# Independent audit of PR73: pair algebra and reflected-prefix boundary Reviewed scientific baseline: `5434d96f0034f842b0229f1d69dbbf3fb7fea270` (merged PR73). Working HEAD at completion of the reading pass: `971c8bd02c834cf31ab98ea2a50060289fbece40`. The intervening diff adds only the six-agent brief, review record and task entry; it changes no mathematical dependency. The coordinator refreshed the remote and reviewed the entire 104-node conclusion register. I read current agent/proposal instructions, README status, the applicable ZERO-GEOMETRY register and freeze, the missing evidence ledger, PR72/73 reports and the four supporting mathematical notes named in the brief. This is an independent paper audit of the specified claims, not a historical certificate replay or independent validation of the external source's computer-assisted curvature theorem. **Wall check: Same open gap.** Closest results: NS100/101, PR72/73. What changes: no arithmetic hypothesis. The exact validation question is whether PR73's mixed form, original pair clock, fixed-prefix asymptotic and full-tail correction have the asserted signs, constants and domains. The dependency is the complete original first-Laguerre target `L1[X](r)>=0` for every real `r`, equivalently positive definiteness of `C`; this is only a necessary first-order part of the original radial-sign program, not the full all-order or RH criterion. ## Verdict and findings No factor, sign or missing-regularity defect was found in those PR73 claims under their stated fixed-prefix and complete-theta hypotheses. In particular, `-8*delta_M^2*r^-6` and the net correction `+8*delta_M^2*r^-6` are correct. Two distinctions should remain explicit in subsequent use: 1. The positive **net correction** includes the cross terms between the prefix and tail. The tail's own first-Laguerre expression is eventually negative; it is not a positive summand. 2. Cancellation of every algebraic boundary term proves rapid frequency decay, not its sign. A direct smooth, positive, strictly log-concave, double-exponentially decaying control below has all odd boundary derivatives zero and nevertheless has a negative first-Laguerre value. It does not preserve the original arithmetic or Jacobi initial-value problem. This is a control on a possible overinterpretation, not a counterexample to PR73 or to the original theta target. ## 1. Fourier and mixed-form constants Take the Fourier convention `X(r)=integral phi(u)*exp(i*r*u) du`, with the complete real even Schwartz kernel `phi`. Symmetrizing the product defining `L1` gives ```text L1[X](r) = (1/2) integral integral (u-v)^2 phi(u) phi(v) exp(i*r*(u+v)) du dv. ``` Replace `v` by `-v`, then set `u=s+t, v=s-t`. The latter substitution has absolute Jacobian `2`, and `(u+v)^2/2=2*s^2`. Consequently ```text L1[X](r) = 4 integral C(t) exp(2*i*r*t) dt = 4 integral C(t) cos(2*r*t) dt, C(t) = integral s^2 phi(s+t) phi(s-t) ds. ``` The complete integrands and the derivatives/moments used here are absolutely integrable. `C` is a real even Schwartz function. Fourier inversion/positive definiteness therefore makes its nonnegative transform equivalent to all finite positive-semidefinite matrices, with no need to weaken the quantifier to sampled frequencies or finitely many shifts. The scaling by `2` covers the entire real frequency line. For the mixed form, fix any finite number of arbitrary real shifts and complex coefficients. Use the inner product linear in its first argument. Putting `s=y-(x_i+x_j)/2` in a matrix entry yields ```text C((x_i-x_j)/2) = integral [y-(x_i+x_j)/2]^2 phi(y-x_i) phi(y-x_j) dy. ``` Expand the square as `(y-x_i)*(y-x_j)+(x_i-x_j)^2/4`. The second contribution after summation is ```text (1/4) [ + - 2*] = (1/2) Re - (1/2)||D1||^2. ``` This independently reproduces PR73's exact formula with `||E||^2` for the first contribution. There is no hidden real-coefficient restriction; Hermitian symmetry is exactly what produces the real part. Coincident shifts cause no problem. The resulting all-coefficient inequality is equivalent to the required positive definiteness, not an established lower bound. Dropping either signed contribution is unjustified. ## 2. Nonlinear clock, derivatives and regularity One can check the clock without the external curvature theorem. Set ```text L=2u, H(L)=exp(L/4)*Theta(exp(L))=(A(L)/pi)^(1/4), H'/H=a/4, a'=(U+a^2-1)/2. ``` The complete theta differential identity gives ```text phi(L/2)=H''(L)-H(L)/16 =H(L)*[U+(3/2)*(a^2-1)]/8 =H(L)*d(L)/8. ``` Here `H>0` and `phi>0` on every finite real clock, so `d>0`. Differentiating its logarithm gives precisely ```text P = a/4 + d'/d, d' = 2*U*(a+chi)+(3*a/2)*(U+a^2-1). ``` Thus `phi'(u)=2*P(2u)*phi(u)` and `P1=V(P)` is an `L` derivative, not a `u` derivative. The given modular initial data give `P(0)=0` directly. With `Pplus=P(2*(s+t))` and `Pminus=P(2*(s-t))`, the independent chain rule is ```text (log K)_t = 2*(Pplus-Pminus), (log K)_tt = 4*(P1plus+P1minus), K_tt = 4*[(Pplus-Pminus)^2+P1plus+P1minus]*K. ``` The plus sign in front of `P1minus` is correct: the minus in `Pplus-Pminus` cancels the derivative of its negative clock. Equivalently, ```text K_tt = phi''(s+t)*phi(s-t) -2*phi'(s+t)*phi'(s-t) +phi(s+t)*phi''(s-t). ``` The last form supplies dominated differentiation without assuming bounds on the separately growing slope variables. For `t` in any compact real interval, complete theta derivative envelopes give an integrable common majorant after multiplication by `s^2`. Consequently the asserted `C''` formula is valid for every finite real `t`, locally uniformly in `t`. No artificial cut or boundary occurs at `s=t` or `s=-t`. The derivation needs the distinguished analytic theta trajectory. It must not replace it by arbitrary branches of the squared Jacobi equation or divide by `chi` at its zero. The supporting source audit already records this distinction. The source's negative real `P1`/curvature information, even if granted, does not fix the sign of the displayed combination or of its oscillatory transform. ## 3. Fixed reflected-prefix asymptotic Let `M>=1` be an integer held fixed, and let `f_M` be PR73's positive half-line sum. Write `a_n=pi*n^2`. Direct differentiation of one term at zero gives ```text b_n = exp(-a_n)*[-4*a_n^3+15*a_n^2-(15/2)*a_n]. ``` For `a>=4`, the bracket divided by `a` is negative: it equals `-23/2` at `4` and its derivative `-8*a+15` is negative thereafter. For every `n>=2`, `a_n>4`. Gaussian summability permits differentiating the full series near zero. Smooth evenness of the complete kernel gives ```text sum_(n>=1) b_n=0, delta_M=f_M'(0)=-sum_(n>M) b_n>0. ``` Thus the reflected prefix is continuous with an upward cusp. Smoothness of each half-line summand is sufficient for the following integrations by parts; smoothness across the reflected cusp is not assumed. For any smooth rapidly decreasing half-line function `h`, repeated integration by parts gives ```text integral h(u)*cos(r*u) du = -h'(0)/r^2 + h'''(0)/r^4 + O(r^-6), integral h(u)*sin(r*u) du = h(0)/r - h''(0)/r^3 + O(r^-5), ``` with the derivatives and decay needed for the displayed remainder. Every weighted derivative of `f_M` has that integrability. Apply these identities separately to `f_M`, `u*f_M` and `u^2*f_M`. In particular, `(u*f_M)''(0)=2*delta_M`, `(u^2*f_M)'(0)=0`, and `(u^2*f_M)'''(0)=6*delta_M`. The resulting three estimates are ```text X_M = -2*delta_M*r^-2 + O_M(r^-4), X_M' = 4*delta_M*r^-3 + O_M(r^-5), X_M'' = -12*delta_M*r^-4 + O_M(r^-6). ``` These are estimates of the differentiated transforms, not derivatives of an uncontrolled remainder. Multiplication gives the coefficients `16-24=-8`, hence ```text L1[X_M] = -8*delta_M^2*r^-6 + O_M(r^-8). ``` Because `delta_M>0`, for every fixed `M` there exists a finite `R_M` such that this is strictly negative for all `r>R_M`. No numerical onset or uniformity in `M` follows from this assertion. Negative heights follow by evenness. The all-height nonnegative-prefix certificate is therefore excluded for exactly this reflected construction. The more general odd-boundary statement also has the correct sign. If the first nonzero odd right derivative has order `2k+1`, put `nu=2k+2` and `c=2*(-1)^(k+1)*h^(2k+1)(0)`. Under the corresponding weighted differentiable remainders, `X=c*r^-nu+...` gives `L1[X]=-nu*c^2*r^(-2*nu-2)+...`. It cannot be applied to a repair whose odd derivatives all vanish, or without the differentiated remainder hypotheses. ## 4. What the full-tail correction actually says Put `T_M=X-X_M`, the transform of the complete omitted reflected tail. The original complete kernel is Schwartz, so `X`, `X'` and `X''` decay faster than every inverse power. For fixed `M` this implies ```text T_M = 2*delta_M*r^-2 + O_M(r^-4), T_M' = -4*delta_M*r^-3 + O_M(r^-5), T_M'' = 12*delta_M*r^-4 + O_M(r^-6). ``` Retaining the cross terms explicitly gives ```text L1[X] = L1[X_M] + L1[T_M] + B_M, B_M = 2*X_M'*T_M' - X_M*T_M'' - T_M*X_M'', L1[T_M] = -8*delta_M^2*r^-6 + O_M(r^-8), B_M = 16*delta_M^2*r^-6 + O_M(r^-8). ``` Therefore PR73's **net** formula is correct: ```text L1[X]-L1[X_M] = 8*delta_M^2*r^-6 + O_M(r^-8). ``` The omitted tail is positive in the real variable, but its own first-Laguerre expression is negative at sufficiently large height. Calling that term a positive remainder would be false. The positive cross term is essential, and the three leading coefficients sum to zero. Smooth evenness in fact cancels the entire algebraic boundary expansion, not only the first `r^-6` contribution. Equivalently, `L1[X]=O_N(r^-N)` for every fixed `N`. This is a valid stronger decay statement, already implicit in the Schwartz observation, but it gives no lower sign bound: it places the remaining expression below every fixed algebraic scale. The fixed-`M` error is not `o(r^-6)`. It also cannot be used without a new uniform estimate for `M=M(r)`. A changing integer prefix must further distinguish evaluations of the fixed-prefix derivative integrals from derivatives of the possibly discontinuous composite `X_{M(r)}(r)`. None of these observations excludes a correctly controlled growing prefix or a globally smooth completion. ## 5. Paper control for an overstrong boundary-cancellation inference This is a separate generic control, not NS101's reciprocal mixture. Its only purpose is to test the inference from perfect boundary cancellation, positivity and real log-concavity to first-Laguerre sign. No theta-specific hypothesis is used in that inference. Define ```text k(u) = [3*exp(-u^2/3)+exp(-(u-1)^2/3)+exp(-(u+1)^2/3)]/sqrt(3*pi), Y(r) = integral k(u)*exp(i*r*u) du = exp(-3*r^2/4)*(3+2*cos r). ``` For `q=3+2*cos r`, direct differentiation gives ```text L1[Y](r) = exp(-3*r^2/2)*[(3/2)*q(r)^2+4+6*cos r], L1[Y](pi) = -(1/2)*exp(-3*pi^2/2)<0. ``` The kernel is smooth, positive and even. The Gaussian mixture identity for the logarithmic second derivative is `(log k)''=-2/3+(4/9)*Var_u(c)`, with centers `c` in `{-1,0,1}`. Their variance is at most `1`, so `(log k)''<=-2/9` everywhere. To retain double-exponential rather than Gaussian real tails, take ```text k_eta(u)=k(u)*exp(-eta*cosh(2*u)), eta>0. ``` Every such kernel is positive, even, real analytic, Schwartz, and has double-exponential tails with all polynomially weighted derivatives. It is strictly log-concave since `(log k_eta)''<=-2/9-4*eta*cosh(2*u)`. All odd derivatives at zero vanish. Dominated convergence against `|u|^j*k(u)`, for `j=0,1,2`, gives `Y_eta^(j)(pi)->Y^(j)(pi)` as `eta` decreases to zero. Therefore there exists `eta_0>0` such that `L1[Y_eta](pi)<0` for every `0 ``` for every finite real shift family and complex coefficient vector, using the original arithmetic trajectory. Boundary cancellation is insufficient for that comparison. The next admissible budget is one bounded paper review of a supplied orbit-specific inequality and its complete transfer; without such an inequality, allocate no scan or certificate-replay budget. This audit itself is complete. No manuscript edit, new research row, commit, push, outreach, numerical scan or symbolic scan was performed. **Final wall check: Same open gap.** The fixed reflected-prefix certificate is excluded; the complete original first-Laguerre inequality, stronger radial-sign target, cofinal signed arithmetic floor, G2 and RH remain open.