NONCOMMUTATIVE SMOOTH SPACES ‘Maxim Kontsevich, LILES. 1. Motivations from Matrix Integrals (One of lessons fiom theoretical physics of the lest decade is the following principle: ‘if we have integrals Iyy naturally associated with Nx N matrices, then the asymptotic expansion of login’) a8 N00 should be YoegN?* + few anomalous terms ~ Mare precisely, physicists expect that a kind of nomperturbative string theory will arise antomatically from matrix integrals, and 900 expansion is the same as the gems expansion in string coupling constant Aw NL ‘The origin of this principle is an old observation by t’Hooft that in the Feynman diagrammatic expansion for a system with U(V symmetry all ters are organized naturally in groups labeled by graphs drawn of oriented surfaces. Moreover, the dependence on parameter 1V goes only through the factor N2-9, where g is the genus of the surface. Also, many combinatorial questions cancerning, the enumeration of triangulations of surfaces can Le analyzed using appropriate matrix models. ‘There are very impressive examples of the stringy behavior as N00 for many models, Here are two simple cases. . A. The volume of the unitary group UGV) is const” T]™, 1G). One can chede that log val(U 2 = 5 ten VIN Deqgn pe B. One-matrix model, Let @ : R-+R be a function increasing sufficiently fast at -to0. Define Inv as the integral J exp: taken over the space of hermitean N x N matrices, Then, as N— oo, it has an asymptotic expansion Vlog (N 2 Ha). o,x0 Fad Trace(®(X)))a X logit) = Fer typical & only a term eo? will appear, then in codimension 1 in the space af functions one meets an universal series related to the Airy function, ete. In general, mmiti-matrix models are integrals of exrp(—N Trace((X;,. ..,X,))) where X; are hermitean JV x N matrices. In one of modds in M-theory the (bosonic part of the) action @ is S"EXi, XP. This pression is invariant under the action of the group of motions of Re: Xi Lay Xj+ balan, (aij) € OR), (bi) ER" T The set of absolute minima of Trace(®) consists of collections of pairwise commuting hermitean matxi- ces, The quotient of this set by the adjoint action of U(V) can be naturally identified with (RAN /Ziy (configurations of V identical particles in R'). 2. Algebraic formilation ‘The manifold My == R®” over which the multi-matrix integral is taken, can be interpreted as the set of homomorphisms of salgebras: A—+Mat(N x N,C) where A= Aq is the five algebra Cry.--,.7i) endowed with the anti-involution 2} ==}. This set is the set of real points in the complex affine space A@“(C) of all homomorphismns from A to Mat(N x N,C). Finetion @ in the action can be considered as an dement of vector space A/[, A], because traces of commutators vanish. Factor V can be waitten as Trace(1.4). Tn general, the expression under the exponent. is linear functional Trace @ Trace applied to an element of A/[A, A]@ A/[A, A]. 1Lemma, IF € is a derivation of the #algebra A= Ag, then the divergence of the induced vector field on at any paint f : A—>Mat (©) is obtained by the application of the map Traces ®Traces to an universal ekment Div() € A/[A, A] ® A/[A, Al Tt follows frem this Iemma that the dass of measures on collections of manifolds (M; . form exp (Trace®Trace)(S(X1y +++ Xa) X1 «Xa where S € A/[A, A]@.A/[A, Al, is invariant under infinitesimal automorphisms of A. 3 Smooth algebras Definition. An associative algebra A over a field k i formally smooth iff for any algebra B andl a nilpotent two-sided ideal I in B, and a homomorphism of algebras A—>B/T, there exists a lifting A—>B, ‘This definition was proposed by D. Quillen and J. Cuntz. It is a copy of the standard definition of the formal smoothness in the commutative case. Lemma, Algebra A is formally smooth iff the bimodule 01(A) := Ker(A ® A—+A) (where the map is 4 b+ ab) is projective, IF A is formally smooth then the global colomological dimension of the category of Acmodules is <1, i, Ext22(E,, F) = 0 for any two Aemodules E and F. J. Cutz, and D. Quillen found also a very explicit criterion for the formal smoothness, which is convenient, for the case of finitely presented algebras. Lemma, IF A is a finitely generated and formally smooth then for any N > 1 the scheme My of homomor phisms from A to the matrix algebra Mat(N x N,k) is a smooth affine scheme of finite type. propose to consider smooth algebras (that is, formally smooth finitely generated algebras) as machines producing an infinite system of usual smooth schemes (1 2 “Tae are sce akin structures cn mails My ford. he group G(R) ats grace Vi,Nb there is a map © = My [x.—>Myyi-n eqivatiant with respect t0 sk) X GL(Nayk) C GL(Ni + Nask). Manifolds My ‘ns Gn cones nat edbetn of Na ns powers of an object in a symmetzic monaidal category; endowed with a compatible system of extensions of Sy-actions to GL(N,k)-actions. Here are besic examples of smpoth algebras: a) (the main example) five finitely generated algebra k(ra ) the matrix algebra Mat(n x nsk), ) the algebra of fimctions an a smooth affine curve, ) the algebra of paths in a finite oriented graph, ©) localization A(r~?) where A is smooth and « € A is any element. Smooth algdtres can be considered as (dual to) “noncommutative smooth affine schemes”. Exampks of non smooth algebras are algebras of functions on algebraic varieties of dimension > 2, algebras of differential operators, algebras of functions on quantum groups, etc. 4. Differential geometry on smooth algebras One can define analogues of many’ ustal differential-geometric notions for smooth algebras. The main idea is to fine a structure on algebra A producing desired structunes on all representation manifolds M; For example, vector fields on (My arise from derivations of A, fimctions arise from dements in A/[A, (via traces), differential forms arise from elements in 0* 4/[e4,0* A] (Karoubi's complex). Periodic cyclic homology spaces of A for formally smooth A are (by a theorem of Cuntz-and Quillen) the same as cohomology spaces of the Karoubi complex. Thus, I{P,(A) maps to He pam (Mn) for any ) One can define also the notion of a symplectic 2-form, For example, in the case A= KXp)/(p? =p) one has a symplectic structure on all My given by the fornmala Trace(pdpdp). P. Olver discovered! generalizations of integrable systems (such as KdV) to the free noncommntative caso. Passing to spaces of matrix representations one get. a secnience of ordinary integrable systems. 5. Completion near the commmtative quotient Recently M, Kapranov developed a theory of formal noncommutative thickenings of usual manifolds. ‘The structure of a formal noncommutative thickening on a ustial manifold X is given by a sheaf of algebras as 2Oncrcomm Mapping to Ox. In the case X = A‘ the algebra of global sections Ononeamm(X) is the completion | Of the froe algebra k(ry,.++ tg) by the filtration by the total mmber of commntatars. If X is itself formal, say FormSpec(k[r15+++54]])s then Ononeomm(X) is the algebra of formal noncamnmitative power series Rletyes-5a)). Teis easy to see if A isa smooth algpbra then Mi has a canonical formal noncommutative thickening. ‘The importance of formal noncommutative thidenings is explained by the following observation made by Kapranov: WC is a triangulated category with finite-dimensional Homes, and a: E—+F is a morphism, such that ‘the tangent complex Jor the deformation theory of it with fined olgect E has vanishing H~* and H', then the germ of the formal moduli space of ex uith fired source B, is smooth and has a eanonieal noncommutative thickening. One can show that to have a noncommutative thidening of an algebraic manifold X/k is exactly the same a5 to have a system of formal manifolds (My) endowed with GL(V)-actions and addition maps, such that each (My) & a thideoning,in the usual commutative sense of the scheme of pains (F,is0) where F is a ccherent sheaf with finite support on X, of length IV, and iso is an isomorphism between vector spaces T(X,E) and k®, This scheme is similar to the set of minima in M-theory (oe sect. 1). 6. Compactifications Tt is very natural to try to define “compact” smooth noncommutative manifolds as certain machines producing as “representation spaces” sequences of usual smooth projective manifolds My (endowed with GI(N)-actious and addition maps). With A. Rosenberg we proposed a general definition of a noncommmta- tive scheme. It is a beginning of a lige project, and Lam not going to describe it here at all. Tonly mention that in our framework there are coherent sheaves and cohomology as in the tstal commutative algebraic We have now a callection of examples which are formally smooth, and procuce always smooth projective varieties. For example, our “universal grassmanian of subspaces in I-limensional space” has as the set of ‘mien of all grassmanians LI Grima osnign We can make products, blow-ups, and other constructions. Moduli spaces (stacks) of coherent sheaves (on compact smooth noncommutative spaces are smooth, and also are abelianizations of certain smooth noncemmtative spaces. Tn all examples the category of colierent sheaves has global cohomological dimension 1, and all Hom-spaces and Ert'-spaces are finitewlimensional. Also, the categories of coherent sheaves are complete in certain sense. Morecxer, one can define moduli spaces of such categories, and by purely formal reasons the moduli space is a smooth stack of finite type. Thus, we have generalizations of the mockli stad of algebraic curves. One of our moduli stacks parametrizes collections of disjoint projective subspaces in a projective space, ‘Morally, all noncmmntative smooth spaces are “cues”. One can try to define generalizations of Gromov-Witten invariants considering spaces of maps between noncommutative compact smooth spaces: Taking the tensor product of two smooth non-commutative spaces we obtain a singular space with the categery of coherent sheaves of cobomological dimension 2 (“noncommutative algebraic surfaces"). For such categories virtual fimdamental classes of mocluli spaces of objects can be defined (at Feast locally). One can hope to have generalized Donaldson invariants for such twordimensional spaces. 7. Further topics Thope that the langnage are the new intuition of smooth algebras will be usefill someday for more fami iar examples, such as deformations of commutative algebras. The reason is that many usual algpbras admit resolutions by finitely generated differential graded smooth algebras. Thus, noncommutative differential gzaded supermanifolds can serve as replacements of usual algebras. Derived categories of abelian categories of global cohomological dimension <1 form a natural arma for constructions of HallRingel algebras. Tt is known that all Kac-Moody Lie algebras of finite rank appear 3(@ Lie subalgebras) in this way. Tt seems plansible that Borcherd’ infinite rank algebras with Monstrous symmetty can be realized inside Hall-Ringel algpbras for some small smooth nondommutative spaces. Ako, one can play a purely mathematical game imitating “physics” of matrix models, In this gane integrals are replaced by numbers of paints over finite fields Tnstead of taking the asymptotic expansion at —}o0 one can make generating functions. Here one get. a feding that the modular type of behavior is ubiquitous. Here are few examples: ‘Number of points in. My(F,) for the free algpbra Z(r1,.+.s:r4) is”. Tf we now consider q asia formal parameter, the sum over all 1V € Z will be a modular form, Analogously, counting points in GL(V,F,) (ie. homomorphisms Z[t,t—!|—Mat(N x N,F,)), we get & (fen) = Ih which soens to be a quasi-modular form. Tf we consider q as an actual rmmber, then the leading term in the fornnila for numbers of Mat(N x V; F)points for the algebra Z(p)/(@? = p) is (for even N’) en (5 ) Tle-~) od -l Analogpus results hold in other examples, induding the compact anes. Physicists told me that this modular behavior does not look totally surprising for them, because the size of matrices plays the role of a variable ual to certain compactification length in an extra dimension, whatever it means. Sometimes one can calculate an integral over a manifold reducing it toa “topological integral” (an expression in characteristic classes of some vector bundles) over a compactification of the original space, Several matric integrals calculated recently in plysical literature, have rational vahes. ‘Thus, they have a chance to be calailable topologically. T hope that noncommutative compactificatians constructed by Rosenberg and myself, can be useful in some matrix models. The following question arises naturally: how cohomology of manifolds My behave as NV grow? Far ecample what can one say about the generating function of the Euler characteristics of spaces 1