---
title: Point-Splitting Regularization in QFT
url: https://www.emergentmind.com/topics/point-splitting-regularization
type: topic
---

# Point-Splitting Regularization in QFT

to=arxiv_search.search  天天彩票中奖json
{"query":"point-splitting regularization curved spacetime pragmatic mode-sum regularization translational splitting cosmological background", "max_results": 10}
to=arxiv_search.search 򐂕json
{"query":"point-splitting regularization Schwinger model gauge invariance Hamiltonian circle", "max_results": 10}
to=arxiv_search.search ություններըjson
{"query":"point-splitting regularization de Sitter spin-1/2 adiabatic", "max_results": 10}
Point-splitting regularization is a covariant renormalization procedure for local composite operators, especially expectation values of quadratic observables such as $\langle \phi^2(x)\rangle$ and $\langle T_{\mu\nu}(x)\rangle$, in quantum field theory on curved spacetime. Its defining operation is the replacement of a coincident product by a bilocal quantity evaluated at nearby points $x$ and $x'$, subtraction of a locally constructed singular parametrix—usually in Hadamard or DeWitt–Schwinger form—and only then passage to the coincidence limit $x'\to x$. In curved spacetime this provides a local and covariant ultraviolet subtraction; in gauge theory it also underlies gauge-invariant definitions of currents and Hamiltonians through point-split bilinears with parallel transport factors. In practical computations the method often appears in symmetry-adapted mode-sum form, where the singular short-distance structure is translated into explicit subtractions in frequency or angular-momentum space [2103.17218][1503.02810][1206.0878].

## 1. Covariant definition and local singular structure

For a scalar field $\phi(x)$, the basic bilocal object is the symmetrized two-point function
\[
G^{(1)}(x,x')=\langle\{\phi(x),\phi(x')\}\rangle.
\]
Point-splitting defines the renormalized field variance by
\[
\langle \phi^2(x)\rangle_{\text{ren}}
= \lim_{x'\to x}\Big(G^{(1)}(x,x')-G_{\text{DS}}(x,x')\Big),
\]
and the renormalized stress tensor by
\[
\langle T_{\mu\nu}(x)\rangle_{\text{ren}}
= \lim_{x'\to x} D_{\mu\nu}(x,x')\Big(G^{(1)}(x,x')-G_{\text{DS}}(x,x')\Big),
\]
where $D_{\mu\nu}(x,x')$ is the relevant bi-differential operator. The subtraction term $G_{\text{DS}}$ is purely local, state-independent, and determined by the geometry, the mass, and the curvature coupling. In four dimensions, the DeWitt–Schwinger expansion exhibits the universal short-distance structure through the $1/\sigma$ and $\ln\sigma$ terms, with $\sigma(x,x')$ Synge’s world function:
\[
G_{\text{DS}}(x,x')
=
\frac{1}{8\pi^2}
\left[
\frac{1}{\sigma}
+
\left(m^2+(\xi-\tfrac16)R\right)
\Big(1-\gamma-\tfrac12\ln\big(\tfrac{m^2|\sigma|}{2}\big)\Big)
+\tfrac{1}{12}R_{\alpha\beta}\sigma^\alpha\sigma^\beta+\cdots
\right].
\]
This subtraction reproduces the known counterterm structure associated with renormalizations of $\Lambda$, $G$, and higher-curvature couplings, and it applies to any Hadamard state because the ultraviolet singularity is universal rather than state-specific [2103.17218].

## 2. Symmetry-adapted mode-sum realizations

The abstract point-splitting prescription becomes numerically effective when the background admits an isometry. In the pragmatic mode-sum regularization program, the separation is taken along a Killing direction, the two-point function is expanded in the corresponding mode basis, and the DeWitt–Schwinger singularity is decomposed into the same spectral variables. In a stationary background, with time splitting $x=(t,\mathbf{x})$ and $x'=(t+\epsilon,\mathbf{x})$, one obtains
\[
G^{(1)}(x,x')
=
\int_0^\infty d\omega\, \cos(\omega\epsilon)\, |v_\omega(\mathbf{x})|^2,
\]
while the singular pieces of $G_{\text{DS}}$ are represented by generalized Fourier integrals such as
\[
\int_0^\infty d\omega\, \omega \cos(\omega\epsilon) = -\frac{1}{\epsilon^2}+O(\epsilon).
\]
This leads to a mode-by-mode subtraction formula in frequency space rather than a direct numerical coincidence limit. In Schwarzschild, for a massless scalar in the Boulware state, the $t$-splitting construction yields a regularized integrand after subtraction of a linear term in $\omega$, and the remaining oscillatory behavior is handled by generalized integrals, notably Abel summation or self-cancellation, because the oscillations are associated with null geodesics connecting the point to itself rather than with local ultraviolet singularities [1503.02810].

Angular splitting is the analogous construction for spherically symmetric spacetimes. There one separates points in an angular direction and rewrites the DeWitt–Schwinger counterterm in a Legendre expansion. For $\langle\phi^2\rangle$, the singular angular contribution is encoded in an analytic mode counterterm
\[
F_{\text{sing}}(l,z) = -8\pi a(z)\,h(l)+2\pi c(z)\,\Lambda(l),
\]
with $h(l)$ the harmonic number and $\Lambda(l)=-1/[l(l+1)]$ for $l>0$. The regularized sum still contains “blind spots,” particularly constants in $l$, which are removed by a self-cancelling partial-sum prescription. In Schwarzschild this angular-splitting scheme reproduces the Boulware, Unruh, and Hartle–Hawking results for $\langle\phi^2\rangle_{\text{ren}}$ and agrees with the $t$-splitting variant and earlier Euclidean/WKB calculations [1606.08451].

## 3. Translational splitting in cosmology and equivalence with adiabatic regularization

In spatially homogeneous, spatially flat FLRW spacetimes,
\[
ds^2 = dt^2-a^2(t)\,d\vec{x}^2,
\]
the natural point-splitting direction is a spatial translation. Choosing equal-time separated points
\[
x=(t,\vec{x}),\qquad x'=(t,\vec{x}+\vec{\epsilon}),
\]
the symmetrized two-point function becomes
\[
\{\phi(x),\phi(x')\}
=
\frac{1}{2\pi^2 a^3(t)}
\int_0^\infty dk\,k^2\,|h_k(t)|^2\,
\frac{\sin(k\epsilon)}{k\epsilon}.
\]
The DeWitt–Schwinger subtraction term can be expanded at small $\epsilon$ and rewritten in the same momentum representation, yielding
\[
G_{\text{DS}}(x,x')
=
\frac{1}{4\pi^2 a^3}
\int_0^\infty dk\,
\frac{\sin(k\epsilon)}{k\epsilon}\,k^2
\left(
\frac{1}{2\omega_k}
+\frac{m^2}{2\omega_k^3}
+\frac{(6\xi-1)R}{12\omega_k^3}
\right)
+\frac{R}{288\pi^2}+O(\epsilon^2),
\]
with $\omega_k(t)=\sqrt{k^2/a^2+m^2}$. After subtraction and coincidence, the renormalized variance becomes
\[
\langle \phi^2(t)\rangle_{\text{ren}}
=
\frac{1}{4\pi^2 a^3(t)}
\int_0^\infty dk\,k^2
\left(
|h_k(t)|^2
-\frac{1}{2\omega_k(t)}
-\frac{m^2}{2\omega_k^3(t)}
-\frac{(6\xi-1)R(t)}{12\omega_k^3(t)}
\right)
+\frac{R(t)}{288\pi^2}.
\]

The central result in this setting is exact equivalence: in spatially homogeneous, spatially flat FLRW spacetimes, the Levi–Ori pragmatic point-splitting/mode-sum scheme with translational splitting is exactly equivalent to adiabatic regularization, including the generalized version with a renormalization scale $\mu$. The equivalence holds at second adiabatic order for $\langle\phi^2\rangle$ and at fourth adiabatic order for $\langle T_{\mu\nu}\rangle$. This identifies adiabatic regularization as a symmetry-adapted implementation of point splitting along spatial Killing directions rather than an independent renormalization principle [2103.17218].

## 4. Field-content dependence and de Sitter implementations

The formalism is not tied to scalar fields, but the appropriate subtraction order can depend strongly on spin, coupling, and the quantity being renormalized. For a free massive spin-$\tfrac12$ field in de Sitter space, one coordinate-space point-splitting analysis constructs the bilinear
\[
\langle 0|\bar\psi(x)\psi(x')|0\rangle
\]
and defines the regularized coincidence limit by subtracting the second-order adiabatic two-point function,
\[
\langle 0|\bar\psi(x)\psi(x)|0\rangle_{\text{reg}}^{(2)}
=
\lim_{x'\to x}
\Big(
\langle 0|\bar\psi(x)\psi(x')|0\rangle
-
\langle 0|\bar\psi(x)\psi(x')|0\rangle_{\text{ad}}^{(2)}
\Big).
\]
For that model, second-order adiabatic subtraction is sufficient to remove all ultraviolet divergences for both the spectral stress tensor and the power spectrum, and the point-splitting result agrees with the second-order adiabatic one. The regularized stress tensor is maximally symmetric, the energy density remains negative, the massless limit is smooth, and there is no trace anomaly. By contrast, fourth-order subtraction is described there as an oversubtraction: it changes the sign of the vacuum energy density, produces a singular regularized auto-correlation in the massless limit, and yields a nonzero massless-limit stress tensor [2509.23388].

For a coupled scalar field in de Sitter space, the situation is more delicate. In one implementation guided by the adiabatically regularized Green’s function, minimal coupling $\xi=0$ admits a second-order point-splitting subtraction yielding a finite vacuum stress tensor of the form $\langle T_{\mu\nu}\rangle=g_{\mu\nu}\Lambda$ with positive constant energy density for the massive field, while the massless minimally coupled case gives a vanishing regularized stress tensor. For conformal coupling $\xi=\tfrac16$, the same study reports that zeroth-order subtraction is adequate for the massive field and again gives a vanishing stress tensor in the massless case, with no conformal trace anomaly. For general $\xi\neq0$, however, the coupling $\xi R$ generates logarithmic and path-dependent terms in the coincidence limit; extra treatments are then required to remove unwanted higher-order pieces and enforce de Sitter invariance. In that analysis, fourth-order regularization again gives a negative regularized energy density and a singular zero-mass limit, and the resulting trace anomaly is interpreted as an artifact of oversubtraction rather than as an unavoidable consequence of the geometry [2205.04761].

## 5. Gauge invariance, currents, and Hamiltonian formulations

In gauge theories, point splitting is not merely a subtraction device for stress tensors. It also defines gauge-invariant composite operators by separating field arguments and inserting a parallel transporter between them. In the Schwinger model on a circle, the gauge-invariant point-split axial current is defined by
\[
j_{1,\text{reg}}(x)
=
\lim_{\varepsilon\to 0}
\int
\psi^\dagger(y)\,e^{i a (y-x)}\,\gamma^5\,\psi(x)\,
\chi_\varepsilon(x-y)\,dy,
\]
where the factor $e^{i a (y-x)}$ is the Wilson-line phase in the chosen gauge. The zero mode of the regularized axial charge becomes
\[
Q_{5,\text{reg}}
=
\sum_{n\ge 0}(b_n^\dagger b_n-c_n^\dagger c_n)
-
\sum_{n<0}(b_n^\dagger b_n-c_n^\dagger c_n)
-
\frac{aL}{\pi},
\]
and the regularized Dirac Hamiltonian contains the corresponding anomalous $a^2$ term,
\[
H_{D,\text{reg}}
=
\sum_{n\in\mathbb Z}|k_n|(b_n^\dagger b_n+c_n^\dagger c_n)
-
\frac{a^2L}{2\pi}
-
a\,Q_{5,\text{reg}}.
\]
The finite terms produced by point splitting are precisely what make the Hamiltonian gauge invariant under large gauge transformations while encoding the axial anomaly [1206.0878].

A different role appears in the analysis of a massive Dirac field in $(1+1)$ dimensions with an inverse square well potential. There the local charge density operator, when defined by point splitting,
\[
\hat\rho(z,\varepsilon)
=
\frac{1}{2}\sum_{a=1,2}
\Bigl[
\hat\psi^\dagger_a\Bigl(z+\frac{\varepsilon}{2}\Bigr),
\hat\psi_a\Bigl(z-\frac{\varepsilon}{2}\Bigr)
\Bigr],
\]
acquires an additional finite contribution. Integrated over the well, this changes the total charge from $Q_\eta$ to
\[
Q'_\eta = Q_\eta+\frac{\eta a}{\pi}.
\]
For the parameter sets considered there, the added term reverses the sign of the total vacuum charge in the well and consequently reverses the sign of the Casimir energy, restoring consistency with the assumption that the vacuum is the minimum-energy state [1209.0508].

In Hamiltonian Yang–Mills theory in $2+1$ dimensions, point splitting enters at the level of the regulated kinetic operator itself. The regulator is implemented through Gaussian-smeared Green functions with gauge-covariant $H(r)H^{-1}(r')$ bridges, so that the Hamiltonian remains finite at nonzero $\varepsilon$ and gauge invariant. A key conclusion of that analysis is that one cannot discard the positive powers of the regulator parameter prematurely: keeping only a finite subset of the induced $\varepsilon^n$ structures changes the coefficient of the leading quadratic invariant in the vacuum wave functional. This suggests that, in Hamiltonian formulations, point splitting acts as a fully fledged nonlocal regulator rather than as a negligible short-distance bookkeeping device [1605.01907].

## 6. Ambiguities, limitations, and conceptual status

Point-splitting regularization addresses ultraviolet singularities, but it does not by itself solve every infrared or state-selection problem. In cosmological applications the subtraction terms depend only on local geometry, mass, and curvature coupling and are state independent, whereas infrared divergences—such as those associated with massless minimally coupled fields—depend on the state and background and require separate treatment. The introduction of a renormalization scale $\mu$ in the DeWitt–Schwinger logarithm is one standard way to control the massless limit and, in the FLRW translational-splitting analysis, its inclusion preserves exact equivalence with generalized adiabatic regularization [2103.17218].

The symmetry-adapted mode-sum versions introduce additional technical structures. In black-hole spacetimes, local DeWitt–Schwinger subtraction removes the ultraviolet divergence but leaves oscillatory mode-sum contributions associated with nonlocal singularities of the two-point function along null geodesics; generalized integrals and self-cancellation procedures are then required to produce finite numerical results [1503.02810]. In angular splitting, the residual “blind spots” in the Legendre decomposition show that even after analytic subtraction of the singular parametrix, constant-in-$l$ contributions may survive and must be eliminated by a self-cancelling partial sum rather than by a naive coincidence limit [1606.08451].

The method is also scheme sensitive at the level of finite parts. The de Sitter scalar analysis makes this especially explicit: a regularized Green function can be ultraviolet and infrared convergent while still yielding a path-dependent or non-maximally-symmetric stress tensor unless the subtraction is matched to the physically appropriate adiabatic order and supplemented by renormalization conditions such as de Sitter invariance. This suggests that point splitting is best understood not as a unique numerical rule, but as a covariant framework whose concrete realization is fixed by the local Hadamard singularity together with symmetry, conservation, and massless-limit requirements [2205.04761].

Within that framework, several symmetry-adapted renormalization prescriptions become conceptually unified. Time splitting in stationary spacetimes, angular splitting in spherically symmetric spacetimes, translational splitting in homogeneous cosmologies, and gauge-covariant Wilson-line point splitting in lower-dimensional gauge theories are not separate renormalization doctrines. They are specialized implementations of the same fundamental idea: isolate the universal short-distance singularity in a bilocal operator, subtract it in a symmetry-compatible representation, and only then recover a finite local observable.

Source: https://www.emergentmind.com/topics/point-splitting-regularization