# Review 9 — positivity notions and the missing implication **Verdict: PASS for the distinctions in the draft; OPEN for the required all-height inequality.** No independent arithmetic estimate follows from the Green operator or a legitimate autocorrelation/Bochner argument. Reviewed baseline: `d93aaf6b8dbe1fcbebf6fff4f6d1f71d993f4208` (PR75), as recorded in the shared brief. I read the current AGENTS review/proposal instructions, this audit's brief and complete derivation, PR74 reviews 01 and 05 on associated kernels and the NS101 control, and PR75's relevant positivity scope. The coordinator owns the full register/dependency review; this is an independent paper review of the positivity implications, not a replay of historical computations or a new scan. **Wall check: Same open gap.** Closest: NS100/101 and PR74/75. What changes: the primitive has legitimate ordinary transforms, but no new implication from its positivity to the complete first-Laguerre sign has been obtained. The original theta arithmetic and distinguished Jacobi IVP remain outside the generic structural argument. ## 1. Separate the positivity assertions Use the draft's convention `Y(r)=integral f(u)*exp(i*r*u) du` and `a=1/2`. For a function `k`, translation positive definiteness means ```text sum_(i,j) c_i*conjugate(c_j)*k(x_i-x_j) >= 0 for every finite list of real x_i and complex c_i. ``` It is not the assertion `k(x)>=0` at each point. | Object or assertion | What is established | What does not follow | | --- | --- | --- | | `G_a(u)=exp(-a*abs(u))/(2*a)` | Pointwise positive; its convolution operator preserves nonnegative functions; it is also a positive self-adjoint operator on `L2`. | Applying it to a pointwise positive input does not make the output's Fourier transform pointwise nonnegative. | | `phi>0`, `f=4*G_a*phi` | `f>0`, with the decay and envelope in (3)-(4). | Neither `f` nor `phi` is thereby proved translation-positive-definite. | | `Y=Fourier(f)` | `Y` is translation-positive-definite as a function of frequency. | `Y(r)>=0`, `L1[Y](r)>=0`, or the corrected inequality (10). | | `A0(t)` | Pointwise positive and translation-positive-definite in `t`. | Positivity of the different weighted kernel `A2`, or of the corrected differential combination, in the translation-positive-definite sense. | | `A2(t)` | Pointwise positive, smooth, integrable with all needed derivatives. | Translation positive definiteness of `A2`. | | Translation positive definiteness of `A2` | Equivalent here to `L1[Y](r)>=0` for all real `r`. | By itself, the corrected inequality for all `abs(r)<1/2`. | | `K=(partial_t^2-1)^2*A2-4*(partial_t^2+1)*A0=256*C` | Exact smooth, rapidly decaying complete expression. It is pointwise positive because the original `phi` is positive. | Translation positive definiteness of `K` or `C`; this is the missing first-Laguerre target. | The `L2` operator positivity of `G_a` is independently transparent from ```text = (1/(2*pi))*integral |Fourier(g)(r)|^2/(a^2+r^2) dr >= 0. ``` This quadratic-form statement must not be confused with pointwise sign after applying the differential inverse `a^2-partial_u^2`. For the frequency-positive-definite assertion about `Y`, no abstract theorem needs to be assumed: ```text sum_(i,j) c_i*conjugate(c_j)*Y(r_i-r_j) = integral f(u)*abs(sum_i c_i*exp(i*r_i*u))^2 du >= 0. ``` The same fact holds for `X` because `phi>0`. Thus requiring both `X` and `Y` to be positive definite in frequency is still a shared property of the NS101 control, not the missing sign estimate. ## 2. The legitimate autocorrelation and its precise limit The substitution `x=s+t` gives ```text A0(t) = integral f(x)*f(x-2*t) dx. ``` Consequently `A0` really is a Gram kernel: its matrices are represented by the translates `f(x-2*t_i)`. Its Fourier transform is ```text Fourier(A0)(omega) = Y(omega/2)^2/2 >= 0. ``` These statements validate the first identity of (8), including the frequency scaling. They do not require theta arithmetic. For `A2`, the extra weight prevents the same fixed-family Gram proof. In fact ```text A2((x_i-x_j)/2) = integral [y-(x_i+x_j)/2]^2*f(y-x_i)*f(y-x_j) dy. ``` The displayed nonnegative weight depends on both matrix indices. It cannot be pulled out as one common nonnegative integration measure. With `f_i(y)=f(y-x_i)` and complex coefficients, define ```text D0=sum_i c_i*f_i, D1=sum_i c_i*x_i*f_i, D2=sum_i c_i*x_i^2*f_i, E=sum_i c_i*(y-x_i)*f_i. ``` Direct expansion gives the same mixed-form obstruction recorded in PR74: ```text sum_(i,j) c_i*conjugate(c_j)*A2((x_i-x_j)/2) = ||E||_2^2 + (1/2)*Re - (1/2)*||D1||_2^2. ``` The last two terms are not an established nonnegative contribution. Replacing this expression by `||E||_2^2` would change the kernel. Equation (8) correctly identifies its Fourier sign: ```text Fourier(A2)(2*r)=L1[Y](r)/4. ``` Because `A2` is pointwise positive, its Fourier transform is itself a positive-definite function of frequency. That statement permits negative pointwise values. It is not the converse assertion that `A2` is positive definite in its translation variable. Fourier inversion and the stated integrability make the latter equivalent to the required pointwise nonnegativity of `L1[Y]`. As a bare logical illustration, `q(r)=3+2*cos(beta*r)` is positive definite (the Fourier transform of three positive point masses) and pointwise positive, but `L1[q](pi/beta)=-2*beta^2<0`. This is only an elementary explanation of the logical distinction; the fully matched smooth primitive control is the existing NS101 family in draft (11)-(12). ## 3. The corrected target has an additional low-frequency obligation Write `p=r^2+1/4`. Equations (7)-(9) give ```text Fourier(K)(2*r)=64*L1[X](r), 16*L1[X](r)=p^2*L1[Y](r)+2*(r^2-1/4)*Y(r)^2. ``` Thus translation positive definiteness of `K` is precisely the all-real first-Laguerre target. It is not a new estimate furnished by the inverse operator. There are two distinct unproved steps if one tries to proceed through `A2` alone: 1. Establish `L1[Y]>=0`, or equivalently translation positive definiteness of `A2`, using something beyond the shared positive-kernel structure. 2. For `abs(r)<1/2`, establish the stronger margin ```text L1[Y](r) >= 2*(1/4-r^2)/(r^2+1/4)^2*Y(r)^2. ``` For `abs(r)>=1/2`, step 1 would suffice; below that threshold it does not. At `abs(r)=1/2` the correction vanishes. At a zero of `Y`, the formula is regular and `L1[Y]=Y'^2>=0`. These special points cannot replace the all-real quantifier. The operator reading reaches the identical restriction: the first term in `Fourier(K)(omega)` has multiplier `(omega^2+1)^2` on `Fourier(A2)`, whereas the second has multiplier `4*(omega^2-1)` on the already nonnegative `Fourier(A0)`. Its negative low-frequency part cannot be dropped. A convolution identity alone supplies neither required margin. ## 4. Matched control and admission decision Draft (11) preserves the positive Green representation, pointwise positivity, the unit exponential tail, the envelope, frequency positive definiteness of `X_beta` and `Y_beta`, pointwise positivity of both `A0_beta` and `A2_beta`, and translation positive definiteness of `A0_beta`. The complete corrected identities remain valid. The recorded PR74 sufficient parameter range nevertheless gives a strict negative value of `L1[X_beta]`. That is the precise reason no implication from this collection of structural properties to the corrected target can be valid. It does not assert that the original `A2` fails positive definiteness, nor that the original corrected target fails. The control changes the single-lattice coefficients and loses the distinguished Jacobi IVP. **Missing edge:** a bound for the complete corrected expression, or an equivalent matrix inequality retaining the mixed terms, at every real height using an original arithmetic/IVP hypothesis absent from the control. No such estimate or method is supplied by these positivity arguments. No new candidate, scan, research row or manuscript version is warranted from this audit. **Final wall check: Same open gap.** The generic structural shortcut is stopped by the existing NS101 matched control; the original arithmetic theta route remains unexcluded.