by a simple line if the vectors form an angle of 120 degrees, by a double line if the angle is 135 degrees and by a triple line if it is 150 degrees. Example. The diagram A, = —. represents two mirrors forming an angle of 120 degrees in the plane. The corresponding reflection group is the symmetry group of an equilateral triangle. 16The Euclidean reflection groups have been classified by Coxeter. Most of them are cristallographic (preserve a lattice). Example. The cristallographic plane reflection groups are J,(p) with P = 2,3,4,6 (Fig.12). The symmetry groups of regular pentagons, hep- tagones, octogones and so on are not cristallographic (they are related to the quasicristalls of physicists). (Fig.12 here) There exists a natural bijection between the set of the cristallographic irreducible Euclidean reflection groups (considered together with the lat- tices) and the set of the simple (complex) Lie algebras (listed above): the reflection groups are the so-called Weyl groups of the algebras. The remaining list of the noncristallographic irreducible Euclidean re- flection groups consists of the series J;(p),p = 5,7,8,9,... and of the two sporadic groups Hs, H.. Hy is the symmetry group of the regular icosahedron in the three-space. A, is the symmetry group of the regular hypericosahedron in the Euclidean four-space. The hypericosahedron has 120 vertices and 600 thetrahedral faces. Since it is not described in the Bourbaki book on Coxeter groups, I give here its construction. Consider the 60 rotations preserving the icosahedron. They form a group. The Spin(3) group SU(2) is the double covering of the rotation group $0(3). The icosahedron rotation group is covered by the binary icosahedron group consisting of 120 elements of SU(2). The group SU(2) is the 3-sphere S* in R*. The 120 elements of the binary icosahedron group are the 120 vertices of the hypericosahedron, which i is their convex hull. The hypericosahedron symmetry group H, acts on the vertices as the product of the left and of the right translations of the binary * group. Hence the Euclidean reflection group H, has 14400 elements. 2.9 Simple Lagrangian Singularities and Euclidean Reflec- tions Groups. The full list of Coxeter Euclidean reflection groups, cristallographic or not, provides an answer to the following problem on Lagrangian mappings in symplectic geometry. Consider a Lagrangian subvariety in the space of a ‘Lagrangian fibration and project it to the base of the fibration along the fibers. Any such projection is called a Lagrangian mapping. Definition. A singularity of a Lagrangian mapping is simple if the set of singularities into which it can be transformed by a small deformation of aw ESthe fibration is finite. Givental Classification Theorem.([23]). There ezists « natural bijec- tion between the simple singularities of Lagrangian mappings of Lagrangian varieties and irreducible Cozeter groups. A variety can have a simple pro- Jection singularity only if it is (locally) a product of a smooth manifold with @ curve. The A,D,E groups correspond to smooth curves (and hence to the pro- Jections of smooth Lagrangian manifolds). The groups with double lines in the Dynkin diagram correspond to the curves with an ordinary double point and to the boundary singularities. The remaining groups correspond to the curve singularities z? = y? (where p = 4,3,3,q—2 for Gz, Hs, Hs, 13(q)). The Coxeter groups and the corresponding cristallic and quasicristal- lic structures are crucial for the study of the geometry of wave fronts and ics in the corresponding problems of symplectic and contact geometry. They control also the asymptotics of the integrals of the short-wave ap- proximations and the ramification of the corresponding special functions in the complex domain. The Coxeter group represents the monodromy group and the cristallographic lattice is the integer homology group generated by the vanishing cicles on a complex hypersurface. To explain the relations between the Coxeter groups and Lagrangian singularities one needs some basic notions of the contact geometry. ‘The relation of the contact geometry to the symplectic geometry is sim- ilar to the relation between the linear algebra and the projective geometry. Any fact in the symplectic geometry can be formulated as a contact ge- ometry fact and vice versa. The calculations are simpler in the symplectic geometry setting, but their geometrical content is better seen in the contact version. The functions and the vector fields of the symplectic geometry are re- placed by the hypersurfaces and the line fields in the contact geometry. The contact geometry is almost unknown in the mathematical physicists comma- nity in spite of the fact that it is the mathematical basis of the geometrical optics, of the Gibbs thermodynamics and of the Huygens general theory of wave propagation and of the optimal control theory. 18.3 Contact Manifolds and Their Legendrian Sub- manifolds. 3.1 Contact Structures. The contact manifolds are the odd-dimensional twins of the symplectic man- ifolds. Each contact manifold has symplectization, which is a symplectic manifold whose dimension exceedes the dimension of the contact manifold by one. ‘The symplectic manifolds have the contactizations whose dimensions ex- ceede by one the dimensions of the symplectic manifolds. Definition. A contact structure on a manifold is a nondegenerate field of tangent hyperplanes. The nondegenericy condition can be formulated in terms of a differential one-form, which locally defines the field as the field of its zeroes. ‘The restriction of the derivative of this form a to each hyperplane a = 0 should be a nondegenerated bilinear form (a linear symplectic structure). ‘The symplectic spaces are even-dimensional. Hence contact structures may exist only on odd-dimensional manifolds. Ifa manifold has a serious reason to be odd-dimensional it usually carries a natural contact structure. Example. A contact element at a point of a smooth manifold is a hyperplane in the tangent space of the manifold at this point (called the contact point of the element). ‘The manifold of all the contact elements on a given base manifold B is called the contact elements bundle’s space PT*B. The space of the contact elements bundle, PT"B, is fibered over the base manifold B. Each fiber is the manifold of contact elements at the same point of the base manifold. These fibers over the base manifold B of dimension n are projective spaces of dimension n — 1 - the projetivized cotangent spaces of B. Thus the contact elements space is the propectivized cotangent bundle space. Its dimension 2n — 1 in always odd. Example. The manifold of contact elements of the plane is three- dimensional: it is diffeomorphic to the open filled torus S* x R?. Definition. The natural contact structure of the manifold M = PT*B of the contact elements on some base manifold B is its tautological struc- ture, namely the field of the preimages of the contact elements under the projection of M to B. Hence the natural contact structure of the manifold M of contact ele- 19ments on B is the field of tangent hyperplanes of M, defined by the following sk ating condition: the velocity of a contact element belongs to the hyper- plane of the field if and only-if the velocity of the point-of contact belongs to the contact element. In other terms the contact element (the skate) may rotate around the point of the contact but may not move transversally to its direction. Theorem. The skating condition defines on the manifold of the contact elements @ contact structure. This example explains the terminology: the manifold of the contact ele- ments is a typical example of a contact manifold. Contact Darboux Theorem. All contact manifolds of the same di- mension are locally contactomorphic. Moreover,any point of a connected contact manifold can be carried to any other by a flow of contactomorhisms. ‘The manifold of the contact elements provides a local model for any contact structure on a manifold of suitable dimension. 3.2 Projective Duality and Legendre Transformation. Consider the manifold of contact elements of the projective space. Theorem. It coincides with the manifold of contact elements of the dual projective space: PIP* = PrP™, Indeed, a contact element of the projective space is a pair consisting of a point of the space and of a hyperplain containing this point. The hyperplane is a point of the dual projective space P** and the point of the original space defines a contact element of the dual space. Hence the manifold of contact elements of a projective space is the total space of two fibrations: the first over the original projective space and the second over the dual one (Fig. 13). Each of these fibrations is the fibration of a manifold of contact elements of some base manifold into the projectivized cotangent spaces of the base manifold. (Fig.13 here) Thus the manifold of contact elements of the projective space has two natural contact structures. The first is the natural contact structure of the manifold of contact elements of the original projective space. The second is the natural contact structure of the manifold of contact elements of the dual projective space. 20‘Theorem. Both contact structures of the manifold of contact elements of the projective space coincide. ‘This theorem is a geometrical statement and it has a geometrical proof with no computations at all. Hint: consider the product of two identical projective spaces and the involution permuting them (I leave to the reader the pleasure to discover this geometrical proof). ‘The projective duality associates to a hypersurface in the projective space the dual hypersurface, formed by the tangent hyperplanes of the orig- inal hypersurface. This dual hypersurface lives in the dual projective space. iFrom the theorem above it follows easily and with no calculations that the dual of the dual hypersurface is the initial hypersurface (at least if both are smooth,for instance for the boundaries of convex bodies). Hint: any nonvertical Legendrian submanifold of the space of contact elements of a manifold is, at least locally, the manifold of tangent planes of a hypersurface in the base manifold, see below the definition and examples of Legendrian submanifolds). The affine or coordinate version of the projective duality is called the Legendre transformation. Thus contact geometry is the geometrical base of the theory of Legendre transformation and hence of all the theories based on it: thermodynamics, optimal control theory ete. 3.3 Manifold of One-Jets of Functions. Working with the projective space one have in mind two ordinary spaces: the affine space (to which one have to add the points at infinity) and the vector space having one more dimension (whose one-dimensional subspaces are the points of the projective space). The contact structure of the manifold of contact elements corresponds to the second approach - to the projectivization of a higher-dimensional eym- plectic manifold, namely of the space of the cotangent bundle. The manifold of contact elements has one dimension less then this even-dimensional phase space. But to obtain an odd number one may as well add one to an even number. ‘Thus we obtain the second very important example of a natural contact structure. Definition. An one-jet of a function is its Taylor polynomial of degree one (i.e. a triple consisting of a point of the manifold where the function is defined, of the value at that point and of the differential of the function at that point). 21 7]The dimension of the space of all one-jets of all functions on a given manifold V of dimension m equals 2m+1. This is an odd number and hence the space of one-jets'T*V x R should carry a natural contact structure, Definition. An one-graph of a smooth function is the submanifold of the manifold of the one-jets of functions, consisting of the one-jets of this function at all points. ‘The dimension of the one-graph of a function on a manifold of dimension ‘Theorem. The tangent spaces of the one-graphs of all functions on a manifold, passing through a given point of the manifold of the one-jets, are contained in a hyperplane tangent to the manifold of one-jets at this point and fill this hyperplane densely. Indeed, in the standard local coordinates (g; for the argument, z for the value, p; for the partial derivatives) in the space of one-jets the hyperplane is given by the equation dz = pydqy + + Padme Theorem. The hyperplanes defined in the preceeding theorem form a con- tact structure. This structure is frequently used as the local normal form of any contact structure.The coordinates (p, q; z) are called Darbouz coordinates. Example. The contact manifolds of the one-jets of the functions on the unit sphere $™ is naturally contactomorphic to the manifold of the cooriented contact elements of the Euclidean space R™* containing the sphere. To obtain from a contact element a one-jet one considers the scalar product with the point of the space at which the element is based as a function on the sphere. Consider the point of the sphere, whose radius- vector is the coorienting normal unit vector of the element. The image of the element in the space of I-jets is the 1-jet of the scalar product function at the point we have constructed. The contactomorphism that we have constructed is essentially the hodo- graph transformation, very usefull in many problems of the symplectic and contact geometries. It is missing in the textbooks that I know. This article might even be the first explicit description of this important contactomor- phism. It is nontrivial and very usefull already in the case m = 1, providing two different geometrical descriptions of the standard contact structure of the filled torus S' x R?. 223.4 Legendrian Submanifolds. Legendrian submanifolds of contact manifolds are the twins of the La- grangian submanifolds in symplectic manifolds. The corresponding Wein- stein principle should say that all interesting objects in contact geometry are Legendrian submanifolds. Definition. An integral submanifold of the hyperplane field of a contact structure is called a Legendrian submanifold, if it has the maximal possible dimension (m in a contact manifold of dimension 2m + 1). Example. The one-graphs of functions are Legendrian submanifolds of the manifold of one-jets of functions. Any nonvertical Legendrian sub- manifold of the manifold of one-jets is (at least locally) the one-graph of a function. The general Legendrian submanifolds of this contact manifold may be considered as generalized functions, in general multivalued. This generaliza- tion is highly productive, for instance one can extend to such generalized functions the Morse inequality, and this way to prove the usual and gener- alized symplectic fixed point theorems. Example. The contact elements, tangent to a given hypersurface in manifold, form a Legendrian submanifold in the manifold of all contact elements. ‘The fiber of the bundle of contact elements is also a Legendrian subman- ifold of the same contact manifold. Interpolating the two cases, consider the contact elements, tangent to any given submanifold of the base manifold. These elements always form a Legendrian submanifold of the projectivized contact bundle space, indepen- dently of the dimension of the submanifold, In many cases it is also true for subvarieti Example. The tangent elements of the semicubical parabola z? = y* form a smooth Legendrian curve in the space of the contact elements of the plane. 3.5 Legendrian Fibrations and Legendrian Singularities. Legendrian mappings are the contact brothers of the Lagrangian mappings of symplectic geometry. Definition. A Legendrian fibration is a fibration whose fibers are Legen- drian. A Legendrian mapping is the projection of a Legendrian submanifold of the space of a Legendrian fibration to the base of the fibration (along its 23,Legendrian fibers). Example. The projectivized cotangent bundle is a Legendrian fibration (sending the manifold of contact elements of a base manifold to the base manifold). ‘This example is universal: any Legendrian fibration is locally contac- tomorphic to this one (in a neighbourhood of a point of the fibred space). The (local) projective structures of the fibers are intrinsically defined by the contact structure and the fibration. Consider a smooth hypersurface in a projective space. Its tangent hy- perplanes define contact elements in the projective space. They form a Leg. endrian submanifold in the manifold of contact elements of the projective space. Project this Legendrian submanifold to the dual projective space along the fibers of the second Legendrian fibration of the manifold of contact ele. ‘ments of the priojective space ( this second fibration sends a contact element to its hyperplane, it is a Legendrian fibration according to the projective du- ality theorem). Thus the projection of a hypersurface onto the variety of its tangent hyperplanes is a Legendrian mapping. In fact this example is (locally) universal: almost any Legendrian map- ing is locally contactomorphic to the mapping of this example. The excep- tions are infinitely degenerate. They will be discussed below. 3.6 Fronts of Legendrian Mappings. The caustic of a Lagrangian mapping in symplectic geometry is a hyper- surface in the base space of a Lagrangian mapping. In the base space of a Legendrian mapping there also is a distinguished hypersurface. Definition. The image of a Legendrian mapping is called its front. As a rule the front is a hypersurface in the base space of a Legendrian fibration. Example. The dual hypersurface of a hypersurface in a projective space is the front of the projection of the manifold of contact elements of the projective space tangent to the hypersurface along the fibers of the second Legendrian fibration (see Fig. 13). Corollary. The graph of the (multivalued) Legendre transformation of ‘@ smouth function is the front of a Legendrian mapping. In fact the last examples provides the local models of almost all Leg- endrian singularities. The exceptional singularities of Legendrian mappingswhich are not contactomorphic to those of these examples form a set of codimension infinity in the space of Legendrian mappings. Example. The Legendrian collapse sends the Legendrian manifold of contact elements tangent to a submanifold which is not a hypersurface (for instance it might be a point) to the base. The image is not a hypersurface (it is the original submanifold and even may be one point). This collapse is an exceptional Legendrian mapping.Under a small generic perturbation of the Legendrian manifold the Legendrian mapping becomes generic. Fig.6 shows what happens to a point front in the plane under some generic perturbations. 3.7 Moving Fronts and the Huygens Principle. The front of a non-exceptional Legendrian mapping is a hypersurface with singularities. The term front originates from the following Example. Consider a hypersurface in an Euclidean or Riemannian space. Suppose, that the hypersurface is cooriented, i.e. equipped with a field of normal vectors. The t-equidistant hypersurface of the given hypersurface is the set of the endpoints of the coorienting normal vectors of lenghts t (in the Riemannian case - of the endpoints of the geodesic segments of lenghts t normal to the hypersurface at their initial points). One may think on the propagation of some perturbation (light, sound, epidemy) with velocity one. If the initial perturbation was bounded by the initial hypersurface, the front of its propagation at time t will be bounded by the tequidistant hypersurface (at least for the case of the propagation from a convex body in an Euclidean space). Huygens Theorem 1. The equidistant hypersurfaces are fronts of Leg- endrian mappings. Indeed, consider the geodesic flow of cooriented contact elements. This one-parameter group of diffeomorphisms g' of the manifold of cooriented - contact elements of the given Euclidean or Riemannian manifold sends each contact element normal to a geodesic to the contact element normal to the same geodesic at the distance ¢ along the geodesic from the original point ‘in the direction defined by the coorientation. Huygens Theorem 2. The geodesic fiow of cooriented contact elements preserves the natural contact structure of the manifold of contact elements. This deep result of contact geometry has been rediscovered too many times under different names in the calculus of variations, optimal control 25and nonlinear programming theories, hence I leave the pleasure to find a geometrical proof to the reader. ‘The first Huygens theorem follows, since the contactomorphism g of the fiow send the Legendrian submanifold of contact elements tangent to the initial hypersurface to the contact elements “tangent” to its t- equidistant, The generic singularities of propagating fronts are thus the same as the singularities of the graphs of the multivalued Legendre transforms of generic smooth functions. Example: one-dimensional fronts. The only generic singularity of one-dimensional front is the semicubical cusp point. (Of course there exist also transversal intersections of different branches of the front, which we here shall not mention). One can see the cusps on the equidistants of an ellipse (Fig. 6) and on the curve, projectively dual to a generic smooth plane curve. The cusp of the dual curve corresponds to the inflection point of the original curve, see Example: fronts in the three-space. The generic singularities are shown in Fig.15: the semicubical cusped edge and the swallowtail surface, studied by Kroneker. (Fig.15 here) ‘Theorem. The generic singularities of fronts of dimension n smaller then siz are simple (have no moduli) and are classified by the simple Lie algebras Ax, Ds, Ex,k Sn +1. In higher dimension moduli (continuos parameters in the classification) ‘occur, but simple Legendrian singularities still correspond to simple A, D, E Lie algebras. 3.8 Euclidean Reflection Groups and Legendrian Singulari- ties. Now we are ready to understand the relation between the fronts and the reflection groups. Definition. The discriminant of an Euclidean reflection group is the hypersurface of singular orbits in the manifold of orbits of the group. Example. Consider the group Az generated by the reflections of the plane in three symmetrical mirrors (Fig.16). A generic orbit consists of si points. It is convinient to think that our plane is the plane 2 +22 + 2: in the three-space. 0 26(Fig.16 here) The group is then generated by the transpositions of two coordinates (reflections in the planes 2; = 2,). It is the group of permutations of the three coordinates. The orbit is an unordered triple of coordinates. The manifold of orbits is the set of polynomials 2° + az +6 whose roots are the coordinates. The singular orbits are the orbits consisting of less elements then the oth- ers. They are are the orbits of the points on the mirrors. The corresponding polynomials have double roots. ‘The polynomials with double roots form a semicubical parabola 4a° + 276? = 0 on the plane of polynomials. This plane curve is the discriminant of the reflection group Az. It has a semicubical singularity. Theorem. The discriminant of an Euclidean reflection group of type A,D or E is the front of a generic Legendrian mapping at a simple sin- gularity. AU the fronts of simple generic Legendrian singularities are lo- cally diffeomorphic to these discriminants (or to the unions of transversal branches having such singularities). 3.9 Caustics of Reflection Groups. Having associated to any Euclidean reflection group a front, one can also associate to it a caustic. Example: the caustic singularity As. The manifold of orbits of the reflection group As (which is the group of symmetries of a thetrahedron) is the space of polynomials 2* + az? + bz +c. We project this space along the ¢ axis to the plane with coordinates (a,b). The discriminant hypersurface is in this example the swallowtail sur face (Fig. 17). It has two types of singularities: the cusped edge and the selfintersection line, (Fig.17 here) ‘The projection of the cusped edge to the plane is a semicubical parabolaIt is the caustic associated to the reflection group As. In the general case one considers any generic projection of the manifold of orbits along any family of curves (the final result does not depend of this choice up to a local diffeomorphism). The higherdimensional cusped edge of the discriminant is the variety of orbits of points belonging to at least two nonorthogonal mirrors. Definition. The caustic of a reflection group is the projection of the cusped edge of the discriminant along a generic family of lines. ar3.10 Caustics and Waves Propagation. In physical terms the caustic is described as the hypersurface swept by the cusped edges of moving fronts (Fig. 18). The propagation of fronts can be described in terms of the space- time as a single hypersurface in space-time - the union of the momentary fronts belonging to different isochrone hypersurfaces t = const. This hypersurface (for the case of a simple singularity corresponding to a reflection group) is locally diffeomorphic to the discriminant of the reflection group. (Fig.18 here) The cusped edges of the momentary fronts sweep the cusped edge of the discriminant hypersurface.Its projection to the physical space along the world lines is the surface in the physical space sweeped by the cusped edges of, the momentary fronts. Thus the generic lines of the mathematical definition of the caustic of a reflection group are the world lines in the space-time of the physical description. The noncristallographical Euclidean reflection groups appear in this the ‘ory as the description of the propagation of fronts in a manifold with a boundary, for instance in the problem of the fastest bypassing of an obstacle bounded by a hypersurface in the ordinary Euclidean space. Example. The fronts in the problem of bypassing of an obstacle bounded by a smooth curve in the Euclidean plane are the Huygens evol- vents (or involutes) of the boundary curve. At a generic inflection point of the boundary they have a peculiar behaviour. The surface, sweeped by these moving evolvents in the three- dimensional space-time, is diffeomorphic to the discriminant of the group of symmetry of the icosahedron (Fig. 19). (Fig. 19 here) The hypericosahedron appears in the description of the bypassing of a generic obstacle in the three-space [24]. 4 First Steps of Symplectic and Contact Topology. 4.1 Lagrangian Intersections and Symplectic Fixed Points. Consider the zero section of the cotangent bundle space of a manifold as Lagrangian submanifold of this symplectic space. We shall intersect this zero section with a neighbouring Lagrangian submanifold. Definition. A Lagrangian submanifold of a cotangent bundle space is ezact, if the action form pdq on it is exact. 28eh Example. Consider the Lagrangian curve p = f(q) on the surface of the cylinder (which is the cotangent bundle space of the circle (qmod2r). This submanifold is exact if and only if the mean value of f is eqal to zero. If the original manifold is simply-connected, every neighbouring La- grangian submanifold is exact. Suppose that the perturbed exact manifold is a section of the cotangent bundle. Then it is the graph of the differential of a function. The intersection points of the perturbed manifold with the orginal zero section are the critical points of this function. Hence Morse theory provides minorations of the numbers of intersection points of such perturbed exact Lagrangian submanifolds with the zero sec- tion of the cotangent bundle (and hence also of their mutual intersections). 4.2 Quasifunctions. ‘The Lagrangian intersection theory is a far-reaching generalization of the Morse minorations to the case of the exact Lagrangian submanifolds of the cotangent bundle space which are not sections ({4], [6], (25]). Definition. A Legendrian submanifold of the manifold of one-jets of functions on a compact manifold is a quasifunction if it can be connected to the one-graph of a function by an isotopy in the class of embedded Legendrian submanifolds. Chekanov Theorem. The projection of a generic quasifunction to the cotangent bundle space intersects the zero section at least b, times, where b, is the sum of the Betti numbers of the original manifold. Remark. The embeddings in the definition of quasifunctions can not be replaced by the immersions. This is clear from Fig. 20. (Fig. 20 here) Both Lagrangian manifolds on the cilinder T*$* are the projections of embedded Legendrian curves in the manifold of one-jets of functions. The left Lagrangian curve intersects the zero section, while the right one does not intersect it. The left curve can be continuously deformed into the right one in the class of projections of closed Legendrian immersed curves. However the evident deformations of this kind contain moments of selfintersections of some intermediate Legendrian curves. ‘The Chekanov theorem shows that such selfintersections are unavoid- able. The left Legendrian curve is a quasifunction, but the right one is not. 29Any path connecting these curves in the infinite-dimensional space of closed Legendrian immersed curves contains curves with selfintersections. ‘The Lagrangian intersections theory is closely related to the symplectic fixed points problem. Indeed, the fixed points are the points of intersection of the graph of a symplectomorphism with the diagonal of the carthesian square of the symplectic manifold. Both the graph and the diagonal are Lagrangian submanifolds of the product. ‘A neighbourhood of the diagonal is symplectomorphic to its cotangent bundle space. The graph is exact (in this neighbourhood) if the symplecto- morphism is exact. Hence the symplectic fixed point theorem follows from the Lagrangian intersections theorem provided that the symplectomorphism ‘moves each point not too far from its original place. 4.3 Neutral Quadratic Forms and Their Perturbations. All the known proofs of the theorems on Lagrangian intersections and sym- plectic fixed points depend on a generalization of Morse theory of geodesics invented by P. Rabinovitz [26]. In the classical Morse theory of geodesics the infinite-dimensional vari- ational problem is approximated by a set of finite-dimensional ones. This approximation is based on the fact that the functional that one wishes to study is a perturbation of a positive definite quadratic form. Moreover, this form becomes steeper and steeper when we travel to the higher harmonics in the Fourier representation of the elements of our infinite- dimensional space, while the perturbation remains small in some sence (at, least is growing slower then the form at high harmonics). This makes it possible to neglect the higher harmonics completely and to deduce information on the critical points in the infinite-dimensional varia- tional problem from the fiaite-dimensional Morse theory of the approximated problem. The new trick, invented by Rabinovitz, is to apply the same reasoning to the quadratic forms wich are not positive definite, but are as far as possible from the positive definiteness, having as much positive squares in the normal form as of negative squares. I shall call them neutral forms (leaving this term with no exact definition). ‘Example. Consider the integral of the action form pdq along the map- pings of the standard circle to the plane with coordinates (p,q). To parametrize the space of such mappings use the Fourier series. Intro- duce the complex notation of the point on the plane z = p+ig and represent 302(t) as the sum of Fourier harmonics aye . The complex coefficients a are the coordinates in our infinite-dimensional space. ‘The integral of the action form is a quadratic form. A simple calculation shows that it is equal to the sum of the terms kAy/2, where A, is the square of the absolute value of a,. Hence the harmonics with positive wave number k contribute positive squares, and those with negative k - negative squares, (as it should be since the integral is the area and the positive k corresponds to a rotation along a circle bounding a positive area). We see also that the quadratic form is steeper in the directions of the higher harmonics, as required. Thus the nonperturbed neutral form has the desired properties. The fact that the perturbations analysis can be reduced to a finite dimensional prob- lem (neglecting higher harmonics corresponding to the large positive and large negative wave numbers) is the infinite- dimensional version of the Hadamard-Perron-Grobman-Hartman-Anosov theorem on the dichotomy of the phase space of a vector field at a saddle stationary point, which is the main fact of the hyperbolical theory of dynamical systems. The Lyapunov idea of structural stability of the attraction, which is the basis of the usual elliptical Morse variational theory, is replaced in the new theory by the structural stability of the neutral saddles of the modern theory of dynamical system: 4.4 Lagrangian Intersections, Floer Homology and Casson Invariant. ‘The developments of the theory of Lagrangian intersections had led Floer to the eight Floer homology groups of the 3-dimensional homology spheres ((27),(28}). The idea behinds that is that the critical points of a function on a man- ifold generate the Morse complex, providing the Betti numbers of the man- ifold. The Lagrangian intersections generate critical points of a functional on an infinite-dimensional manifold. Under certain conditions one can associate to these critical points a gen- ‘eralized Morse complex and its homology. To a homology 3-sphere one can associate some Lagrangian intersections, and the corresponding nhomology an invariant of the homological sphere. In the classical Morse theory the dimension of the cycle associated to a critical point is the negative inertia index of the second differential of the function. In the Floer theory both the positive and negative indices are 31infinite and the dimension is defined only modulo eight. Floer homology is only defined for three-manifolds. One may specu- late that there should exist higher-dimensional versions, which might be invariants of contact manifolds of dimension 4n — 1 rather than of smooth manifolds. Example. The Euler characteristic of Floer homology is the Casson invariant of a homological 3-sphere (it counts the representations of the fundamental group with appropriete signs). ‘The knot of the eritical point of a holomorphic function is the intersection of its critical level hypersurface with a small sphere centered at the cr point. J. Wahl and W. Newmann [29] have discovered an astonishing relation between the the topology of critical points of holomorphic functions and the theory of Floer homology. ‘Theorem For any weigted-homogenous function of three complez vari- ables, whose knot is a homological 3-sphere, the Casson invariant is (up to a ‘universal multiplier) equal to the signature of the Milnor fiber. ‘The Milnor fiber of a function is the part of the nonsingular level hyper- surface of the function inside the small ball centered at the critical point. It is a four-dimensional manifold with boundary. The boundary is diffeomor- phic to the knot of the critical point. Example. For the function 2 + y* + z* the knot is the Poincare’ do- decahedron space, which is a homological 3-sphere. ‘The Milnor fiber is a four-manifold bounded by a dodecahedron space ‘The intersection form on its two- dimensional homology is negative definite and the signature is equal to -8. This form is, with a minus sign, the restriction of the Euclidean scalar product to the lattice, generated by the vectors with scalar square 2, whose angles are defined by the Dynkin diagram Ey. Remark. The link of the Ey singularity in five variables tajtety +0 is homeomorphic but not diffeomorphic to the 7-sphere. This smooth man- ifold is one of the 28 Milnor’s ezotic apheres, The above simple equation for the Milnor sphere had been discovered by E-Brieskorn. Unlike the Casson invariant, the signature is defined for any singular- ity.One may ask whether the Casson invariant and Floer homology can be defined in this situation. 32‘The knot has a natural contact structure and the Milnor fiber a sym- plectic structure. Its homology is generated by the vanishing cycles, which are Lagrangian spheres. ‘The intersection’ form is related to the linking of their Legendrian representatives in the knot manifold. ‘The Morse complex, corresponding to this situation, has yet to be con- structed. 4.5 The Characteristic Class Entering the Quantization Con- ditions. The existence of symplectic topology was first appreciated by the mathe. matical community when the characteristic class entering the quantization conditions had been discovered. I had christened it the Maslov class in (30). In its simplest version this class is a one-dimensional cohomology class of a Lagrangian submanifold of the cotangent bundle space of a manifold. The case where the manifold is R” and the Lagrangian submanifold lives in the standard linear symplectic phase space R* is very instructive, even for n= 1 (curves in the phase plane). The Maslov class associates to any closed curve on a Lagrangian sub- manifold an integer number (its Maslov indez), which depends only on the homology class of the curve. ‘The Maslov index of a curve is the intersection index of the curve with the caustic of the projection of the Lagrangian submanifold to the base space of the fibration (to the configuration space). It is well defined, because the caustic is a codimension one cycle on the Lagrangian manifold and has a natural coorientation. I shall not give here the definition of this coorientation (see [30]). The reader may try to guess the general definition from the example in Fig. 21. (Fig.21 here) In physical terms the Maslov index describes the well-known effect of the loss of a quarter of a wave at a caustic. This correction to the short wave approximation produces the 1/2 correction term in the Bohr-Sommerfeld quantization conditions. ‘These conditions describe the possible energy levels in a one- dimensional quantum system in terms of the area, bounded by the corresponding phase curve on the phase plane. In some units (called quanta of action and depending on the Plank con- stant value) the area (the integral of the action form) should be an integer, if the correction is not taken into account. 33‘The correction due to the existence of the caustic is 1/4 for each crossing of a critical point of the projéction in the direction defined by the natural coorientation. Hence the total correction is 1/4 multiplied by the Maslov index of the phase curve. Since this index is equal to two, according to Fig.21, the number of quanta of action inside the n-th energy level should be n + 1/2. This is the quantization condition. It was known to physicists that something similar happens in higher dimensions, at least in classically integrable systems. Maslov had tried to formulate it mathematically and had recognized the topological nature of the correction term as well as the importance of the Lagrangian submanifolds this context. By the way he had christened these manifolds Lagrangian because the so- called Lagrangian brackets ( the 18 century way to describe the symplectic structure) vanish there. In the first work on symplectic topology {3] I was, calling them aull- submanifolds (while Sophus Lie, according to Klein, had initially used the term mad manifolds for all isotropic manifolds). Using the short-waves approximation Maslov had proved in his thesis [31] that his coorientation rule defines a cohomology class modulo four. The work [30] explaining why it is an integer-valued class was initially written as a referee report. Remark, The short waves or quasiclassical approximation is usually called now WKB(1)-method in physics, by the names of quantum mechanics people using it. It seems that the method was first published by F. Carlini (1817) [32]. A detailed exposition of the method (later used by Stokes, Kelvin and many others in the 19th century) was published by Jacobi (see vol.7 of his Collected Works,pp 175-245; I am endebted to Prof. S. Graffi for these references). The Maslov index of a closed curve on a Lagrangian submanifold in R* can be defined as the number of complete rotations of the square of the determinant of a unitary matrix. Consider the complex space C” with its usual Hermitean structure. The real part of the Hermitean structure is an Euclidean structure in R?", and its imaginary part is a symplectic structure. Both the Euclidean and the symplectic structures are invariant under the multiplication by i. The La- grangian subspaces are the real n-subspaces orthogonal to their images under the multiplication by i. Each orthonormal frame in a Lagrangean subspace defines an Hermitean orthonormal frame in the complex space. Consider the unitary operator crysending the fixed standard coordinate unitary frame to the choosen orthonor- mal frame in a Lagrangian subspace. Theorem. The square of this determinant is independent on the choice of the frame in a given Lagrangian subspace. Indeed, the change of this frame is multiplying the unitary operator by an orthogonal real one. The determinant of a real orthogonal operator is +1 or -1. Hence the square of the determinant remains unchanged when the frame changes. ‘The tangent planes of the Lagrangian submanifold along a given curve are Lagrangian planes. Theorem. The Maslov indez of a curve on a Lagrangian submanifold of the Euclidean phase space is equal to the number of rotations of the square of the determinant, corresponding to the tangent planes of the Lagrangian submanifold along the curve. 4.6 Lagrangian and Legendrian Characteristic Classes. The Maslov class, dual to the caustic of a Lagrangian mapping, can be extended to a more general category of real vector bundles whose complexi- fications are trivialized [3]. There exist others natural generalizations of the preceeding constructions. Definition. The Lagrangian Grassmannian manifold is the manifold of the Lagrangian subspaces of a symplectic real vector-space. ‘This manifold is the homogenous space U(n)/O(n). Each cohomology class of the Lagrangian Grassmannian defines a charactristic class on the ‘Lagrangian submanifolds of R?", Indded, any Lagrangian submanifold of the vector-space is sent to the Lagrangian Grassmannian by the Gauss mappings (associating to a point of a Lagrangian submanifold the direction of its tangent plane). The pull-back of a class of the Grassmannian is the characteristic class of the Lagrangian submanifold. Example. The Maslov class is the pull-back of the universal Maslov class- the 1-dimensional cohomology class of the Lagrangian Grassmannian, induced from the unit circle by the mapping det”. The cohomology ring of the Lagrangian Grassmannian is well known. ‘The multiplicative generators in dimension 4k +1 are related to the Pontr- Jagin classes by the Bott periodicity shift. The 1-dimensional class is dual to the set of critical points of a La- grangian mapping, This set is generically stratified accordingly to the classes 35of singularities of caustics of different codimensions: cusps or cusped edges, swallowtails, ete. One can construct higher-dimensional characteristic classes dual to higher codimension Lagrangian or Legendrian singularities. These construc- tions provide a lot of geometric information on the coexistence and interre- lations of singularities of caustics and wavefronts. Example. The number of the swallowtails on a generic caustic of a com- pact Lagrangian 3-manifold or on a generic front of a compact 2-dimensional Legendrian manifold is even. Indeed, a swallowtail is an end point of the selfintersection line of a caustic or of a front, while at any other singular point on the selfintersection line there meet an even number of rays. I refer to [33] for dozens of less trivial examples. Since in dimensions higher than 6 we still have no explicit description of the natural stratifi- cation of the space of Lagrangian or Legendrian singularities, the combina- torics related to the Lagrangian and Legendrian singularities defining higher characteristic classes is an immense challenge. ‘The recent progress in a similar problem in knot theory, due to V. A. Vassiliev and M. E. Kontsevich, shows that the complicated combinatorics might hide rather simple and universal algebraic structures. In fact this progress in knot theory is the byproduct of the previous works on the global theory of the Lagrangian and Legendrian singularities. 4.7. Lagrangian and Legendrian Cobordisms. Consider a Lagrangian submanifold immersed into the space of cotangent bundle of a base manifold with boundary. One can define the Lagrangian boundary of the Lagrangian submanifold, which is a Lagrangian submani- fold immersed into the cotangent bundle space of the boundary of the base manifold: the projection of the intersection of the Lagrangian manifold with the boundary of the phase space along the characteristics of this boundary. Physically an immersed Lagrangian submanifold represents the short- spproximation to a wave state inside the base manifold. If the base manifold has a boundary, the wave state inside the base manifold determines the state on the boundary (think of the light inside a room and on its walls). Hence it is natural that the inside Lagrangian manifold defines a Lagrangian boundary which is an immersed Lagrangian submanifold having one dimension less. 36‘One may also imagine the restriction of a (multivalued) function to the boundary of its domain. Definition. Two immersed Lagrangian submanifolds (of the cotangent bundle space over B are (cylindrically) Lagrangian cobordant, if they form the Lagrangian boundary of a Lagrangian immersed submanifold, transver- sally intersecting the boundary of the cotangent bundle over the cilindre Bx (0,1). In the oriented version, as usually, dl = L,— Lg, where the minus means the reversal of the orientation, see Fig.22. (Fig.22 here) Example. Two oriented closed curves immersed into the plane are Lagrange (cylindrically, orientably) cobordant if and only if they have the same Maslov index and bound equal areas (have equal action integrals). As usually, one defines the addition as the disjoint union, and form the (commutative) semigroup from cobordism classes. This semigroop is in fact. a group. Example. The group of Lagrangian (oriented cylindrical) cobordism classes of plane curves is Z+ R. ‘The characteristic numbers of cobordant objects are equal. Hence one can use the numbers of point singularities of Lagrangian mappings to distin- guish cobordism classes or use the information on the classes to understand the adjacencies of singularities. ‘There exists a similar theory in contact geometry for Legendrian sin- gularities and cobordisms. In this case one may simply consider the fronts cobordisms, defined as cobordisms of stratified varieties - they represent faitfully the Legendrian cobordisms of immersed Legendrian subvarieties in the contact spaces. ‘Theorem. The Legendrian (oriented cylindrical) cobordism group of Legendrian curves immersed into the standard contact $-space is isomorphi- cal to the group of integers Z. The generator is the class of the Legendrian curve whose front is the bow-tie curve, shown in Fig. 23. (Fig. 23 here) I leave to the reader the pleasure to find the geometrical proof of this fact: one can really decompose any front having no vertical tangents into bow-ties. Our fronts have no vertical tangents, since our 3-space with coordinates (P,4;2) catties the standard Darboux contact structure dz = pdg. ‘The front lives on the plane with coordinates z and g. It is nonvertical since its inclination p is finite. 37‘The number of the bow-ties in the decomposition can be read immediatly from the given generic front. Recall that a generic front has only cusp singularities and that our fronts are coriented (since they have no vertical tangents). Choose the coorienting one-form dz — pdg. Definition. The cusp on an oriented and cooriented front is positive (negative) if the orienting motion leaves the cusp point in the direction to the side where the coorienting one-form is positive (negative). Example. Both cusp points of the bow-tie front in Fig.23 are positive. For the bow-tie with the opposite orientation (but same coorientation) the cusps are negative. ‘Theorem. The difference between the number of positive and the num- ber of negative cusps of an oriented and cooriented front is an invariant of the oriented Legendrian cobordisms. Hence the number of the standard bow-ties to which the front is cobor- dant is equal to one half of the above difference which is easily computable. Remark. The difference between the numbers of positive and negative cusps is called the Maslov indes of the front (or of the Legendrian curve). It is indeed the Maslov index of a curve on a Lagrangian submanifold of the Lagrangian fibration space (to obtain these objects one should symplectize the given Legendrian and contact objects). ‘The Maslov index is the only oriented Legendre cobordism invariant of the oriented and cooriented fronts in the plane. Theorem. The nonoriented Legendrian curves immersed into the stan- dard 3-space form a trivial cobordism group. Indeed, the bow-tie cobordism to the void curve is shown in Fig.24. The Legendrian surface, represented by this sequence of transformations of the front, is a Moebius band, bounding a circle (whose projection is the bow- tie). Since any front is cobordant to several bow-ties, it is also cobordant to zero. (Fig.24 here) ‘The Legendrian cobordism groups are simpler then the Lagrangian ones, since the invariants are discrete (unlike the area in the Lagrangian case). I only quote from the book of M. Audin [34] the ‘Theorem. The Legendrian oriented cobordism groups of the standard Euclidean spaces form a graded anticommutative ring. This ring (tensorized by the rational numbers to kill the torsion) has one free multiplicative gen- erator in each degree of the form 4k +1. 38The nonoriented Legendrian cobordism groups form the ring Zafts, 28,2115] (one generator of each odd degree except the degrees of the form 2 ~ 1). ‘The Legendrian cobordism rings formed by the Legendrian submanifolds of the manifolds ST*R™ of the cooriented contact elements of the Euclidean spaces are naturally isomorphic to the rings of the above theorem (describ- ing the Legendrian submanifolds of the spaces of 1-jets of functions on the Euclidean spaces). 4.8 Lagrange Embeddings and Inclusions. ‘A circle can be embedded into the plane and hence there exists an embedded Lagrangian torus in the standard phase space R‘. However this embedding is not exact: the integral of the action form is a multivalied function on the torus. Gromov Theorem. No compact smooth manifold admits an ezact La- grangian embedding into the standard symplectic space R?". This theorem has a simple topological proof for all surfaces different from the torus, for which the theorem is really very hard and was proved only recently by the methods of the theory of (pseudo)holomorphic functions of complex variable. ‘This technique has been introduced into symplectic topology by Gromov [6] and the most striking results in symplectic and contact topology have ‘been up to now obtained only by this method, which seems to be foreign to the subject. Ite strange force is a cousin of the well-known power of the variational approach to the elliptical PDE’ Definition. A Lagrangian inclusion is a smooth mapping of a manifold into a symplectic manifold, which is a Lagrangian embedding in a neigh- bourhood of almost every point (and which hence induce the zero 2-form from the symplectic structure). Example. The conormal bundle space of a subvariety of the config- uration space is the variety of the cotangent vectors of the configuration space, vanishing on the tangent vectors of the subvariety. The Givental open umbrella singularity is the singularity of the conormal bunle space of 2 semicubical parabola in the plane. ‘The Givental singularity is (the image of) a Lagrangian inclusion (of the plane). Topologically this surface in R* is nonsingular. But it has one point of nonsmoothness. 39Indeed, the Givental surface is parametrized by two parameters (1 along the parabola and s across it), namely = tg = P,P = ~38t, Pr = 2s. Theorem (Givental, Ishikawa, Zakaljukin). The only singularities of the generic Lagrangian inclusions of a surface into the 4-space are the transversal selfintersections and the Givental singularities. These points are stable, Theorem (Givental [35]. Any orientable surface of genus g > 1 ad- mits a Lagrangian inclusion into the standard symplectic Euclidean 4-space with 2g — 2 Givental open umbrella singularities and no selfentersections (topologically this inclusion is an embedding). ‘Any nonorientable surface of negative genus —4k admits a Lagrangian embedding into the standard symplectic Euclidean 4-space (with no singular- ities at all). ‘The exact Lagrangian inclusion of the real projective plane into R¢ is represented in fig. 25 by its front. (Fig. 25 here) This front is a surface in R°. This surface has a semicubical cusped edge and a selfintersection line, as the fronts of smooth Legendrian varieties have, but they meet at three points. The singularities of the front at these points are called folded umbrellas. At a neighbourhood of a folded umbrella the front can be described by the normal form y? = 22 (Fig. 26). (Fig. 26 here) As the ordinary Cayley - Whitney umbrella y? = 22?, the folded umbrella contains the handle (z = y = 0, < 0), shown in Fig.26 as a dotted line. The handle is not included into the real front. Both umbrellas are not too usefull in case of a rain, as it is clear from Fig.26. The Cayley-Whitney umbrellas are the only singularities of the generic mappings of the surfaces into the three-space (besides the transversal cross- ings of two or even three branches). ‘The mapping in the neighbourhood of such a singular point can be writ- ten in the Whitney normal form (z = u,y = uv,z =v), where u and v are the local coordinates on the surface. The folded umbrella is a standard element of many singulariti plectic and contact geometries. Example. The tangent lines of a generic smooth curve in the Euclidean 3-space sweep a surface. The original curve is the cusped edge of semicubical in sym- 40type of this surface. At the points where the torsion of the curve vanishes the surface is locally diffeomorphic to the folded umbrella. These folded umbrellas and double points obstruct the smooth La- grangian embeddings of surfaces into the Euclidean space. One can not eliminate them deforming the embedding. ‘Two folded umbrellas may replace one double point (see Fig.27). (Fig. 27 here) Each part of Fig.27 represents a front in the three-space. Each front consists of two surfaces 2 = f(z,y), z = —f(z,y) connected along the cusped edges, where z = 0. The shape of the graph of function f is shown by the level lines and by the directions of the fastest descent, ‘The Lagrangian submanifold corresponding to the left front has no self- intersections. Indeed, the planes, tangent to the front at the two points with egal (z, y) are nowhere parallel,since they are symmetric with respect to the plane z = 0 and are not horizontal, the function having no critical points outside the cusped edge. ‘This cusped edge contains two folded umbrellas and the corresponding Lagrangian submanifold of R¢ has two open umbrellas. ‘The Lagrangian submanifold, corresponding to the right front, has one selfintersection point (corresponding to the critical point of f). It has no Givental open umbrellas, 4.9 Lagrangian and Legendrian knots. A smooth manifold may have many different embeddings into an Euclidean space. Two smooth embeddings are called isotopical if they belong to the same connected component of the space of embeddings. The classification of these components is a generalization of the knot theory, studying the embeddings of a circle into the 3-space. ‘An equivalent (and even in a sence more close to the reality) definition of a knot starts from a standard straight line in the 3-space. The knot is represented by an embedding, which differs from the standard one only in a finite ball (Fig. 28). (Fig.28 here) Definition. A Lagrangian knot type is a connected component of the space of Lagrangian embeddings of R" into the standard symplectic space 2%, differing from the embedding of a fixed Lagrangian plane only inside finite balls. Problem ([4]). Do there ezist nontrivial Lagrangian knot types? 41The trivial type is the one containing the standard Lagrangian plane embedding. Theorem (Luttinger, Eliashberg, Polterovich, (86]). All La- grangian knots in the standard symplectic 4-space are trivial in the sence of differential topology. In other words, any Lagrangian perturbation of a Lagrangian 2-plane which does not move the points outside some ball is unknotted: one can connect the plane with the perturbed surface by a continuous deformation in the class of smooth (not necessarily Lagrangian) embeddings. ‘Moreover, this deformation might even be realized by a time-dependent flow, fixing the points outside some ball and transforming the plane into the perturbed surface at time 1. If this flow were Hamiltonian, all Lagrangian knots would be trivial. The theorem above means that nontrivial Lagrangian knots, if they do exist, are purely Lagrangian- topologically they are trivial. Whether higher-dimensional Lagrangian knotes may be topologically nontrivial is unknown. ‘A Legendrian curve in the standard contact RS or S* may be considered as a knot in the usual sence and every knot has such Legendrian represen- tatives. However the type of the knot in the usual sence does not define its Legendrian type (the component of the space of Legendrian embeddings). ‘Example. The Legendre curve whose front is the bow-tie is in the same trivial class of ordinary knots as the “lips”curve (Fig. 29). But they are different as Legendrian knots in R°. (Fig.29 here) Indeed, the “lips tie is not. The Maslov index, distinguishing the Legendrian cobordism classes, is not the only invariant of Legendrian isotopies of curves, even if they are unknotted in the ordinary sence. Definition. The Bennequin invariant of a Legendrian curve in R° is the intersection index of an oriented surface bounded by the curve with a curve obtained from the given one by a small shift in the direction orthogonal to the contact plane. Example. The Bennequin invariant of the curve shown in Fig. 30 is ba1+s4+2=1 (Fig. 30 here) Bennequin Theorem ([37]). The value of the Bennequin invariant on an unknotted curve in the standard contact R° is positive. curve is Legendrian cobordant to zero, while the bow- 42Remark. In asence the Bennequin invariant (which might be defined for higher-dimensional Legendrian submanifolds of topologically trivial contact manifolds) is a quadratic form. “For’instance, this selfiinking number is multiplied by 4 if the curve is covered twice. The Bennequin inequality claims that this form is positive definite. Bennequin had used this inequality to prove that some twisted contact structure in R° is exotic. Namely he had constructed for the twisted struc- ture a Legendrian curve, unknotted in the ordinary sence, whose Bennequin invariant (depending on the contact structure) is not positive. Eliashberg-Fraser Theorem ([38]). The Maslov indez and the Ben- nequin invariant are the only invariants of Legendrian isotopies of the topo- logically unknotted Legenrdian curves in the stanard contact space R°. The fronts of the Legendrian curves representing all the classes are shown in Fig.30.To obtain all the classes one has to consider both orientations of the fronts with s,t = 0,1,.... 4.10 Contact and Symplectic Structures. ‘The first example of an exotic contact structure in R® has been constructed by Bennequin. At present all such structures are known. Theorem (Eliashberg, [39]). The complete list of nonequivalent con- tact structures on R® is countable. ‘The list itself can be found in [39]. The theorem schould be compared with the fact (due to the same author) that there exists a continuous family of pairwise noneqivalent contact structures on the filled torus S* x R?. The contact structures on a closed manifold are rigid: any small defor- mation is trivial (J. Gray (40]). It means that the deformed structure is equivalent to the initial one (one can obtain it from the initial one by a diffeomorphism, closed to the identity.) ‘The homotopy type of the hyperplane field, defining the contact struc- ture, is of course an invariant of the structure. ‘Any closed 3-manifold has a contact structure (J.Martinet, [41]). ‘The fields of tangent hyperplanes on S* are classified by the integers. Indeed, this sphere is a group (the Spin(3) or SU(2) group). We identify all the tangent spaces using the left translations on the group. The field is then described by a mapping 5° + $?. The homotopy classes are labeled by the Hopf invariants of these mappings. ‘Theorem (Eliashberg, [42]). Every class of plane fields on S* con- tains (up to isotopy) ezactly one contact structure, with the ezception of the 43zero class, which contains ezactly two contact structures. V.L.Ginzburg [43] had defined the natural contact cobordisms groups.He had proved that all the contact cobordism groups of the contact manifolds of dimensions 4k +3 are trivial while the contact cobordism group of the contact manifolds of dimension of the form 4k + 1 consists of two elements. Example. The contact circle is not contact cobordant to zero. It means that a compact oriented surface bounded by a circle has no line fields transversal to the boundary circle. The union of two circles is contact, cobordant to zero, since there exists a line field on a cilinder transversal to both boundary circles. In the symplectic cobordism theory of Ginzburg the situation is quite different. The invariants of symplectic cobordism are the integrals of the differential forms, which are the products of the Chern forms and of the exterior powers of the eymplectic structure. Hence the cobordism groups are not finitely-generated. Example. The only cobordism invariants of symplectic structures on surfaces are the symplectic area and the Euler characteristic (which of course are additive cobordism invariant 4.11 Existence of Symplectic Topology. Most natural questions in symplectic and contact topology are yet unan- swered, but one thing is firmly established: the symplectic and contact topology do exist. This statement is not a constatation of mathematical activity in these domains, but a technical result: a highly nontrivial theorem. The definitions of the symplectic and of the contact structures use the differentiable structure of the underlying smooth manifold. The topological symplectic or contact geometry should be invariant under the homeomor- phisms preserving the structure. Example. A closed curve dividing the 2-sphere into two parts of equal areas intersects the equator at least at two points. This is true (for the Lebesgue area) for the nonsmooth Jordan curves. Hence this fact belongs to the topological symplectic geometry. Similar examples show that the toplogical volume-preserving geometry does exist. However the symplectic case is more delicate, when the dimension is higher then 2. Instead of the patological topological ob jects one may consider the limits (in the C° topology where no convergence of the derivatives is supposed) of smooth objects (Lagrangian submanifolds, symplectomorphisms e.t.c.). The 44problem arises whether these limits preserve the properties of the smooth original objects. Of course, the limit should not-be too bad. For instance, the limit of a sequence of Lagrangian curves might be the whole phase plane, which of course should not be considered as an honest topological Lagrangian sub- manifold. ‘The resulting topological theory would preserve the traces of the sym- plectic (and not just volume-preserving) geometry only if the topological lim- its will not change their symplectical status provided that they are smooth submanifolds or mappings. ‘Theorem (Eliashberg - Gromov, (6]). A diffeomorphism of a com- pact symplectic manifold is a symplectomorhism, provided that it is the C°- limit of symplectomorphisms. ‘Theorem (Laudenbach-Sikorav, 1993). A smooth embedding of a closed manifold into the standard symplectic space is Lagrangian provided that it is the C® mit of Lagrangian embeddings of the same closed manifold. Similar results hold for the contactomorphisms and for the Legendrian submanifolds. In spite of the recent proof of these long awaited theorems, the real construction of the topological symplectic and contact geometries (even of the PL- version, which is evidently important in optimal control theory) is yet to be achieved. Remark. I use the names topological symplectic and contact geometries for the studies of the symplectic and contact properties of nonsmooth ob- Jects, reserving the names symplectic and contact topologies for the study of the discrete invariants of smooth objects in the symplectic and contact manifolds. I think, moreover, that topology is the stady of the discrete invariants of various objects in ail the geometries, rather then the study of the homeo- morphisms invariants. (Geometry, as everyone knows, is the study of those properties of different objects in the space where a group is acting, which are invariant with respect to this group action). Ina recent American book I have read that (differential) geometry is the art of making no mistakes in long calculations. This definition seems to be too restrictive. 454.12 The Symplectic world. Symplectic and contact geometries are of course differential geometries of manifolds with some additional structures. Some rather natural axioms had led E. Cartan to a small list of natural geometries of this kind, associated to the simple (pseudo)-groups of diffeomorphisms. ‘The Cartan list of simple pseudo-groups contains real and complex differ- ential and volume-preserving geometries, symplectic and contact geometries and few conformal versions of the preceeding geometries. This list is some- what similar (and closely related to) the Killing list of simple Lie algebras discussed above (see Fig.10). It is well known in the theory of Lie algebras that practically any fact of matrix theory can be reformulated in coordinate-free terms of the so-called root theory (which is a natural extension of the theory of the eigenvalues). The roots of the linear group form the simplest Dynkin diagram Ay. Once formulated in terms of the roots the result becomes meaningfull for the other Dynkin diagramms. This way one obtains at least a conjecture valid for all simple Lie algebras. In most cases one can then prove the conjecture (sometimes slightly correcting it). ‘This method unifies the geometries of finite-dimensional simple Lie al- gebras. It seems that a similar unification might be usefull in the infinite dimensional case of the simple diffeomorhiems groups. {From this point of view the symplectic, contact or holomorphic ge ‘ometries and topologies should be considered as the sisters of the ordinary geometry and topology rather then as the parts of it, the same way as in the theory of simple Lie groups one consider the orthogonal group as the sister of the linear group rather then as its subgroup (in spite of the fact that the orthogonal group és the group of linear transformations preserving an additional structure). ‘One may thus imagine that quite a few of the notions of the ordinary geometry and topology of manifold might have parallels in the symplectic, contact, complex and other geometries. Example. The following list of parallel objects in the real and in the complex geometries is well known (see e.g.(4]): 46Ro O(n), SO(n) U(n), SU(n) Tm ™ K(x,0) K(x,1) Zz RP" CPm Stiefel-Whitney classes Chern classes S'=RPi 9 S'=R/Z st Morse theory Picard-Lefachetz theory symmetric group braid group boundary ramified covering. orientation of R" element of x1(U(n)) It is clear that the complexification of the “homology” is not at all ho- mology theory in the sence of homological algebra, but some algebraically completly different object (yet to be discovered). Similarly to the complexification, the symplectizations and contactiza- tions of the usual geometrical objects may be very different from the original objects. There yet exist no axioms for the mathematical operations of this rank (like quantization, superization, symplectization).. However one may guess that in many cases the symplectization can be achieved by the following procedure. One starts from some object in an ordinary manifold. One symplectize it, considering the cotangent bundle space and associating to the original object some kind of its prolongation (the conormal bundle space of a submanifold, the action of a diffeomorphism on cotangent vectors,e.t.c.). Finally one can try to generalize the properties of the resulting objects to make them symplectically invariant. ‘This way one obtains in the preceeding examples the Lagrangian mani- folds or varieties starting from the ordinary manifolds or varieties, the Hamil- tonian vector fields from the ordinary vector fields and so on. It seems that the Morse number and the Lagrangian intersection theory have something to do with the symplectizations of the Euler characteristic and of the Lefschetz fixed points theorem of ordinary topology, while La- grangian and Legendrian cobordism theories might be the symplectization and the contactization of the usual cobordisms. Most of the branches of mathematics (from the theory of PDEs to the calculus of variations, from the theory of group representations to number theory) have been symplectized or are under symplectization these years. In this brave new symplectic world the old concrete theories live a new 47life in the family of the cousins whith whom they were separated before the symplectization (similarly to the ellipse, the parabola and the hyperbola, which were separated before the projectivization). One starts from some concrete theory. Say, one considers the elementary theory of the Legendre transformation or the geometry of the equidistant surfaces of a surface in the Euclidean 3-space, or the classical theory of pedal surfaces (also called the derivative surfaces). The derivative of a hypersurface in the Euclidean space is the variety, formed by the feets of the perpendicular from the origin to the tangent plane of the given hypersurface (Fig.29). The primitive hypersurface of a given hypersurface is hence the enveloppe of the hyperplanes, orthogonal to the radius- vectors of the points of a given hypersurface. (Fig.29 here) In symplectic (or better contact) geometry all these hypersurfaces show their true face: they are the fronts of Legendrian mappings. It follows, for instance, that the singularities of the Legendre transforms, of the equidis- tants, of the derivatives and of the primitive hypersurfaces coincide. Hence it is sufficient to study any of these objects to master all the others. Riemannian geometry is a special case of the symplectic geometry of pairs of hypersurfaces in symplectic manifolds. This remark allows one to use the experience and intuition of the Riemannian geometry and even of the elementary geometry of the surfaces in the usual Euclidean 3-space to derive results in symplectic and contact geometries of pairs of hypersurfaces. The resulting theory has usefull applications to other problems in the initial 3-space, in particularly to the variational problems with one-sided constrains, holonomic or not, and to the optimal control problems. ‘The power of the symplectic and contact geometries depends on the unification of many apparently disjoint branches of mathematics, provided bby these theories. It is comparable to the unification of most branches of mathematics provided by the linear algebra or, in other terms, by the projective geometry. Some 19th century mathematicians were objecting to the tendency of projectivization of the affine and Euclidean geometries. Cayley had finally settled the problem, proclaiming “Projective Geometry is All Geometry” In the same sence one might now say “Symplectic Geometry is All Ge- ometry”, but I would prefer to formulate it in a more geometrical form: Contact Geometry is All Geometry. 48[17] VA. Arnold, A.B. Givental. Symplectic geometry. In: Encyclopaedia of Math. Sciences, Dynamical Systems 4. Springer, 1990, pp. 4-136. [18] Vil. Amold. On one problem of Liouville, concerning integrable problems of dynamics, Siberian Math.Journ., 4:2 (1963), 471-474. [19] VL Arnold. Normal forms for functions near degenerate critical points, the Weyl groups As,DsyEy and Lagrangian singularities. Funct. ‘Anal. and Its Appl., 6:4 (1972), 254-272. [20] V.L. Arnold. Critical points of smooth functions. In: Proceedings of the International Congress of Mathematicians, Vancouver 1974, 1, 19-39. [21] V. Batyrev. Dual polyhedra and mirror symmetry for Calabi- Yau hypersurfaces in toric varieties. Journ.Alg.Geom., 3 (1994). [22] LG. Scherbak. Focal set of a surface with boundary and caustics of groups generated by reflections By, Ce, Fs. Funct. Anal. and Its Appl., 18:1 (1984), 84-85. {23] A. B. Givental. Singular Lagrange varieties and their Lagrange map- pings. In: Ttogi Nauki VINITI, 38 (1988), 56-112, transl. as Journ.of Sov. Math., 52:4 (1990), 3246-3278. 24] 0. P. Scherbak. Wave fronts and reflection groups. Russian Math Surveys, 43:3 (1988), 149-194. [25] Yu. V. Chekanov. Legandrova teorija Morsa. Uspekhi Mat. Nauk, 42:4 (1987), 139-141. (Uspekhi are in general translated as Russian Math. ‘Surveys, but some papers contain something new and hence are not trans- lated!) F. Laudenbach, J.-C. Sikorav. Persistence d’intersections avec la sec- tion nulle au course d’une isotopie hamiltonienne dans un fibre’ cotangent. Invent. Math., 82:2 (1985), 349-358. [26] P. Rabinovitz. Critical points of indefinite functionals and peri- odie solutions of differential equations. In: Proceedings of the International ‘Congress of Mathematicians, Helsinki 1978, pp.791-796 (Ed.: Acad.Sci. Fen- nica, Helsinki 1980). [27] A. Floer. An instanton invariant for $-manifolds. Comm. Math. Phys., 118:2 (1988), 215-240. [28] M. Atiyah. New invariants of $- and 4- manifolds. In: The math- ematical heritage of Hermann Weyl. Durham, N.C., 1987. Sympos. Pure ‘Math., 48, Amer. Math. Soc., Providence, RI, 1988, pp. 285-289. [29] W. Newmann, J. Wahl. Casson invariant of links of singularities. Comment. Math. Helvetici, 65 (1990), 58-78. {30] V. 1. Amold. On the characteristic class entering in quantization conditions. Funct. Anal.and Its Appl., 1:1 (1967), 1-14. 50{31] V. P. Maslov. Theorie des perturbations et me’thodes asympto- tiques. Thesis, Moscow State Univ., 1965. Paris, Dunod 1972. (32] F. Carlini. Richerche sulla:convergenza della serie che serve alla soluzione del Problema di Keplero. Milano 1817; Schumacher Astronomische Nachrichten, 28:665, 257-270, 30:709-712, 197-254. [33] V. A. Vasiliev. Lagrange and Legendre characteristic classes. Gor- don and Breach, 1988. [34] M. Audin. Cobordisms dimmersions Lagrangiennes et Legendri- ennes. Traveaux en Cours,20. Hermann,1987. {35] A. B. Givental. Lagrangian imbeddings of surfaces and the open Whitney umbrella., Funct. Anal. and Its Appl., 20:3 (1986), 35-41. [36] ¥. Eliashberg, L. Polterovich. Unknottedness of Lagrangian surfaces in symplectic 4-manifolds. Preprint 1993, 9pp. [37] D. Bennequin. Entrelacements et equations de Pfaff. Aste’risque, 107-108 (1983), 83-161. [a8] Y. Eliashberg. Legendrian and transversal knots in tight contact S-manifolds. In: Topological Methods in Modern Mathematics, J.Milnor’s 60th birthday volume. Publish or Perish, Houston 1993, pp. 171-193. [39] Y. Eliashberg. Classification of contact structures on R°, Duke Math. Journ., International Math. Res. Notices, 1993 N3, 87-91. [40] J. W. Gray. Some global properties of contact structures. Ann. of Math., 2:69 (1959), 421-450. [41] J. Martinet. Formes de contact sur les varie'te’s de dimension 3. Lect. Notes in Math., 209 (1971), 142-163. [42] Y. Eliashberg. Contact 3-manifolds twenty years since J. Martinet’s work. Annales de lInst. Fourier, 42:1-2 (1992), 165-192. [43] V. L. Ginzburg. Calculation of contact and symplectic cobordism groups. Topology, 31:4 (1992), 757-762. [44] B. Fortune, A.Weinstein. A symplectic fized point theorem for com- lez projective spaces. Bull. Amer. Math. Soc., 12:1 (1985), 128-130. [45] A.B.Givental. Nonlinear generalization of the Maslov indez. In: Singularity theory and its applications, V.LArnold Editor, Advances in So- viet Mathematics, 1, AMS, Providence R.L.,1990, 71-103. 51Figures Captions. Fig.1. The symplectic camel problem. Fig.2. The symplectic packing problem. Fig.3. An exact symplectomorphism of a torus. Fig.4. The vertices of a plane curve. Fig.5. The first caustic of a point on a surface. Fig.6. The eqidistant curves of an ellipsis. Fig.7. The nested tori of an integrable system. Fig.8. The caustic of an ellipsis. Fig.9. The swallowtail, the purse and the piramide. Fig.10.The Dynkin diagrams. Fig.11.The caustic of the group Fy. Fig.12.The cristallographic and noncristallographic plane reflection group: Fig-13.The projective duality. Fig.14.The cusp dual to an inflection. Fig.15.The generic singularities of a front in the 3-space. Fig.16.The mirrors, the orbit space and the discriminant of the triangle symmetry group Aa. Fig-17.The caustic of the reflection group Ay as the projection of the cusped edge of its discriminant. Fig.18.The sweeping of a caustic by the cusped edges of the fronts. Fig.19.The discriminant surface of the icosahedron symmetry group Hs and the evolvents of a curve with an inflection point. Fig-20. A quasifunction and a nonquasifunction. Fig.21. The Maslov index. Fig.22. An (oriented cilindrical) Lagrangian cobordism. Fig.23. The bow-tie. Fig.24. The (nonoriented) Legendrian cobordism of the bow-tie to zero. Fig.25. An exact Lagrangian inclusion of the projective plane into the Aspace. Fig.26. The folded umbrella and the Whitney- Cayley umbrella. Fig.27. The Givental handles. Fig.28. A knot on an infinite string. Fig.29. The fronts of two different purely Legendrian knots (knots, which are unknotted in the usual sence). Fig.30. The classification of the purely Legendrian knots in R°. Fig.31. The derivative and the primitive of a hypersurface. 52Fig. QD | Fig.2 BEES |Fig.Ronuect Ineo tf Tavigeuce wt mw the confi bon a. (Secgeste gu i fk 12a) curs edagee te jBouudlazs (et & placed Dall the i cucuesGuwed ak hy Uvex o the sulfa are truge [bm the cout Oct ead ¢ Ue. seckio, Leyes The wf aefct ceusp rh foc eh focal LaineFro.9 ole Ap Se Be - Soe De eee Ce uh oo Ee — %\ ae Ee ca) Ga ied ye i ae. x|prey prep feAa\ reffection space or@it space Fig (6the & AcAst upstaté thowkd prejector8it Space = = space-time,tu. phase space ured agra ng@an sug, n 5, t projection the. Lagrang, ke gase congéguration manifoed, SON a“ tho. Sulit Fig. 24 she thy: acrows showin “ewes of the, ets ¢ &OC a= aint > p= Sea st B= aindt + 2. Aim 2t Fig. 25 OC Sor BLAGamers He stiruyt {vis Hush) BS open um@reffas point Feg QE Symplectic space ee a a swamantola Sg 28Mastoy indeac=2 Massey index =() Gow— tie dips t=2 4=6 Mastev index = 28=42 Bennequin inviactaut Ba4rar2t = ALes . M4 all s ingkecti onal é Cuor tangent fing . are aud < eriginag cucve orig ty cucu Oits the & age £41 unch sithogc