WorksPhysics and Astronomy
Anomaly inflow for simplicial Kähler fermions
Abstract
We study massive Kähler Dirac fermions propagating on finite triangulations of various three dimensional manifolds with one or more boundaries. To probe the anomaly structure of these models we decompose the Kähler field into two reduced Kähler fields and choose boundary conditions that ensure that any edge modes are necessarily massless. We then show how to factorize the Pfaffian of this system into a product of bulk and boundary sectors each of which can separately pick up a phase under a global transformation of the fermions that simultaneously changes the sign of the mass. These phases are topological, depending only on the index of the lattice Kähler operator in that sector. For the boundary sector a non-trivial phase signals the presence of a 't Hooft anomaly. We study cases where the boundary is anomalous but an inflow from the bulk cancels its phase. Canceling this anomaly for a non-orientable boundary in the absence of the bulk recovers the known classification of topological superconductors. We also give a continuum interpretation of these results in terms of inflow from a gravitational Chern Simons theory which is induced by integrating out massive bulk Kähler Dirac fermions. In the presence of a boundary the latter gives rise to Wen-Zee terms that mediate the anomaly inflow mechanism.
Introduction
The Kähler-Dirac equation furnishes an alternative to the Dirac equation for describing fermions [1]. The Kähler-Dirac field is composed of a set of antisymmetric tensors or -forms that are coupled via an action that involves the exterior derivative and its adjoint. In four flat Euclidean dimensions it is nevertheless equivalent to four degenerate copies of the Dirac equation with the spinors constructed from the -forms [2, 3]. However, the Kähler-Dirac equation, since it involves tensors and not spinors, differs from the Dirac equation in curved space. It can be formulated on any manifold without requiring a spin structure [4] and its global properties are also different with the zero modes of the massless theory corresponding to harmonic forms [5]. Thus the Kähler-Dirac operator possesses zero modes on compact positive curvature manifolds - something that is impossible for Dirac fermions [6, 7].
From a lattice perspective it is particularly interesting as it can be discretized without creating fermion doubles [2, 3, 5] - lattice modes which have no place in the continuum theory and that do not decouple in the continuum limit [8]. This renders it attractive in comparison to the Dirac equation [9, 10]. It also possesses an exact onsite symmetry which, as we will see, suffers from a mixed anomaly with gravity even in the discrete theory [6, 7, 11]. Finally, in flat space the theory is equivalent to the well known staggered fermion construction [2, 3, 12–14] which has been used very successfully in studies of lattice QCD [15].
In this paper we will show that the bulk-boundary correspondence or anomaly inflow mechanism that is familiar from the continuum [16–19] also survives in a discrete setting for simplicial Kähler-Dirac fermions and allows us to re-derive a known mod 2 anomaly [11, 20, 21] of Kähler-Dirac fermions. Furthermore, in the continuum we can interpret the inflow as arising from a gravitational Chern Simons theory induced by integrating out massive Kähler-Dirac fermions [22].
Our focus will be on anomaly inflow arising from Kähler-Dirac fermions propagating on a triangulation of a three dimensional bulk geometry that contains one or more boundary components but in the end it will be easy to generalize the construction to other geometries and higher dimensions.
Kähler-Dirac fermions on the 2-sphere
We start by reviewing the anomaly for massless Kähler-Dirac fermions propagating on a triangulation of the 2-sphere [6, 7, 11]. The Euclidean massless lattice action is given by
where the lattice or simplicial Kähler-Dirac field consists of a set of fields living on -simplices in the triangulation with independent of in Euclidean space. The lattice Kähler-Dirac operator is given by
where is an incidence matrix which is only non-zero when a given simplex lies in the boundary of the -simplex in which case it takes the value according to the orientation of the boundary simplex [5, 23]. This action is invariant under the symmetry:
The operator distinguishes even simplex fields from odd simplex fields and anticommutes with . If we integrate over the fields we obtain a determinant. Under a transformation the measure is not invariant but transforms according to a Jacobian of the form . Equivalently we retain the original measure and imagine the fermion operator varying as . The partition function hence transforms
where we have used the result the Euler characteristic with for the sphere [5, 6]. This anomaly is linked to the existence of exact zero modes of which are dictated by the index theorem [19]
with the number of zero modes with . The symmetry is hence broken to - the partition function is not invariant under the classical symmetry - there is a mixed -gravitational anomaly - here computed exactly in the lattice theory [7, 11].
However, the presence of these zero modes makes the previous discussion rather formal since they force . One can add a (real) mass to lift these zeros leading to
This representation shows that the non-zero modes give a real, positive definite contribution to the determinant and the phase is hence carried entirely by the zero modes. It is thus topological depending only on the Euler characteristic. .
Rather than adding a mass one can absorb these zero modes by inserting an operator with charge four - a four fermion operator - into the definition of the partition function so that the result of the fermion integration is the product of the non-zero eigenvalues. This operator insertion explicitly breaks the symmetry to .
Since the massless Kähler-Dirac action can be decomposed into two independent pieces using the projectors where
and the action is
Setting say one arrives at a reduced Kähler-Dirac action which describes a system with half as many degrees of freedom. This decomposition of a Kähler-Dirac field into reduced fields is analogous to the decomposition of a Dirac spinor into Weyl spinors. If we define
then the full action can also be written
In this representation the projection to a single reduced Kähler-Dirac field is equivalent to setting eg. .
To construct a non-zero partition function for a reduced Kähler-Dirac field requires the insertion of a fermion bilinear with charge two. However, the resultant partition function will then change sign under a transformation – the reduced Kähler-Dirac field propagating on possesses a ’t Hooft anomaly. To try to cancel this anomaly we can try to realize the two dimensional surface as the boundary of a bulk three dimensional triangulation carrying a massive Kähler-Dirac field. By careful choice of boundary condition we can then build an anomaly free model with massless edge modes and gapped bulk modes in which the phase arising from a transformation of the massless boundary modes cancels against a corresponding phase from the bulk partition function. The main thrust of this paper will be to demonstrate examples of this anomaly inflow mechanism for simplicial Kähler-Dirac fermions.
Kähler-Dirac fermions on the 3-ball
We will consider Kähler-Dirac fermions on a triangulation of the 3-ball . Let us divide the simplices defining this geometry into a set that reside exclusively on the boundary of the 3-ball which we label and the remaining simplices which form a complementary set in the bulk labeled . To facilitate analysis we will decompose the field into the two reduced fields and discussed in section II. Let us further impose the boundary condition that the field vanishes on :
In this situation the bulk contains one full massive Kähler-Dirac field while the boundary field is a massless reduced Kähler-Dirac fermion. With the Kähler-Dirac field split into bulk and boundary components the action can be written in block form as
The matrix can then be factorized as
where
and
After integration over the fermions one obtains the Pfaffian of which takes the factorized form ()
Substituting for and one finds
At large mass this operator has the expansion
The presence of the massive bulk propagator ensures that the lightest modes of the system are exponentially localized to the surface in a manner similar to domain wall fermions [24, 25]. Indeed, in the limit of infinite bulk mass the boundary action becomes that of a reduced Kähler-Dirac field with the properties
These results show that no boundary mass is induced via coupling to the bulk for the single boundary field . Indeed, it is not hard to see that , like , possesses two zero modes. It follows that the boundary Pfaffian, and hence the free vacuum partition function of the complete system vanishes
As in the discussion in section II this simply instructs us to insert a bilinear fermion operator to absorb these zero modes in the Grassmann integral and thereby obtain a modified partition function which is non-zero.
However, for the moment let us ignore these zero modes and examine the result of a transformation on the fields. This will be useful later when we compare to a continuum perturbative calculation. The reduced components transform as
To achieve an invariance of the action we will also assume that the mass can be treated as a spurion field that also rotates:
One can then think of the fermion operator as transforming and the corresponding boundary and bulk Pfaffians behave as
Notice that only a single field contributes to the boundary Pfaffian. The bulk Euler characteristic can be found from the relation
Thus the phases of both bulk and boundary cancel under a general rotation. This is the anomaly inflow mechanism.
Once zero modes are included we know we should consider only operators that absorb those modes and restrict ourselves to the subgroup of . In this case we see that the modified boundary Pfaffian changes sign under a transformation signaling a ’t Hooft anomaly. This is then canceled by a corresponding minus sign arising from the transformation of the bulk Pfaffian.
Notice also that under the subgroup the mass can be taken real which guarantees the bulk operator can always be inverted. It also means that the spectrum of non-zero modes is paired and, as observed earlier, the anomalous phase arises only from the topologically protected zero modes.
Kähler-Dirac Fermions on
It is interesting to ask whether one can construct similar models where the boundary is replaced by its non-orientable cousin . Since there is no closed 3-manifold whose boundary is we instead consider the geometry which comprises a finite interval terminated by two boundaries. Following the logic of § 3 we choose boundary conditions that leave one reduced Kähler-Dirac fermion on each boundary. Since the Euler character of the bulk is one it generates a bulk contribution to the phase which cancels the factor of coming from the boundaries. We will decompose the field into four reduced components where we have set to achieve reduced boundary fermions. The action can be written
where is given by
Setting
this can be written
As before, this admits the Schur factorization where
with
Integrating out the fermions once again shows that the Pfaffian of the full system is simply
Plugging eqn. (27) into eqn. (30) one finds
where
In a manner similar to the case the effective boundary operator is corrected as a result of integrating out the bulk. For large mass, the presence of the bulk propagator ensures that the light modes are exponentially localized to one of the two boundaries. These boundaries are now coupled via the bulk but the magnitude of this coupling decreases exponentially with the length .
As in our earlier construction the effective boundary operator is antisymmetric, an even function of , and anticommutes with . It does not behave like a mass term and does not lift the boundary zero modes. To render the partition function non-zero one must again insert a single fermion mode on each boundary. Under a transformation the boundary Pfaffian transforms by a factor as one would expect for two reduced fields living on . Conversely the Pfaffian for the single massive bulk Kähler-Dirac field also transforms by a factor of since . Thus the anomalous phase associated to the boundary fermions is canceled by the phase transformation of the bulk under . Anomaly inflow again. To cancel the anomaly for each boundary independent of the bulk will require multiples of four copies of the Kähler-Dirac field.
In the flat space continuum limit each reduced Kähler-Dirac field on the boundary describes 2 Majorana spinors [7] so canceling this anomaly for each boundary separately would require multiples of eight Majorana spinors. This agrees with the classification of topological superconductors [33] [26, 27, 28, 29, 30, 31, 32] and the cancellation of a Dai-Freed anomaly [34, 35, 36] for chiral fermion parity [37, 38].
Effective Bulk Continuum Action
Manifolds with no boundary
Since the anomaly is topological and independent of the triangulation it should survive into the continuum. Indeed if we start with a massive Kähler-Dirac fermion in two dimensions we can argue that the phase of the mass can be rotated away at the expense of adding a gravitational theta term which is just the Euler density
The anomalous phase encountered under a transformation then corresponds to the shift . It is then interesting to ask whether there is a corresponding continuum theory that naturally generates the corresponding three dimensional bulk phase.
If we integrate out continuum Dirac fermions in three dimensions in the presence of a background gauge field we obtain a topological Chern-Simons term [39, 40]. Integrating out Kähler-Dirac fermions leads to a similar result except that the emergent Chern-Simons term is associated with a topological gravity theory [22].
To understand how this arises let us first recast the continuum two dimensional Kähler-Dirac field in terms of a matrix field [2, 3] where
where are (Euclidean) Dirac gamma matrices. Each column of this matrix can be interpreted as a Dirac spinor and these spinors can be rotated into each other under a flavor symmetry. It is then a straightforward exercise to show that the Kähler-Dirac action can be written as
The matrix fermions carry both Lorentz and flavor indices and this action is naively invariant under both an (left acting) (Euclidean) Lorentz rotation and an (right acting) flavor transformation. However, if we are to retain the original Kähler-Dirac interpretation in terms of p-forms the Lorentz symmetry must be locked to a subgroup of the flavor symmetry with the matrix fermion transforming as under Lorentz transformations. As an aside we remark that the operator introduced in the form representation can be realized as and so is sometimes called a twisted chiral symmetry.
Let us see how to generalize this construction to three dimensions. A three dimensional Kähler-Dirac fermion field consists of eight complex degrees of freedom corresponding to the set of p-forms in three dimensions (). A matrix representation of this field fermion containing these forms requires using four dimensional gamma matrices together with a projection that reduces the total number of degrees of freedom from .
For this projector to commute with one needs to introduce new 3d gamma matrices
and then write
where is the 3d Kähler-Dirac field. For simplicity let us work in Euclidean chiral basis. Then
And the massive Kähler-Dirac action reduces to two independent Dirac actions each describing a doublet of fermions with opposite signs of the mass
As a set of Dirac fermions the system has a internal flavor symmetry corresponding to independent rotations of these doublets plus an additional flavor symmetry corresponding to which is locked to axial symmetry via the projection. Let us switch on a background gauge field for the symmetry.
where are (Euclidean) Dirac matrices in the flavor space. In a chiral basis this can be written
where
where and . The rationale for the notation will become obvious later. Integrating out the fermions in the limit yields an effective action which is the difference of two Chern Simons terms (the relative minus sign reflects the opposite signs of the mass in the two Dirac systems)
where
and . 1 In fact this can be rewritten keeping the invariance explicitly as
As described earlier the Kähler-Dirac interpretation of the fermion matrix in terms of p-forms requires that the Lorentz group be locked to a (sub)group of the flavor symmetry. Thus the components of the gauge field corresponding to a subgroup of the flavor symmetry can be identified with the background spin connection . 2 The remaining generators of transform as a vector under this and are then naturally associated with the background frame . 3 Indeed, the CS action written in terms of and (replacing the latter by to keep dimensionless) yields the Einstein Hilbert action with positive cosmological constant in Palatini-Cartan form [41, 42]
where is the curvature of the spin connection. Notice that this is not a generic Einstein-Hilbert action because the relative coefficients of the two terms are fixed so that the action has an hidden Spin(4) gauge symmetry corresponding to the Euclidean de Sitter group [43, 44, 45].
In summary, we find that the continuum effective action arrived at after integrating out a massive Kähler-Dirac fermion on a three dimensional fixed background space takes the form of a gravitational CS term [22].
Manifolds with boundary
The preceding analysis does not take into account boundaries or interfaces. To understand anomaly inflow effects one needs to prescribe a set of boundary conditions for the gauge fields associated with such surfaces. To figure out what these should be we can go back to the fermion description. Let’s assume for simplicity that the 3d geometry in question corresponds the half space . This means that the appropriate fermion mass profile is given by
In this background the edge mode solutions take the form where and
These are the usual conditions for Dirac fermions. The Kähler-Dirac character of the field implies an additional constraint that . Putting these two constraints together we find that which is just the condition that the boundary fermion be a reduced Kähler-Dirac field where as in the lattice constructions. The only surviving symmetries on the boundary must commute with the boundary projector . Because of the locking of Lorentz to flavor for Kähler-Dirac fermions this is the same as . Beside this flavor symmetry the only other good boundary symmetry corresponds to two dimensional rotations with generator . Thus the surviving boundary symmetry is Spin(2)Spin(2). On this residual symmetry naturally leads to the following boundary conditions on the gauge fields.4
Notice that the generator of the symmetry restricted to the boundary is also . Thus
where we have employed the edge mode condition . Thus bulk flavor rotations with generator generate the symmetry for the boundary fermions. The corresponding field is then a gauge field for this symmetry that extends the symmetry into the bulk. With these boundary conditions the CS action eqn. (48) (including the factor of i) can be rewritten
The additional terms are similar to the leading term in eqn. (54) but vanish on the boundary. The leading piece is a Wen-Zee term [46] and consists of a coupling of the 2d spin connection with the internal gauge field . Under the gauge variation it yields, after an integration by parts,
Thus the bulk fermions contribute a phase factor to the partition function under a rotation. This cancels against the phase arising directly from the edge modes which, as we have seen, corresponds to reduced Kähler-Dirac fermions. This is the continuum perturbative version of the lattice inflow story described earlier ignoring boundary zero modes. Thus the ’t Hooft anomaly of the massless boundary fermions can be canceled by anomaly inflow from a Chern-Simons theory in the form of a Wen-Zee term associated with the boundary gauge symmetry [22]. Clearly, from the earlier lattice discussion, in order to see the non-perturbative anomaly one needs to also consider non-orientable surfaces.
Conclusions
We have studied anomaly inflow for massive simplicial Kähler-Dirac fermions propagating in a triangulated three dimensional bulk space equipped with a boundary. We decompose the Kähler-Dirac field into two reduced fields and choose boundary conditions that ensure that while the bulk fermions are gapped the edge modes are described by a reduced field and remain massless.
On a 3-ball geometry the edge modes living on experience a ’t Hooft anomaly which makes the associated edge partition function change sign under a transformation. This is compensated by a minus sign coming from the determinant of the massive bulk modes giving a concrete lattice realization of anomaly inflow or the bulk-boundary correspondence.
In the continuum the local boundary condition that projects the Kähler-Dirac to a reduced field is sufficient to render the bulk fermion operator self-adjoint since the corresponding boundary term vanishes. This is to be contrasted with Dirac fermions where a similar local chiral boundary condition must be replaced by a non-local APS condition for the bulk operator to be self adjoint. It is the appearance of this non-local boundary condition that lies at the heart of the measure problem for chiral fermions. The fact that a local boundary condition can be employed in the case of Kähler-Dirac fermions is the reason that there is no need to construct an overlap operator for Kähler-Dirac fermions. It is also related to the absence of the eta invariant in the Gauss Bonnet identity relating the index of the Kähler-Dirac operator to the integral of the Euler density as discussed in [19].
It is natural to ask whether what happens when the orientable boundary is replaced by its non-orientable cousin . In fact the simplest 3-geometry that is anomaly free corresponds to , with two such boundaries each carrying a single reduced field. In this case the Pfaffian associated with each boundary changes by a factor of under a transformation giving a net boundary contribution of that is canceled by the sign change of the bulk Pfaffian. Clearly to cancel the boundary anomaly completely requires multiples of four reduced Kähler-Dirac fermions. Locally each such reduced field corresponds to two Majorana fermions so this cancellation condition matches the usual topological classification of SPT phases [47, 48, 49, 50] [7, 11, 21].
These features are independent of the precise triangulation used and hence topological in nature. Indeed, we have shown that integrating out massive Kähler-Dirac fermions in the continuum produces an action that corresponds to a gravitational Chern-Simons theory. In the presence of a boundary the gauge symmetry of this Chern Simons theory is broken down to a subgroup which contains a symmetry that can be identified with the twisted chiral symmetry of the boundary. Indeed, the resultant bulk action then contains a Wen-Zee term that couples this gauge field to the spin connection at the boundary [46]. The gauge variation of this term then cancels the ’t Hooft anomaly of the edge fermions.
In the continuum theory one can also replace the flavor symmetry of the bulk Kähler-Dirac fermions with . If one integrates out the Kähler-Dirac fermions in the corresponding background gauge field one again obtains a gravitational Chern Simons (CS) theory but now one whose equations of motion yield (Euclidean) anti-de Sitter space [43, 44]. The corresponding symmetry acting on the boundary fields becomes a transformation which can be interpreted as a twisted Weyl symmetry acting on the Kähler-Dirac fermions.
The construction can be easily generalized to higher dimensions. To achieve naturally massless edge modes one needs to employ boundary conditions that select for reduced Kähler-Dirac fermions. To have a well defined anomaly inflow problem one needs to balance this requirement with the topology of the bulk space so that the total anomaly vanishes. For example the arguments of § 4 can be lifted to a geometry where applying the same arguments leads to the conclusion that copies of the system are needed to cancel the boundary ’t Hooft anomaly. Locally this implies that sixteen Majorana fermions are needed which matches the usual spin- anomaly cancellation condition.
The corresponding continuum story also generalizes – integrating out massive Kähler-Dirac fermions in five dimensions generates a 5d Chern Simons term whose gauge field can be decomposed into a 5d spin connection and frame corresponding to a a topological gravity theory whose symmetry is the Euclidean de Sitter group [45]. In the presence of a boundary or interface this CS theory contains a generalization of the Wen-Zee term coupling a gauge field for a (twisted) chiral symmetry of the interface fermions to the four dimensional Euler density.
Clearly, Kähler-Dirac fermions are capable of probing interesting aspects of geometry. In this paper we have examined how the usual bulk boundary correspondence or anomaly inflow mechanism works for such fermions. One of the advantages of Kähler-Dirac fermions is that this analysis can be performed directly on a finite lattice with no need to take a continuum limit. One might hope that they can therefore serve as a useful analog toy model for understanding the chiral fermion problem. In fact the connection to chiral fermions is stronger than this – if one decomposes the minimal anomaly free theory of reduced Kähler-Dirac fermions into spinors one discovers the Pati-Salam GUT [11]. More concretely, one finds in the continuum that the global symmetries and matter representations of the theory match those of Pati-Salam - a chiral gauge theory containing the Standard Model. Furthermore, as we have seen, it has a natural lattice formulation that preserves certain key symmetries and their anomalies. However a successful approach to (a class of) chiral lattice gauge theories along these lines would necessitate gauging certain offsite symmetries - the shift symmetries of staggered fermions. This is a very non-trivial problem and indeed it is not clear that a solution is even possible. Nevertheless, first steps in this direction can be found in [20].
Acknowledgments
SC was supported by DOE grant DE-SC0009998. He acknowledges helpful discussions with Scott Collier, Jay Hubisz, Ho Tat Lam and Alex Maloney. Anthropic’s Claude AI was used to help generate references and check algebra.
References
- [33]J. Kaidi, J. Parra-Martinez, and Y. Tachikawa, SciPost Phys. 9, 010 (2020), with a mathematical appendix by A. Debray, arXiv:1911.11780.DOI
- [34]X. Dai and D. S. Freed, J. Math. Phys. 35, 5155 (1994), arXiv:hep-th/9405012.DOI
- [35]K. Yonekura, JHEP 09, 022 (2016), arXiv:1607.01873.DOI
- [36]I. García-Etxebarria and M. Montero, JHEP 08, 003 (2019), arXiv:1808.00009.
- [37]D. Delmastro, D. Gaiotto, and J. Gomis, JHEP 11, 142 (2021), arXiv:2101.02218.DOI
- [38]D. Tong and C. Turner, arXiv e-prints (2019), arXiv:1906.07199.DOI
- [39]A. J. Niemi and G. W. Semenoff, Phys. Rev. Lett. 51, 2077 (1983).
- [40]A. N. Redlich, Phys. Rev. D 29, 2366 (1984).
- [41]A. Palatini, Rend. Circ. Mat. Palermo 43, 203 (1919).
- [42]S. W. MacDowell and F. Mansouri, Phys. Rev. Lett. 38, 739 (1977), [Erratum: Phys. Rev. Lett. 38, 1376 (1977)].
- [43]A. Achúcarrо and P. K. Townsend, Phys. Lett. B 180, 89 (1986).
- [44]E. Witten, Nucl. Phys. B 311, 46 (1988).
- [45]J. Zanelli, arXiv e-prints (2005), arXiv:hep-th/0502193.arxiv.org/abs/hep-th/0502193
- [46]X. G. Wen and A. Zee, Phys. Rev. Lett. 69, 953 (1992).
- [47]V. Ayyar and S. Chandrasekharan, Phys. Rev. D 93, 081701 (2016), arXiv:1511.09071.DOI
- [48]S. Catterall, JHEP 01, 121 (2016), arXiv:1510.04153.DOI
- [49]D. Tong, JHEP 07, 001 (2022), arXiv:2104.03997.DOI
- [50]J. Wang and Y.-Z. You, Symmetry 14, 1475 (2022), arXiv:2204.14271.DOI