SU-4252-774 PreprinttypesetinJHEPstyle-HYPERVERSION Lattice Supersymmetry and Topological Field Theory 3 0 0 2 n Simon Catterall a J Department of Physics, Syracuse University, Syracuse, NY 13244, USA 8 E-mail: [email protected] 2 1 v Abstract: It is known that certain theories with extended supersymmetry can be dis- 8 cretized in such a way as to preserve an exact fermionic symmetry. In the simplest model 2 0 of this kind,weshowthat thisresidualsupersymmetricinvarianceis actually aBRSTsym- 1 metry associated with gauge fixing an underlying local shift symmetry. Furthermore, the 0 3 starting lattice action is then seen to be entirely a gauge fixing term. The corresponding 0 continuum theory is known to be a topological field theory. We look, in detail, at one / t a example - supersymmetric quantum mechanics which possesses two such BRST symme- l - tries. In this case, we show that the lattice theory can be obtained by blocking out of the p e continuum in a carefully chosen background metric. Such a procedure will not change the h Ward identities corresponding to the BRST symmetries since they correspond to topolog- : v ical observables. Thus, at the quantum level, the continuum BRST symmetry is preserved i X in the lattice theory. Similar conclusions are reached for the two-dimensional complex r Wess-Zumino model and imply that all the supersymmetric Ward identities are satisfied a exactly on the lattice. Numerical results supporting these conclusions are presented. Keywords: Lattice, Supersymmetry, Topological. Contents 1. Introduction 1 2. Simple Example 2 3. Supersymmetric Quantum Mechanics 4 4. Continuum Topological Field Theories 5 5. Relation Between Lattice and Continuum Theories 7 6. Ward Identities in Supersymmetric Quantum Mechanics 11 7. Complex Wess-Zumino Model in Two Dimensions 13 8. Discussion 16 1. Introduction Supersymmetric field theories are interesting both from phenomenological and theoretical pointsofview-theirimprovedU.Vbehavioroffersthehopeofresolvingthegaugehierarchy problem and they arise naturally as low energy limits of string and M-theory. Most of the interesting physics of such theories lies in non-perturbative regimes. Discretization on a space-time lattice appearstoprovideanaturalway to studysuchtheories and considerable effort has gone into formulating such lattice supersymmetric theories [1]. Unfortunately supersymmetry is typically broken at the classical level in such models. At the quantum level theabsenceof such asymmetry leads toan effective action containing relevant SUSY- violating interactions. To achieve a supersymmetric continuum limit then necessitates fine tuning the couplings to each of these operators. In most situations this is prohibitively difficult. In an previous paper [2] we advocated a different approach – try to preserve a subset of the full supersymmetry in the lattice model. A similar approach has also been adopted in [3]. Numerical simulations of two models where this idea can be implemented explicitly lend strong support to the idea that preserving some subset of the continuum supersym- metry transformations can indeed protect the lattice model from the dangerous radiative corrections that generically plague discretizations of supersymmetric field theories [2, 4]. In this paper we offer a proof of this statement for the case of supersymmetric quantum mechanics and the two-dimensional complex Wess-Zumino model. Indeed, our result is much stronger - by careful choice of the lattice action we can show that there are no cut-off – 1 – effects in the lattice supersymmetric Ward identities even for supersymmetries which are broken at the classical level. We will argue that the reason that these lattice models are well behaved is that they are related to Witten-type continuum topological field theories [5]. Such theories are con- structed using a nilpotent symmetry formed from elements of the original supersymmetry. A key feature of such theories is that they contain observables whose expectation values are independent of the metric. After the partition function itself, the simplest examples of such observables are the Ward identities corresponding to the nilpotent symmetry. This metric independence plays a crucial role in allowing us to establish a direct link between the continuum and lattice theories and allows the latter to possess rather remarkable prop- erties. In the next section we introduce a simple lattice model which exhibits an exact fermionic symmetry and show that this symmetry is actually a BRST symmetry following from fixing a local topological symmetry. We then show how this topological model can be used to describe a lattice regularized theory of supersymmetric quantum mechanics [4]. In this latter case we show that the model is naturally associated with two independent BRST symmetries. In thecontinuum these two symmetries exhaustthe original supersym- metries and show that that supersymmetric quantum mechanics can indeed be viewed as a topological field theory. For convenience we include a brief summary of the main features of such topological field theories. We then show how this continuum topological structure allows us to derive the lattice theory by integrating out the continuum fields in a carefully chosen background geometry. More precisely, we consider metrics which are continuous functions of a param- eter β such that in the limit β a “lattice” structure is induced in the model. For → ∞ any finite value of β we can perform an associated β-dependent change of variables in the continuum model which preserves the topological symmetries. Hence such a procedure ensures that any topological observable is independent of β. We show that indeed, a local, lattice theory is approached in the limit of large β. Furthermore, our construction then guarantees that the resulting lattice theory retains an element of supersymmetry at the quantum level. ThesameanalysisappliedtothecomplexWessZuminomodelintwodimensionsallows us to write down lattice actions with respect to which all supersymmetric Ward identities are satisfied exactly. Numerical results confirming these conclusions are presented. Finally we discuss the prospects for extending these ideas to more realistic models. 2. Simple Example Consider the simple model discussed in [2] 1 ∂N S = N2(x)+ψ iψ (2.1) 2α i i∂x j j which admits a fermionic symmetry δx = ψ ξ i i – 2 – δψ = N ξ i i δψ = 0 i Here, N (x) is an arbitrary function of the scalar field x and ψ , ψ are real, independent i i i i grassmann variables. Notice that this transformation is nilpotent on-shell. Indeed, if we introduce an auxiliary (commuting) field B we can define a new action i 1 ∂N S′ = αB2+N B +ψ iψ (2.2) −2 i i i i∂x j j If we integrate over B (along the imaginary axis) we just recover the original action S. i This new action also has an invariance δ x = ψ ξ (1) i i δ ψ = B ξ (1) i i δ ψ = 0 (1) i δ B = 0 (1) i Thesignificanceofthesubscriptlabelingtheaction andsymmetryvariation willbebecome apparent later. Itis easy to seethat this new transformation is nilpotent off-shell δ2 Φ = 0 (1) ′ for any of the fields Φ = ψ,ψ,x,B . More importantly it is clear that the new action S { } is nothing but the variation of another function 1 ′ S ξ = δ ψ N αB (1) i i− 2 i (cid:18) (cid:18) (cid:19)(cid:19) Thus we recognize our original invariance as a BRST invariance and our original action as nothing but a gauge fixing term! The topological origins of the lattice theory are made more manifest when it is realized that the local gauge symmetry which is being fixed to generate the BRST invariance is nothing but an arbitrary shift in the scalar field x . Imagine a classical theory depending on a scalar field x with trivial classical action i i S(x) = 0. Clearly this theory is invariant under a huge local symmetry - namely arbitrary shifts in the scalar field x x +ǫ i i i → To quantize we need to pick a gauge. One simple way to do is is to impose N (x) = 0 i where N (x) is some arbitrary function of the field x . Then the partition function will be i i ∂N i Z = dx δ(N )det i i ∂x Z (cid:18) j(cid:19) Ifwerepresentthedeterminantusinganticommutingghostsandintroduceamultiplierfield B for the delta function we recover our simple model eqn. 2.1 in α = 0 gauge. The usual i theorem associated with quantization of gauge theories allows us to relax this Landau-like gauge to a Feynman-like gauge with α non-zero without changing the expectation values of gauge invariant quantities. Notice that the physical ‘fermions’ of the SUSY theory are to be identified with the ghosts of the gauge fixed scalar field theory. – 3 – There is another simple way to see the the model written down in eqn. 2.1 exhibits an unusual symmetry. If we imagine performing a change of variables in the path integral defining this theory according to η = N (x) the Jacobian of this transformation cancels i i the fermionic determinant and the partition function factorizes into a product of gaussians [6] −η2 Z = dηie 2α i Y This resultant partition function is trivially invariant under the same local shift symmetry discussed earlier. A transformation of this type which cancels off the fermion determinant is called a Nicolai map and it should now be clear that the existence of a local Nicolai map can be attributed to the presence of a topological symmetry [5]. We now turn to the simplest application of these ideas - supersymmetric quantum mechanics 3. Supersymmetric Quantum Mechanics (1) ′ Imagine now choosing the function N (x) = N (x) corresponding to an action S = S i i (1) where N(1)(x) = DSx +P′(x) i ij j i In this expression DS represents the symmetric difference operator and P′(x) is some i arbitrary polynomial in the lattice field x . To be concrete we can imagine a model with a i single interaction coupling g of the form P′ = mx +mWx +gx3 (3.1) i ij j i Notice that we are also free to add a Wilson mass term mW to the potential to eliminate lattice doubles associated with the choice of lattice derivative operator DS. In this case we recognize our simple model as a lattice regularized version of super- symmetric quantum mechanics - a model well known to possess a topological field theory interpretation [5]. We discuss some of the generic features of such theories in the next section but suffice it to say here that such theories possess observables whose expectation values are metric independent. Clearly, on the lattice, there is no notion of continuum metric and so in a strict sense the lattice model cannot be said to be topological. However, the fact that the action is a BRST variation of a local function of the lattice fields clearly imposes strong restrictions on the quantum theory. Indeed we shall see that in this model certain symmetries which are broken at the level of the classical lattice action are restored in the full quantum theory. Notice also that the partition function of this model (in α = 0 gauge) just reduces to an integral over the set of field configurations satisfying the gauge condition. In the case of supersymmetric quantum mechanics this is just the moduli space of classical solutions to the equation of motion. On a circle these classical solutions are just a finite set of points ′ x = x where P (x ) = 0. Thus the partition function just reduces to a sum over the i c c critical points of the potential P(x). Notice that this solution is independent of the lattice – 4 – cut-off – indeed it is the same result one would have gotten for the analogous continuum model. Furthermore we can consider another gauge condition which corresponds to the same (2) classical moduli space N (x) = N (x) with i i N(2)(x) = DSx P′(x) i ij j − WecanconstructtheactionS followingfromthisgaugeconditionbythesameprocedure (2) and furthermore after exchanging the roles of ghost ψ and antighost ψ we can easily see it only differs from S by the addition of a simple operator C(g) (1) S = S 2C (2) (1) − The operator C = DSx P′(x) and corresponds to the integral of a total derivative term in ij j i ′ the continuum. On the lattice it will be non-zero if P (x) contains nonlinear powers of the field x which is the case for an interacting model with g = 0. 6 The variation of the fields under this second (nilpotent) BRST symmetry is δ x = ξψ (2) i i ′ δ ψ = ξ B 2P (x) (2) i i− i δ(2)ψi = 0(cid:0) (cid:1) ′′ δ B = 2ξP ψ (2) i i InthecontinuumwhereS = S theactionofsupersymmetricquantummechanicswould (1) (2) then possess two BRST invariances. On the lattice if we choose S as action we no longer (1) have δ asasymmetry(except forafreetheory)andviceversa. Thusattheclassicallevel (2) discretization on a lattice necessarily breaks one of the continuum symmetries. However, we will see that this symmetry is restored at the quantum level with lattice Ward identities corresponding to both δ and δ being satisfied exactly for arbitrary lattice spacing. (1) (2) Indeed we will prove that there exists a one parameter family of lattice actions in which all of the BRST Ward identities are satisfied with no cut-off effects. This feature is crucially dependent on the existence of this topological symmetry. 4. Continuum Topological Field Theories Inthissectionwegiveabriefreviewofsomeofthegeneralfeaturesofcontinuumtopological field theories [5, 7]. Such theories are formulated on a n-dimensional manifold equipped withametricg . OnthisspacethereexistsasetoffieldsΦandanactionS(Φ). Typically µν these theories contain operators or topological observables O (Φ) with the property that β their expectation values are metric independent δ O ...O = 0 δgµν h β1 β2i One way to guarantee this corresponds to the case in which there exists a nilpotent sym- metry δ such that δO = 0, T = δG β µν µν – 5 – The energy-momentum tensor T = δ S(Φ). These conditions lead to the following µν δgµν expressions for the expectation values of topological variables. δ O ...O = DΦδ O ...O G e−S(Φ) = 0 δgµν h β1 β2i − β1 β2 µν Z (cid:16) (cid:17) Herewehave alsoassumedthatthemeasureisinvariantunderthenilpotentsymmetryand that the observables themselves do not contain the metric explicitly. Theories constructed in this way are called Witten type or cohomological topological quantum field theories. Typically the nilpotent symmetry δ is realized as a BRST symmetry arising from gauge fixing some underlying local shift symmetry. In such theories S(Φ) = δΛ(Φ) This latter result is very important as it guarantees that topological observables and the partition function itself can be computed exactly in the semi-classical limit. To see this introduce a parameter ǫ playing the role of Planck’s constant in the definition of a topolog- ical expectation value and examine the variation of that expectation value under variation of ǫ ∂ 1 ∂ǫ hOβ1...Oβ2i = ǫ2 DΦδ(Oβ1...Oβ2Λ(Φ))e−1ǫS = 0 Z Thus topological observables may becomputed exactly in the semi-classical approximation ǫ 0. This semi-classical exactness may be translated into an independenceof topological → observables on coupling. To see this consider an action of the form S = S (Φ)+gΦn 0 where the quadratic terms are contained in S (Φ) and we allow for a generic interaction 0 term. If we rescale the fields using Φ √ǫΦ it is easy to see that topological observables → computed in an ensemble with Planck constant ǫ and coupling g is equivalent to the same observable computed in the ensemble ǫ = 1 and g′ = gǫn/2−1 The limit ǫ 0 now corre- → ′ sponds to g 0 in the latter ensemble and hence the expectation value can be computed → in the free field limit. For the case O = 1 this implies that the partition function itself is independent of g. This, of course, is also the property enjoyed by models with a local Nicolai map and makes plausiblethe conjecture that models possessinga local Nicolai map contain within them a topological symmetry. A trivial set of topological observables correspond to operators of the form ′ O = δO While trivial in a true topological sense (their expectation value vanishes trivially on ac- count of the nilpotent nature of δ) they will be of crucial importance in the parent super- symmetric theory since they yield the supersymmetric Ward identities. – 6 – 5. Relation Between Lattice and Continuum Theories Insection3wediscussedalatticetheoryofsupersymmetricquantummechanicsandshowed that itwas possibletochoose anaction which reproducedthecontinuum partition function exactly (up to a parameter independent multiplicative constant). The simplest way to see this utilizes the gauge α = 0 in which the partition function reduces to a sum over the ′ critical points of the potential P (x) = 0. The solution of this equation is identical in both lattice and continuum theories. Actually, the gauge α = 0 can be used for an arbitrary topological observable and implies that the expectation values for all such observables only depend on the properties of the classical solution to the field equations. This prompts us to guess that it should be possible to forge an explicit connection between the lattice and continuum theories useful for the computation of such observables. We will now show that indeed this is the case. One standard way to relate a lattice theory to an underlyingcontinuum theory derives from therenormalization group. Thelattice fieldat somepointis constructed by averaging thecontinuumfieldoveraneighborhoodofthatpoint. Thisaveragingorblockingprocedure maybeaccomplishedbyconvolvingthecontinuumfieldwithablockingfunction. Typically, the precise shape of the blocking function is not important for long distance physics. The lattice or block field which results from this operation is usually a non-local function of the continuum fields. However, this need not be the case for a topological field theory. If we are only concerned with the computation of topological observables we are at liberty to block the continuum fields in an arbitrary background metric. If this metric is then chosen carefully we can arrange for the block fields to be related to the continuum fields in a completely local manner. Let us examine this in the case of supersymmetric quantum mechanics. The bosonic part of the continuum action for an arbitrary one-dimensional metric written in terms of the einbein e(t) takes the form (we have integrated out the auxiliary field B(t)) 2 1 dx ′ S = dte(t) +P (x) e(t) dt Z (cid:20) (cid:21) Define now a scalar block field xB(t) in the continuum as a convolution over the original − field x(t) using a blocking function B (t) β xB(t) = dt′e(t′)B− t t′ x(t′) (5.1) β − Z (cid:0) (cid:1) − We will choose the blocking function B (t) to be β 1 − B (t) = [L (t+δ) L (t a+δ)] β 2a β − β − We require that the function L (t) tend to the step function for large β β lim L (t) =θ(t) β β→∞ – 7 – − This choice of B (t) ensures that for large β the blocked field at point t contains contri- β butions from all points within a cell defined by δ t (a δ). Furthermore, we will − ≤ ≤ − require the parameter δ 0+ at the end of the calculation. A concrete example of such a → function is given by L (t) = tanhβt. To capture the structure of the lattice theory we will β choose an associated “lattice” metric given by e(t) = e (t) where β N a ′ e (t) = L (t na) β A(β) β − n=1 X where the sum runs over a finite set of N points with “lattice spacing” a. The constant A(β) is chosen so that dte (t) = Na β Z andweassumethecontinuumtheoryisdefinedoveracirclewithcircumferenceNa. Notice that dL ′ β lim L = lim = A δ(t) β→∞ β β→∞ dt L where A = A( ) is just some numerical coefficient. Returning to eqn. 5.1 we can now L ∞ compute the block field explicitly N a xB(t) x(na)B−(t na) for large β ≃ A(β) β − n=1 X In the limit β this relation yields the result → ∞ N 1 lim xB(t) = x(na) [θ(t na+δ) θ(t (n+1)a+δ)] β→∞ 2AL − − − n=1 X That is, the continuum block field xB(t) is constant within each unit cell of the lattice changing its value only on passing from one cell to the next. To proceed further it is necessary to compute its derivative. dxB(t) N a dB− β x(na) (t na) dt ≃ A(β) dt − n=1 X N 1 ′ ′ x(na) L (t na+δ) L (t (n+1)a+δ) ≃ 2A(β) β − − β − n=1 X (cid:2) (cid:3) Notice that this derivative does indeed vanish in the limit β for any point within a → ∞ cell. To compute the action evaluated on a block configuration in this background we need to compute 1 dxB. For any point within the cell δ+na t (n+1)a δ the leading eβ dt − ≤ ≤ − contribution at large β is seen to be 1 dxB 1 (f (z)[2x(na) x((n+1)a) x((n 1)a)]+[x((n+1)a) x((n 1)a)]) β e dt ≃ 2aA(β) − − − − − β – 8 – where ′ ′ L (a z) L (z) f (z) = β − − β β ′ ′ L (z)+L (a z) β β − ! with z = t na and we have set δ to zero for simplicity. This in turn reduces to − 1 dxB N 1 lim = [x(na) x((n 1)a)][θ(t (n 1/2)a) θ(t (n+1/2)a)] β→∞ eβ(t) dt 2aAL − − − − − − n=1 X Thusthederivativeofablockfunctioninsuchabackgroundgeometryatlargeβ isconstant now within a unitcell of the dual lattice. Theseproperties allow us to compute the integral of an arbitrary function of the block field xB(t) and its first derivatives. An example is the bosonic part of the continuum action S . This becomes B N 2 1 lim S = a D− xB +P′(xB) (5.2) β→∞ B n=1 (cid:20)2ALa nm m n (cid:21) X where the bosonic action now only depends on the blocked fields at the lattice points and we use the obvious notation xB(na) xB. The most striking thing about this block action ≡ n S is that it coincides with the bosonic part of the lattice action S discussed earlier if B (1) we identify the lattice field as the blocked continuum field evaluated on a lattice point1. Notice also that the lattice theory we arrive at in this manner automatically incorporates an r =1 Wilson mass term to remove potential lattice doublers (since DS m = D−). W − − Notice that the backward difference operator D arises directly from our choice of − blocking function B . If we had made the equally valid choice β 1 B+ = [L (x+a δ) L (x δ)] β 2a β − − β − − we would have arrived at a block action of the same form as in eqn. 5.2 but with D replaced by the forward derivative D+. Generalization to include the fermionic sector is straightforward and leads to the con- clusion that the total action when evaluated on block configurations as defined by eqn. 5.1 goes over into the full lattice action S described in section 3. (1) This similarity between the blocked continuum theory at large β and the lattice model describedinsection3isintriguingbutbyitselfdoesnotyetguaranteetheexactequivalence of continuumandlattice theories. Tocompletethecorrespondencebetween continuumand lattice theories, and to show that for topological observables the continuum expectation values are just equal to their lattice counterparts, we still need to argue that only block fields need to be taken into account inside continuum path integrals as β . We offer → ∞ two arguments for this. First consider the continuum boson kinetic term in terms of the original fields 2 1 dx S = dt K e dt Z β (cid:18) (cid:19) 1a finite rescaling of the field is also needed to make the correspondence exact - but this in turn is equivalent to a rescaling of thecoupling which will not affect topological observables – 9 –