Regulated Jacobians and Measure Variation

A fermion transformation can leave the classical action invariant and still change the quantum functional integral because a Grassmann measure transforms with an inverse determinant. In a finite basis that statement is exact. In a field theory, however, its logarithm is an infinite trace and has no meaning until the field space, operator domain, regulator, and preserved background symmetries have been declared.

For a gauge-covariant heat-kernel regulator on a closed Euclidean spin four-manifold, the regulated trace has a local short-time limit. Its integrated chiral supertrace is an index, while its local coefficient supplies a covariant anomaly representative. A different compatible prescription or a local counterterm can move that representative. Only the part that survives the admissible-counterterm test is an anomaly class.

This page first calibrates every sign on a massless Dirac axial rotation and then isolates the covariant measure density for one Weyl fermion. It does not identify that density with the consistent variation of a chiral determinant; that conversion, anomaly cancellation, descent, and global phases are handled on later pages.

Required background. What Is an Anomaly? supplies the admissible-counterterm test and the warning that a Jacobian is a diagnostic rather than the final verdict. Changes of Variables and Regulated Jacobians supplies exact finite-regulator changes of variables, determinant direction, and Berezin integration. Canonical Quantization of the Free Dirac Field supplies fermionic normalization, charge, and massless degrees of freedom; the Euclidean spectral measure used below is defined here.

Helpful background. Fredholm and Dirac Index Theorems and Zero-Mode Counting supplies the heat-kernel supertrace, nonzero-mode pairing, and twisted Dirac index used without proof.

A finite fermion basis gives an exact Berezinian

Start with two independent sets of $N$ Grassmann coefficients. In an orthonormal Euclidean spinor basis,

$$\psi_N(x)=\sum_{n=1}^{N}a_n\phi_n(x), \qquad \bar\psi_N(x)=\sum_{n=1}^{N}\bar a_n\phi_n^\dagger(x),$$

and inherit the site’s paired ordering:

$$\mathcal D_N(\bar\psi,\psi) = \mathrm d\bar a_N\,\mathrm da_N \cdots \mathrm d\bar a_1\,\mathrm da_1,$$

normalized so that

$$\int\mathcal D_N(\bar\psi,\psi)\, a_1\bar a_1\cdots a_N\bar a_N=1.$$

If the coefficient maps are

$$a'_m=C_{mn}a_n, \qquad \bar a'_m=\bar a_n\,\bar C_{nm},$$

then Berezin integration gives

$$\mathcal D_N(\bar\psi',\psi') =J_N\,\mathcal D_N(\bar\psi,\psi), \qquad J_N=\frac{1}{\det C\,\det\bar C}.$$

The inverse determinant is not a continuum subtlety: for one odd variable, $a'=ca$ and $\int\mathrm da'\,a'=1$ force $\mathrm da'=c^{-1}\mathrm da$. The general finite Berezinian is derived in Zinn-Justin 2021, §§ 1.6.4–1.6.5, pp. 10–12, eqs. (1.55)–(1.63).

For a local axial rotation of a Dirac field,

$$\psi'=e^{i\alpha(x)\Gamma_5}\psi, \qquad \bar\psi'=\bar\psi e^{i\alpha(x)\Gamma_5},$$

the two infinitesimal matrices agree:

$$\begin{aligned} C_{mn} &=\delta_{mn}+iB_{mn}+O(\alpha^2), \\ B_{mn} &=\int_X\mathrm d\operatorname{vol}_g\, \phi_m^\dagger\alpha\Gamma_5\phi_n, \qquad \bar C=C. \end{aligned}$$

Consequently,

$$\log J_N=-2i\operatorname{Tr}_N B+O(\alpha^2).$$

The factor two belongs to this Dirac axial transformation: both $\psi$ and $\bar\psi$ rotate with the same sign. For a vector transformation their matrices are inverse and the two finite determinants cancel. A one-Weyl gauge measure is different again; it will be treated as a chiral difference, not by halving this formula. Fujikawa’s original observation was precisely that the axial transformation has a nontrivial fermion-measure Jacobian Fujikawa 1979, pp. 1195–1198.

The continuum trace must be regulated

Taking $N\to\infty$ formally turns the preceding result into

$$\log J\overset{\mathrm{formal}}{=} -2i\int_X\mathrm d\operatorname{vol}_g\, \alpha(x) \sum_n\phi_n^\dagger(x)\Gamma_5\phi_n(x).$$

The sum is a coincident-point trace of the identity with a chirality insertion. It is not a number, and writing it as $\operatorname{tr}(\Gamma_5)\delta(0)$ does not define it.

Fix the continuation from the site’s mostly-minus convention by $t=-i\tau$ and

$$\gamma_E^4=\gamma_M^0, \qquad \gamma_E^k=-i\gamma_M^k, \qquad \epsilon_E^{1234}=+1.$$

The Hermitian Euclidean matrices obey $\{\gamma_E^\mu,\gamma_E^\nu\}=2\delta^{\mu\nu}$. Define

$$\Gamma_5 =-\gamma_E^1\gamma_E^2\gamma_E^3\gamma_E^4 =\gamma_{5,M}, \qquad P_\pm=\frac{1\pm\Gamma_5}{2}.$$

Let $X$ now be a closed oriented Riemannian spin four-manifold, and let $E\to X$ be a Hermitian bundle with unitary connection. The twisted operator

$$\mathcal D_E =\gamma_E^\mu\nabla_\mu^E$$

is anti-Hermitian on its closed domain. Thus $H_E=i\mathcal D_E$ is self-adjoint and

$$H_E^2=\mathcal D_E^\dagger\mathcal D_E=-\mathcal D_E^2$$

is a nonnegative elliptic operator. For a smooth damping profile $r$ with $r(0)=1$ and sufficiently rapid decay, define

$$\boxed{ \delta_\alpha\log J_\Lambda =-2i\operatorname{Tr} \left[ \alpha\Gamma_5\,r\!\left(\frac{H_E^2}{\Lambda^2}\right) \right] }.$$

The heat-kernel choice is $r(u)=e^{-u}$. A sharp finite spectral projection also produces an exact finite Grassmann measure, but is nonlocal. A smooth heat kernel instead regulates the trace in a local, background-covariant way; it is not itself a literal finite-dimensional measure. For a local $\alpha$, even a finite transformation can leave a retained spectral subspace, so the finite calculation must use the projected infinitesimal matrix before the cutoff is removed.

Regulator compatibility is a mathematical condition, not a slogan. If a background gauge transformation acts by $U_g$, a regulator chosen to preserve that symmetry must obey

$$H_E[A^g]^2 =U_gH_E[A]^2U_g^{-1}.$$

Then its regulated trace transforms covariantly. A regulator can instead be chosen to preserve a different set of Ward identities; the displaced local terms must then be tracked rather than silently discarded. The mode calculation and background-covariant cutoff are developed in Bilal 2008, §§ 3.1–3.4, arXiv v1, pp. 8–12, eqs. (3.4)–(3.22), Open PDF. A modern family of convergent functional traces and its regulator conditions is given in Cohen, Lu, and Zhang 2023, §§ 3.2–3.4, arXiv v1, pp. 12–21, eqs. (3.13)–(3.51), Open PDF.

The Jacobian is one term in a complete change-of-variables identity. For an insertion $\mathcal O$ and Euclidean weight $e^{-S_E}$,

$$0= \left\langle \delta_\alpha\mathcal O -\mathcal O\,\delta_\alpha S_E +\mathcal O\,\delta_\alpha\log J_\Lambda \right\rangle,$$

provided the transformation preserves the integration domain and produces no unaccounted contour or boundary term. A contact term can therefore come from $\delta\mathcal O$, an explicit breaking from $\delta S_E$, and an anomaly candidate from the measure. None may be inferred from the Jacobian alone.

The heat-kernel trace becomes a local density

For the heat regulator, introduce the diagonal supertrace density

$$\mathcal K_\Lambda(x) = \operatorname{tr}_{S\otimes E} \left\langle x\left| \Gamma_5e^{-H_E^2/\Lambda^2} \right|x\right\rangle.$$

Heat-kernel asymptotics make this expression local. With the chirality, orientation, and curvature conventions just fixed,

$$\boxed{ \lim_{\Lambda\to\infty} \mathcal K_\Lambda(x)\,\mathrm d\operatorname{vol}_g = \left[ \widehat A(TX)\operatorname{ch}(E) \right]_4 }.$$

The equality is the local index density statement, understood in the short-time asymptotic or distributional sense appropriate to the smeared trace. Its proof uses the heat coefficient of a Laplace-type operator and is owned by the linked Mathematical Methods page. The general local heat-kernel expansion is given in Vassilevich 2003, § 2.2, arXiv v3, pp. 14–17, eqs. (2.18)–(2.37), Open PDF; its use in the regulated chiral Jacobian is explained in Vassilevich 2003, § 7.2, arXiv v3, pp. 68–71, eqs. (7.23)–(7.38), Open PDF.

The integrated supertrace has a stronger, profile-independent statement. Define

$$n_\pm =\dim\ker\!\left( \mathcal D_E^\pm: \Gamma(S_\pm\otimes E)\longrightarrow\Gamma(S_\mp\otimes E) \right).$$

If $r(0)=1$, then

$$\begin{aligned} \operatorname{Tr} \left[ \Gamma_5r\!\left(\frac{H_E^2}{\Lambda^2}\right) \right] &=n_+-n_- \\ &=\operatorname{ind}\mathcal D_E^+. \end{aligned}$$

Every nonzero eigenmode is paired with an opposite-chirality partner, while the zero modes remain unpaired. This is why a constant axial rotation has the exact finite phase

$$J(\alpha)= \exp\!\left[-2i\alpha\, \operatorname{ind}\mathcal D_E^+ \right].$$

It is important not to reverse the logic: as $\Lambda\to\infty$ the regulator does not become a pointwise zero-mode projector. The local density comes from the full short-time spectrum; only the integrated supertrace reduces to the kernel by spectral pairing. The zero-mode/index argument is given in Bilal 2008, § 3.7, arXiv v1, pp. 15–17, eqs. (3.34)–(3.43), Open PDF and Vassilevich 2003, § 7.3, arXiv v3, pp. 72–74, eqs. (7.39)–(7.46), Open PDF.

As a normalization check, take the Euclidean action

$$S_E=\int_X\mathrm d\operatorname{vol}_g\, \bar\psi(\mathcal D_E+m)\psi.$$

Use the Euclidean axial current

$$j_{5,E}^\mu =\bar\psi\gamma_E^\mu\Gamma_5\psi.$$

For the local axial rotation above,

$$\delta_\alpha S_E =-i\int_X\mathrm d\operatorname{vol}_g\, \alpha \left( \nabla_\mu j_{5,E}^\mu -2m\bar\psi\Gamma_5\psi \right).$$

At separated points, the regulated Ward identity therefore gives

$$\nabla_\mu j_{5,E}^\mu =2m\bar\psi\Gamma_5\psi +2\mathcal K, \qquad \mathcal K\,\mathrm d\operatorname{vol}_g =\left[\widehat A(TX)\operatorname{ch}(E)\right]_4.$$

Insertions add their contact terms. The sign and factor two here calibrate a Dirac axial current; they are not yet a one-Weyl gauge anomaly.

A four-dimensional Weyl fermion exposes both background terms

Specialize to a complex line bundle $L^q$ with $q\in\mathbb Z$ and a real nondynamical $U(1)$ connection $a$, normalized so a unit charge has holonomy $e^{i\oint a}$. Its curvature $f$ is global, while $f=\mathrm da$ holds only in a local trivialization. Set

$$c_1=\frac{f}{2\pi}, \qquad \nabla^{L^q}=\nabla-iqa.$$

Use the Pontryagin convention

$$p_1(TX) =-\frac{1}{8\pi^2} \operatorname{tr}_{\mathrm{vec}} (\mathcal R\wedge\mathcal R), \qquad \widehat A(TX)=1-\frac{p_1(TX)}{24}+\cdots.$$

The index density for a charge-$q$ Dirac operator is

$$\left[ \widehat A(TX)e^{qc_1} \right]_4 = \frac{q^2}{2}c_1^2 -\frac{1}{24}p_1(TX).$$

This is the density $\mathcal K$ in the axial Dirac identity. A chiral gauge measure requires a different insertion. For one positive-chirality Weyl fermion, let

$$\psi_+\longmapsto e^{iq\lambda}\psi_+, \qquad \bar\psi_-\longmapsto\bar\psi_-e^{-iq\lambda}.$$

The chiral operator maps between opposite chiralities rather than defining an endomorphism of one of them. Use the doubled elliptic Dirac operator to regulate the two independent Weyl integration factors. Their parity-even traces cancel, leaving the chiral difference $P_+-P_-=\Gamma_5$. Define the covariantly regulated local insertion by

$$\boxed{ \begin{aligned} \mathcal I^{\mathrm{cov}}_4(q) &=\lim_{\Lambda\to\infty} \operatorname{tr}_{S\otimes L^q} \left\langle x\left| q\Gamma_5e^{-H_E^2/\Lambda^2} \right|x\right\rangle \mathrm d\operatorname{vol}_g \\ &=q\left[ \widehat A(TX)e^{qc_1} \right]_4 \\ &=\frac{q^3}{2}c_1^2 -\frac{q}{24}p_1(TX). \end{aligned} }$$

This formula is an index-normalized covariant local measure density. It is not the logarithmic variation of a globally defined chiral determinant, nor is it automatically the consistent variation $\delta_\lambda W$ of a single effective action. Constructing that phase, enforcing Wess–Zumino consistency, and relating the consistent and covariant currents require a different prescription or a local Bardeen–Zumino current shift. In these conventions the pure $U(1)^3$ coefficient of the covariant density is three times the consistent coefficient. The mixed linear-in-$q$ coefficients agree in the standard representative that preserves diffeomorphism and local-Lorentz covariance and places the mixed variation in the $U(1)$ Ward identity; a local counterterm can redistribute that mixed representative without changing its class. The coefficient comparison and its counterterm ceiling are explicit in Cohen, Lu, and Zhang 2023, §§ 3.4 and 4.2, arXiv v1, pp. 18–21 and 24–27, eqs. (3.36)–(3.51) and (4.18)–(4.20), Open PDF. This is a flat-four-dimensional comparison, not a boundary or global theorem. It is included as a normalization warning, not a derivation. Neither coefficient is obtained by blindly halving the Dirac axial result.

The characteristic-form normalization and the simultaneous gauge and gravitational background terms follow from Álvarez-Gaumé and Vázquez-Mozo 2024, §§ 2–3, arXiv v2, pp. 4–8, eqs. (6)–(11), (17)–(19), and (22), Open PDF. Their anomaly polynomial is the next degree in the same index expansion; this page uses only the regulated four-form insertion. The regulated chiral difference and the failure of an anomalous covariant current to satisfy Bose symmetry or Wess–Zumino integrability are exhibited in Fujikawa 1994, § 3, arXiv v2, pp. 11–16, eqs. (3.1), (3.3), and (3.10), Open PDF. The distinction between a gauge-covariant prescription and a Wess–Zumino-consistent one is also made explicit in Cohen, Lu, and Zhang 2023, § 3.4, arXiv v1, pp. 18–21, eqs. (3.36)–(3.51), Open PDF.

Three immediate checks fix the interpretation:

• Reversing Euclidean chirality exchanges $P_+$ and $P_-$ and reverses the whole density.
• A Dirac pair can be written as positive-chirality charges $q$ and $-q$; both the cubic and linear terms cancel in the gauge measure.
• The axial current of that same Dirac field is a different transformation: its gauge term is proportional to $q^2$ and does not cancel.

The next page sums these densities over representations and connects them to triangle amplitudes. No cancellation criterion beyond these checks is being claimed here.

Regulator and counterterms choose the representative

Four objects that are often conflated must be kept separate:

object	precise status
finite retained-mode Berezinian $J_N$	exact after the ordered basis, projection, and transformation map are fixed
formal continuum trace	undefined before a domain and regulator are supplied
regulated local density	a prescription-dependent representative constrained by the symmetries preserved by the regulator
anomaly class	the surviving equivalence class modulo admissible local counterterms

Suppose two local, UV-admissible prescriptions preserve the same required background data and differ by a local functional $C[b]$. Their effective actions and Ward representatives satisfy

$$W_2[b]=W_1[b]+C[b], \qquad \mathfrak A_2(\lambda;b) =\mathfrak A_1(\lambda;b)+\delta_\lambda C[b].$$

The local expression has moved, but the counterterm quotient has not. If an allowed $C$ cancels the full variation while preserving every other required identity, the candidate was removable. If no such $C$ exists, the class is a genuine anomaly. This criterion is stated compactly in Bilal 2008, § 6.2, arXiv v1, pp. 42–43, eq. (6.7), Open PDF.

This also explains why there is no regulator-independent assignment of every local term to “the measure.” A local field redefinition or counterterm can move a contribution among the Jacobian, action, composite-current definition, and source functional. The full transformed functional integral is the invariant comparison. Cohen, Lu, and Zhang make this separation explicit by constructing convergent regulator families and their associated local counterterms Cohen, Lu, and Zhang 2023, §§ 2.1–2.3 and 4.2, arXiv v1, pp. 4–9 and 24–27, eqs. (2.7)–(2.23) and (4.15)–(4.20), Open PDF.

The conclusion is one-way. Compatible local regulators can change the representative by admissible local terms under the stated locality and renormalization hypotheses. This does not say that every regulator is compatible, that a regulator preserving one current preserves all currents, or that local counterterms detect finite and torsion phases.

Zero modes and global information set the stopping point

For the massless chiral block $\mathcal D_E^+:\mathcal H_+\to\mathcal H_-$, set

$$K_+=\ker\mathcal D_E^+, \qquad K_-=\ker\mathcal D_E^- \cong\operatorname{coker}\mathcal D_E^+.$$

Choose ordered orthonormal bases $(u_1,\ldots,u_{n_+})$ of $K_+$ and $(v_1,\ldots,v_{n_-})$ of $K_-$, together with the orthogonal splittings

$$\mathcal H_+=K_+\oplus K_+^\perp, \qquad \mathcal H_-=K_-\oplus K_-^\perp.$$

The symbol $\det'\mathcal D_E^+$ refers only to the invertible restriction

$$\mathcal D_E^{+\prime}: K_+^\perp\longrightarrow K_-^\perp.$$

Expanding the zero modes as $\psi_{+,0}=a_i u_i$ and $\bar\psi_{-,0}=\bar b_jv_j^\dagger$, fix their remaining orientation by

$$\omega_0 = \mathrm d\bar b_{n_-}\cdots\mathrm d\bar b_1\, \mathrm da_{n_+}\cdots\mathrm da_1.$$

If either $K_+$ or $K_-$ is nonzero, the massless uninserted Gaussian determinant vanishes. The partition function without zero-mode-saturating insertions may therefore be zero, so neither $\log Z$ nor $Z'/Z$ nor a determinant phase follows from the naive Gaussian formula. Correlators are nonzero only when their insertions saturate every required $a_i$ and $\bar b_j$; the transformation of the oriented wedge $\omega_0$ produces the index selection rule. A nonzero mass can lift these zero modes, so this statement is not a claim about $\mathcal D_E+m$ at generic $m$.

This finite zero-mode orientation is compatible with the constant-parameter Jacobian, but it does not settle the global problem. As backgrounds vary, the kernel can jump, so the chosen complements, ordered bases, and primed determinant must be rechecked. The local heat coefficient controls only the local infinitesimal or perturbative anomaly. This page makes no claim about finite holonomies or torsion phases; those questions belong to the global-anomaly and determinant-line pages.

The closed-manifold assumption was also essential. With a boundary, the operator domain and boundary condition are part of the regulator. A chiral rotation can change a bag projector rather than map the integration domain to itself, and the heat expansion can acquire genuine boundary coefficients. This is shown for local bag conditions in Marachevsky and Vassilevich 2004, § 2, arXiv v1, pp. 3–5, eqs. (6)–(17), and §§ 5.2–6, pp. 12–14, eqs. (57)–(59), Open PDF. That result has specific assumptions, including a restricted four-dimensional boundary geometry; it is not a universal boundary formula.

For a bounded or noncompact problem, the correct stopping test is therefore:

1. declare a self-adjoint elliptic domain on a bounded region, or an $L^2$/Fredholm domain with explicit falloff and asymptotic data on a noncompact space;
2. verify that every ordinary heat trace used is trace class, or supply the relative-trace subtraction needed for a noncompact problem;
3. verify that the transformation preserves the domain, or transform the boundary data as part of the problem;
4. include boundary heat coefficients, flux terms, zero-mode saturation, and contour data; and
5. only then apply the local-counterterm quotient.

The corrected technical extension of Fujikawa’s method to smooth regulator classes, even dimensions, gravity, and the zero-frequency sector is Fujikawa 1980, pp. 2848–2858, together with its 1980 erratum.

Common pitfalls

Treating the formal trace as a calculation. Neither $\sum_n\phi_n^\dagger\Gamma_5\phi_n$ nor $\operatorname{tr}(\Gamma_5)\delta(0)$ is defined. The regulated operator, domain, and symmetry properties are part of the result.

Calling the local heat density “the anomaly.” It is a regulated representative. One must still include action, sources, contact and boundary terms, and then quotient by admissible local counterterms.

Halving a Dirac axial anomaly to obtain a Weyl gauge anomaly. The axial Dirac factor two comes from equal rotations of $\psi$ and $\bar\psi$. A Weyl gauge determinant uses a chiral difference and has an additional consistent-versus-covariant distinction.

Saying only zero modes create the local density. Zero modes determine the integrated supertrace after nonzero-mode cancellation. The pointwise local coefficient is a short-time spectral asymptotic involving all modes.

Ignoring the integration domain. On a boundary, a transformation can change the boundary condition or carry charge. That is not the same change-of-variables problem as a redundancy preserving a fixed domain.

Assuming a local calculation clears global anomalies. Heat-kernel coefficients detect perturbative local data. They do not by themselves test large transformations, determinant-line holonomy, or torsion.

Check your understanding

1. Let $a'=ca$ for one Grassmann variable. Determine the transformed measure from $\int\mathrm da'\,a'=1$.

   Solution

   Write $\mathrm da'=k\,\mathrm da$. Then $1=\int k\,\mathrm da\,ca=kc$, so $k=c^{-1}$. This is the one-variable origin of the inverse determinant in a fermion measure.

2. Why is $\operatorname{Tr}[\Gamma_5r(H_E^2/\Lambda^2)]$ independent of the damping profile when $r(0)=1$?

   Solution

   The nonzero spectrum is paired between opposite chiralities, so every nonzero contribution cancels at the same value of $H_E^2$. Only zero modes remain, and each is weighted by $r(0)=1$. The answer is therefore $n_+-n_-=\operatorname{ind}\mathcal D_E^+$.

3. Add the positive-chirality $U(1)$ densities for charges $q$ and $-q$. What cancels, and what does this not say about an axial rotation?

   Solution

   Both $q^3c_1^2/2$ and $-q\,p_1/24$ are odd in $q$, so the two gauge densities cancel. The axial Dirac rotation acts with the same sign on the two independent Grassmann factors and has a gauge term proportional to $q^2$; it is a different symmetry and need not cancel.

4. Two compatible prescriptions give $\mathfrak A_2-\mathfrak A_1=\delta_\lambda C$ for an admissible local $C$. Have they found different anomaly classes?

   Solution

   No. They have chosen different representatives of the same class. If one representative is completely canceled by an allowed counterterm while all required identities remain intact, the class itself is trivial.

5. Why can the closed-manifold heat-kernel formula not simply be reused with a boundary?

   Solution

   A boundary condition is part of the operator domain and may not be preserved by the transformation. Boundary heat coefficients and Ward fluxes can also survive. The domain, transformation law, and boundary contributions must therefore be checked before applying the counterterm test.

What to carry forward

• Perturbative Chiral and Gauge Anomalies turns the local density into representation traces, triangle amplitudes, and cancellation conditions.
• Consistent and Covariant Anomalies constructs the current and counterterm shift that relate the two local representatives.
• Wess–Zumino Consistency and Descent derives the integrability condition obeyed by a consistent anomaly.
• Anomaly Polynomials and Inflow develops the higher-degree index polynomial and bulk cancellation mechanism.
• Global and Torsion Anomalies supplies large-transformation, determinant-holonomy, and torsion-sensitive tests.
• Fredholm and Dirac Index Theorems and Zero-Mode Counting proves the supertrace and index statements used here.
• Spin, Gauge, and Gravitational-Anomaly Responses owns the corresponding deployment on curved backgrounds.

References

• Álvarez-Gaumé, Luis, and Miguel Á. Vázquez-Mozo. “Anomalies and the Green–Schwarz Mechanism.” In Handbook of Quantum Gravity, edited by Cosimo Bambi, Leonardo Modesto, and Ilya L. Shapiro, 2241–2284. Singapore: Springer, 2024. DOI. Open PDF, arXiv v2.
• Bilal, Adel. “Lectures on Anomalies.” LPTENS-08/05, arXiv:0802.0634v1, 2008. Stable record. Open PDF.
• Cohen, Timothy, Xiaochuan Lu, and Zhengkang Zhang. “Anomalies from the Covariant Derivative Expansion.” Physical Review D 107 (2023): 116015. DOI. Open PDF, arXiv v1.
• Fujikawa, Kazuo. “Path-Integral Measure for Gauge-Invariant Fermion Theories.” Physical Review Letters 42, no. 18 (1979): 1195–1198. DOI.
• Fujikawa, Kazuo. “Path Integral for Gauge Theories with Fermions.” Physical Review D 21, no. 10 (1980): 2848–2858; erratum, Physical Review D 22 (1980): 1499. DOI. Erratum.
• Fujikawa, Kazuo. “Generalized Pauli–Villars Regularization and the Covariant Form of Anomalies.” Nuclear Physics B 428, nos. 1–2 (1994): 169–188. DOI. Open PDF, arXiv v2.
• Marachevsky, Valery N., and Dmitri V. Vassilevich. “Chiral Anomaly for Local Boundary Conditions.” Nuclear Physics B 677, nos. 1–2 (2004): 535–552. DOI. Open PDF, arXiv v1.
• Vassilevich, Dmitri V. “Heat Kernel Expansion: User’s Manual.” Physics Reports 388, nos. 5–6 (2003): 279–360. DOI. Open PDF, arXiv v3.
• Zinn-Justin, Jean. Quantum Field Theory and Critical Phenomena. 5th ed. Oxford: Oxford University Press, 2021. DOI.