Skip to content

Exterior and Graded Algebra, Grassmann Variables, and Berezin Integration

Exterior algebra packages antisymmetric tensors, while a Z2\mathbb Z_2-grading packages the distinction between even and odd elements. Grassmann variables are odd generators: exchanging two of them changes the sign, and every finite collection generates a finite-dimensional algebra. Berezin integration is coefficient extraction on that algebra. With a fixed ordering convention, complex Grassmann Gaussian integrals produce determinants and real antisymmetric Gaussian integrals produce Pfaffians.

Required background. Direct Sums, Tensor Products, and Index Structure supplies tensor powers, quotient constructions, and antisymmetrization.

Every sign on this page follows from an explicit order. Products of odd elements, derivatives, integration measures, and source factors cannot be reordered silently. The QFT-facing formulas are finite-dimensional regulated identities; a continuum “fermion measure” requires additional analytic and physical control.

Let VV be a finite-dimensional vector space over F=R\mathbb F=\mathbb R or C\mathbb C. Its tensor algebra is

T(V)=k=0Vk,V0=F.T(V) = \bigoplus_{k=0}^{\infty}V^{\otimes k}, \qquad V^{\otimes0}=\mathbb F.

The exterior algebra is the quotient

Λ(V)=T(V)vv:vV.\Lambda(V) = \frac{T(V)} {\langle v\otimes v:v\in V\rangle}.

The image of vwv\otimes w is written vwv\wedge w. The quotient relation implies

vv=0,vw=wv.v\wedge v=0, \qquad v\wedge w=-w\wedge v.

It is graded by degree:

Λ(V)=k=0dimVΛk(V),\Lambda(V) = \bigoplus_{k=0}^{\dim V}\Lambda^k(V),

and homogeneous elements satisfy

αβ=(1)pqβα,αΛp(V),βΛq(V).\alpha\wedge\beta = (-1)^{pq}\beta\wedge\alpha, \qquad \alpha\in\Lambda^p(V), \quad \beta\in\Lambda^q(V).

If e1,,ene_1,\ldots,e_n is a basis of VV, then

ei1eik,1i1<<ikn,e_{i_1}\wedge\cdots\wedge e_{i_k}, \qquad 1\leq i_1<\cdots<i_k\leq n,

is a basis of Λk(V)\Lambda^k(V). Therefore

dimΛk(V)=(nk),dimΛ(V)=2n.\dim\Lambda^k(V)=\binom nk, \qquad \dim\Lambda(V)=2^n.

The top exterior power Λn(V)\Lambda^n(V) is one dimensional. If F=R\mathbb F=\mathbb R, a nonzero top vector determines an orientation together with a scale; positive rescaling preserves the orientation and negative rescaling reverses it. If F=C\mathbb F=\mathbb C, it determines a complex volume element, not an additional real orientation.

Reducing exterior degree modulo two gives a Z2\mathbb Z_2-graded algebra,

Λ(V)=Λ0ˉ(V)Λ1ˉ(V).\Lambda(V) = \Lambda^{\bar0}(V) \oplus \Lambda^{\bar1}(V).

For a homogeneous element aa, write a{0,1}|a|\in\{0,1\} for its parity. The graded commutator is

[a,b]gr=ab(1)abba.[a,b]_{\mathrm{gr}} = ab-(-1)^{|a||b|}ba.

It becomes an ordinary commutator if either element is even and an anticommutator if both are odd. In a supercommutative algebra such as an exterior algebra,

[a,b]gr=0[a,b]_{\mathrm{gr}}=0

for homogeneous aa and bb.

The same rule controls products in a graded tensor product:

(ab)(cd)=(1)bc(ac)(bd).(a\otimes b)(c\otimes d) = (-1)^{|b||c|} (ac)\otimes(bd).

The sign appears because bb must pass through cc. This Koszul sign rule is the reliable way to track signs in long expressions: count every exchange of odd factors. A systematic account of this graded convention appears in Deligne and Morgan 1999, §§ 1.1 and 1.10–1.11.

Grassmann generators θ1,,θn\theta_1,\ldots,\theta_n obey

θiθj=θjθi,θi2=0.\theta_i\theta_j=-\theta_j\theta_i, \qquad \theta_i^2=0.

Every function of finitely many generators is a finite polynomial,

F(θ)=I{1,,n}FIθi1θiI,F(\theta) = \sum_{I\subseteq\{1,\ldots,n\}} F_I\,\theta_{i_1}\cdots\theta_{i_{|I|}},

with the indices in each monomial put in a chosen increasing order. Exponentials terminate whenever their exponent has positive Grassmann degree.

Grassmann variables are algebraic odd numbers, not fermionic creation or field operators. Both use anticommutation, but operators act on a state space, whereas Grassmann generators belong to a supercommutative coefficient algebra.

This page uses left Grassmann derivatives, defined by

Lθjθi=δij\frac{\partial^L\theta_j}{\partial\theta_i} = \delta_{ij}

and the graded Leibniz rule

Lθi(FG)=(LFθi)G+(1)FF(LGθi)\frac{\partial^L}{\partial\theta_i}(FG) = \left( \frac{\partial^L F}{\partial\theta_i} \right)G +(-1)^{|F|} F \left( \frac{\partial^L G}{\partial\theta_i} \right)

for homogeneous FF. For example,

Lθ1(θ1θ2)=θ2,\frac{\partial^L}{\partial\theta_1} (\theta_1\theta_2) = \theta_2,

whereas

Lθ2(θ1θ2)=θ1.\frac{\partial^L}{\partial\theta_2} (\theta_1\theta_2) = -\theta_1.

Right derivatives are equally valid but obey a different displayed Leibniz rule. Mixing left- and right-derivative formulas without translating their signs is a common source of errors.

For homogeneous FF, Berezin integration by parts follows from the graded Leibniz rule:

dθ(LFθ)G=(1)FdθF(LGθ).\int d\theta\, \left(\frac{\partial^L F}{\partial\theta}\right)G = -(-1)^{|F|} \int d\theta\, F\left(\frac{\partial^L G}{\partial\theta}\right).

Indeed, dθθL(FG)=0\int d\theta\,\partial_\theta^L(FG)=0 because a Grassmann derivative has no θ\theta coefficient left for the integral to extract.

Berezin integration is coefficient extraction

Section titled “Berezin integration is coefficient extraction”

For one Grassmann variable, Berezin integration is the linear operation fixed by

dθ1=0,dθθ=1.\int d\theta\,1=0, \qquad \int d\theta\,\theta=1.

Thus, if F(θ)=a+bθF(\theta)=a+b\theta,

dθF(θ)=b.\int d\theta\,F(\theta)=b.

It agrees with left differentiation for one variable and is translation invariant:

dθF(θ+η)=dθF(θ)\int d\theta\,F(\theta+\eta) = \int d\theta\,F(\theta)

for an independent odd η\eta.

For several variables, fix

dnθ=dθndθ1d^n\theta = d\theta_n\cdots d\theta_1

and normalize

dθndθ1θ1θn=1.\int d\theta_n\cdots d\theta_1\, \theta_1\cdots\theta_n = 1.

The integral extracts the coefficient of the ordered top monomial θ1θn\theta_1\cdots\theta_n. Reversing either the generator order or the measure order contributes the sign of the corresponding permutation. The differentials are odd and anticommute. With the declared left derivatives, iterated integration means

dθndθ1F=LθnLθ1F,\int d\theta_n\cdots d\theta_1\,F = \frac{\partial^L}{\partial\theta_n} \cdots \frac{\partial^L}{\partial\theta_1}F,

where the rightmost derivative acts first.

These algebraic integration rules originate in Berezin’s construction; see Berezin 1966, Chapters 1–2.

For an invertible linear change of odd variables

θ=Mθ,MGL(n,F),\theta'=M\theta, \qquad M\in GL(n,\mathbb F),

the measure transforms oppositely to an ordinary commuting measure:

dnθ=(detM)1dnθ.d^n\theta' = (\det M)^{-1}d^n\theta.

For one variable this follows at once from θ=aθ\theta'=a\theta with a0a\neq0 and the normalization dθθ=1\int d\theta'\,\theta'=1. The many-variable determinant follows by antisymmetry. In a mixed even–odd change of variables, the corresponding object is the Berezinian, or superdeterminant; its general theory lies beyond the finite odd Gaussian calculations needed here.

Complex Grassmann Gaussians give determinants

Section titled “Complex Grassmann Gaussians give determinants”

Introduce two independent sets of odd generators

ψ1,,ψN,ψˉ1,,ψˉN.\psi_1,\ldots,\psi_N, \qquad \bar\psi_1,\ldots,\bar\psi_N.

The bar is a conventional label for the paired variables; in Berezin integration, ψˉ\bar\psi and ψ\psi are independent. Fix the paired measure

D(ψˉ,ψ)=dψˉNdψNdψˉ1dψ1\mathcal D(\bar\psi,\psi) = d\bar\psi_N\,d\psi_N \cdots d\bar\psi_1\,d\psi_1

so that

D(ψˉ,ψ)ψ1ψˉ1ψNψˉN=1.\int\mathcal D(\bar\psi,\psi)\, \psi_1\bar\psi_1\cdots\psi_N\bar\psi_N = 1.

This is the preceding convention for the ordered generator list (ψ1,ψˉ1,,ψN,ψˉN)(\psi_1,\bar\psi_1,\ldots,\psi_N,\bar\psi_N).

For an N×NN\times N matrix AA with commuting entries,

D(ψˉ,ψ)exp(ψˉiAijψj)=detA.\boxed{ \int\mathcal D(\bar\psi,\psi)\, \exp(-\bar\psi_iA_{ij}\psi_j) = \det A }.

The exponential is a finite polynomial. Its top-degree coefficient is the antisymmetrized sum over permutations that defines detA\det A. For one pair, the convention is visible without any general argument:

dψˉdψeaψˉψ=dψˉdψ(1aψˉψ)=dψˉdψ(1+aψψˉ)=a.\begin{aligned} \int d\bar\psi\,d\psi\, e^{-a\bar\psi\psi} &= \int d\bar\psi\,d\psi\, (1-a\bar\psi\psi)\\ &= \int d\bar\psi\,d\psi\, (1+a\psi\bar\psi) =a. \end{aligned}

If AA is invertible and η,ηˉ\eta,\bar\eta are independent odd sources that anticommute with ψi,ψˉi\psi_i,\bar\psi_i and with one another, translation invariance gives

D(ψˉ,ψ)exp(ψˉAψ+ηˉψ+ψˉη)=detAexp(ηˉA1η).\boxed{ \begin{aligned} &\int\mathcal D(\bar\psi,\psi)\, \exp\left( -\bar\psi A\psi +\bar\eta\psi +\bar\psi\eta \right)\\ &\qquad = \det A\, \exp\left( \bar\eta A^{-1}\eta \right). \end{aligned} }

The sign is fixed by completing the square in the stated order:

ψˉAψ+ηˉψ+ψˉη=(ψˉηˉA1)A(ψA1η)+ηˉA1η.\begin{aligned} &-\bar\psi A\psi+\bar\eta\psi+\bar\psi\eta\\ &\quad = -(\bar\psi-\bar\eta A^{-1}) A (\psi-A^{-1}\eta) +\bar\eta A^{-1}\eta. \end{aligned}

If AA is singular, the determinant vanishes and A1A^{-1} does not exist. Insertions can saturate the associated zero modes, but the invertible-source formula cannot simply be reused. For a parallel derivation with explicit left-derivative and Jacobian conventions, see Coleman, n.d., Appendices 12B–12D, PDF.

Single-set antisymmetric Grassmann Gaussians give Pfaffians

Section titled “Single-set antisymmetric Grassmann Gaussians give Pfaffians”

Let θ1,,θ2m\theta_1,\ldots,\theta_{2m} be odd generators and let AA be a 2m×2m2m\times2m real or complex antisymmetric matrix with commuting entries. With

d2mθ=dθ2mdθ1,d^{2m}\theta = d\theta_{2m}\cdots d\theta_1,

the Gaussian identity is

d2mθexp(12θiAijθj)=Pf(A).\boxed{ \int d^{2m}\theta\, \exp\left( \frac12\theta_iA_{ij}\theta_j \right) = \operatorname{Pf}(A) }.

This is an algebraic identity; any Majorana reality condition belongs to the physical application, not to the Berezin formula.

For two variables,

A=(0aa0)A= \begin{pmatrix} 0&a\\ -a&0 \end{pmatrix}

gives

12θiAijθj=aθ1θ2,\frac12\theta_iA_{ij}\theta_j = a\theta_1\theta_2,

so the integral is aa, fixing the Pfaffian sign convention. In general,

Pf(A)2=detA.\operatorname{Pf}(A)^2=\det A.

The square does not determine the Pfaffian sign: the ordering of the Grassmann variables, equivalently the orientation of the top exterior power, is part of the definition.

Complex paired variables produce a determinant because ψˉ\bar\psi and ψ\psi are independent sets. A single set with an antisymmetric quadratic form produces a Pfaffian. This is the finite-dimensional algebra behind the determinants associated with Dirac fermions and the Pfaffians associated with Majorana-type quadratic forms; compare Zinn-Justin 2002, Chapter 1, §§ 1.5–1.7, pp. 6–15.

At a lattice, mode cutoff, or other finite regulator, a quadratic fermion action has the form

SF=ψˉAψ.S_F=\bar\psi A\psi.

Its Euclidean Berezin integral is exactly

ZF=D(ψˉ,ψ)eSF=detA.Z_F = \int\mathcal D(\bar\psi,\psi)e^{-S_F} = \det A.

With sources, the inverse matrix A1A^{-1} appears in ηˉA1η\bar\eta A^{-1}\eta and generates the regulated two-point kernel. This is the fermionic counterpart of a bosonic Gaussian, but the determinant power is inverted: a convergent complex bosonic Gaussian is proportional to (detA)1(\det A)^{-1}, while the paired Grassmann Gaussian equals detA\det A under the normalization above.

The finite identity does not by itself define

det(differential operator)\det(\text{differential operator})

in the continuum. Boundary conditions, zero modes, the regulator, phases, and renormalization all matter. A transformation of infinitely many Grassmann variables can also acquire a regulated Jacobian with physical content. The regulated free-fermion application, including source and propagator conventions, is developed on Grassmann Functional Integrals for Free Fermions.

Moving odd factors without a sign. Every exchange of two homogeneous odd objects contributes 1-1. Sources, differentials, and odd derivatives participate in the same sign rule.

Leaving the integration order implicit. A multiple Berezin integral is an oriented coefficient extraction. Reversing two differentials reverses the answer.

Treating ψˉ\bar\psi as an ordinary complex conjugate during integration. The paired variables are algebraically independent. Reality conditions and contours belong to the physical construction, not to the finite Berezin identity.

Using the bosonic Jacobian rule. For θ=Mθ\theta'=M\theta, the odd measure transforms with (detM)1(\det M)^{-1}, not detM\det M.

Confusing determinants and Pfaffians. Independent paired variables with ψˉAψ\bar\psi A\psi give detA\det A. One set of variables with an antisymmetric quadratic form gives Pf(A)\operatorname{Pf}(A).

Writing an inverse in the presence of zero modes. If detA=0\det A=0, the source formula with A1A^{-1} is undefined. The zero modes must be separated and saturated or otherwise treated.

Promoting a regulated identity to a continuum measure. A finite Grassmann algebra has no convergence problem because every expansion terminates. An infinite-dimensional functional integral introduces new questions that finite nilpotence does not answer.

  1. Using left derivatives, compute

    Lθ1(θ1θ2θ3),Lθ2(θ1θ2θ3).\frac{\partial^L}{\partial\theta_1} (\theta_1\theta_2\theta_3), \qquad \frac{\partial^L}{\partial\theta_2} (\theta_1\theta_2\theta_3).
    Solution

    The first derivative removes the first generator without a sign:

    Lθ1(θ1θ2θ3)=θ2θ3.\frac{\partial^L}{\partial\theta_1} (\theta_1\theta_2\theta_3) = \theta_2\theta_3.

    To reach θ2\theta_2, the odd derivative passes one odd factor, so

    Lθ2(θ1θ2θ3)=θ1θ3.\frac{\partial^L}{\partial\theta_2} (\theta_1\theta_2\theta_3) = -\theta_1\theta_3.
  2. Let

    A=(abcd).A= \begin{pmatrix} a&b\\ c&d \end{pmatrix}.

    Expand the two-pair Grassmann Gaussian and verify that its integral is adbcad-bc.

    Solution

    Only the degree-four term contributes. Writing X=ψˉiAijψjX=\bar\psi_iA_{ij}\psi_j, the relevant term is X2/2X^2/2 in eXe^{-X}. Reordering it to ψ1ψˉ1ψ2ψˉ2\psi_1\bar\psi_1\psi_2\bar\psi_2 gives coefficient adbcad-bc. Therefore

    dψˉ2dψ2dψˉ1dψ1eψˉAψ=adbc.\int d\bar\psi_2\,d\psi_2\, d\bar\psi_1\,d\psi_1\, e^{-\bar\psi A\psi} = ad-bc.
  3. For a 4×44\times4 antisymmetric matrix, use the Gaussian definition to show

    Pf(A)=A12A34A13A24+A14A23.\operatorname{Pf}(A) = A_{12}A_{34} -A_{13}A_{24} +A_{14}A_{23}.
    Solution

    Since

    12θiAijθj=i<jAijθiθj,\frac12\theta_iA_{ij}\theta_j = \sum_{i<j}A_{ij}\theta_i\theta_j,

    the top-degree term comes from one half of the square of this expression. The three pairings of {1,2,3,4}\{1,2,3,4\} are (12)(34)(12)(34), (13)(24)(13)(24), and (14)(23)(14)(23). Reordering each product to θ1θ2θ3θ4\theta_1\theta_2\theta_3\theta_4 gives signs +,,++,-,+, respectively. Coefficient extraction yields the displayed formula.

  4. If θ=Mθ\theta'=M\theta with MGL(n,F)M\in GL(n,\mathbb F), verify the inverse-determinant measure rule directly from the top monomial.

    Solution

    Antisymmetry gives

    θ1θn=(detM)θ1θn.\theta'_1\cdots\theta'_n = (\det M)\theta_1\cdots\theta_n.

    To keep the normalized integral of the primed top monomial equal to one, the primed measure must contribute the inverse factor:

    dnθ=(detM)1dnθ.d^n\theta' = (\det M)^{-1}d^n\theta.