THE ODD FERMION

DANIEL S. FREED, MICHAEL J. HOPKINS, AND CONSTANTIN TELEMAN

Abstract. In this short note we use the geometric approach to (topological) field theory to address
the question: Does an odd number of quantum mechanical fermions make sense?
arXiv:2401.04223v1 [hep-th] 8 Jan 2024

Old questions: Does an odd number of quantum mechanical fermions make sense? And, if so,
what is the dimension of the Hilbert space of states?
Answers: Yes, it makes sense as an anomalous theory. The standard definition of ‘dimension’
√
gives 0, and the “twisted dimension” for a single fermion is 2.
Key points:
• The theory is anomalous: it is a 1-dimensional field theory relative to the 2-dimensional
Arf theory.
• The theory is topological, so computations are easily carried out.
• The state space is a module over the Clifford algebra CℓC 1 ; ‘dimension’ and ‘twisted dimen-
sion’ are relative concepts—relative to the Clifford algebra action.
• The tensor product construction of a theory of N1 + N2 fermions works as it should.
This question has garnered some interest over the years, and the theory of an odd number of
fermions has sometimes been deemed inconsistent. See [DGG], [W, §2.1], and [SS, footnote 1] for
recent comments; the latter contains several additional references to the literature. That the state
space is to be regarded as a module over CℓC1 already appears in [F1, Remark 5.10].
We thank Nati Seiberg for his perspective. His lecture at the November, 2023 annual meeting of
the Simons Collaboration on Global Categorical Symmetries inspired us to write this note.

Quantum mechanics of N fermions

Let N be a positive integer, which we presently assume is odd. Consider the quantum mechanical
theory of N free massless spinor fields. It is a 1-dimensional theory which, after Wick rotation, has
background fields a Riemannian metric and spin structure. The fluctuating field is a spinor field
with values in a real vector space of dimension N . For N even, one can work out this theory as a
standard exercise; see [F2, FP16] for example.

Remark 1. We assume from the beginning that the spinor field is fermionic, or odd (anticommut-
ing). Therefore, its quantization generates fermionic states. Furthermore, Bose-Fermi statistics are
encoded in a Z/2Z-grading on the state space, and algebras of observables are also Z/2Z-graded;

Date: November 28, 2023.

This material is based upon work supported by the National Science Foundation under Grant Number DMS-
2005286 and by the Simons Foundation Award 888988 as part of the Simons Collaboration on Global Categorical
Symmetries. This work was performed in part at Aspen Center for Physics, which is supported by National Science
Foundation grant PHY-2210452.
1
2 D. S. FREED, M. J. HOPKINS, AND C. TELEMAN

they are superalgebras. Alternatively, one can consider the system in which the spinor field is
bosonic and there are no fermionic states; see the discussion of a lattice version in [SS, §1.1].

Let W be an N -dimensional real inner product space. The spin structure on a spin Riemannian
1-manifold X is given as a double cover Xe → X. A spinor field ψ on X is a function ψ̃ : X
e → ΠW
e → X. Here ΠW is the parity-reversed vector space,
that is odd under the deck transformation of X
1
i.e., W as an odd vector space. The lagrangian density is

1
(2) L = ⟨ψ, DX ψ⟩ dµX ,
2

where µX is the Riemannian measure on X and DX is the Dirac operator on X. The latter is
DX ψ = ∂ ψ̃ for ∂ the positively oriented unit norm vector field on X. e
Compute canonical quantization on Y = pt as follows. First, solve the classical equations of
motion on R × Y to obtain a symplectic vector space, which in this case is odd. Here the classical
equation asserts that ψ is locally constant, so the vector space of solutions is canonically isomorphic
to ΠW . After complexification, classical observables form the algebra Sym• ΠWC∗ . This is the finite
V1
dimensional exterior algebra of WC∗ with the Z/2Z-grading that makes WC∗ = WC∗ odd. Second,
form the deformation quantization using the Poisson bracket derived from the lagrangian (2). The
result is the Clifford algebra Cliff(WC∗ ) for the complexified dual inner product; the Clifford algebra
carries its usual Z/2Z-grading. Third, quantum states are defined by producing an irreducible
Z/2Z-graded module M over Cliff(WC∗ ). Here is where the parity of N shows up. If N is even,
then the complex algebra Cliff(WC∗ ) of observables is the full endomorphism algebra of M . In this
case M is a Z/2Z-graded vector space of dimension

(3) dim M = 2N/2−1 2N/2−1 ;

the notation ‘b|f ’ indicates the dimensions of the even (Bose) and odd (Fermi) homogeneous sub-
spaces of a Z/2Z-graded vector space. For N odd the action of Cliff(WC∗ ) on M has a nontrivial
commutant:2 the subalgebra of End(M ) consisting of operators that commute with every element
of Cliff(WC∗ ) is isomorphic to CℓC
1 , the complex Clifford algebra on a single odd generator δ.

Example 4 (N = 1). Let W = R with its standard inner product, so Cliff(WC∗ ) = CℓC1 is the
complex superalgebra generated by an odd element γ. An irreducible module is M = C1|1 with
γ = ( 01 10 ). The commutant is generated by δ = 0i −i


0 .

Key point: One must consider the state space as a module over the commutant. Explanation:
In this case the algebra of complex observables A is the commutant of the commutant B inside the
full endomorphism algebra End(M ), so is identified as EndB (M ), the algebra of endomorphisms
of M as a B-module. Typically, one must complete the module M to obtain a Hilbert space of
states, but in this case M is finite dimensional and no completion is necessary.
1More precisely, ΠW = Π ⊗ W , where Π is the canonical odd line.
2As always in super algebra, the commutant is taken in the Z/2Z-graded sense: odd elements a1 , a2 commute if
a2 a1 = −a1 a2 .
 THE ODD FERMION 3

Remark 5. An extra “spectator fermion” has appeared in many previous works, for example [tH,
LW, K]. It plays a similar role to the commutant, but the construction in the previous paragraph
applies generally and leads to the commutant as part of the anomaly.

Next, we compute (heuristically, using a regularization) the path integral on X = S 1 . For

simplicity take N = 1, as in Example 4; the general case is the N th power of the computation here.
There are two spin structures: bounding (antiperiodic, Neveu-Schwarz) and non-bounding (periodic,
Ramond). The relevant one for us is the bounding spin structure. Let X be a Riemannian circle of
length L. The path integral of the Lagrangian (2) is the pfaffian pfaff DX of the Dirac operator. It
is naturally an element of the real Pfaffian line Pfaff DX . That line carries a natural inner product,
and the norm square of pfaff DX is the determinant of DX . The set of eigenvalues of DX is

 
2πin 1
(6) :n∈Z+ .
L 2

The computation of the product of eigenvalues is

Q Q
  n n
Y 2πi Y Y n̸=0 n̸=0
(7) n = n= n= Q = Q = 2.
L n 2n
n∈Z+1/2 n∈Z+1/2 n odd n even n̸=0
n̸=0

We use ζ-functions to regulate the infinite products. In the first and second steps we use that the
ζ-regularized cardinality of Z, and so too of Z + 1/2, is zero. In the last step we use the corollary
that the ζ-regularized cardinality of Z̸=0 is −1. The pfaffian is the square root of (7), determined
here up to sign.

Remark 8.
(1) The path integral is independent of the length L: the theory is topological.
(2) That the state space M is a module over CℓC 1 signals that the theory is anomalous, as
confirmed by the fact that the path integral is an element of a line rather than a number.
(3) In the next section we use a canonical element of Pfaff DX to identify the pfaffian pfaff DX
√
as a number: 2.
(4) For the nonbounding spin structure the Dirac operator has a 1-dimensional kernel, and
hence the Z/2Z-graded Pfaffian line is odd in the sense of super vector spaces. The nonzero
kernel also implies that the pfaffian element is zero. In fact, this is a consequence of the
oddness of the Pfaffian line: in the category of super vector spaces every homomorphism
C → Π from an even line to an odd line is zero.

N fermions as a boundary topological field theory

In the Wick-rotated setting an anomaly is a once-categorified invertible field theory, and the
anomalous theory is defined relative to it. Both theories are defined on the same bordism category,
so in particular have the same dimension. It often happens that the anomaly extends to a full
theory in one higher dimension. See [F3] for discussion and motivation.
4 D. S. FREED, M. J. HOPKINS, AND C. TELEMAN

Figure 1. The CℓC

1 -module M

Fix N to be odd. We work with topological field theories defined on spin manifolds and taking
values in the Morita 2-category of complex superalgebras. The preceding implies that the anomaly
theory of N fermions assigns the algebra3 CℓC 1 to a point and the Pfaffian line of Dirac to a
circle. The resulting once-categorified 1-dimensional invertible theory α is the truncation of the 2-
dimensional invertible “Arf theory”; see [Gu, DeGu] for example. The partition function on a closed
spin 2-manifold W is α(W ) = (−1)Arf W , where Arf W is the Arf invariant of the spin structure:
0 for even spin structures and 1 for odd spin structures. The quantum theory of N fermions is a
boundary theory of α; the combined theory (α + boundary) assigns the left CℓC 1 -module M to the
bordism depicted in Figure 1. Green depicts components of the boundary colored by the module
theory, and the red arrow tells if an uncolored boundary component is incoming or outgoing.
In a spin theory the usual dimension of the value of the theory on a codimension one closed
manifold Y is computed by taking Cartesian product of Y with the nonbounding S 1 : the spin
double cover of the nonbounding S 1 is the product double cover, so it computes the trace of the
identity map in whatever category we are in. Furthermore, this trace lives in the Hochschild
homology of the value of Y . In our case, for Y = pt the Arf theory on the Cartesian product
1
evaluates to α(Snonbounding ) = Π, where Π is the odd line. As mentioned in Remark 8(4), an odd
line has no nonzero elements. Hence the usual dimension is 0.

Remark 9. This is the ‘usual dimension’, or categorical dimension, in the category of CℓC 1 -modules.
b|f
By comparison, the ‘usual dimension’ in the category of super vector spaces is dim C = b − f .
This is oft referred to as ‘trace (−1)F ’. The ‘twisted dimension’ we are after is the categorical trace
of the grading operator, which for the super vector space Cb|f equals b + f . We now compute
‘twisted dimension’ in the category of CℓC 1 -modules.

Figure 2. The twisted dimension of M

1
Consider the Cartesian product of the bordism in Figure 1 with the bounding circle Sbounding .
The theory (α + boundary) computes the supertrace of the grading operator on M —what we are
1
calling the twisted dimension of M —which lives in the Hochschild cohomology α(Sbounding ) of the
C
algebra A = Cℓ1 . The Hochschild cohomology is an algebra, so has a canonical element: the unit
for multiplication. Since the Arf theory α is invertible, the Hochschild cohomology is a line, and
so the unit is a basis element. We use this basis element to define the twisted dimension as a
3TheClifford algebra is invertible in the Morita 2-category of superalgebras. This is equivalent to its being central
simple. See [Wa, De].
 THE ODD FERMION 5

number, the value of the theory on Figure 2. Compute the square of this number by evaluating
the squared theory α⊗2 on Figure 2. Observe that α⊗2 (pt) = CℓC C ∼
1 ⊗ Cℓ1 = Cℓ2 , and the square
C

of (α + boundary) assigns the left Cℓ2 -module M ⊗ M to the bordism in Figure 1. Observe that
C

every finite dimensional A = CℓC 1 -module is free, so has the form M = A ⊗ V for an even vector
space V . For N fermions (N odd) we have dim V = 2(N −1)/2 . The Morita equivalence CℓC 2 ≃ C
identifies CℓC
2 -modules with super vector spaces; under this equivalence M ⊗ M corresponds to the
2
super vector space V ⊗ V ⊗ A. By Remark 9, the twisted dimension of M ⊗ M is 2(dim V ) . This
is the square of the twisted dimension of M , hence the twisted dimension of M equals
√ √
(10) 2 dim V = 2(N −1)/2 2.

√
For a single fermion N = 1, this twisted dimension equals 2.

Remark 11.
(1) Tensor products work as they should. For example, the tensor product of the state spaces in
two theories with N odd is the tensor product of two CℓC C
1 -modules, which is a Cℓ2 -module.
As above, apply the Morita equivalence CℓC2 ≃ C to identify Cℓ2 -modules with super vector
C

spaces. The twisted dimension of the resulting super vector space is 1/2 the product of the
twisted dimensions of the CℓC 1 -modules.
(2) There are variants of this story on pin manifolds. In that case the anomaly theory is more
involved; see [DeGu].

References

[De] Pierre Deligne, Notes on spinors, Quantum Fields and Strings: a course for mathematicians, Vol. 1, 2
(Princeton, NJ, 1996/1997), Amer. Math. Soc., Providence, RI, 1999, pp. 99–135. 4
[DeGu] Arun Debray and Sam Gunningham, The Arf-Brown TQFT of pin− surfaces, Topology and quantum
theory in interaction, Contemp. Math., vol. 718, Amer. Math. Soc., Providence, RI, 2018, pp. 49–87.
arXiv:1803.11183. 4, 5
[DGG] Diego Delmastro, Davide Gaiotto, and Jaume Gomis, Global anomalies on the Hilbert space, JHEP 11
(2021), 142, arXiv:2101.02218 [hep-th]. 1
[F1] Daniel S Freed, Anomalies and invertible field theories, Proc. Symp. Pure Math, Proc. Sympos. Pure Math.,
vol. 88, Amer. Math. Soc., Providence, RI, 2014, pp. 25–45. arXiv:1404.7224. 1
[F2] Daniel S. Freed et. al., Homework, pp. 609–717, American Mathematical Society, Providence, RI, 1999. in
Quantum Fields and Strings: A Course for Mathematicians. Vol. 1. eds P. Deligne, P. Etingof, D. S. Freed,
L. C. Jeffrey, D. Kazhdan, J. W. Morgan, D. R. Morrison, E. Witten. 1
[F3] Daniel S. Freed, What is an anomaly?, arXiv:2307.08147. 3
[Gu] Sam Gunningham, Spin Hurwitz numbers and topological quantum field theory, Geom. Topol. 20 (2016),
no. 4, 1859–1907, arXiv:1201.1273. 4
[K] Alexei Kitaev, Unpaired Majorana fermions in quantum wires, Physics-Uspekhi 44 (2001), no. 10S, 131,
arXiv:cond-mat/0010440. 3
[LW] Jaehoon Lee and Frank Wilczek, Algebra of Majorana Doubling, Phys. Rev. Lett. 111 (2013), no. 22,
226402, arXiv:1307.3245 [cond-mat.supr-con]. 3
[SS] Nathan Seiberg and Shu-Heng Shao, Majorana chain and Ising model – (non-invertible) translations, anom-
alies, and emanant symmetries, arXiv:2307.02534 [cond-mat.str-el]. 1, 2
6 D. S. FREED, M. J. HOPKINS, AND C. TELEMAN

[tH] Gerard ’t Hooft, Naturalness, chiral symmetry, and spontaneous chiral symmetry breaking, NATO Sci. Ser.
B 59 (1980), 135–157. 3
[W] Edward Witten, Anomalies and Nonsupersymmetric D-Branes, arXiv:2305.01012 [hep-th]. 1
[Wa] C. T. C. Wall, Graded Brauer groups, J. Reine Angew. Math. 213 (1963/1964), 187–199. 4

Harvard University, Department of Mathematics, Science Center Room 325, 1 Oxford Street,
Cambridge, MA 02138
Email address: dafr@math.harvard.edu

Harvard University, Department of Mathematics, Science Center Room 325, 1 Oxford Street,
Cambridge, MA 02138
Email address: mjh@math.harvard.edu

Department of Mathematics, University of California, 970 Evans Hall #3840, Berkeley, CA 94720-
3840
Email address: teleman@berkeley.edu