Anomalies via descent equations and the Atiyah-Singer index theorem

Intuition Beginner

A symmetry in physics is a rule that says some quantity never leaks away: it is conserved. Electric charge is the famous example. You can shuffle charges around, but the total never changes. Such conservation laws are the backbone of how physicists reason about the world.

An anomaly is a symmetry that survives in the classical theory but quietly breaks once you account for the quantum world. The conservation law that looked airtight develops a small leak, and the size of the leak is fixed, not adjustable. You cannot tune it away. It is set by the underlying geometry of the fields.

Think of a spinning top that ought to keep pointing the same way forever. In an idealized world it does. In the real world, tiny unavoidable effects make it slowly drift. The drift is not a mistake in your setup; it is a forced consequence of the situation. A quantum anomaly is that forced drift for a conservation law.

Visual Beginner

Picture a pipe carrying a current that is supposed to stay constant from one end to the other. A conservation law says: whatever flows in, flows out. An anomaly is a hidden side-tap on the pipe through which a precise, computable trickle escapes, even though you sealed every visible joint.

The remarkable part is that the leak rate is not random. It is dictated by the background field the current sits in. Change the field and the leak changes in a fixed proportion. This rigidity is why anomalies became a precision tool rather than a nuisance: the leak counts something geometric.

Worked example Beginner

Imagine counting two kinds of particles, left-handed and right-handed, that a classical rule says should always balance. Start with 5 of each, so the difference is 0. Now switch on a background field for a moment.

The quantum rule says the difference between left and right counts changes by a fixed amount tied to the field. Suppose the field is tuned so the change is exactly 2. After the field acts, you have 6 left-handed and 4 right-handed: the difference is now 6 − 4 = 2, matching the prediction.

You did nothing by hand; the field forced the imbalance. If you doubled the field strength in the right way, the change would be 4 instead of 2.

What this tells us: the anomaly converts a background field into a definite, whole-number mismatch between two populations. That whole number is the seed of the deep link to counting solutions of an equation, which is what the index theorem measures.

Check your understanding Beginner

Formal definition Intermediate+

Let $A$ be a connection on a principal $G$-bundle over an even-dimensional manifold, with curvature $F=dA+A^2$ valued in the Lie algebra $\mathfrak{g}$, and let $c$ denote the ghost field of the BRST construction 03.07.31, a Grassmann-odd $\mathfrak{g}$-valued generator of ghost number one. Write $s$ for the BRST differential, an odd nilpotent derivation acting by $sA=-dc-[A,c]$ and $sc=-c^2$, where $c^2=\frac{1}{2}[c,c]$ ^{[tong anomalies]}.

Form the gauge-invariant invariant polynomial of degree $n+1$ in the curvature, the relevant piece of the Chern character,

$$P_{2n+2}(F)=\mathrm{Str}\left(F^{n+1}\right),$$

a closed $(2n+2)$-form by the Chern-Weil homomorphism 03.06.06. Being closed and built from $F$, it is locally exact:

$$P_{2n+2}(F)=d{\omega }_{2n+1}(A,F),$$

and the $(2n+1)$-form ${\omega }_{2n+1}$ is the Chern-Simons form (the transgression of $P$). Because $P_{2n+2}$ is gauge-invariant, its descendant ${\omega }_{2n+1}$ is invariant only up to a total derivative, so its BRST variation is exact:

$$s{\omega }_{2n+1}=-d{\omega }_{2n}^1(c,A,F).$$

The $2n$-form ${\omega }_{2n}^1$, linear in the ghost $c$ and therefore of ghost number one, is the consistent anomaly in dimension $2n$.

A consistent anomaly is a local $2n$-form functional $\mathcal{A}=\int {\omega }_{2n}^1$ of ghost number one satisfying the Wess-Zumino consistency condition $s\mathcal{A}=0$, taken modulo BRST-exact terms $s(\text{local counterterm})$. A non-example worth flagging: a quantity that fails $s\mathcal{A}=0$ is not a candidate anomaly at all, because consistency is forced by the abelian nature of successive symmetry variations; and a quantity of the form $s\Lambda$ is removable by a counterterm, so it represents the zero class. The genuine anomalies are the nonzero classes of $H^1(s)$ at ghost number one and form degree $2n$.

The full descent tower reads $$ 0 = s,P_{2n+2}, \qquad s,\omega_{2n+1} + d,\omega_{2n}^{1} = 0, \qquad s,\omega_{2n}^{1} + d,\omega_{2n-1}^{2} = 0, ;\ldots, $$ with ${\omega }_{2n-1}^2$ of ghost number two encoding the Schwinger term in the current-algebra commutator ^{[tong anomalies]}.

Key theorem with proof Intermediate+

Theorem (Stora-Zumino descent). Let $P=\mathrm{Str}(F^{n+1})$ be the degree-$(2n+2)$ Chern-Weil polynomial of a connection $A$ with curvature $F$, and let $s$ be the BRST differential with $sA=-Dc$ and $sc=-c^2$, anticommuting with the exterior derivative $d$ in the sense $sd+ds=0$. Then there exist local forms ${\omega }_{2n+1}^0,{\omega }_{2n}^1,{\omega }_{2n-1}^2,\dots$ of bidegree (form, ghost) such that

$$P=d{\omega }_{2n+1}^0,s{\omega }_{2n+1-k}^k+d{\omega }_{2n-k}^{k+1}=0(k\ge 0),$$

and the ghost-number-one term ${\omega }_{2n}^1$ satisfies the Wess-Zumino condition $s\int {\omega }_{2n}^1=0$ up to a total derivative.

Proof. The polynomial $P=\mathrm{Str}(F^{n+1})$ is closed: $dP=(n+1)\mathrm{Str}((DF)F^n)=0$ by the Bianchi identity $DF=0$ and the invariance of the symmetrised trace. It is also BRST-invariant, $sP=0$, because $s$ acts on $A$ as an infinitesimal gauge transformation with the anticommuting parameter $c$, and $P$ is gauge-invariant by Chern-Weil 03.06.06.

Closedness on a region with vanishing de Rham cohomology in degree $2n+2$ yields a primitive ${\omega }_{2n+1}^0$ with $P=d{\omega }_{2n+1}^0$, the Chern-Simons form obtained by the standard transgression: interpolate $A_t=tA$, $F_t=d{A}_t+A_t^2$, and set $$ \omega_{2n+1}^{0} = (n+1)\int_0^1 ! dt ,\operatorname{Str}\big(A,F_t^{,n}\big). $$

Apply $s$ to $P=d{\omega }_{2n+1}^0$. Using $sP=0$ and the anticommutation $sd=-ds$, $$ 0 = s,P = s,d,\omega_{2n+1}^{0} = -,d,(s,\omega_{2n+1}^{0}). $$ Hence $s{\omega }_{2n+1}^0$ is closed, and on the same acyclic region it is exact: there is a $2n$-form ${\omega }_{2n}^1$, linear in $c$ and of ghost number one, with $$ s,\omega_{2n+1}^{0} = -,d,\omega_{2n}^{1}. $$

Iterate. Applying $s$ to this equation and using $s^2=0$ together with $sd=-ds$, $$ 0 = s^2,\omega_{2n+1}^{0} = -,s,d,\omega_{2n}^{1} = d,(s,\omega_{2n}^{1}), $$ so $s{\omega }_{2n}^1$ is closed and equals $-d{\omega }_{2n-1}^2$ for some ghost-number-two form. The recursion $s{\omega }_{2n+1-k}^k+d{\omega }_{2n-k}^{k+1}=0$ continues until the form degree reaches zero.

Finally, integrate the $k=1$ relation $s{\omega }_{2n}^1=-d{\omega }_{2n-1}^2$ over the closed $2n$-manifold. The right side is a total derivative, so $\int s{\omega }_{2n}^1=-\int d{\omega }_{2n-1}^2=0$. Therefore $s\int {\omega }_{2n}^1=0$, which is the Wess-Zumino consistency condition. $\square$

Bridge. The descent tower builds toward the cohomological classification of anomalies, in which ${\omega }_{2n}^1$ is a representative of a ghost-number-one BRST class and the obstruction to gauge invariance is its nonvanishing in $H^1(s)$; it appears again in the index theorem, where the top polynomial $P_{2n+2}$ is the index density of a Dirac operator on a manifold two dimensions higher, so the anomaly inflow from $2n+2$ to $2n$ dimensions is the index theorem read along the descent; this construction generalises the abelian Adler-Bell-Jackiw anomaly to arbitrary compact gauge groups and even dimensions, and the index-density origin is dual to Fujikawa's statement that the path-integral measure transforms by the same density; the bridge is the single invariant polynomial $\mathrm{Str}(F^{n+1})$ feeding three faces at once — the index, the consistent anomaly, and the Chern-Simons term 03.06.06.

Exercises Intermediate+

Lean formalization Intermediate+

lean_status: none — the Chern-Weil transgression apparatus and the BRST bicomplex required to state the descent equations intrinsically are not present in Mathlib.

-- Pseudocode only: the anomaly descent tower is not yet available.
axiom Form : ℤ → ℤ → Type*          -- forms graded by (form degree, ghost number)
axiom d  {p q : ℤ} : Form p q → Form (p+1) q       -- exterior derivative
axiom s  {p q : ℤ} : Form p q → Form p (q+1)       -- BRST differential

axiom d_sq  {p q : ℤ} (ω : Form p q) : d (d ω) = 0
axiom s_sq  {p q : ℤ} (ω : Form p q) : s (s ω) = 0
axiom sd_anticommute {p q : ℤ} (ω : Form p q) :
    s (d ω) = - d (s ω)

axiom chernWeil : Form (2*n+2) 0                    -- Str(F^{n+1}), closed, s-closed
axiom chernSimons : Form (2*n+1) 0                  -- ω with d ω = chernWeil
axiom anomaly : Form (2*n) 1                        -- ω with s chernSimons = - d ω

-- Wess-Zumino consistency: s (∫ anomaly) = 0
axiom wessZumino : s anomaly = - d (0 : Form (2*n-1) 2) → True

The missing formalization work includes the bigraded complex of forms carrying both a de Rham degree and a ghost-number grading, the two anticommuting differentials $d$ and $s$, the Chern-Weil invariant polynomial and its transgression, the solution of the descent tower, and the identification of the top polynomial with the Atiyah-Singer index density of a twisted Dirac operator.

Advanced results Master

The chiral (axial) current $j_5^{\mu }=\bar{\psi }{\gamma }^{\mu }{\gamma }^5\psi$ of a Dirac fermion coupled to an abelian gauge field is conserved at the classical level. At one loop it acquires the Adler-Bell-Jackiw anomaly $$ \partial_\mu j^\mu_5 = \frac{1}{16\pi^2},\varepsilon^{\mu\nu\rho\sigma}F_{\mu\nu}F_{\rho\sigma} = \frac{1}{4\pi^2},F\wedge F . $$ The right-hand side is precisely the four-form $P_4=\frac{1}{8{\pi }^2}\mathrm{tr}(F\wedge F)$, the second Chern form, evaluated as a density. The integrated anomaly $\int {\partial }_{\mu }{j}_5^{\mu }=2\nu$ equals twice the instanton number $\nu =\frac{1}{8{\pi }^2}\int \mathrm{tr}(F\wedge F)$, an integer ^{[tong anomalies]}.

This integer is the index of the Dirac operator. By the Atiyah-Singer index theorem [03.07.05 background; Atiyah-Singer 1968], for a twisted Dirac operator $D_A$ on a compact even-dimensional spin manifold, $$ \operatorname{ind}(D_A) = n_+ - n_- = \int_M \hat{A}(R),\operatorname{ch}(F), $$ the difference between positive- and negative-chirality zero modes. On four-manifolds the relevant term is ${\int }_M{\mathrm{ch}}_2(F)=\frac{1}{8{\pi }^2}{\int }_M\mathrm{tr}(F\wedge F)$, so the integrated axial anomaly counts the net chirality of fermion zero modes. The anomaly is the local index density: the pointwise version of $\hat{A}(R)\mathrm{ch}(F)$ restricted to the degree matching the spacetime dimension.

Fujikawa's derivation makes this identification manifest at the level of the path integral. Under an infinitesimal chiral rotation $\psi \mapsto e^{i\alpha {\gamma }^5}\psi$, the fermionic measure $\mathcal{D}\psi \mathcal{D}\bar{\psi }$ is not invariant: it picks up a Jacobian $\exp (-2i\int \alpha \mathcal{J}(x))$, where $\mathcal{J}(x)=\lim \mathrm{tr}{\gamma }^5{e}^{-D_A^2/M^2}$ is the regularised trace of ${\gamma }^5$ over the Dirac spectrum. The heat-kernel expansion of this trace produces exactly $\frac{1}{32{\pi }^2}{\epsilon }^{\mu \nu \rho \sigma }\mathrm{tr}(F_{\mu \nu }{F}_{\rho \sigma })$, the local index density. The anomalous divergence is the Jacobian, and the Jacobian is the index density ^{[tong anomalies]}.

The descent reading and the index reading are two faces of one object. The consistent anomaly ${\omega }_{2n}^1$, the BRST descendant of the Chern-Simons form ${\omega }_{2n+1}^0$ whose differential is the invariant polynomial $P_{2n+2}$, lives in dimension $2n$; the index density $P_{2n+2}$ lives in dimension $2n+2$. The phenomenon of anomaly inflow connects them: a gauge anomaly localised on a $2n$-dimensional boundary or defect is cancelled by the variation of a Chern-Simons term in the $(2n+1)$-dimensional bulk, whose differential is the $(2n+2)$-dimensional index density. The descent equations are the differential-form bookkeeping of this inflow, and the Atiyah-Singer theorem supplies the topological content of the top rung.

The distinction between the consistent anomaly ${\omega }_{2n}^1$, which satisfies the Wess-Zumino condition $s\mathcal{A}=0$ by construction, and the covariant anomaly, which transforms covariantly but fails consistency, is resolved by a local counterterm (the Bardeen-Zumino polynomial). They differ by a BRST-exact piece, so they represent the same physics measured in two bases. For gravitational anomalies the same machinery applies with the curvature two-form $R$ of the tangent bundle replacing $F$ and the polynomial built from $\hat{A}(R)$; these occur only in dimensions $4k+2$ and were classified geometrically by Alvarez-Gaumé and Witten and by Alvarez-Gaumé and Ginsparg ^{[Alvarez-Gaumé-Ginsparg 1985]}.

Synthesis. A single gauge-invariant invariant polynomial $P_{2n+2}=\mathrm{Str}(F^{n+1})$ governs the entire subject: it is closed and BRST-closed, its transgression is the Chern-Simons form ${\omega }_{2n+1}^0$ with $d{\omega }_{2n+1}^0=P_{2n+2}$, and the BRST variation $s{\omega }_{2n+1}^0=-d{\omega }_{2n}^1$ produces the consistent anomaly ${\omega }_{2n}^1$ of ghost number one. This descent tower builds toward the cohomological definition of an anomaly as a nonzero class in $H^1(s)$ obeying the Wess-Zumino consistency condition $s\int {\omega }_{2n}^1=0$ 03.07.31; it appears again in the Atiyah-Singer index theorem, where $P_{2n+2}$ is the index density $\hat{A}(R)\mathrm{ch}(F)$ of a twisted Dirac operator on a manifold two dimensions higher, so the integrated axial anomaly equals the net chirality $\mathrm{ind}(D_A)=n_{+}-n_{-}$. The Fujikawa Jacobian is dual to this statement, computing the same density as the regularised trace $\mathrm{tr}{\gamma }^5{e}^{-D_A^2/M^2}$ of the fermion measure. Anomaly inflow identifies the $2n$-dimensional boundary anomaly with the variation of a bulk Chern-Simons term whose differential is the index density. Putting these together, the abelian ABJ result ${\partial }_{\mu }{j}_5^{\mu }=\frac{1}{4{\pi }^2}F\wedge F$ generalises to every compact group and even dimension as one rung of a ladder whose top is a characteristic number 03.06.06, and the central insight is that the obstruction to gauge invariance, the count of Dirac zero modes, and the second Chern form are three readings of the one invariant polynomial.

Full proof set Master

Proposition (abelian descent and the two-dimensional anomaly). Let $A$ be an abelian connection, $F=dA$, and $P_4=F^2$. Then ${\omega }_3^0=AdA$, ${\omega }_2^1=cF$, and the Wess-Zumino condition $s{\int }_{{\Sigma }_2}{\omega }_2^1=0$ holds on a closed surface ${\Sigma }_2$.

Proof. Closedness $d{P}_4=2dFF=0$ holds since $dF=ddA=0$. The primitive is ${\omega }_3^0=AdA$ because $d(AdA)=dAdA-Ad(dA)=(dA{)}^2=F^2=P_4$. Apply the abelian BRST variation $sA=-dc$, $sc=0$: $$ s,\omega_3^0 = s(A,dA) = (sA),dA - A,d(sA) = (-dc),dA - A,d(-dc) = -dc,dA, $$ where $d(-dc)=0$. Now $-dcdA=-d(cdA)$, so $s{\omega }_3^0=-d{\omega }_2^1$ with ${\omega }_2^1=cdA=cF$. Integrating $s{\omega }_2^1$ over a closed ${\Sigma }_2$: $s{\omega }_2^1=s(cF)=(sc)F-c(sF)=0$ since $sc=0$ and $sF=s(dA)=-d(sA)=d(dc)=0$. Hence $s{\int }_{{\Sigma }_2}{\omega }_2^1=0$ outright, a fortiori up to a total derivative. $\square$

Proposition (index-anomaly equality in four dimensions). For a twisted Dirac operator $D_A$ on a closed spin four-manifold $M$, the integrated abelian axial anomaly equals twice the second Chern number, and this is $2\mathrm{ind}(D_A)$ in the relevant normalisation.

Proof. The anomalous divergence is ${\partial }_{\mu }{j}_5^{\mu }=\frac{1}{16{\pi }^2}{\epsilon }^{\mu \nu \rho \sigma }{F}_{\mu \nu }{F}_{\rho \sigma }$. Integrating over $M$, $$ \int_M \partial_\mu j^\mu_5;d^4x = \frac{1}{16\pi^2}\int_M \varepsilon^{\mu\nu\rho\sigma}F_{\mu\nu}F_{\rho\sigma},d^4x = \frac{1}{4\pi^2}\int_M \operatorname{tr}(F\wedge F) = 2,c_2[F], $$ using ${\mathrm{ch}}_2=\frac{1}{8{\pi }^2}\mathrm{tr}(F\wedge F)$ for the second Chern character. The Atiyah-Singer theorem gives $\mathrm{ind}(D_A)={\int }_M\hat{A}(R)\mathrm{ch}(F)$; on a spin four-manifold with $\hat{A}=1-\frac{1}{24}{p}_1+\cdots$ the curvature-quadratic term picks out ${\int }_M{\mathrm{ch}}_2(F)=c_2[F]$ for the gauge part. Hence the integrated axial anomaly equals $2c_2[F]$, twice the Dirac index in this normalisation, and is an even integer. $\square$

Proposition (consistency is automatic from descent). Any anomaly ${\omega }_{2n}^1$ arising as the ghost-one descendant of a transgressed invariant polynomial satisfies the Wess-Zumino consistency condition.

Proof. By the descent theorem, $s{\omega }_{2n}^1+d{\omega }_{2n-1}^2=0$. Integrate over a closed $2n$-manifold $M$: ${\int }_{M}s{\omega }_{2n}^1=-{\int }_{M}d{\omega }_{2n-1}^2=0$ by Stokes, since $M$ has no boundary. Therefore $s{\int }_M{\omega }_{2n}^1=0$, the Wess-Zumino condition. Consistency is a structural consequence of the anomaly descending from a closed invariant polynomial, not an extra assumption. $\square$

Connections Master

• BRST cohomology and Faddeev-Popov-ghost quantisation 03.07.31 — the consistent anomaly is a ghost-number-one BRST cohomology class, and the Wess-Zumino consistency condition $s\mathcal{A}=0$ is exactly BRST-closedness; the ghost $c$ of the Faddeev-Popov construction is the variable in which the anomaly is linear, so anomaly theory is the degree-one chapter of the cohomology whose degree-zero chapter is the space of physical observables.

• Chern-Weil homomorphism 03.06.06 — the entire descent tower is generated by a single Chern-Weil invariant polynomial $\mathrm{Str}(F^{n+1})$, whose closedness and gauge-invariance are the input to transgression; the anomaly is the obstruction to gauge-invariance of the transgressed Chern-Simons form, so anomalies are a physical manifestation of the Chern-Weil construction one degree down.

• Yang-Mills action 03.07.05 — anomalies are obstructions to consistently quantising a Yang-Mills theory coupled to chiral fermions; a gauge theory whose anomaly class is nonzero is inconsistent, which is why the symmetric structure constant $d_{abc}=\frac{1}{2}\mathrm{tr}(T_a\{T_b,T_c\})$ must vanish in any sensible chiral gauge theory, constraining the allowed fermion content.

• Electromagnetism as a U(1) Yang-Mills theory 03.07.29 — the original Adler-Bell-Jackiw anomaly is the abelian case, where ${\partial }_{\mu }{j}_5^{\mu }\propto F\wedge F$ with $F=dA$ the Maxwell field strength, giving the neutral-pion decay rate as the simplest physical consequence of the descent ladder.

Historical & philosophical context Master

The anomaly was discovered in 1969 by Adler and independently by Bell and Jackiw, who found that the axial-vector current of spinor electrodynamics is not conserved at one loop and that the triangle diagram produces a finite, regularisation-independent term proportional to ${\epsilon }^{\mu \nu \rho \sigma }{F}_{\mu \nu }{F}_{\rho \sigma }$ ^{[Adler 1969] [Bell-Jackiw 1969]}. Wess and Zumino formulated the consistency condition on anomalous Ward identities in 1971, identifying the algebraic constraint that any candidate anomaly must satisfy ^{[Wess-Zumino 1971]}. The topological origin became explicit through the descent equations of Stora and Zumino in the 1983 Cargèse and Les Houches lectures, which recast the consistency condition as BRST-closedness and traced the anomaly to the transgression of a Chern-Weil polynomial ^{[Zumino 1983] [Stora 1983]}. Fujikawa supplied the path-integral derivation in 1979, showing the anomaly is the Jacobian of the fermionic measure under a chiral rotation, computed as a regularised spectral trace ^{[Fujikawa 1979]}. The bridge to the Atiyah-Singer index theorem of 1968, in which the integrated anomaly counts the net chirality of Dirac zero modes through $\mathrm{ind}(D_A)=\int \hat{A}(R)\mathrm{ch}(F)$, was developed and synthesised by Alvarez-Gaumé and Ginsparg, who gave the unified geometric account of gauge and gravitational anomalies in even dimensions ^{[Atiyah-Singer 1968] [Alvarez-Gaumé-Ginsparg 1985]}.

Bibliography Master

@article{Adler1969,
  author  = {Adler, Stephen L.},
  title   = {Axial-Vector Vertex in Spinor Electrodynamics},
  journal = {Physical Review},
  volume  = {177},
  pages   = {2426--2438},
  year    = {1969}
}

@article{BellJackiw1969,
  author  = {Bell, J. S. and Jackiw, R.},
  title   = {A PCAC Puzzle: $\pi^0 \to \gamma\gamma$ in the $\sigma$-Model},
  journal = {Nuovo Cimento A},
  volume  = {60},
  pages   = {47--61},
  year    = {1969}
}

@article{WessZumino1971,
  author  = {Wess, J. and Zumino, B.},
  title   = {Consequences of Anomalous Ward Identities},
  journal = {Physics Letters B},
  volume  = {37},
  pages   = {95--97},
  year    = {1971}
}

@article{Fujikawa1979,
  author  = {Fujikawa, Kazuo},
  title   = {Path-Integral Measure for Gauge-Invariant Fermion Theories},
  journal = {Physical Review Letters},
  volume  = {42},
  pages   = {1195--1198},
  year    = {1979}
}

@article{AtiyahSinger1968,
  author  = {Atiyah, M. F. and Singer, I. M.},
  title   = {The Index of Elliptic Operators: I},
  journal = {Annals of Mathematics},
  volume  = {87},
  pages   = {484--530},
  year    = {1968}
}

@incollection{Zumino1983,
  author    = {Zumino, Bruno},
  title     = {Chiral Anomalies and Differential Geometry},
  booktitle = {Relativity, Groups and Topology II (Les Houches 1983)},
  publisher = {North-Holland},
  year      = {1984}
}

@incollection{Stora1984,
  author    = {Stora, Raymond},
  title     = {Algebraic Structure and Topological Origin of Anomalies},
  booktitle = {Progress in Gauge Field Theory (Carg\`ese 1983)},
  publisher = {Plenum Press},
  year      = {1984}
}

@article{AlvarezGaumeGinsparg1985,
  author  = {Alvarez-Gaum\'e, Luis and Ginsparg, Paul},
  title   = {The Structure of Gauge and Gravitational Anomalies},
  journal = {Annals of Physics},
  volume  = {161},
  pages   = {423--490},
  year    = {1985}
}

@article{AlvarezGaumeWitten1984,
  author  = {Alvarez-Gaum\'e, Luis and Witten, Edward},
  title   = {Gravitational Anomalies},
  journal = {Nuclear Physics B},
  volume  = {234},
  pages   = {269--330},
  year    = {1984}
}

@book{Sternberg2012,
  author    = {Sternberg, Shlomo},
  title     = {Curvature in Mathematics and Physics},
  publisher = {Dover Publications},
  year      = {2012}
}

@book{BertlmannAnomalies1996,
  author    = {Bertlmann, Reinhold A.},
  title     = {Anomalies in Quantum Field Theory},
  publisher = {Oxford University Press},
  year      = {1996}
}