Skip to content

With the Lorentzian source term +Jϕ+\int J\phi, a normalized generating functional packages vacuum time-ordered correlation functions into one object: each functional derivative inserts a field and contributes one factor of ii, so the nn-point function is (i)n(-i)^n times the nnth derivative at zero source. This conclusion is meaningful only after the state, ordering, boundary prescription, regulator, source sign, and normalization have been fixed.

This page gives both the operator and regulated path-integral definitions, derives the insertion rule, and evaluates the functional exactly for the free real scalar. The result is an in–out/Feynman object. Retarded response, closed-time-path expectation values, connected functions, Wick factorization, effective actions, and Schwinger–Dyson identities require additional constructions developed elsewhere.

Required background. Gaussian Fields and Sources supplies the regulated completion of the square and the distinction between an inverse kernel and a formal continuum expression.

Helpful background. Characteristic Functions, Moments, Cumulants, and Generating Functionals supplies the parallel finite-dimensional language of moments and cumulants.

Work in four-dimensional Minkowski spacetime and let JJ be an external c-number test source. Write

Jϕd4xJ(x)ϕ(x),SJ[ϕ]=S[ϕ]+Jϕ.J\cdot\phi \equiv \int \mathrm d^4x\,J(x)\phi(x), \qquad S_J[\phi]=S[\phi]+J\cdot\phi.

The test-source language matters because quantum fields and their correlation functions are distributions. Pointwise functional derivatives are shorthand for the corresponding smeared identities.

For a normalized vacuum Ω|\Omega\rangle, the operator definition is

Zop[J]=ΩTexp ⁣(iJϕ^)ΩΩΩ,Zop[0]=1.Z_{\mathrm{op}}[J] = \frac{ \langle\Omega| \mathrm T\exp\!\left(iJ\cdot\widehat\phi\right) |\Omega\rangle }{\langle\Omega|\Omega\rangle}, \qquad Z_{\mathrm{op}}[0]=1.

The symbol T\mathrm T is part of the definition. Differentiation will expose this ordering; it does not create it.

In a regulated functional description, let CF\mathcal C_F denote the integration cycle together with the vacuum/Feynman boundary prescription. Define

Z[J]=CFDϕexp ⁣(iS[ϕ]+iJϕ),\mathcal Z[J] = \int_{\mathcal C_F}\mathcal D\phi\, \exp\!\left(iS[\phi]+iJ\cdot\phi\right),

and, when Z[0]0\mathcal Z[0]\ne0,

Z[J]Z[J]Z[0],Z[0]=1.Z[J] \equiv \frac{\mathcal Z[J]}{\mathcal Z[0]}, \qquad Z[0]=1.

The continuum notation summarizes a regulated construction and its proposed limit; it is not a flat infinite-dimensional Lebesgue measure. Matching Z[J]Z[J] with the operator functional requires the same action, regulator, field normalization, state, ordering, boundary data, cycle, and zero-source normalization. Weinberg 1995, § 9.1, pp. 378–383 derives the fixed-endpoint phase-space integral by time slicing and shows how ordered insertions become a time-ordered product. The additional regulator, state, cycle, field-normalization, and zero-source matching conditions are imposed here.

Source differentiation inserts ordered fields

Section titled “Source differentiation inserts ordered fields”

Expand the ordered exponential as a formal power series, or as a convergent finite-regulator expression where justified:

TeiJϕ^=n=0inn!r=1n[d4xrJ(xr)]×T ⁣{ϕ^(x1)ϕ^(xn)}.\begin{aligned} \mathrm T e^{iJ\cdot\widehat\phi} &= \sum_{n=0}^{\infty}\frac{i^n}{n!} \int\prod_{r=1}^{n} \left[\mathrm d^4x_r\,J(x_r)\right] \\ &\qquad\times \mathrm T\!\left\{ \widehat\phi(x_1)\cdots\widehat\phi(x_n) \right\}. \end{aligned}

Using

δJ(y)δJ(x)=δ(4)(xy),\frac{\delta J(y)}{\delta J(x)} =\delta^{(4)}(x-y),

each derivative removes one source and leaves one ordered field insertion. Hence

δnZ[J]δJ(x1)δJ(xn)J=0=inΩT ⁣{ϕ^(x1)ϕ^(xn)}Ω.\begin{aligned} &\left. \frac{\delta^n Z[J]} {\delta J(x_1)\cdots\delta J(x_n)} \right|_{J=0} \\ &\qquad= i^n \langle\Omega| \mathrm T\!\left\{ \widehat\phi(x_1)\cdots\widehat\phi(x_n) \right\} |\Omega\rangle. \end{aligned}

Therefore the normalized vacuum nn-point function is

G(n)(x1,,xn)=ΩT ⁣{ϕ^(x1)ϕ^(xn)}Ω=(i)nδnZ[J]δJ(x1)δJ(xn)J=0.\boxed{ \begin{aligned} G^{(n)}(x_1,\ldots,x_n) &= \langle\Omega| \mathrm T\!\left\{ \widehat\phi(x_1)\cdots\widehat\phi(x_n) \right\} |\Omega\rangle \\ &= \left.(-i)^n \frac{\delta^nZ[J]} {\delta J(x_1)\cdots\delta J(x_n)} \right|_{J=0}. \end{aligned} }

This derivation, including the plus-source convention and division by the zero-source vacuum amplitude, is given in Schwartz 2014, § 14.3, pp. 261–262. Schwartz calls the unnormalized functional Z[J]Z[J] and retains an explicit factor Z[0]1Z[0]^{-1}; the normalized notation on this page absorbs that factor. A source convention with iJϕ-iJ\cdot\phi instead would reverse every odd-derivative sign; the formula cannot be imported without translating the source.

Away from zero source, normalization by the source-dependent functional is also required. Define

OJ=CFDϕO[ϕ]eiS[ϕ]+iJϕZ[J].\langle\mathcal O\rangle_J = \frac{ \displaystyle \int_{\mathcal C_F}\mathcal D\phi\, \mathcal O[\phi]e^{iS[\phi]+iJ\cdot\phi} }{\mathcal Z[J]}.

Then

1Z[J]δnZ[J]δJ(x1)δJ(xn)=inTϕ(x1)ϕ(xn)J.\frac{1}{Z[J]} \frac{\delta^nZ[J]} {\delta J(x_1)\cdots\delta J(x_n)} = i^n \left\langle \mathrm T\phi(x_1)\cdots\phi(x_n) \right\rangle_J.

Omitting Z[J]1Z[J]^{-1} would give an unnormalized insertion. At J=0J=0, the chosen normalization makes this factor one.

For two fields, the sign check is especially useful. With

DF(x,y)ΩTϕ(x)ϕ(y)Ω,D_F(x,y) \equiv \langle\Omega| \mathrm T\phi(x)\phi(y) |\Omega\rangle,

one has

δ2ZδJ(x)δJ(y)J=0=i2DF(x,y)=DF(x,y).\left. \frac{\delta^2Z} {\delta J(x)\delta J(y)} \right|_{J=0} =i^2D_F(x,y) =-D_F(x,y).

Multiplying by (i)2=1(-i)^2=-1 returns DFD_F. These two minus signs are an efficient round-trip test of the declared convention.

The exact regulated free-scalar functional

Section titled “The exact regulated free-scalar functional”

At a finite regulator, collect the retained real field variables into a vector ϕ\phi. After integration by parts, write the free action with its damping prescription as

S0,ϵ[ϕ]=12ϕTPϵϕ,Pϵ=PiϵB,P=PT,B=BT>0,S_{0,\epsilon}[\phi] =-\frac12\phi^{\mathsf T}P_\epsilon\phi, \qquad P_\epsilon=P-i\epsilon B, \qquad P=P^{\mathsf T}, \qquad B=B^{\mathsf T}>0,

and suppose PϵP_\epsilon is invertible on the declared regulated space. Then PϵP_\epsilon and its inverse are symmetric. Set

Gϵ=Pϵ1.G_\epsilon=P_\epsilon^{-1}.

The finite-dimensional completion of the square is exact:

12ϕTPϵϕ+JTϕ=12(ϕGϵJ)TPϵ(ϕGϵJ)+12JTGϵJ.\begin{aligned} -\frac12\phi^{\mathsf T}P_\epsilon\phi +J^{\mathsf T}\phi &= -\frac12 (\phi-G_\epsilon J)^{\mathsf T} P_\epsilon (\phi-G_\epsilon J) \\ &\qquad+ \frac12J^{\mathsf T}G_\epsilon J. \end{aligned}

Provided the contour may be translated by GϵJG_\epsilon J, the remaining Gaussian integral is the same source-independent factor in numerator and denominator. Thus

Z0,ϵ[J]=exp ⁣(i2JTGϵJ).Z_{0,\epsilon}[J] = \exp\!\left( \frac{i}{2}J^{\mathsf T}G_\epsilon J \right).

Take the Feynman boundary-value limit on the same regulated family and define

GF=limϵ0+Gϵ,GF=PF1=iDF.G_F =\lim_{\epsilon\to0^+}G_\epsilon, \qquad G_F=P_F^{-1}=iD_F.

With

JGFJd4xd4yJ(x)GF(x,y)J(y),J\cdot G_F\cdot J \equiv \int\mathrm d^4x\,\mathrm d^4y\, J(x)G_F(x,y)J(y),

the exact normalized free functional is

Z0[J]=exp ⁣(i2JGFJ)=exp ⁣(12JDFJ).\boxed{ Z_0[J] = \exp\!\left( \frac{i}{2}J\cdot G_F\cdot J \right) = \exp\!\left( -\frac12J\cdot D_F\cdot J \right). }

For the translation-invariant vacuum,

DF(xy)=d4p(2π)4ieip(xy)p2m2+i0.D_F(x-y) = \int\frac{\mathrm d^4p}{(2\pi)^4} \frac{i\,e^{-ip\cdot(x-y)}} {p^2-m^2+i0}.

With the site’s Fourier convention, PF=+m2P_F=\Box+m^2 acts as the Feynman boundary-value operator, so

PFGF=I,PFDF=iI.P_FG_F=I, \qquad P_FD_F=-iI.

The delta-normalized inverse is GF=iDFG_F=iD_F, not DFD_F itself. The free completion and Gaussian source functional are developed in Schwartz 2014, §§ 14.3.1–14.3.2, pp. 262–263; dividing by its zero-source value gives the normalized formula used here. The vacuum boundary deformation and resulting +i0+i0 Feynman denominator are derived in Schwartz 2014, § 14.4.1, pp. 265–266. Comparison with other scalar kernels belongs to Scalar Propagators, Ordered Correlators, and Sources.

Two further derivatives expose the difference between a source-deformed raw moment and its centered part. Define

ϕˉJ(x)i1Z0[J]δZ0[J]δJ(x)=d4yGF(x,y)J(y).\bar\phi_J(x) \equiv -i\,\frac{1}{Z_0[J]} \frac{\delta Z_0[J]}{\delta J(x)} = \int\mathrm d^4y\,G_F(x,y)J(y).

Then

PFϕˉJ=J,P_F\bar\phi_J=J,

with Feynman rather than retarded boundary data, and

Tϕ(x)ϕ(y)J=DF(x,y)+ϕˉJ(x)ϕˉJ(y).\left\langle \mathrm T\phi(x)\phi(y) \right\rangle_J = D_F(x,y) +\bar\phi_J(x)\bar\phi_J(y).

At J=0J=0, the centered free vacuum gives exactly DFD_F. The Gaussian functional also contains all higher free moments, but the general pairing theorem and fermionic signs belong to Wick’s Theorem and Free Gaussian Factorization.

Each line below detects a different category of error.

CheckCorrect resultWhat a failure means
Engineering dimensionIn four dimensions, [ϕ]=1[\phi]=1 and [J]=3[J]=3, so JϕJ\cdot\phi is dimensionlessThe source normalization or spacetime measure is inconsistent
Zero sourceZ[0]=1Z[0]=1The vacuum factor was not divided out, or the numerator and denominator are different problems
Two derivatives$\delta^2Z/\delta J(x)\delta J(y)_0=-D_F(x,y)$
Inverse kernelPFGF=IP_FG_F=I and PFDF=iIP_FD_F=-iIDFD_F and GF=iDFG_F=iD_F were conflated, or the boundary prescription changed
Nonzero sourceZ1δnZ=inTϕnJZ^{-1}\delta^nZ=i^n\langle\mathrm T\phi^n\rangle_JThe source-deformed insertion was left unnormalized
Euclidean cross-checkZE[J]=exp(12JCEJ)Z_E[J]=\exp(\tfrac12J\cdot C_E\cdot J) and ordinary derivatives generate momentsLorentzian factors were copied into a positive Euclidean Gaussian, or conversely

Zero-source normalization has a limited but exact job. It cancels a source-independent vacuum factor or Gaussian determinant only when the numerator and denominator use the same action, kernel, regulator, domain, measure, contour, and boundary data. It does not remove regulator dependence from DFD_F, cancel determinant ratios between different theories, repair a zero mode, construct an interacting continuum measure, or turn a complex Lorentzian functional into a probability distribution.

The adjective “generating” does not mean that one source functional produces every correlator by changing notation.

Desired objectDefining extra structureIs it generated by the Z[J]Z[J] above?
Vacuum time-ordered correlatorT\mathrm T ordering and Feynman/in–out boundary dataYes
Wightman functionA specified unsorted operator orderNo
Retarded responseA commutator multiplied by future-support step functionsNo
Euclidean Schwinger functionA Euclidean functional and its own source and continuation conventionsNo
In–in expectation valueA closed time path, initial state or density operator, and doubled branch sourcesNo

In particular, ϕˉJ=GFJ\bar\phi_J=G_FJ is generally an in–out mean field and may be complex. It is not the retarded solution produced by an initial-value source. The distinctions among Feynman, Wightman, retarded, and advanced kernels are developed on Scalar Propagators, Ordered Correlators, and Sources; the in–out/in–in distinction belongs to In–Out versus In–In Expectation Values.

The derivatives of ZZ give full ordered moments, including disconnected products. The logarithm W=ilogZW=-i\log Z and the moment–cumulant separation are the subject of Connected Correlators and Cumulants. This page does not use the free Gaussian formula as an interacting identity, derive Wick’s theorem, perform a Legendre transform, or derive a Schwinger–Dyson hierarchy.

Differentiation does not choose the ordering. The time-ordering symbol in the operator definition and the Feynman boundary construction in the path integral choose it. Differentiation only inserts fields into that already-defined object.

A normalized functional is not a probability distribution. In Lorentzian signature its weight is oscillatory and its values may be complex. The equation Z[0]=1Z[0]=1 is a vacuum-normalization statement, not positivity.

The Feynman inverse is not an arbitrary inverse. Retarded, advanced, and Feynman kernels can invert the same local differential expression while satisfying different boundary conditions. Replacing one by another changes the generated object.

Connectedness is not automatic. Derivatives of ZZ generate full moments. Taking a logarithm and deriving the connected partition relations is a separate step.

Use the stated criteria to locate any error in source normalization, differentiation factors, signs, or boundary data.

ModePromptA satisfactory responseRepair or continuation
RetrievalDefine normalized Z[J]Z[J] and name the data hidden by a bare functional-integral symbolGives the zero-source ratio and names the regulator, state, ordering, boundary prescription, cycle, source sign, and normalizationGaussian Fields and Sources
DerivationRecover the factor multiplying the nnth source derivativeExpands eiJϕe^{iJ\cdot\phi}, obtains ini^n insertions, and solves for the correlator with (i)n(-i)^nSource differentiation inserts ordered fields
Sign checkDifferentiate Z0=eJDFJ/2Z_0=e^{-J\cdot D_F\cdot J/2} twice at J=0J=0Obtains DF-D_F from differentiation and then DFD_F after multiplication by (i)2(-i)^2The exact free-scalar functional
Failure modeSet one eigenvalue of PϵP_\epsilon to zero before taking the inverseSays that neither GϵG_\epsilon nor the displayed Gaussian normalization exists on the full space and names a declared projection, mass deformation, or separate zero-mode treatmentGaussian Fields and Sources
TransferFor a symmetric invertible finite matrix GG, start from Z(J)=eiJTGJ/2Z(J)=e^{iJ^{\mathsf T}GJ/2}Finds iZ1Z/J=GJ-iZ^{-1}\partial Z/\partial J=GJ and the raw second moment iG+(GJ)(GJ)T-iG+(GJ)(GJ)^{\mathsf T}; with G=iDG=iD, this becomes D+(GJ)(GJ)TD+(GJ)(GJ)^{\mathsf T}The exact free-scalar functional
HandoffDecide where to compute a connected function or a causal expectation valueSends connected moments to the next leaf and causal expectation values to the in–in/closed-time-path chapter rather than relabeling ZZConnected Correlators and Cumulants and In–Out versus In–In Expectation Values
  • Schwartz, Matthew D. Quantum Field Theory and the Standard Model. Cambridge: Cambridge University Press, 2014. DOI.
  • Weinberg, Steven. The Quantum Theory of Fields, Volume I: Foundations. Cambridge: Cambridge University Press, 1995. First edition; 2005 paperback, 2012 printing consulted. DOI.