---
title: Getzler's Rescaling Argument
url: https://www.emergentmind.com/topics/getzler-s-rescaling-argument
type: topic
---

# Getzler's Rescaling Argument

Getzler’s rescaling argument is a local method for extracting the small-time contribution of a Dirac-type operator to the heat-kernel supertrace, and hence to the local index density. In the formulation relevant to the local equivariant index theorem, one works in normal coordinates near a fixed point, rescales both the base variables and the Clifford variables, and isolates a universal “model operator” whose heat kernel can be computed explicitly. In the conformal and equivariant setting treated in "Noncommutative geometry, conformal geometry, and the local equivariant index theorem" [1210.2032], this mechanism is combined with an equivariant version of Greiner’s approach to the heat kernel asymptotic; in later work it is recast in adiabatic groupoid, pseudodifferential, and singular-geometric frameworks [1607.07039] [2303.04013] [2508.01982].

## 1. Geometric setting and local normal form

The standard setup starts with an \(n\)-dimensional Riemannian manifold \((M,g)\), spin (or spin\(^c\)), with spinor bundle \(S \to M\) and Clifford action \(c(v)\). Let \(D\) be the Dirac operator acting on \(C^\infty(M,S)\). In local coordinates \(x=(x^1,\dots,x^n)\) near a fixed point \(x_0\), one chooses Riemannian normal coordinates so that at \(x_0\),
\[
g_{ij}(x_0)=\delta_{ij}, \qquad \partial_k g_{ij}(x_0)=0.
\]
In a parallel trivialization of \(S\) over these coordinates, \(D\) takes the form
\[
D = c(e_i)\bigl(\partial_{x^i}+A_i(x)\bigr),
\]
where \(A_i(x)\) is a matrix of \(1\)-forms vanishing to first order at \(x_0\), and \(\{e_i\}\) is the orthonormal frame at \(x_0\) [1210.2032].

A closely related formulation uses normal coordinates centered at a point \(y\in M\) together with a synchronous frame of the spinor bundle \(S\), so that at \(x=0\) all connection one-forms vanish and the Riemann curvature enters only at first order in \(x\). In that setting, one writes local operators as finite sums of monomials in \(x\), \(\partial_x\), and Clifford generators \(c_j=c(dx^j)\), preparing the operator for a graded analysis [2201.00379].

The same local normal-form principle persists in later adaptations. In the pseudodifferential formulation, \(D=\sum_{i=1}^n e^i\!\cdot\nabla_{e_i}\) acts on the \(\mathbb Z_2\)-graded complex spinor bundle \(\Sigma M=\Sigma^+M\oplus\Sigma^-M\), and \(D^2\) is treated as a geometric second-order operator whose coefficients in normal geodesic trivialisation are universal polynomials in the jets of the metric of a precise “Gilkey order” compatible with the total differential order [2303.04013].

## 2. Rescaling, grading, and the principal model

The defining step is the introduction of a small parameter \(t \to 0\) and the rescaled variables
\[
x'=\frac{x}{\sqrt t}, \qquad \psi'=\frac{\psi}{\sqrt t},
\]
where \(\psi\) denotes a local Clifford generator, or equivalently the pullback by the dilation map \(\delta_t:x\mapsto \sqrt t\,x\). In one standard grading convention, one sets
\[
\deg(x^i)=+1,\qquad \deg(\psi^i)=+1,\qquad \deg(\partial_{x^i})=-1.
\]
Any differential operator \(P\) is then expanded as a formal series
\[
P=\sum_{k=-m}^{\infty}P^{(k)},
\]
where \(P^{(k)}\) is homogeneous of Getzler-degree \(k\). The highest-degree piece \(P^{(-m)}\) is the “model operator” [1210.2032].

Another standard exposition introduces the Getzler order \(\ord_G\) on monomials by
\[
\ord_G(x^j)=-1,\qquad \ord_G(\partial_{x^j})=+1,\qquad \ord_G(c_j)=+1,
\]
extends it additively to products, and defines the filtration
\[
F^k=\{\,P:\ord_G(P)\le k\}.
\]
In this language, \(\ord_G(\nabla^S)\le 1\), \(\ord_G(D)\le 1\), and \(\ord_G(D^2)\le 2\) [2201.00379]. This suggests that the literature uses different sign conventions for the grading, while preserving the same mechanism: the rescaling isolates the leading homogeneous component.

A particularly explicit scaling map is
\[
\delta_\varepsilon(x^j)=\varepsilon x^j,\qquad
\delta_\varepsilon(\partial_{x^j})=\varepsilon^{-1}\partial_{x^j},\qquad
\delta_\varepsilon(\psi^j)=\varepsilon^{1/2}\psi^j.
\]
If
\[
P=\sum_{\alpha,I}a_{\alpha,I}(x)\,\partial_x^\alpha\,\psi^I,
\]
then
\[
\delta_\varepsilon P\delta_\varepsilon^{-1}
=
\sum_{\alpha,I}
\varepsilon^{-|\alpha|+\tfrac12|I|}
a_{\alpha,I}(\varepsilon x)\,\partial_x^\alpha\,\psi^I.
\]
If \(\ord_G(P)=k\), then
\[
\delta_\varepsilon P\delta_\varepsilon^{-1}
=
\varepsilon^{-k}\bigl(\sigma_G(P)(x,\partial_x,\psi)\bigr)
+O(\varepsilon^{-k+1}),
\]
where \(\sigma_G(P)\) is the principal Getzler symbol obtained by keeping only monomials of total weight \(k\) and freezing all smooth coefficients at \(x=0\) [2201.00379].

## 3. The rescaled square of the Dirac operator

The operator to be analyzed is usually the square of the Dirac operator, or the square plus curvature terms in the equivariant setting. One conjugates by the dilation and defines
\[
P_t:= t\,(\delta_t)^*D^2(\delta_t)^{-1}.
\]
In local normal coordinates one finds
\[
P_t=
-\,g^{ij}(\sqrt t\,x)\,\partial_{x^i}\partial_{x^j}
+t\,c(F)(\sqrt t\,x)
+\text{lower-order terms},
\]
where \(F\) is the curvature matrix of the spin connection. The metric coefficients expand as
\[
g^{ij}(\sqrt t\,x)=
\delta^{ij}
+\frac{t}{3}R^i{}_k{}^j{}_\ell(x_0)x^kx^\ell
+O(t^{3/2}),
\]
and similarly
\[
A_i(\sqrt t\,x)=t^{1/2}A_{i,(1)}(x)+O(t).
\]
After collecting powers of \(t\) and using the Getzler grading, the leading piece as \(t\to 0\) is the harmonic oscillator in \(\mathbb R^n\) with quadratic potential coming from the curvature [1210.2032].

In the concise form used in the same account, the highest-degree part is
\[
H:=
-\sum_i \partial_{x^i}^2
+\tfrac14\,R_{ij}(x_0)x^ix^j
+\sum_{i<j}c(e_i)c(e_j)\,R_{ij}(x_0).
\]
In the explicit graded expansion of a geometric Dirac operator \(D=C^j(\partial_{x^j}+A_j(x))\), one instead writes
\[
D_\varepsilon
=
\varepsilon\,\delta_\varepsilon D\delta_\varepsilon^{-1}
=
D_{(-1)}+\varepsilon D_{(0)}+\cdots,
\]
with
\[
D_{(-1)}=c^j\partial_{x^j},\qquad
D_{(0)}=\tfrac14\,x^kR_{kj\ell m}(0)c^\ell c^m.
\]
Squaring gives
\[
H_\varepsilon
=
\varepsilon^2\,\delta_\varepsilon D^2\delta_\varepsilon^{-1}
=
H_{(-2)}+\varepsilon H_{(-1)}+\cdots,
\]
where \(H_{(-2)}\) is the bosonic–fermionic harmonic oscillator with curvature-coupling [2201.00379].

The pseudodifferential formulation replaces the heat-operator viewpoint by a rescaled family
\[
\widetilde P_\lambda^{\rm Ge}
=
\lambda^m\,f_\lambda^\sharp\Bigl(U_\lambda^\sharp\bigl(c^g(P)\bigr)\Bigr),
\]
with \(f_\lambda:x\mapsto \lambda x\) and \(U_\lambda\) rescaling wedge-degree. For \(P=D^2\), the limit
\[
P_{\rm model}=\lim_{\lambda\to 0}\widetilde P_\lambda^{\rm Ge}
\]
exists for rescalable operators and yields a differential operator on \(\mathbb R^n\); in fact for \(D^2\) one recovers the harmonic-oscillator–Clifford model [2303.04013].

## 4. Heat-kernel asymptotics and the local index density

Let \(K_t(x,y)\) be the heat kernel of \(D^2\). In normal coordinates and after rescaling one shows
\[
K_t(x_0,x_0)\sim (4\pi t)^{-n/2}\sum_{j=0}^{\infty} t^j a_j(x_0).
\]
Under Getzler’s rescaling the only contribution to \(a_{n/2}(x_0)\) comes from the trace of the heat kernel of the model operator \(H\) at the origin, and one obtains
\[
a_{n/2}(x_0)=(2\pi i)^{-n/2}\Pf\!\bigl(R(x_0)\bigr).
\]
This is exactly the local index density [1210.2032].

In the explicit graded approach, the passage from the full operator to the model operator is implemented by Duhamel expansion or by Mehler’s formula:
\[
e^{-tH}
=
\delta_\varepsilon^{-1}\exp\bigl(-t\,\varepsilon^{-2}H_\varepsilon\bigr)\delta_\varepsilon
\sim
\delta_\varepsilon^{-1}\exp\!\bigl(-t\,H_{(-2)}\bigr)\delta_\varepsilon
+O(\varepsilon).
\]
At the diagonal, taking supertrace gives
\[
\Str\bigl(k_t(x,x)\bigr)
\sim
(4\pi t)^{-n/2}
\Bigl[\Str\bigl(e^{-t\,H_{(-2)}(0,0)}\bigr)\Bigr]
+O(t^{-(n/2)+1}),
\]
and the coefficient \(a_{n/2}(x)\) is the usual local index density [2201.00379].

For twisted Dirac operators, the leading term is described by the Lichnerowicz identity
\[
\mathfrak D_x^2=\nabla^*\nabla+\tfrac14\,\kappa_x+c(F_{W/S}),
\]
and the diagonal of the model heat kernel produces the “\(\widehat A\)-polynomial” in the curvature \(R_x\) together with the exponential of the twisting curvature:
\[
a_0(x)=(2\pi i)^{-n/2}\,\widehat A(R_x)\wedge e^{F_{W/S}(x)},
\]
viewed as an \(\End(W_x)\)-valued density [1607.07039].

The pseudodifferential account expresses the same local density through the \(\mathbb Z_2\)-graded Wodzicki residue. Scott’s theorem gives
\[
\Index(D^+)=-\tfrac12\,\sres\bigl(\log(D^2)\bigr),
\]
and the localisation formula identifies the local residue density with the logarithm of the model operator. A direct calculation of the \((-n)\)-homogeneous symbol of \(\log(\widetilde D^2_{\rm model})\) yields
\[
(2\pi i)^{-n/2}\,\widehat A\bigl(R(p)\bigr),
\]
showing that the symbol-calculus and heat-kernel formulations recover the same local index form [2303.04013].

## 5. Equivariant form and the Greiner parametrix

In the equivariant index problem, one replaces \(D^2\) by the infinitesimal equivariant Laplacian
\[
D^2+L_X,
\]
and applies Getzler’s rescaling to the total operator. The Greiner approach constructs a Volterra parametrix for the heat operator
\[
\partial_t + D^2 + L_X
\]
in which one sees, after Getzler rescaling, that only the leading homogeneous part contributes to the supertrace. The same harmonic-oscillator computation yields the equivariant Chern character form [1210.2032].

This is the mechanism used in the proof of the local equivariant index theorem cited in [1210.2032]: the computation of the CM cocycle of an equivariant Dirac spectral triple is obtained from a new proof of the local equivariant index theorem of Patodi, Donelly-Patodi and Gilkey, and that proof is obtained by combining Getzler’s rescaling with an equivariant version of Greiner’s approach to the heat kernel asymptotic. The same source states that it is believed that this approach should hold in various other geometric settings [1210.2032].

A closely related extension appears on Lie manifolds via the adiabatic groupoid. There the rescaling is built into a “rescaled bundle” \(\mathcal E\to G_{\mathrm{ad}}\), and if \(v\in A_x\), then under the identification \(\Exp(tv)\approx g\) near the unit, the Clifford multiplication rescales as
\[
c_t(v):=t^{1/2}c(v).
\]
The family of Dirac operators \(\mathfrak D_t:=t^{1/2}\mathfrak D\) on the rescaled bundle keeps track of the order of vanishing in \(t\), and the functional calculus \(\Psi_D\) produces heat kernels as smooth sections over the adiabatic groupoid [1607.07039].

## 6. Variants, modifications, and singular geometries

The core structure of the argument survives substantial reformulation. One variant emphasizes the grading technique itself. In that presentation, Getzler’s rescaling is described as a filtration argument on local differential operators, with the principal Getzler symbol extracted by conjugation with \(\delta_\varepsilon\) and passage to the limit \(\varepsilon\to 0\). This version is adapted to compute the leading terms of asymptotic expansions of traces of heat kernels in other situations [2201.00379].

A second variant is entirely pseudodifferential. Instead of analyzing \(e^{-tD^2}\) directly, one analyzes the symbol of \(\log(D^2)\), the graded Wodzicki residue, and the limit of rescaled operators acting on differential forms. The paper characterizes this as a localisation formula for the \(\mathbb Z_2\)-graded Wodzicki residue of the logarithm of a class of differential operators, and when applied to complex powers of the square of a Dirac operator, it expresses the index in terms of a local density involving the logarithm of the Getzler rescaled limit of its square [2303.04013].

A third extension treats manifolds with a Lie structure at infinity. There the proof of the local index theorem relies on a rescaling technique similar in spirit to Getzler’s rescaling, formulated on the adiabatic groupoid, combined with a renormalized supertrace defined on a suitable class of regularizing operators. The result is a renormalized local index formula
\[
\operatorname{index}D
=
\int_M (2\pi i)^{-n/2}\,\mathrm{Str}\bigl(\widehat A(R)\wedge e^{F_{W/S}}\bigr)
\]
for a Cl\((A)\)-Dirac operator on a spin Lie manifold [1607.07039].

The argument has also been generalized to incomplete cusp edge spaces. Near a boundary fibration \(Z\to \partial M\to Y\), the metric degenerates like
\[
g=dx^2+x^{2k}g_Z+\phi^*g_Y,\qquad k\ge 2.
\]
The heat space requires a second, quasi-homogeneous blow-up of order \(k-1\), producing coordinates
\[
x',\qquad
\mathcal T=\frac{\sqrt t}{(x')^k},\qquad
\mathcal S=\frac{x-x'}{(x')^k},\qquad
\eta=\frac{y-y'}{(x')^k},\qquad
z,z'.
\]
Under the horizontal Getzler rescaling, the principal part of \(\tau^2\slashed D^2\) at the new front face becomes
\[
\delta_{x'}\bigl(\tau^2\slashed D^2\bigr)\delta_{x'}^{-1}
=
-\tfrac12\,\mathcal T\partial_{\mathcal T}
+\mathcal T^2\bigl(-\partial_{\mathcal S}^2-\Delta_\eta+\slashed D_{Z,y}^2\bigr)
+(x')^{k-1}\mathcal T^2L_{k-1}+O\bigl((x')^k\bigr),
\]
and the leading normal solution is the product of a Euclidean heat kernel and the fibre heat kernel. The resulting supertrace reproduces the interior local index density and yields a boundary contribution given by the Bismut–Cheeger eta-form [2508.01982].

These formulations indicate that Getzler’s rescaling argument is not restricted to one presentation. What remains invariant is the strategy: introduce a filtration compatible with Clifford degree and differential order, conjugate by a one-parameter scaling, isolate a universal constant-coefficient or polynomial-coefficient model, and compute the surviving local term in the supertrace or residue. In the classical closed-manifold case this local term is the Pfaffian or \(\widehat A\)-density; in equivariant, groupoid, and singular settings it continues to encode the local index contribution, sometimes together with additional eta-form terms [1210.2032] [1607.07039] [2508.01982].

Source: https://www.emergentmind.com/topics/getzler-s-rescaling-argument