Skip to content

Observable normalizations

Before comparing two numbers, identify what was held fixed: the operator, its source, the gauge group, the integration measure and the background. A factor of two in a current coefficient can be a notation change. Replacing an ordinary trace by a normalized trace inside the action can change the coupling. This reference makes those distinctions explicit; the linked lessons supply the complete derivations.

For Euclidean sources collectively denoted by ss, use

Z[s]=∫Dϕ e−S[ϕ;s],W[s]=−log⁡Z[s],F[s]=Re⁡W[s]=−log⁡∣Z[s]∣.(RF10.1)Z[s]=\int\mathcal D\phi\,e^{-S[\phi;s]},\qquad \mathcal W[s]=-\log Z[s],\qquad F[s]=\operatorname{Re}\mathcal W[s]=-\log|Z[s]|. \tag{RF10.1}

The complex logarithm uses a continuous local branch where Z≠0Z\ne0. A normalized expectation value includes division by the same ZZ. A sphere partition function is not, merely by being called a partition function, a thermal trace or an integer count.

Five different counting and generating objects

Section titled “Five different counting and generating objects”

Here Γ=(−1)F\Gamma=(-1)^{\mathsf F} denotes fermion parity. It is distinct from a flavor charge and from the free energy FF.

ObjectDefinition and essential dataWhat it measures
Witten index in quantum mechanicsI(β)=Tr⁡HΓe−βHI(\beta)=\operatorname{Tr}_{\mathcal H}\Gamma e^{-\beta H}, with H={Q,Q†}/2H=\{Q,Q^\dagger\}/2, a specified Hilbert space and operator domainsA signed count of zero-energy states when the heat traces and pairing argument are valid
Three-dimensional superconformal indexA graded trace on the physical Hilbert space on S2S^2, using a chosen conformal supercharge and charges commuting with itA weighted signed count of its cohomology, including states of nonzero conformal dimension
Hilbert seriesHR(t)=∑n≥0(dim⁡Rn)tnH_{\mathcal R}(t)=\sum_{n\ge0}(\dim\mathcal R_n)t^n for a specified nonnegatively graded ring with finite-dimensional graded piecesDimensions of graded ring components; no fermion-parity sign is inserted in this definition
Sphere partition functionThe path integral on a specified supersymmetric sphere background, with a fixed contour, source prescription and countertermsA generating function for integrated responses and normalized insertions
Topologically twisted indexA supersymmetric trace with the spatial R-symmetry flux and circle spin/R-holonomy specified; flavor fluxes and fugacities are additional dataA signed count in the twisted background, which differs from the radial superconformal background

For the site’s Witten index, the upper component is bosonic. Its positive-frequency supersymmetric oscillator has I=+1I=+1. Tong assigns the opposite parity to the upper component, so that example has the opposite sign in his convention. Also, I=0I=0 does not imply that there are no supersymmetric states: one even and one odd zero-energy state cancel. Noncompact continuous spectra require additional care; subtracting two divergent traces is not the definition above. Tong, section 1.2.2

The three-dimensional N=2\mathcal N=2 radial index uses the dimensionless conformal generator DD, the physical superconformal charge RR and a sphere rotation J3J_3. Choose qq with charges (D,R,J3)=(1/2,1,−1/2)(D,R,J_3)=(1/2,1,-1/2). Then

Hq={q,q†}=D−R−J3≥0,I(x,za)=Tr⁡H(S2)Γe−βHqxD+J3∏azaFa.(RF10.2)H_q=\{q,q^\dagger\}=D-R-J_3\ge0,\qquad \mathcal I(x,z_a)=\operatorname{Tr}_{\mathcal H(S^2)} \Gamma e^{-\beta H_q}x^{D+J_3}\prod_a z_a^{F_a}. \tag{RF10.2}

The flavor charges FaF_a commute with qq and q†q^\dagger; DD alone does not. The trace includes the gauge projection and required dynamical magnetic sectors. Background flavor fluxes, if present, are fixed separately. The regulator has β>0\beta>0 here; the parameter in IDX3 (2.1) is βsource=−β\beta_{\rm source}=-\beta. Convergence of the fugacity expansion is part of the prescription.

By contrast, the simplest twisted S2×S1S^2\times S^1 trace uses R-flux −1-1, periodic circle spin structure, and the corresponding trivial circle R-holonomy. With no rotational fugacity its cohomology trace has the insertion Γ∏ayaFa\Gamma\prod_a y_a^{F_a}. Changing the spin/R-holonomy choice changes that insertion; changing the R-charge lattice changes which fields are globally defined. The same manifold name S2×S1S^2\times S^1 therefore does not identify this trace with (RF10.2). CK19 (6.1)–(6.5)

A Hilbert series is a third operation. For example, with deg⁡X=1\deg X=1, R=C[X]/(X2)\mathcal R=\mathbb C[X]/(X^2) has HR=1+tH_{\mathcal R}=1+t because 1,X1,X are independent and XX is nilpotent. Its reduced ring is C\mathbb C, with series 11, although both have the same single vacuum point. Neither series determines the Witten index without a Hilbert space, parity and supercharge. See chiral rings and vacuum points.

Current coefficients: an actual factor-of-two translation

Section titled “Current coefficients: an actual factor-of-two translation”

At separated points in a three-dimensional conformal theory, define Iμν(x)=δμν−2xμxν/∣x∣2I_{\mu\nu}(x)=\delta_{\mu\nu}-2x_\mu x_\nu/|x|^2 and write ⟨Jμ(x)Jν(0)⟩=CJIμν(x)/∣x∣4\langle J_\mu(x)J_\nu(0)\rangle=C_J I_{\mu\nu}(x)/|x|^4. Two sources attach a coefficient called τ\tau to different prefactors:

τP16π2(δμν∂2−∂μ∂ν)1∣x∣2=τCT32π2(δμν∂2−∂μ∂ν)1∣x∣2.(RF10.3)\frac{\tau_P}{16\pi^2} (\delta_{\mu\nu}\partial^2-\partial_\mu\partial_\nu)\frac1{|x|^2} = \frac{\tau_{CT}}{32\pi^2} (\delta_{\mu\nu}\partial^2-\partial_\mu\partial_\nu)\frac1{|x|^2}. \tag{RF10.3}

The left convention is P (8.3); the right is CT (2.2), with its separate contact term omitted here. For the same normalized current, equality requires τCT=2τP\tau_{CT}=2\tau_P. Direct differentiation gives

(δμν∂2−∂μ∂ν)∣x∣−2=4Iμν(x)∣x∣−4,CJ=τP4π2=τCT8π2.(RF10.4)(\delta_{\mu\nu}\partial^2-\partial_\mu\partial_\nu)|x|^{-2} =4I_{\mu\nu}(x)|x|^{-4},\qquad C_J=\frac{\tau_P}{4\pi^2}=\frac{\tau_{CT}}{8\pi^2}. \tag{RF10.4}

For the physical R-current of a free three-dimensional N=2\mathcal N=2 chiral multiplet, whose scalar has R-charge 1/21/2, τP=1/4\tau_P=1/4, so τCT=1/2\tau_{CT}=1/2 and CJ=1/(16π2)C_J=1/(16\pi^2). This is a translation of notation, not a rescaling of JJ.

The current-correlator lesson fixes the accompanying stress-tensor convention. In that convention CT=3τP/(2π2)C_T=3\tau_P/(2\pi^2) for the superconformal R-current. The specified squashing family gives

d2Fdb2∣b=1=π22τP.(RF10.5)\left.\frac{d^2F}{db^2}\right|_{b=1}=\frac{\pi^2}{2}\tau_P. \tag{RF10.5}

This uses the full P background, the physical R-current and the stated supersymmetric conformal subtraction. It is not a formula for an arbitrary metric-only deformation or a trial R-current. The free chiral check is F′′(1)=π2/8F''(1)=\pi^2/8. A change of the parameter bb also changes the numerical Hessian by the chain rule.

The normalized four-dimensional circular loop

Section titled “The normalized four-dimensional circular loop”

For the fundamental half-BPS circle in N=4\mathcal N=4 super Yang–Mills, set λ=g2N\lambda=g^2N and M=2πℓaM=2\pi\ell a, where ℓ\ell is the sphere radius and aa the constant Hermitian localization scalar. The finite-NN matrix expectation is

WU(λ,N)=∫dM e−2NTr⁡M2/λ N−1Tr⁡eM∫dM e−2NTr⁡M2/λ,WSU=e−λ/(8N2)WU.(RF10.6)W_U(\lambda,N)= \frac{\int dM\,e^{-2N\operatorname{Tr}M^2/\lambda} \,N^{-1}\operatorname{Tr}e^M} {\int dM\,e^{-2N\operatorname{Tr}M^2/\lambda}},\qquad W_{SU}=e^{-\lambda/(8N^2)}W_U. \tag{RF10.6}

Tr⁡\operatorname{Tr} is the ordinary fundamental trace. The first integral is over all Hermitian matrices; the SU(N)SU(N) answer instead integrates traceless matrices, at the same gg and λ\lambda. This includes the scalar coupling of the half-BPS loop, not just gauge holonomy. It is an exact finite-NN formula for this observable, not a planar approximation. The finite-NN lesson connects the original proposal to localization and derives the center factor. The trace distinction appears explicitly in DG (5.1).

For four-dimensional pure SU(2)SU(2) N=2\mathcal N=2 theory, this site uses the first paper’s electric and magnetic special coordinates a,aDa,a_D:

Zγ=nea+nmaD,Mγ≥2 ∣Zγ∣,τeff=daDda.(RF10.7)Z_\gamma=n_ea+n_ma_D,\qquad M_\gamma\ge\sqrt2\,|Z_\gamma|,\qquad \tau_{\rm eff}=\frac{da_D}{da}. \tag{RF10.7}

The W boson has ne=1n_e=1 and the smooth basic monopole has nm=1n_m=1. A fundamental Wilson probe has electric label 1/21/2; it is not dynamical matter in the pure theory. The coordinates have mass dimension one. The bound is saturated by an existing BPS state; it does not assert the existence or stability of every charge pair. These are exact two-derivative low-energy data, and in general aD≠τeffaa_D\ne\tau_{\rm eff}a. The charge and period dictionary explains the conversion to the algebraic central charge Z=2Zγ\mathcal Z=\sqrt2 Z_\gamma. SW1 (4.3)–(4.6)

Use compact U(N)k×U(N)−kU(N)_k\times U(N)_{-k} with no additional quotient, positive integer kk, and the massless round-sphere prescription of the small-rank calculation: H=−i/ℓH=-i/\ell, A=V=C=0A=V=C=0, with no flavor masses or Fayet–Iliopoulos sources. The Cartan measure includes 1/(N!)21/(N!)^2, and the dimensionless variables s,ts,t have phase eiπk(∑si2−∑ti2)e^{i\pi k(\sum s_i^2-\sum t_i^2)}. The exact normalization obeys

Z(0,k)=1,Z(1,k)=14k,F(N,k)=−log⁡Z(N,k).(RF10.8)Z(0,k)=1,\qquad Z(1,k)=\frac1{4k},\qquad F(N,k)=-\log Z(N,k). \tag{RF10.8}

In the exact Fermi-gas formulation, Z(1,k)=Tr⁡ρZ(1,k)=\operatorname{Tr}\rho; the positive operator ρ\rho is not normalized to unit trace. Its Fredholm determinant generates the canonical coefficients Z(N,k)Z(N,k). That auxiliary particle-number generating function is not a supersymmetric index. Large-NN phase-volume and Airy approximations retain this measure convention but are not substitutes for the exact finite-NN coefficients. FG (3.3), (3.7), (3.9), (3.24)

For any one source parameter ss, differentiating (RF10.1) gives

W′′(s)=⟨S′′(s)⟩−⟨S′(s)S′(s)⟩c.(RF10.9)\mathcal W''(s)=\langle S''(s)\rangle- \langle S'(s)S'(s)\rangle_c. \tag{RF10.9}

The explicit second derivative includes quadratic source couplings, often called seagulls. Thus an integrated response is not obtained by integrating only the separated two-point function. The source-response lesson also fixes the supersymmetric path through auxiliary sources when a mass is varied.

A source-only local term Ict[s]I_{\rm ct}[s] multiplies ZZ by e−Icte^{-I_{\rm ct}} and shifts W\mathcal W by IctI_{\rm ct}. It cancels between numerator and denominator of a fixed normalized insertion. A separate finite counterterm on the line can still redefine a Wilson-loop observable. Specify that defect prescription too.

For a compact, unit-charge flavor source on a spin three-manifold, an allowed integer Chern–Simons counterterm shifts the contact coefficient κ\kappa by an integer; this is not an arbitrary continuous change of the separated coefficient CJC_J. The charge lattice and spin assumption matter. CT (1.3)–(1.7)

In four-dimensional N=2\mathcal N=2 superconformal theories a different ambiguity occurs. For the renormalized supersymmetric sphere partition function, locally on the space of exactly marginal couplings τI\tau^I,

ZS4=eK/12,K⟼K+f(τ)+f‾(τˉ),GIJˉ=12∂I∂Jˉlog⁡ZS4.(RF10.10)Z_{S^4}=e^{K/12},\qquad K\longmapsto K+f(\tau)+\overline f(\bar\tau),\qquad G_{I\bar J}=12\partial_I\partial_{\bar J}\log Z_{S^4}. \tag{RF10.10}

Here KK is a Kähler potential on coupling space, not on the Seiberg–Witten vacuum space. The metric GG is unchanged by this local holomorphic ambiguity. Its operator normalization uses the source term π−2∫δτIOI+conjugate\pi^{-2}\int\delta\tau^I O_I+\text{conjugate} and ⟨OI(x)O‾Jˉ(0)⟩=GIJˉ/∣x∣8\langle O_I(x)\overline O_{\bar J}(0)\rangle=G_{I\bar J}/|x|^8. See GI (1.1)–(2.3) and universal observables.

1. Replace the trace in the Gaussian action

Section titled “1. Replace the trace in the Gaussian action”

Define tr⁡=Tr⁡/N\operatorname{tr}=\operatorname{Tr}/N. Someone replaces Tr⁡\operatorname{Tr} by tr⁡\operatorname{tr} in the action of (RF10.6), retaining 2N/λ2N/\lambda, and uses the correctly normalized insertion tr⁡eM\operatorname{tr}e^M. Has division by ZZ repaired the change? Find the coupling actually being evaluated.

Solution

The new action is (2/λ)Tr⁡M2(2/\lambda)\operatorname{Tr}M^2, whereas the intended action is (2N/λ)Tr⁡M2(2N/\lambda)\operatorname{Tr}M^2. It therefore evaluates WU(λeff,N)W_U(\lambda_{\rm eff},N) with λeff=Nλ\lambda_{\rm eff}=N\lambda. A normalized ratio cancels a constant measure factor, not a change in the distribution. To keep the same theory, use (2N2/λ)tr⁡M2(2N^2/\lambda)\operatorname{tr}M^2. As a check, the Gaussian contraction gives WU=1+λ/8+O(λ2)W_U=1+\lambda/8+O(\lambda^2); the mistaken action gives 1+Nλ/8+O(λ2)1+N\lambda/8+O(\lambda^2).

Replace the physical R-current by J′=2JRJ'=2J_R and choose its source B′B' so that B′J′=BJRB'J'=BJ_R. What happens to CJC_J and to the coefficient called τP\tau_P? Can the new coefficient be inserted unchanged into the stress-tensor relation?

Solution

B′=B/2B'=B/2, while ⟨J′J′⟩=4⟨JRJR⟩\langle J'J'\rangle=4\langle J_RJ_R\rangle. Hence CJ′=4CJC'_J=4C_J and, if the prefactor in (RF10.3) is retained, τP′=4τP\tau'_P=4\tau_P. This is different from the factor-two source-notation translation in (RF10.3). The physical stress tensor has not changed. The relation CT=3τP/(2π2)C_T=3\tau_P/(2\pi^2) assumes R-charge normalization in which the supercharges have unit magnitude; J′J' does not obey that normalization. For compact sources, the connection periods and charge lattice must also be translated, not silently held fixed while dividing BB by two.

3. Continue a mass away from the real axis

Section titled “3. Continue a mass away from the real axis”

On the round sphere with H=−i/ℓH=-i/\ell, A=V=C=0A=V=C=0, take a constant flavor scalar mm, auxiliary field DB=im/ℓD_B=im/\ell, vanishing flavor connection, and x=ℓmx=\ell m. An integer level-nn supersymmetric flavor counterterm gives Zn(x)=eiπnx2Z0(x)Z_n(x)=e^{i\pi n x^2}Z_0(x). Determine its effect on F=−log⁡∣Z∣F=-\log|Z| for real xx, and then for x=u+ivx=u+iv with u,vu,v real. Does equality of absolute values at real mass establish equality of complex partition functions?

Solution

The bosonic term in P (6.28) is −nDBm/(2π)-nD_Bm/(2\pi). Multiplication by the sphere volume 2π2ℓ32\pi^2\ell^3 gives Ict=−iπnx2I_{\rm ct}=-i\pi n x^2, confirming the phase in the question. For real xx its modulus is one, so Fn=F0F_n=F_0 although Wn=W0−iπnx2\mathcal W_n=\mathcal W_0-i\pi n x^2 locally. For complex xx, ∣eiπnx2∣=e−2πnuv|e^{i\pi n x^2}|=e^{-2\pi n uv} and Fn−F0=2πnuvF_n-F_0=2\pi n uv. Real-mass absolute values therefore miss the phase and do not determine the analytically continued absolute values. The statement compares two declared integer-counterterm prescriptions; it does not license arbitrary changes to the global flavor-source normalization.