Reflection Positivity and Transfer-Matrix Criteria

A Euclidean lattice measure supports a positive-state transfer interpretation when it is reflection positive with respect to a declared time reflection and satisfies the additional regularity and translation assumptions needed for reconstruction. For the nearest-neighbor scalar, the one-time-step kernel factors into positive multiplication operators and a positive Gaussian convolution, producing a positive self-adjoint transfer operator. Spatial improvement is often compatible with this construction; extended couplings across the reflection plane in Euclidean time can spoil it, producing negative spectral weights or extra unphysical poles even when the small-momentum dispersion looks improved.

Required background. Scalar Lattice Actions and Difference Operators supplies the quadratic action and boundary-sensitive adjoints. Reflection Positivity within Osterwalder–Schrader Reconstruction defines the foundational positivity condition and its reconstruction role.

Helpful background. Self-Adjointness, Extensions, and Unitary Evolution explains why positivity and self-adjointness are operator-domain statements.

Time reflection and the positive-time algebra

Choose a distinguished Euclidean-time coordinate $t=na_t$. A link reflection about the plane between slices $0$ and $1$ acts on sites by

$$\vartheta(t,\mathbf x)=(a_t-t,\mathbf x).$$

For real scalar fields, define the antilinear action on a functional $F$ by

$$(\Theta F)[\phi]=F[\phi\circ\vartheta]^*.$$

Let $\mathcal A_+$ be the algebra of bounded polynomial or suitable cylindrical functionals supported strictly on the positive-time half-lattice. Reflection positivity is the inequality

$$\langle(\Theta F)F\rangle_E\ge0 \qquad\text{for every }F\in\mathcal A_+.$$

The support restriction matters: positivity of ordinary expectation values, positivity of the Boltzmann weight, and positivity of a covariance matrix are not substitutes for this reflected quadratic form. Site reflection and link reflection lead to related but not identical factorizations, especially when interactions or gauge links lie on the reflection plane.

Reflection and regulator conventions. We use a flat Euclidean lattice, real scalar fields, link reflection in the distinguished time direction, and the weight $e^{-S_E}$. The site-wide conventions apply. The reflection plane, allocation of variables on that plane, temporal boundary condition, and support algebra are local data. “Positive transfer matrix” means a positive self-adjoint operator on the reconstructed finite-lattice Hilbert space, not a proof of the full continuum Wightman theory.

Reflection positivity turns $\mathcal A_+$ into a pre-Hilbert space with sesquilinear form

$$(F,G)_{\mathrm{OS}} =\langle(\Theta F)G\rangle_E.$$

One quotients null vectors and completes. A one-step positive-time translation then induces a contraction that can be represented, under suitable assumptions, by a transfer operator $T$. If $T$ is positive and self-adjoint, one may write $T=e^{-a_tH}$ with $H\ge0$ after normalizing the vacuum energy. The full reconstruction requires more than this single inequality; the Euclidean axioms and reconstruction hypotheses are developed in Osterwalder and Schrader 1973, pp. 83–112 and their corrected continuation Osterwalder and Schrader 1975, pp. 281–305.

One-step kernel for the nearest-neighbor scalar

Keep space discretized and collect all scalar values on time slice $n$ into a vector $\varphi_n$. Write the action as

$$S_E=\sum_n\left[ \frac{a_s^{d-1}}{2a_t} \lVert\varphi_{n+1}-\varphi_n\rVert^2 +\frac{a_t}{2}\bigl(U(\varphi_{n+1})+U(\varphi_n)\bigr) \right],$$

where

$$U(\varphi)=a_s^{d-1}\sum_{\mathbf x}\left[ \frac12\sum_i(\nabla_i^+\varphi_{\mathbf x})^2 +\frac12m_0^2\varphi_{\mathbf x}^2 +V_{\mathrm{int}}(\varphi_{\mathbf x}) \right].$$

The symmetric split assigns half of the within-slice potential to each adjacent step. The transfer kernel is

$$T(\varphi',\varphi) =\mathcal N \exp\!\left[ -\frac{a_s^{d-1}}{2a_t}\lVert\varphi'-\varphi\rVert^2 -\frac{a_t}{2}U(\varphi') -\frac{a_t}{2}U(\varphi) \right].$$

Symmetry under $\varphi\leftrightarrow\varphi'$ makes the integral operator self-adjoint with the flat slice measure, assuming $U$ is real and the kernel defines a bounded or suitably controlled operator. Positivity is stronger than the pointwise statement $T(\varphi',\varphi)>0$. Factorize

$$T=M_{e^{-a_tU/2}}\,C\,M_{e^{-a_tU/2}},$$

where $M$ is multiplication and $C$ is Gaussian convolution. For any square-integrable wavefunctional $\Psi$,

$$\langle\Psi,C\Psi\rangle =\int\frac{\mathrm d^M\pi}{(2\pi)^M} \widetilde c(\pi)\, \lvert\widetilde\Psi(\pi)\rvert^2\ge0,$$

because the Fourier transform of the Gaussian, $\widetilde c(\pi)$, is nonnegative. Hence $C\ge0$ and $T\ge0$. This finite-dimensional factorization is the minimal analytic benchmark. It also shows which ingredients are essential: real action, nearest-neighbor temporal difference, symmetric potential allocation, and a positive slice measure.

For one spatial site with $U(\varphi)=m_0^2\varphi^2/2$, the kernel is Gaussian and diagonalization recovers a lattice harmonic oscillator. Its energy gap $E_a$ satisfies

$$4\sinh^2\!\left(\frac{a_tE_a}{2}\right)=a_t^2m_0^2,$$

the zero-spatial-momentum case of the cutoff dispersion derived on the propagator page. Agreement between kernel diagonalization and the propagator pole is an independent normalization check.

Positive spectral weights

If $T=e^{-a_tH}$ acts on a positive Hilbert space and $\mathcal O$ corresponds to an operator on a time slice, then for integer $n>0$

$$C(n a_t) =\langle0|\mathcal O,T^n\mathcal O^\dagger|0\rangle =\sum_k \lvert\langle k|\mathcal O^\dagger|0\rangle\rvert^2 e^{-n a_t(E_k-E_0)}.$$

Every coefficient is nonnegative. This implies useful diagnostics:

• $C(t)\ge0$ for a Hermitian operator with the stated quantum numbers;
• the effective energy $a_tE_{\mathrm{eff}}(n)=\log[C(n)/C(n+1)]$ approaches the lowest contributing gap from a mixture governed by positive weights; and
• Hankel matrices $C_{ij}=C(t_i+t_j)$ constructed from reflected supports are positive semidefinite.

These consequences depend on the operator and reflection assignment. Gauge-variant fields, fermionic insertions, derivative operators odd under reflection, and subtracted correlators require the corresponding conjugation and sign structure; testing scalar positivity formulas on them is invalid.

The relationship among Euclidean reflection, transfer evolution, the physical constraint space, and later Hamiltonian limits is summarized below. Inspect where positivity enters and where Gauss-law or local-Hilbert restrictions add separate conditions.

Reflection positivity supports a finite-regulator transfer interpretation, while constraint enforcement, Hamiltonian-domain control, observable matching, and continuum equivalence remain separate requirements. Schematic, not to scale.

How an apparently improved action can fail

Suppose a temporal kinetic term includes next-nearest slices to cancel an $O(a_t^2)$ dispersion error. Its Fourier kernel may have the form

$$K(p_0)=m^2+c_1\bigl(1-\cos(a_tp_0)\bigr) +c_2\bigl(1-\cos(2a_tp_0)\bigr).$$

Appropriate $c_1,c_2$ can improve the small-$p_0$ expansion. Yet the cross-reflection part now couples more than one pair of slices. The simple one-step Gaussian factorization no longer applies, and the propagator can acquire additional poles with negative residues or non-real transfer energies. A smaller leading Taylor coefficient is therefore insufficient evidence for a positive spectral theory.

One can expose the failure without attempting full reconstruction:

1. factor the action across the chosen reflection plane and test whether the cross term is a sum of $B_\alpha\Theta B_\alpha$ with nonnegative coefficients;
2. locate every pole of the free propagator in the complex energy plane;
3. decompose the time correlator into exponentials and inspect the signs and reality of residues; and
4. test reflected Gram matrices on a finite basis of positive-time functionals.

Passing a finite basis is necessary but not sufficient for the universal inequality. Failing one basis element is decisive.

Gauge actions have their own reflection assignments for oriented links and plaquettes. Wilson’s plaquette action admits a positive transfer construction under standard conditions Lüscher 1977, pp. 283–292 and Osterwalder and Seiler 1978, §§2–3, while arbitrary extended loops require renewed analysis. The explicit gauge Hamiltonian and Gauss-law projection belong to Transfer Matrices between Euclidean and Hamiltonian QFT.

What the criterion does not prove

Reflection positivity at fixed $a$ does not by itself prove:

• existence of an infinite-volume limit;
• existence or uniqueness of a continuum limit;
• restoration of Lorentz symmetry;
• locality and positivity of a proposed improved continuum observable;
• equivalence of Euclidean and Hamiltonian regulators after independent truncations; or
• the full hypotheses of a constructive reconstruction theorem.

Conversely, a finite-$a$ action that violates reflection positivity may still approach a unitary continuum theory if the violating states remain at the cutoff and decouple. That is a stronger continuum claim: it requires explicit scale separation and positive renormalized observables across several spacings. One should not call the finite regulator positive while making that argument.

Reflection positivity controls one logical step, whereas the target statement also depends on cutoff, volume, and reconstruction limits. The diagram below keeps those axes separate; inspect the crossed routes before treating a positive finite-spacing kernel as a completed continuum construction.

Each limit names the observable and held-fixed physical quantities. The crossed routes show two common noncommuting or misidentified sequences: a zero-mode-sensitive massless limit and $a\to0$ at fixed site count, which shrinks rather than enlarges the box. The diagram is schematic and not to scale.

Observable-level validation checklist

Before claiming a transfer interpretation, verify:

• the reflection plane, positive-time algebra, conjugation, and boundary conditions are explicit;
• the action and measure are invariant under the reflection;
• the cross-plane couplings admit a nonnegative factorization or an applicable theorem;
• the one-step kernel is self-adjoint and positive as an operator, not merely pointwise;
• free poles have real nonnegative energies and the expected residue signs;
• reflected Gram matrices pass analytic and numerical tests;
• temporal improvement and extended loops have been checked separately from spatial improvement;
• the claimed spectral weights use operators with the correct reflection parity; and
• continuum conclusions are stated only after volume, tuning, symmetry, and cutoff limits are controlled.

Exercises

1. Positive Gaussian convolution. Let $(C\Psi)(x)=\int\mathrm dy\,e^{-\alpha(x-y)^2}\Psi(y)$ with $\alpha>0$. Show that $C$ is a positive operator.

Solution

Fourier transformation gives $\widetilde{C\Psi}(k)=\sqrt{\pi/\alpha}\,e^{-k^2/(4\alpha)}\widetilde\Psi(k)$. Plancherel’s identity then yields

$$\langle\Psi,C\Psi\rangle =\int\frac{\mathrm dk}{2\pi} \sqrt{\frac\pi\alpha} e^{-k^2/(4\alpha)} \lvert\widetilde\Psi(k)\rvert^2\ge0.$$

Multiplication on both sides by a real positive function preserves positivity, establishing the factorized transfer-kernel result.

2. A reflected Gram test. If $C(n)=A e^{-nE_1}+B e^{-nE_2}$ with $E_2>E_1>0$, form the $2\times2$ matrix $M_{ij}=C(i+j)$ for $i,j=1,2$. Show that $A,B\ge0$ imply $M\ge0$ and explain how $B<0$ can be detected.

Solution

Write $M=A v_1v_1^{\mathsf T}+B v_2v_2^{\mathsf T}$ with $v_k=(e^{-E_k},e^{-2E_k})$. Nonnegative $A,B$ make it a sum of positive rank-one matrices. If $B<0$, the determinant is

$$\det M=AB\,e^{-2(E_1+E_2)} \bigl(e^{-E_1}-e^{-E_2}\bigr)^2<0,$$

so one eigenvalue is negative.

What you can now do

You should now be able to state the lattice reflection-positivity inequality, construct the nearest-neighbor scalar transfer kernel, and distinguish a positive finite-regulator interpretation from a full reconstruction or continuum claim. Continue with Bare Parameters, Tuning Conditions, and Continuum Targets to specify which physical theory a sequence of such regulated transfer systems is meant to approach.

References

• Lüscher, Martin. “Construction of a Selfadjoint, Strictly Positive Transfer Matrix for Euclidean Lattice Gauge Theories.” Communications in Mathematical Physics 54, no. 3 (1977): 283–292. doi:10.1007/BF01614090.
• Osterwalder, Konrad, and Robert Schrader. “Axioms for Euclidean Green’s Functions.” Communications in Mathematical Physics 31, no. 2 (1973): 83–112. doi:10.1007/BF01645738.
• Osterwalder, Konrad, and Robert Schrader. “Axioms for Euclidean Green’s Functions II.” Communications in Mathematical Physics 42, no. 3 (1975): 281–305. doi:10.1007/BF01608978.
• Osterwalder, Konrad, and Erhard Seiler. “Gauge Field Theories on a Lattice.” Annals of Physics 110, no. 2 (1978): 440–471. doi:10.1016/0003-4916(78)90039-8.