Introduction
Large-N N N unitary matrix integrals often admit several descriptions that appear quite different when expressed in their original variables. A small-source expansion organizes the answer into rational coefficient functions. A spectral construction gives an algebraic relation for a mass or source derivative of the free energy. Differential recursions relate successive expansion coefficients, whereas functional transformations connect expressions obtained in different parameter regions. The purpose of the present account is to place these structures in a common framework for two closely related models: the Gross–Witten–Wadia (GWW) model [1 , 2 ] in a fixed determinant-charge sector and the source-free one-dimensional QCD model in the joint large-N N N , large-N f N_f N f limit (for source-deformed Polyakov-loop extensions, see Refs. [3 , 4 ] ).
We organize the comparison in fixed determinant-charge sectors, represented by U ( N ) U(N) U ( N ) integrals. The S U ( N ) SU(N) S U ( N ) partition functions are reconstructed by summing over these sectors. Earlier large-N N N results for the S U ( N ) SU(N) S U ( N ) GWW integral and the one-dimensional QCD partition function are given in Refs. [5 , 6 ] , respectively.
Determinant deformations of GWW provide a related line of work. Russo analyzed the planar phases of fermion-inspired U ( N ) U(N) U ( N ) deformations [7 ] ; Russo and Tierz studied characteristic-polynomial insertions using Toeplitz determinants and Fisher–Hartwig asymptotics [8 ] . Santilli and Tierz established exact relations between determinant-deformed unitary ensembles and other random matrix ensembles and showed that their large-N N N phase diagrams can differ [9 ] . Malik studied the spectral density and phase structure of QCD-inspired U ( N ) U(N) U ( N ) and S U ( N ) SU(N) S U ( N ) models with linear sources and determinant terms [10 ] . Malik and Ahmed examined a complex unitary potential with Fisher–Hartwig singularities [11 ] . The comparison here centers on the algebraic generating functions at fixed charge, their projective symmetries, and the differential recursions of their coefficients.
The algebraic curves organize the paper. § 3 and § 4 start from the planar resolvent and the one-cut condition, obtaining the quartic and the sextic. The projective coordinates of § 2 make their parameter symmetries explicit, while the coefficient identities and generating PDEs are established on the curves. § 5 uses the QCD sextic to derive the small-charge recurrence, its generating equation, and the pyramid with GWW and Narayana boundaries. § 6 then explains how the GWW boundary also arises as a scaling limit of the QCD integral and curve. The finite-N N N calculation in § 4.3 tests the chosen branch independently of these derivations.
The central observation is that the relevant parameter transformations become elementary when the parameters are represented by homogeneous triples. For GWW the triple is [ u : h : i ] [u : h : i] [ u : h : i ] , where u = q / N u = q/N u = q / N and h h h is the source. For one-dimensional QCD it is [ k : k + 1 : k u ] [k : k + 1 : ku] [ k : k + 1 : k u ] , where k = N f / N k = N_f/N k = N f / N and u = q / N f u = q/N_f u = q / N f . Permutations and sign changes of these coordinates induce rational, and sometimes complex, transformations in the usual affine parameters. Their apparent complexity is largely a consequence of restoring an affine normalization after performing a simple homogeneous operation.
Geometrically, the quartic and sextic describe algebraic coverings of the corresponding parameter charts. The signed permutations act on the base coordinates and transport roots between fibers; in QCD the mass is a separate coordinate. For GWW a permutation can change the sign of the orientation covariant, while the normalized QCD root W W W is invariant on its transported sheet. The physical germ is selected by boundary data of the integral. This distinction is why projective equivalence alone does not assert equality of separately selected thermodynamic branches.
This projective viewpoint also clarifies what it means for a differential equation to respect a symmetry. A group transformation can permute several component operators even when a particular combination of them is preserved up to a scalar factor. Conversely, a covariant transformation law for one complete operator does not determine how all possible decompositions of that operator behave. The distinction is visible already in GWW. There, the difference of two operators can change sign under a duality transformation, preserving its homogeneous equation, while the two component operators are exchanged. It is therefore natural to seek a comparable formulation in one-dimensional QCD rather than infer the operator structure solely from an invariant partial differential equation.
A second issue concerns differential order. The algebraic degree of a generating function, the total order of a partial differential equation, and the order of an ordinary differential equation obtained by fixing external parameters are separate quantities. In particular, an algebraic function of degree six can satisfy a mixed partial differential equation of order two without admitting a second-order ordinary differential equation in the mass alone. Similarly, replacing a free energy by a suitable first derivative can lower the order of its partial differential equation, whereas an invertible change of coordinates preserves the order of a given differential operator. These distinctions are essential when comparing the explicit third-derivative GWW raising recursion with the second-order parameter constraints that arise in QCD.
We retain the intermediate normalizations needed to move between the free energy, its algebraic derivative, and its projective generating function. This is particularly useful for interpreting earlier GWW recursions, whose generating function differs from the regular free energy by a sign and an elementary quadratic term. For QCD, a specific logarithmic subtraction converts an affine transformation of the free energy into a homogeneous transformation law. In those variables, the permuted operator triple has a short expression in Euler derivatives. The presentation concerns analytic fixed-sector branches before any sector summation or charge extremization. Polynomial identities are established at the level of the algebraic curve, and therefore hold for every regular local root of the relevant polynomial. A physical branch requires additional information from the matrix integral, including its small-source boundary condition and the continuation of the saddle contour. These qualifications enter at the points where they affect the interpretation of a formula. They do not restrict the purely algebraic or operator identities derived below.
The comparison is organized around the following results.
The one-cut resolvent gives the GWW quartic and QCD sextic (§ 3 and § 4 ). The GWW raising recursion is the reference case; the Toeplitz calculation is an independent test of the QCD branch.
The projective S 4 S_4 S 4 maps act on both curves. The induced operator action is a literal permutation in homogeneous coordinates and involves normalization factors in the physical charts.
The QCD row recurrence (109) and generating PDE (111) yield the explicit column reconstruction (113) . The first column is GWW, and the leading resummed boundary is Narayana at all orders. The collapsed slice admits a separate exact tree count.
The many-flavor scaling relates the matrix integrals, the curves and the coefficient expansions (§ 6 ). The limit needs transverse coordinates and does not identify the particular operator triples component by component.
Sector integrals, normalizations, and projective coordinates
Fixed-charge integrals and algebraic derivatives
We use the GWW sector integral
D q G ( N , h ) = ∫ U ( N ) d U ( det U ) q exp [ N h 2 Tr ( U + U − 1 ) ] , u = q N . (1) D_q^{\mathrm{G}}(N,h)=\int_{U(N)}\mathrm{d}U\,(\det U)^q\exp\left[\frac{Nh}{2}\operatorname{Tr}\left(U+U^{-1}\right)\right],\qquad u=\frac{q}{N}. \tag*{(1)} D q G ( N , h ) = ∫ U ( N ) d U ( det U ) q exp [ 2 N h Tr ( U + U − 1 ) ] , u = N q . (1)
The symmetry D − q G = D q G D_{-q}^{\mathrm{G}}=D_q^{\mathrm{G}} D − q G = D q G allows us to choose u ≥ 0 u\ge0 u ≥ 0 on the real fixed-sector branch. For algebraic continuation, u u u will be treated as a complex parameter. The planar free energy and its source derivative are
Ψ G ( u , h ) = lim N → ∞ N − 2 log D q G , g ( u , h ) = h 2 ∂ h Ψ G = x ∂ x Ψ G , x = h 2 . (2) \Psi_{\mathrm{G}}(u,h)=\lim_{N\to\infty}N^{-2}\log D_q^{\mathrm{G}},\qquad g(u,h)=\frac{h}{2}\partial_h\Psi_{\mathrm{G}}=x\partial_x\Psi_{\mathrm{G}},\qquad x=h^2. \tag*{(2)} Ψ G ( u , h ) = N → ∞ lim N − 2 log D q G , g ( u , h ) = 2 h ∂ h Ψ G = x ∂ x Ψ G , x = h 2 . (2)
The exact finite-N N N determinant
D q G ( N , h ) = det 1 ≤ i , j ≤ N I q + i − j ( N h ) (3) D_q^{\mathrm{G}}(N,h)=\det_{1\le i,j\le N} I_{q+i-j}(Nh) \tag*{(3)} D q G ( N , h ) = 1 ≤ i , j ≤ N det I q + i − j ( N h ) (3)
provides an independent definition and fixes the normalization of the small-source expansion. Its role here is to identify the sector and its boundary data; the planar algebraic description can be developed without expanding this determinant in representations.
For source-free one-dimensional QCD we consider
D q Q ( N , N f , h ) = ∫ U ( N ) d U ( det U ) q det ( 1 + h U ) N f det ( 1 + h U − 1 ) N f . (4) D_q^{\mathrm{Q}}(N,N_f,h)=\int_{U(N)}\mathrm{d}U\,(\det U)^q\det(1+hU)^{N_f}\det(1+hU^{-1})^{N_f}. \tag*{(4)} D q Q ( N , N f , h ) = ∫ U ( N ) d U ( det U ) q det ( 1 + h U ) N f det ( 1 + h U − 1 ) N f . (4)
The two determinants are retained exactly. The parameters in the Veneziano limit are
k = κ = N f N , u = q N f , v = k u = q N . (5) k=\kappa=\frac{N_f}{N},\qquad u=\frac{q}{N_f},\qquad v=ku=\frac{q}{N}. \tag*{(5)} k = κ = N N f , u = N f q , v = k u = N q . (5)
The QCD charge u u u thus differs from the GWW charge by a factor k k k . This distinction will be maintained explicitly in the scaling limit. For positive physical charge the source-free determinant has the saturation range 0 ≤ u ≤ 1 0 \le u \le1 0 ≤ u ≤ 1 , whereas the GWW sector has no corresponding upper bound.
The static mass factor has been removed from (4) . With h = e − m h=e^{-m} h = e − m , the free energy normalized by N N f NN_f N N f and the determinant free energy normalized by N 2 N^2 N 2 are related by
Ψ Q = lim N − 2 log D q Q , S ( k , u ; h ) = − log h + Ψ Q ( k , u ; h ) k . (6) \Psi_{\mathrm{Q}}=\lim N^{-2}\log D_q^{\mathrm{Q}}, \qquad S(k,u;h)=-\log h+\frac{\Psi_{\mathrm{Q}}(k,u;h)}{k}. \tag*{(6)} Ψ Q = lim N − 2 log D q Q , S ( k , u ; h ) = − log h + k Ψ Q ( k , u ; h ) . (6)
We define
G Q = x ∂ x Ψ Q , R = G Q k − 1 2 = h 2 S h . (7) G_{\mathrm{Q}}=x\partial_x\Psi_{\mathrm{Q}}, \qquad R=\frac{G_{\mathrm{Q}}}{k}-\frac{1}{2}=\frac{h}{2}S_h. \tag*{(7)} G Q = x ∂ x Ψ Q , R = k G Q − 2 1 = 2 h S h . (7)
The additive 1 / 2 1/2 1/2 is a normalization effect of the static mass factor. For example, ∂ m S = − 2 R \partial_m S=-2R ∂ m S = − 2 R in the convention of (6) . Selecting a sector or combining several sectors is a subsequent operation.
In what follows, the term algebraic generating function refers to g g g , G Q G_{\mathrm{Q}} G Q , R R R , or an explicitly normalized version of these derivatives. The free energies themselves contain logarithms. Their algebraic derivatives determine them up to functions of the external parameters, which must be fixed by the sector normalization. This integration freedom is also responsible for additive terms in the functional transformations of the free energy.
Homogeneous triples and the finite projective group
A homogeneous triple represents the same projective point after a common nonzero rescaling,
[ a : b : c ] = [ λ a : λ b : λ c ] . (8) [a:b:c]=[\lambda a:\lambda b:\lambda c]. \tag*{(8)} [ a : b : c ] = [ λa : λb : λ c ] . (8)
In the chart a = 1 a=1 a = 1 , let p = b / a p=b/a p = b / a and q = c / a q=c/a q = c / a . The three elementary permutations of the homogeneous coordinates act as
( p , q ) → b c ( q , p ) , ( p , q ) → a b ( 1 / p , q / p ) , ( p , q ) → a c ( p / q , 1 / q ) . (9) (p,q)\xrightarrow{bc}(q,p), \qquad(p,q)\xrightarrow{ab}(1/p,q/p), \qquad(p,q)\xrightarrow{ac}(p/q,1/q). \tag*{(9)} ( p , q ) b c ( q , p ) , ( p , q ) ab ( 1/ p , q / p ) , ( p , q ) a c ( p / q , 1/ q ) . (9)
The denominators in these transformations arise from the restoration of the chart. They are the elementary origin of the fractional-linear relations encountered in coefficient expansions.
The two models admit the triples
[ a : b : c ] G = [ u : h : i ] , [ a : b : c ] Q = [ k : k + 1 : k u ] . (10) [a:b:c]_{\mathrm{G}}=[u:h:i], \qquad[a:b:c]_{\mathrm{Q}}=[k:k+1:ku]. \tag*{(10)} [ a : b : c ] G = [ u : h : i ] , [ a : b : c ] Q = [ k : k + 1 : k u ] . (10)
Their different dependence on h h h is significant. The GWW source is a coordinate of the triple and can be exchanged with the charge by a projective transformation. In QCD the mass remains an additional variable, while the triple consists of flavor and charge data.
Together with independent sign changes, permutations generate
Γ = ( ( Z 2 ) 3 / Z 2 ) × S 3 ≃ ( Z 2 ) 2 × S 3 ≃ S 4 . (11) \Gamma=((\mathbb{Z}_2)^3/\mathbb{Z}_2)\times S_3 \simeq(\mathbb{Z}_2)^2\times S_3 \simeq S_4. \tag*{(11)} Γ = (( Z 2 ) 3 / Z 2 ) × S 3 ≃ ( Z 2 ) 2 × S 3 ≃ S 4 . (11)
The quotient removes the common sign, which is a projective rescaling. The resulting group has order 24. On the three squared coordinates A 1 = a 2 A_1=a^2 A 1 = a 2 , A 2 = b 2 A_2=b^2 A 2 = b 2 , and A 3 = c 2 A_3=c^2 A 3 = c 2 , sign changes act trivially, and only the quotient S 3 S_3 S 3 permutes the coordinate labels. This is the permutation action that will exchange the differential operators.
This finite action should be distinguished from invariance under the entire projective linear group. It should also be distinguished from a Galois group over a fixed coefficient field: the transformations in (11) change external parameters, whereas a Galois action fixes the prescribed coefficient field. A relation between these two structures would require a separate analysis.
Symmetric invariants and orientation
The elementary symmetric polynomials in A i A_i A i are
e 1 = A 1 + A 2 + A 3 , e 2 = A 1 A 2 + A 1 A 3 + A 2 A 3 , e 3 = A 1 A 2 A 3 . (12) e_1 = A_1 + A_2 + A_3,\qquad e_2 = A_1A_2 + A_1A_3 + A_2A_3,\qquad e_3 = A_1A_2A_3. \tag*{(12)} e 1 = A 1 + A 2 + A 3 , e 2 = A 1 A 2 + A 1 A 3 + A 2 A 3 , e 3 = A 1 A 2 A 3 . (12)
On e 1 ≠ 0 e_1 \ne0 e 1 = 0 we use the rational projective invariants
r = e 2 e 1 2 , s = e 3 e 1 3 . (13) r = \frac{e_2}{e_1^2},\qquad s = \frac{e_3}{e_1^3}. \tag*{(13)} r = e 1 2 e 2 , s = e 1 3 e 3 . (13)
Indeed, A i / e 1 A_i/e_1 A i / e 1 are the roots of
λ 3 − λ 2 + r λ − s = 0. (14) \lambda^3 - \lambda^2 + r\lambda- s = 0. \tag*{(14)} λ 3 − λ 2 + r λ − s = 0. (14)
The normalized orientation covariant and its square are
d = ( A 1 − A 3 ) ( A 2 − A 3 ) ( A 1 − A 2 ) e 1 3 , d 2 = Δ = r 2 − 4 r 3 − 4 s + 18 r s − 27 s 2 . (15) d = \frac{(A_1-A_3)(A_2-A_3)(A_1-A_2)}{e_1^3},\qquad d^2 = \Delta= r^2 - 4r^3 - 4s + 18rs - 27s^2. \tag*{(15)} d = e 1 3 ( A 1 − A 3 ) ( A 2 − A 3 ) ( A 1 − A 2 ) , d 2 = Δ = r 2 − 4 r 3 − 4 s + 18 r s − 27 s 2 . (15)
An odd permutation reverses the sign of d d d , while r r r , s s s , and Δ \Delta Δ remain unchanged. This orientation information is needed in the depressed GWW quartic. The QCD sextic, in contrast, can be written entirely in the symmetric invariants. Passing from the ordered coordinates to r r r , s s s is locally invertible away from the corresponding ramification loci, but it identifies points related by permutations. Consequently an invariant description can conceal an operator permutation that is transparent in homogeneous coordinates.
The GWW algebraic generating function and its differential system
A rational coordinate for the fixed-charge branch
The GWW quartic, its projective transformations and the associated hierarchy of differential recursions are discussed in Ref. [12 ] . Here we give a short resolvent derivation in the same form as the QCD calculation in § 4.1 : a rational spectral function, a one-cut double-zero condition, and elimination to an algebraic relation for the free-energy derivative.
The algebraic curve can be obtained from a one-cut planar construction. For nonzero charge, the saddle contour generally moves into the complex eigenvalue plane. Combining the Haar measure with the determinant insertion gives the effective potential
V ( z ) = − h 2 ( z + 1 z ) − ( u − 1 ) log z . (16) V(z) = -\frac{h}{2}\left(z+\frac{1}{z}\right)-(u-1)\log z. \tag*{(16)} V ( z ) = − 2 h ( z + z 1 ) − ( u − 1 ) log z . (16)
Let ω ( z ) \omega(z) ω ( z ) be the planar resolvent, normalized by ω ( z ) = z − 1 + m 1 z − 2 + O ( z − 3 ) \omega(z) = z^{-1} + m_1z^{-2} + O(z^{-3}) ω ( z ) = z − 1 + m 1 z − 2 + O ( z − 3 ) at infinity. For a single cut with endpoints a ∗ a_* a ∗ , b ∗ b_* b ∗ , set ρ 2 = a ∗ b ∗ \rho^2 = a_*b_* ρ 2 = a ∗ b ∗ and choose the square root R ( z ) = ( z − a ∗ ) ( z − b ∗ ) R(z) = \sqrt{(z-a_*)(z-b_*)} R ( z ) = ( z − a ∗ ) ( z − b ∗ ) so that R ( z ) ∼ z R(z) \sim z R ( z ) ∼ z and R ( 0 ) = − ρ R(0) = -\rho R ( 0 ) = − ρ . Matching the second-order pole at the origin and the constant at infinity gives
V ′ ( z ) − 2 ω ( z ) = − h 2 z 2 ( z + 1 ρ ) R ( z ) . (17) V'(z)-2\omega(z)=-\frac{h}{2z^2}\left(z+\frac{1}{\rho}\right)R(z). \tag*{(17)} V ′ ( z ) − 2 ω ( z ) = − 2 z 2 h ( z + ρ 1 ) R ( z ) . (17)
If S ∗ = a ∗ + b ∗ S_* = a_* + b_* S ∗ = a ∗ + b ∗ , the matching conditions are
h 2 ( ρ − S ∗ 2 ρ 2 ) = 1 − u , S ∗ = 2 ρ − 4 ( 1 + u ) h . (18) \frac{h}{2}\left(\rho-\frac{S_*}{2\rho^2}\right)=1-u,\qquad S_*=\frac{2}{\rho}-\frac{4(1+u)}{h}. \tag*{(18)} 2 h ( ρ − 2 ρ 2 S ∗ ) = 1 − u , S ∗ = ρ 2 − h 4 ( 1 + u ) . (18)
Eliminating S ∗ S_* S ∗ produces the rational parameterization
h = 2 ρ [ 1 + u + ( u − 1 ) ρ 2 ] 1 − ρ 4 , g = h 4 ( ρ + 1 ρ ) − 1 + ρ 4 2 ( 1 + ρ 2 ) 2 . (19) h=\frac{2\rho[1+u+(u-1)\rho^2]}{1-\rho^4},\qquad g=\frac{h}{4}\left(\rho+\frac{1}{\rho}\right)-\frac{1+\rho^4}{2(1+\rho^2)^2}. \tag*{(19)} h = 1 − ρ 4 2 ρ [ 1 + u + ( u − 1 ) ρ 2 ] , g = 4 h ( ρ + ρ 1 ) − 2 ( 1 + ρ 2 ) 2 1 + ρ 4 . (19)
The second formula follows from the first moment of the resolvent and the sector identity m 1 − m − 1 = − 2 u / h m_1-m_{-1}=-2u/h m 1 − m − 1 = − 2 u / h . The chosen small-source branch has ρ ∼ h / [ 2 ( 1 + u ) ] \rho\sim h/[2(1+u)] ρ ∼ h / [ 2 ( 1 + u )] .
For completeness, define the moments by m j = ∫ ρ ( ξ ) ξ j d ξ m_j=\int\rho(\xi)\xi^j\,\mathrm{d}\xi m j = ∫ ρ ( ξ ) ξ j d ξ . Planar factorization gives
ω ( z ) 2 − V ′ ( z ) ω ( z ) + f G ( z ) = 0 , f G ( z ) = − h 2 z − h m − 1 2 z 2 . (20) \omega(z)^2-V'(z)\omega(z)+f_{\mathrm{G}}(z)=0,\qquad f_{\mathrm{G}}(z)=-\frac{h}{2z}-\frac{hm_{-1}}{2z^2}. \tag*{(20)} ω ( z ) 2 − V ′ ( z ) ω ( z ) + f G ( z ) = 0 , f G ( z ) = − 2 z h − 2 z 2 h m − 1 . (20)
Indeed, inserting the rational force V ′ = − h / 2 + ( 1 − u ) / z + h / ( 2 z 2 ) V'=-h/2+(1-u)/z+h/(2z^2) V ′ = − h /2 + ( 1 − u ) / z + h / ( 2 z 2 ) into f G = ∫ [ V ′ ( z ) − V ′ ( ξ ) ] ρ ( ξ ) / ( z − ξ ) d ξ f_{\mathrm{G}}=\int[V'(z)-V'(\xi)]\rho(\xi)/(z-\xi)\,\mathrm{d}\xi f G = ∫ [ V ′ ( z ) − V ′ ( ξ )] ρ ( ξ ) / ( z − ξ ) d ξ and matching the 1 / z 1/z 1/ z term at infinity yields this expression. The source derivative and the scaling Ward identity give
g = h 4 ( m 1 + m − 1 ) , m 1 − m − 1 = − 2 u h , h m − 1 = 2 g + u . (21) g=\frac{h}{4}(m_1+m_{-1}),\qquad m_1-m_{-1}=-\frac{2u}{h},\qquad hm_{-1}=2g+u. \tag*{(21)} g = 4 h ( m 1 + m − 1 ) , m 1 − m − 1 = − h 2 u , h m − 1 = 2 g + u . (21)
Thus the squared spectral function has a quartic numerator:
( V ′ − 2 ω ) 2 = Q G ( z ) 4 z 4 , (22) (V'-2\omega)^2=\frac{Q_{\mathrm{G}}(z)}{4z^4}, \tag*{(22)} ( V ′ − 2 ω ) 2 = 4 z 4 Q G ( z ) , (22)
Q G ( z ) = [ − h z 2 + 2 ( 1 − u ) z + h ] 2 + 8 h z 3 + 8 ( 2 g + u ) z 2 . Q_{\mathrm{G}}(z)=[-hz^2+2(1-u)z+h]^2+8hz^3+8(2g+u)z^2. Q G ( z ) = [ − h z 2 + 2 ( 1 − u ) z + h ] 2 + 8 h z 3 + 8 ( 2 g + u ) z 2 .
The one-cut ansatz (17) makes this numerator h 2 ( z + ρ − 1 ) 2 ( z − a ∗ ) ( z − b ∗ ) h^2(z+\rho^{-1})^2(z-a_*)(z-b_*) h 2 ( z + ρ − 1 ) 2 ( z − a ∗ ) ( z − b ∗ ) ; hence it has a double zero. More precisely, direct elimination gives the exact identity
Disc z Q G = 2 20 h 4 P 4 ( g ; u , h ) . (23) \operatorname{Disc}_z Q_{\mathrm{G}}=2^{20}h^4P_4(g;u,h). \tag*{(23)} Disc z Q G = 2 20 h 4 P 4 ( g ; u , h ) . (23)
The polynomial P 4 P_4 P 4 is displayed in (25) below. This supplies the same spectral-discriminant construction that gives the QCD sextic. Its physical local branch is fixed by g = u / 2 + h 2 / [ 4 ( 1 + u ) ] + O ( h 4 ) g=u/2+h^2/[4(1+u)]+O(h^4) g = u /2 + h 2 / [ 4 ( 1 + u )] + O ( h 4 ) ; the discriminant equation alone does not select that branch.
For u , h > 0 u,h>0 u , h > 0 , the real parameter 0 < ρ < 1 0<\rho<1 0 < ρ < 1 is particularly useful:
∂ h ∂ ρ ∣ u = 2 { u ( 1 + ρ 2 ) 3 + ( 1 − ρ 2 ) 3 } ( 1 − ρ 4 ) 2 > 0. (24) \left.\frac{\partial h}{\partial\rho}\right|_u=\frac{2\{u(1+\rho^2)^3+(1-\rho^2)^3\}}{(1-\rho^4)^2}>0. \tag*{(24)} ∂ ρ ∂ h u = ( 1 − ρ 4 ) 2 2 { u ( 1 + ρ 2 ) 3 + ( 1 − ρ 2 ) 3 } > 0. (24)
It therefore labels a unique solution within this one-cut family for every positive pair ( u , h ) (u,h) ( u , h ) . The algebraic discussion below concerns that family and its analytic continuations. A comparison with all other possible complex-contour saddles would be needed to establish a global dominance statement.
The quartic and its projective form
Elimination of ρ \rho ρ in (19) gives the monic quartic
0 = P 4 ( g ; u , h ) = g 4 + 1 + u 2 − h 2 2 g 3 + ( 1 − u ) 2 + 2 ( u 2 − 4 ) h 2 + h 4 16 g 2 + − 4 u 2 ( 1 + u 2 ) + ( u 2 − 1 ) h 2 + 5 h 4 32 g − u 2 ( 1 + u 2 ) 2 64 − u 2 ( 3 u 2 − 5 ) h 2 64 + ( 1 − 12 u 2 ) h 4 256 − h 6 64 . (25) \begin{aligned}
0=P_4(g;u,h)={}&g^4+\frac{1+u^2-h^2}{2}g^3\\
&+\frac{(1-u)^2+2(u^2-4)h^2+h^4}{16}g^2\\
&+\frac{-4u^2(1+u^2)+(u^2-1)h^2+5h^4}{32}g\\
&-\frac{u^2(1+u^2)^2}{64}-\frac{u^2(3u^2-5)h^2}{64}
+\frac{(1-12u^2)h^4}{256}-\frac{h^6}{64}. \tag*{(25)}
\end{aligned} 0 = P 4 ( g ; u , h ) = g 4 + 2 1 + u 2 − h 2 g 3 + 16 ( 1 − u ) 2 + 2 ( u 2 − 4 ) h 2 + h 4 g 2 + 32 − 4 u 2 ( 1 + u 2 ) + ( u 2 − 1 ) h 2 + 5 h 4 g − 64 u 2 ( 1 + u 2 ) 2 − 64 u 2 ( 3 u 2 − 5 ) h 2 + 256 ( 1 − 12 u 2 ) h 4 − 64 h 6 . (25)
The boundary condition
g = u 2 + h 2 4 ( 1 + u ) + O ( h 4 ) (26) g=\frac{u}{2}+\frac{h^2}{4(1+u)}+O(h^4) \tag*{(26)} g = 2 u + 4 ( 1 + u ) h 2 + O ( h 4 ) (26)
identifies the desired local root. Both the polynomial and the boundary condition are needed: the polynomial alone specifies all four branches.
For [ a : b : c ] = [ u : h : i ] [a:b:c]=[u:h:i] [ a : b : c ] = [ u : h : i ] , the symmetric polynomials become
e 1 = u 2 + h 2 − 1 , e 2 = u 2 h 2 − u 2 − h 2 , e 3 = − u 2 h 2 . (27) e_1=u^2+h^2-1,\qquad e_2=u^2h^2-u^2-h^2,\qquad e_3=-u^2h^2. \tag*{(27)} e 1 = u 2 + h 2 − 1 , e 2 = u 2 h 2 − u 2 − h 2 , e 3 = − u 2 h 2 . (27)
Introduce the centered variable and the unnormalized orientation factor
τ = − 4 g − 1 + u 2 − h 2 2 , δ = ( 1 + u 2 ) ( 1 + h 2 ) ( u 2 − h 2 ) . (28) \tau=-4g-\frac{1+u^2-h^2}{2},\qquad\delta=(1+u^2)(1+h^2)(u^2-h^2). \tag*{(28)} τ = − 4 g − 2 1 + u 2 − h 2 , δ = ( 1 + u 2 ) ( 1 + h 2 ) ( u 2 − h 2 ) . (28)
The quartic then takes the homogeneous form
0 = τ 4 + 12 e 2 − e 1 2 2 τ 2 + 4 δ τ + e 1 4 16 + e 1 2 e 2 2 − 3 e 2 2 + 18 e 1 e 3 . (29) 0=\tau^4+\frac{12e_2-e_1^2}{2}\tau^2+4\delta\tau+\frac{e_1^4}{16}+\frac{e_1^2e_2}{2}-3e_2^2+18e_1e_3. \tag*{(29)} 0 = τ 4 + 2 12 e 2 − e 1 2 τ 2 + 4 δ τ + 16 e 1 4 + 2 e 1 2 e 2 − 3 e 2 2 + 18 e 1 e 3 . (29)
Every term has homogeneous degree eight when τ \tau τ has degree two in a , b , c a,b,c a , b , c . Substitution of (28) into the left-hand side of (29) gives exactly 256 P 4 256P_4 256 P 4 . The projective form therefore preserves the complete algebraic relation, including its normalization.
Under an odd permutation, δ \delta δ and τ \tau τ both change sign. After restoring c = i c=i c = i , the elementary transpositions give
( u ′ , h ′ ) τ ′ a b ( h , u ) − τ a c ( − 1 / u , i h / u ) τ / u 2 b c ( i u / h , − 1 / h ) τ / h 2 (30) \begin{array}{c|c|c}
& (u',h') & \tau'\\
ab & (h,u) & -\tau\\
ac & (-1/u,ih/u) & \tau/u^2\\
bc & (iu/h,-1/h) & \tau/h^2 \tag*{(30)}
\end{array} ab a c b c ( u ′ , h ′ ) ( h , u ) ( − 1/ u , ih / u ) ( i u / h , − 1/ h ) τ ′ − τ τ / u 2 τ / h 2 (30)
These are pure permutations of the homogeneous coordinates followed by rescaling. Additional sign changes give alternative representatives, such as ( 1 / u , i h / u ) (1/u,ih/u) ( 1/ u , ih / u ) . In terms of the original generating function,
g a b ′ = − g − 1 4 , g a c ′ = g − h 2 / 4 u 2 , g b c ′ = g + ( 1 + u 2 − h 2 ) / 4 h 2 . (31) g'_{ab}=-g-\frac{1}{4},\qquad g'_{ac}=\frac{g-h^2/4}{u^2},\qquad g'_{bc}=\frac{g+(1+u^2-h^2)/4}{h^2}. \tag*{(31)} g ab ′ = − g − 4 1 , g a c ′ = u 2 g − h 2 /4 , g b c ′ = h 2 g + ( 1 + u 2 − h 2 ) /4 . (31)
The first relation already shows why an algebraic symmetry need not preserve a physical root: it maps a positive real g g g to a negative one. Equations (30) and (31) should therefore be interpreted with the root transported along the parameter transformation, rather than reselected independently at the image point.
Dividing by the appropriate homogeneous powers yields a description in projective invariants. Set w = τ / e 1 w=\tau/e_1 w = τ / e 1 and d = δ / e 1 3 d=\delta/e_1^3 d = δ / e 1 3 . Then
w 4 + 12 r − 1 2 w 2 + 4 d w + 1 16 + r 2 − 3 r 2 + 18 s = 0. (32) w^4+\frac{12r-1}{2}w^2+4dw+\frac{1}{16}+\frac{r}{2}-3r^2+18s=0. \tag*{(32)} w 4 + 2 12 r − 1 w 2 + 4 d w + 16 1 + 2 r − 3 r 2 + 18 s = 0. (32)
The scalar w w w is odd under a transposition, but ζ = d w \zeta=dw ζ = d w is invariant. Its polynomial is
0 = ζ 4 + 12 r − 1 2 Δ ζ 2 + 4 Δ 2 ζ + ( 1 16 + r 2 − 3 r 2 + 18 s ) Δ 2 . (33) 0=\zeta^4+\frac{12r-1}{2}\Delta\zeta^2+4\Delta^2\zeta+\left(\frac{1}{16}+\frac{r}{2}-3r^2+18s\right)\Delta^2. \tag*{(33)} 0 = ζ 4 + 2 12 r − 1 Δ ζ 2 + 4 Δ 2 ζ + ( 16 1 + 2 r − 3 r 2 + 18 s ) Δ 2 . (33)
On Δ = 0 \Delta=0 Δ = 0 , this particular invariant variable loses information about w w w . The oriented equation (32) , or the original quartic, supplies a regular chart when the invariant normalization degenerates.
An explicit coefficient recursion and the regular free energy
The quartic branches obey the differential identity
h g h h h = u g u u u . (34) h g_{hhh}=u g_{uuu}. \tag*{(34)} h g hhh = u g uuu . (34)
It is a linear partial differential equation in two parameters, and its small-source expansion gives a particularly simple raising relation. Write
g = u 2 + h 2 4 − ∑ n ≥ 0 R n ( u ) ( h 2 ) 2 n + 2 , R 0 = u 1 + u . (35) g=\frac{u}{2}+\frac{h^2}{4}-\sum_{n\geq0}R_n(u)\left(\frac{h}{2}\right)^{2n+2},\qquad R_0=\frac{u}{1+u}. \tag*{(35)} g = 2 u + 4 h 2 − n ≥ 0 ∑ R n ( u ) ( 2 h ) 2 n + 2 , R 0 = 1 + u u . (35)
Comparison of the coefficient of h 2 n h^{2n} h 2 n in (34) yields
R n ( u ) = u ∂ u 3 R n − 1 ( u ) n ( n + 1 ) ( 2 n + 1 ) , n ≥ 1. (36) R_n(u)=\frac{u\,\partial_u^3R_{n-1}(u)}{n(n+1)(2n+1)},\qquad n\geq1. \tag*{(36)} R n ( u ) = n ( n + 1 ) ( 2 n + 1 ) u ∂ u 3 R n − 1 ( u ) , n ≥ 1. (36)
The right-hand side acts only on the known previous coefficient. Thus (36) is an explicit differential raising formula, a property that will be useful in assessing the QCD recursions. The corresponding signed numerator triangle is OEIS A336110 [13 ] ; see also Ref. [12 ] for the differential recursion.
The first coefficients are
R 0 = u 1 + u , R 1 = u ( 1 + u ) 4 , (37) R_0=\frac{u}{1+u},\qquad R_1=\frac{u}{(1+u)^4}, \tag*{(37)} R 0 = 1 + u u , R 1 = ( 1 + u ) 4 u , (37)
R 2 = − 2 u ( u − 1 ) ( 1 + u ) 7 , R 3 = u ( 5 u 2 − 14 u + 5 ) ( 1 + u ) 10 . (38) R_2=-\frac{2u(u-1)}{(1+u)^7},\qquad R_3=\frac{u(5u^2-14u+5)}{(1+u)^{10}}. \tag*{(38)} R 2 = − ( 1 + u ) 7 2 u ( u − 1 ) , R 3 = ( 1 + u ) 10 u ( 5 u 2 − 14 u + 5 ) . (38)
They give
g = u 2 + h 2 4 ( 1 + u ) − u h 4 16 ( 1 + u ) 4 + u ( u − 1 ) h 6 32 ( 1 + u ) 7 − u ( 5 u 2 − 14 u + 5 ) h 8 256 ( 1 + u ) 10 + O ( h 10 ) . (39) g=\frac{u}{2}+\frac{h^2}{4(1+u)}-\frac{u h^4}{16(1+u)^4}+\frac{u(u-1)h^6}{32(1+u)^7}-\frac{u(5u^2-14u+5)h^8}{256(1+u)^{10}}+O(h^{10}). \tag*{(39)} g = 2 u + 4 ( 1 + u ) h 2 − 16 ( 1 + u ) 4 u h 4 + 32 ( 1 + u ) 7 u ( u − 1 ) h 6 − 256 ( 1 + u ) 10 u ( 5 u 2 − 14 u + 5 ) h 8 + O ( h 10 ) . (39)
The powers of 2 originate from the source normalization in (1) . Integrating (35) gives
Ψ G = u log h + C ( u ) + h 2 4 − ∑ n ≥ 0 R n ( u ) n + 1 ( h 2 ) 2 n + 2 . (40) \Psi_{\mathrm G}=u\log h+C(u)+\frac{h^2}{4}-\sum_{n\geq0}\frac{R_n(u)}{n+1}\left(\frac{h}{2}\right)^{2n+2}. \tag*{(40)} Ψ G = u log h + C ( u ) + 4 h 2 − n ≥ 0 ∑ n + 1 R n ( u ) ( 2 h ) 2 n + 2 . (40)
The function C ( u ) C(u) C ( u ) is fixed by the matrix integral and cannot be recovered from the mass derivative alone.
To compare with an operator formulation written directly for the regular free energy, define
P ( u , h ) = h 2 4 − [ Ψ G − u log h − C ( u ) ] = ∑ n ≥ 0 P n ( u , h ) , (41) P(u,h)=\frac{h^2}{4}-[\Psi_{\mathrm G}-u\log h-C(u)]=\sum_{n\geq0}P_n(u,h), \tag*{(41)} P ( u , h ) = 4 h 2 − [ Ψ G − u log h − C ( u )] = n ≥ 0 ∑ P n ( u , h ) , (41)
P n = R n ( u ) n + 1 ( h 2 ) 2 n + 2 , g = u 2 + h 2 4 − h 2 P h . (42) P_n=\frac{R_n(u)}{n+1}\left(\frac{h}{2}\right)^{2n+2},\qquad g=\frac{u}{2}+\frac{h^2}{4}-\frac{h}{2}P_h. \tag*{(42)} P n = n + 1 R n ( u ) ( 2 h ) 2 n + 2 , g = 2 u + 4 h 2 − 2 h P h . (42)
For example,
P 0 = u h 2 4 ( 1 + u ) , P 1 = u h 4 32 ( 1 + u ) 4 , P 2 = − u ( u − 1 ) h 6 96 ( 1 + u ) 7 . (43) P_0=\frac{u h^2}{4(1+u)},\qquad P_1=\frac{u h^4}{32(1+u)^4},\qquad P_2=-\frac{u(u-1)h^6}{96(1+u)^7}. \tag*{(43)} P 0 = 4 ( 1 + u ) u h 2 , P 1 = 32 ( 1 + u ) 4 u h 4 , P 2 = − 96 ( 1 + u ) 7 u ( u − 1 ) h 6 . (43)
The sign and the factor n + 1 n+1 n + 1 distinguish P n P_n P n from the coefficient functions in the algebraic derivative. Keeping this dictionary explicit avoids an apparent disagreement between equivalent recursions.
Let
K + = ∂ h 2 ( h ∂ h ) , R + = u ∂ u 3 , D = u ∂ u + h ∂ h , (44) K_{+}=\partial_h^2(h\partial_h), \qquad R_{+}=u\partial_u^3, \qquad D=u\partial_u+h\partial_h, \tag*{(44)} K + = ∂ h 2 ( h ∂ h ) , R + = u ∂ u 3 , D = u ∂ u + h ∂ h , (44)
and introduce
Δ 0 = D ( D − 1 ) ( D − 2 ) = u 3 ∂ u 3 + 3 u 2 h ∂ u 2 ∂ h + 3 u h 2 ∂ u ∂ h 2 + h 3 ∂ h 3 . (45) \begin{aligned}
\Delta_0 &= D(D-1)(D-2) \\
&=u^3\partial_u^3+3u^2h\partial_u^2\partial_h+3uh^2\partial_u\partial_h^2+h^3\partial_h^3. \tag*{(45)}
\end{aligned} Δ 0 = D ( D − 1 ) ( D − 2 ) = u 3 ∂ u 3 + 3 u 2 h ∂ u 2 ∂ h + 3 u h 2 ∂ u ∂ h 2 + h 3 ∂ h 3 . (45)
The notation Δ 0 \Delta_0 Δ 0 separates this operator from the projective discriminant Δ ( r , s ) \Delta(r,s) Δ ( r , s ) . In particular, K + = h ∂ h 3 + 2 ∂ h 2 K_{+}=h\partial_h^3+2\partial_h^2 K + = h ∂ h 3 + 2 ∂ h 2 ; the second term is part of its definition. Equation (36) is equivalent to
R + P n = K + P n + 1 = ( 2 n + 3 ) ( 2 n + 4 ) 2 h 2 P n + 1 . (46) R_{+}P_n=K_{+}P_{n+1}=\frac{(2n+3)(2n+4)^2}{h^2}P_{n+1}. \tag*{(46)} R + P n = K + P n + 1 = h 2 ( 2 n + 3 ) ( 2 n + 4 ) 2 P n + 1 . (46)
Summation over n n n gives
( R + − K + ) P = − u 1 + u , ( Δ 0 − K + ) P = − u 1 + u . (47) (R_{+}-K_{+})P=-\frac{u}{1+u}, \qquad(\Delta_0-K_{+})P=-\frac{u}{1+u}. \tag*{(47)} ( R + − K + ) P = − 1 + u u , ( Δ 0 − K + ) P = − 1 + u u . (47)
Consequently ( Δ 0 − R + ) P = 0 (\Delta_0-R_{+})P=0 ( Δ 0 − R + ) P = 0 . Exchanging R + R_{+} R + and Δ 0 \Delta_0 Δ 0 changes the sign of this last operator and preserves its homogeneous equation. This is a simple example of how a nontrivial permutation of components can coexist with the preservation of the complete equation.
For the convention [ A , B ] = A B − B A [A,B]=AB-BA [ A , B ] = A B − B A , homogeneity gives
[ K + , D ] = 2 K + , [ R + , D ] = 2 R + , [ Δ 0 , D ] = 0 , (48) [K_{+},D]=2K_{+}, \qquad[R_{+},D]=2R_{+}, \qquad[\Delta_0,D]=0, \tag*{(48)} [ K + , D ] = 2 K + , [ R + , D ] = 2 R + , [ Δ 0 , D ] = 0 , (48)
and hence
[ Δ 0 − R + , D ] = − 2 R + , [ Δ 0 − K + , D ] = − 2 K + . (49) [\Delta_0-R_{+},D]=-2R_{+}, \qquad[\Delta_0-K_{+},D]=-2K_{+}. \tag*{(49)} [ Δ 0 − R + , D ] = − 2 R + , [ Δ 0 − K + , D ] = − 2 K + . (49)
These identities identify the Euler-operator structure of the recursions. A connection with conformal field theory would require additional structure beyond these homogeneity relations.
A homogeneous third-order operator triple
The projective nature of the differential system is most transparent for the centered algebraic function τ \tau τ . Extend it from the chart c = i c=i c = i to a homogeneous function of degree two in a , b , c a,b,c a , b , c , and define
O 1 = a ∂ a 3 , O 2 = b ∂ b 3 , O 3 = c ∂ c 3 . (50) O_1=a\partial_a^3, \qquad O_2=b\partial_b^3, \qquad O_3=c\partial_c^3. \tag*{(50)} O 1 = a ∂ a 3 , O 2 = b ∂ b 3 , O 3 = c ∂ c 3 . (50)
The full system is
O 1 τ = O 2 τ = O 3 τ . (51) O_1\tau=O_2\tau=O_3\tau. \tag*{(51)} O 1 τ = O 2 τ = O 3 τ . (51)
Permutations of the homogeneous coordinates permute these operators literally. A sign change of a coordinate leaves the corresponding operator unchanged, since the coordinate and its third derivative supply four sign factors in total.
To recover the affine system, use homogeneity to write c ∂ c τ = ( 2 − D ) τ c\partial_c\tau=(2-D)\tau c ∂ c τ = ( 2 − D ) τ . It follows that
c 3 ∂ c 3 τ = ( 2 − D ) ( 1 − D ) ( − D ) τ = − D ( D − 1 ) ( D − 2 ) τ . (52) c^3\partial_c^3\tau=(2-D)(1-D)(-D)\tau=-D(D-1)(D-2)\tau. \tag*{(52)} c 3 ∂ c 3 τ = ( 2 − D ) ( 1 − D ) ( − D ) τ = − D ( D − 1 ) ( D − 2 ) τ . (52)
Since c 2 = − 1 c^2=-1 c 2 = − 1 on the chosen chart, O 3 τ = Δ 0 τ O_3\tau=\Delta_0\tau O 3 τ = Δ 0 τ . The affine polynomial subtracted in (28) is annihilated by all three third-order operators. Thus (51) becomes
u g u u u = h g h h h = Δ 0 g . (53) u g_{uuu}=h g_{hhh}=\Delta_0g. \tag*{(53)} u g uuu = h g hhh = Δ 0 g . (53)
Both independent differences have been checked by exact polynomial reduction modulo (25) . The third pairwise difference is linearly dependent on the first two, as expected:
( O 1 − O 2 ) + ( O 2 − O 3 ) + ( O 3 − O 1 ) = 0. (54) (O_1-O_2)+(O_2-O_3)+(O_3-O_1)=0. \tag*{(54)} ( O 1 − O 2 ) + ( O 2 − O 3 ) + ( O 3 − O 1 ) = 0. (54)
A permuted triple therefore need not provide three independent equations. The role of the triple is to exhibit the coordinate symmetry of the differential system.
An additional second-order invariant equation
The algebraic function also satisfies differential identities of a form different from the raising recursion. Write (32) as
w 4 + a w 2 + b w + c = 0 , (55) w^4+aw^2+bw+c=0, \tag*{(55)} w 4 + a w 2 + b w + c = 0 , (55)
where
a = 12 r − 1 2 , b = 4 d , c = 1 16 + r 2 − 3 r 2 + 18 s . (56) a=\frac{12r-1}{2},\qquad b=4d,\qquad c=\frac{1}{16}+\frac{r}{2}-3r^2+18s. \tag*{(56)} a = 2 12 r − 1 , b = 4 d , c = 16 1 + 2 r − 3 r 2 + 18 s . (56)
On a generic chart with a b ≠ 0 ab\ne0 ab = 0 , introduce
α = b 2 a 3 , β = c a 2 , Y G = b w a 2 . (57) \alpha=\frac{b^2}{a^3},\qquad\beta=\frac{c}{a^2},\qquad Y_G=\frac{bw}{a^2}. \tag*{(57)} α = a 3 b 2 , β = a 2 c , Y G = a 2 b w . (57)
All three quantities are invariant under the signed-permutation action. They satisfy the canonical quartic
Y G 4 + α Y G 2 + α 2 Y G + α 2 β = 0 , (58) Y_G^4+\alpha Y_G^2+\alpha^2Y_G+\alpha^2\beta=0, \tag*{(58)} Y G 4 + α Y G 2 + α 2 Y G + α 2 β = 0 , (58)
and the second-order partial differential equation
4 α 2 Y α α + 3 α 2 Y α β + 2 α β Y β β − 2 α Y α + 2 Y = 0 , Y = Y G . (59) 4\alpha^2Y_{\alpha\alpha}+3\alpha^2Y_{\alpha\beta}+2\alpha\beta Y_{\beta\beta}-2\alpha Y_\alpha+2Y=0,\qquad Y=Y_G. \tag*{(59)} 4 α 2 Y α α + 3 α 2 Y α β + 2 α β Y β β − 2 α Y α + 2 Y = 0 , Y = Y G . (59)
This relation can be derived without using the particular GWW parameterization. For a root f ( A , B , C ) f(A,B,C) f ( A , B , C ) of f 4 + A f 2 + B f + C = 0 f^4+Af^2+Bf+C=0 f 4 + A f 2 + B f + C = 0 , implicit differentiation gives f A C = f B B f_{AC}=f_{BB} f A C = f B B , while weighted homogeneity gives 2 A f A + 3 B f B + 4 C f C = f 2Af_A+3Bf_B+4Cf_C=f 2 A f A + 3 B f B + 4 C f C = f . Differentiating the latter relation with respect to C C C , eliminating f A C f_{AC} f A C , and fixing A = 1 A=1 A = 1 yields
2 f B B + 3 B f B C + 4 C f C C + 3 f C = 0. (60) 2f_{BB}+3Bf_{BC}+4Cf_{CC}+3f_C=0. \tag*{(60)} 2 f B B + 3 B f B C + 4 C f C C + 3 f C = 0. (60)
The substitution α = B 2 \alpha=B^2 α = B 2 , β = C \beta=C β = C , Y = B f Y=Bf Y = B f gives (59) . An independent polynomial-remainder calculation verifies the same identity against (58) .
The map from ( r , s ) (r,s) ( r , s ) to ( α , β ) (\alpha,\beta) ( α , β ) has Jacobian
det ∂ ( α , β ) ∂ ( r , s ) = − 4096 ( 27 s − 1 ) 2 ( 12 r − 1 ) 6 . (61) \det\frac{\partial(\alpha,\beta)}{\partial(r,s)}=-\frac{4096(27s-1)^2}{(12r-1)^6}. \tag*{(61)} det ∂ ( r , s ) ∂ ( α , β ) = − ( 12 r − 1 ) 6 4096 ( 27 s − 1 ) 2 . (61)
and is locally nonsingular generically. Nevertheless, (59) is an additional differential identity, rather than a coordinate transform that lowers the order of (34) . The two statements have different content. In particular, the existence of (59) does not replace the explicit third-derivative raising formula (36) by an explicit second-derivative raising formula.
The integrated free energy and its real branch
The rational coordinate (19) also makes the free-energy integration elementary. Define A = 1 + u + ( u − 1 ) ρ 2 A=1+u+(u-1)\rho^2 A = 1 + u + ( u − 1 ) ρ 2 . With logarithms continued consistently along the chosen branch, one obtains
Ψ G = u log ρ + u 2 2 log ( 1 + ρ 2 ) − u 2 + 1 2 log A + 1 2 log ( 1 − ρ 2 ) + 2 u 1 − ρ 2 + u + 1 1 + ρ 2 − 1 ( 1 + ρ 2 ) 2 + u 2 2 log u − 3 u 2 . (62) \begin{aligned}
\Psi_G={}&u\log\rho+\frac{u^2}{2}\log(1+\rho^2)-\frac{u^2+1}{2}\log A+\frac{1}{2}\log(1-\rho^2)\\
&+\frac{2u}{1-\rho^2}+\frac{u+1}{1+\rho^2}-\frac{1}{(1+\rho^2)^2}+\frac{u^2}{2}\log u-\frac{3u}{2}. \tag*{(62)}
\end{aligned} Ψ G = u log ρ + 2 u 2 log ( 1 + ρ 2 ) − 2 u 2 + 1 log A + 2 1 log ( 1 − ρ 2 ) + 1 − ρ 2 2 u + 1 + ρ 2 u + 1 − ( 1 + ρ 2 ) 2 1 + 2 u 2 log u − 2 3 u . (62)
Direct differentiation gives ∂ ρ Ψ G = ( 2 g / h ) ∂ ρ h \partial_\rho\Psi_G=(2g/h)\partial_\rho h ∂ ρ Ψ G = ( 2 g / h ) ∂ ρ h . The integration constant is fixed by the leading term of the finite-N N N sector integral,
D q G ( N , h ) = ( N h 2 ) N q ∏ j = 0 N − 1 j ! ( q + j ) ! [ 1 + O ( h 2 ) ] , q ≥ 0. (63) D_q^G(N,h)=\left(\frac{Nh}{2}\right)^{Nq}\prod_{j=0}^{N-1}\frac{j!}{(q+j)!}\left[1+O(h^2)\right],\qquad q\ge0. \tag*{(63)} D q G ( N , h ) = ( 2 N h ) N q j = 0 ∏ N − 1 ( q + j )! j ! [ 1 + O ( h 2 ) ] , q ≥ 0. (63)
Its planar limit gives
C ( u ) = 3 u 2 − u log 2 + u 2 2 log u − ( 1 + u ) 2 2 log ( 1 + u ) (64) \mathcal{C}(u)=\frac{3u}{2}-u\log2+\frac{u^2}{2}\log u-\frac{(1+u)^2}{2}\log(1+u) \tag*{(64)} C ( u ) = 2 3 u − u log 2 + 2 u 2 log u − 2 ( 1 + u ) 2 log ( 1 + u ) (64)
in (40) . An elementary primitive without this parameter-dependent constant would describe a different sector normalization.
At zero charge the familiar two GWW branches are recovered as boundary limits of the parameterization:
Ψ G ( 0 , h ) = { h 2 / 4 , 0 ≤ h ≤ 1 , h − 1 2 log h − 3 4 , h ≥ 1. (65) \Psi_{\mathrm{G}}(0,h)=
\begin{cases}
h^2/4, & 0\le h\le1,\\
h-\frac{1}{2}\log h-\frac{3}{4}, & h\ge1.
\end{cases} \tag*{(65)} Ψ G ( 0 , h ) = { h 2 /4 , h − 2 1 log h − 4 3 , 0 ≤ h ≤ 1 , h ≥ 1. (65)
For h > 1 h>1 h > 1 , the limit is u ↓ 0 u\downarrow0 u ↓ 0 together with ρ ↑ 1 \rho\uparrow1 ρ ↑ 1 ; substitution of u = 0 u=0 u = 0 at fixed ρ < 1 \rho<1 ρ < 1 retains only the weak branch. This illustrates how the same algebraic curve can contain different physical boundary limits.
The quartic discriminant is
Disc G P 4 = u 2 h 2 2 16 [ ( 1 − u 2 − h 2 ) 3 − 27 u 2 h 2 ] 3 . (66) \operatorname{Disc}_{G} P_4=\frac{u^2h^2}{2^{16}}\left[(1-u^2-h^2)^3-27u^2h^2\right]^3. \tag*{(66)} Disc G P 4 = 2 16 u 2 h 2 [ ( 1 − u 2 − h 2 ) 3 − 27 u 2 h 2 ] 3 . (66)
The nontrivial component in the positive quadrant is u 2 / 3 + h 2 / 3 = 1 u^{2/3}+h^{2/3}=1 u 2/3 + h 2/3 = 1 . If a = u 2 / 3 a=u^{2/3} a = u 2/3 on this curve, the colliding algebraic root and the continued physical root are
g c o l l = 1 − 3 a 4 , g p h = − ( 1 − a ) 2 ( 2 a + 1 ) 4 + 1 2 ( a 2 − a + 1 ) 3 / 2 . (67) g_{\mathrm{coll}}=\frac{1-3a}{4},\qquad
g_{\mathrm{ph}}=-\frac{(1-a)^2(2a+1)}{4}+\frac{1}{2}(a^2-a+1)^{3/2}. \tag*{(67)} g coll = 4 1 − 3 a , g ph = − 4 ( 1 − a ) 2 ( 2 a + 1 ) + 2 1 ( a 2 − a + 1 ) 3/2 . (67)
They coincide only at ( u , h ) = ( 0 , 1 ) (u,h)=(0,1) ( u , h ) = ( 0 , 1 ) . Thus the vanishing of the polynomial discriminant does not identify an additional physical singularity of the positive-charge branch by itself.
For completeness, the fixed-sector expansion at u = 0 u=0 u = 0 and 0 < h < 1 0<h<1 0 < h < 1 can be written in terms of
a ( h ) = log 1 + 1 − h 2 h − 1 − h 2 (68) a(h)=\log\frac{1+\sqrt{1-h^2}}{h}-\sqrt{1-h^2} \tag*{(68)} a ( h ) = log h 1 + 1 − h 2 − 1 − h 2 (68)
as
Ψ G = h 2 4 − a ( h ) u + u 2 2 log u − [ 3 4 + 1 4 log ( 1 − h 2 ) ] u 2 + O ( u 3 ) . (69) \Psi_{\mathrm{G}}=\frac{h^2}{4}-a(h)u+\frac{u^2}{2}\log u-\left[\frac{3}{4}+\frac{1}{4}\log(1-h^2)\right]u^2+O(u^3). \tag*{(69)} Ψ G = 4 h 2 − a ( h ) u + 2 u 2 log u − [ 4 3 + 4 1 log ( 1 − h 2 ) ] u 2 + O ( u 3 ) . (69)
The nonanalytic term u 2 log u u^2\log u u 2 log u is a property of this fixed-sector boundary expansion; its thermodynamic interpretation requires a separate branch and sector selection.
A conjectural description-tree refinement
The numerator polynomials in the GWW recursion admit a concrete conjectural interpretation by the description trees of type ( 2 , 2 ) (2,2) ( 2 , 2 ) of Cori and Schaeffer [14 ] , counted by OEIS A000309 [15 ] .
Let T s \mathcal{T}_s T s be this family with s ≥ 1 s\ge1 s ≥ 1 edges: each leaf has label 2, each nonroot internal vertex v v v has an integer label 2 ≤ ℓ ( v ) ≤ 2 + ∑ w c h i l d o f v ℓ ( w ) 2\le\ell(v)\le2+\sum_{w\ \mathrm{child\ of}\ v}\ell(w) 2 ≤ ℓ ( v ) ≤ 2 + ∑ w child of v ℓ ( w ) , and the root attains the upper bound. Call an internal vertex maximal when this upper bound is attained, and put ν ( τ ) = # { leaves } + # { maximal internal vertices } \nu(\tau)=\#\{\text{leaves}\}+\#\{\text{maximal internal vertices}\} ν ( τ ) = # { leaves } + # { maximal internal vertices } ; the root therefore always contributes. With the row indexing R s ( u ) = u ∑ n = 1 s T ( s , n ) u n − 1 / ( 1 + u ) 3 s + 1 R_s(u)=u\sum_{n=1}^{s}T(s,n)u^{n-1}/(1+u)^{3s+1} R s ( u ) = u ∑ n = 1 s T ( s , n ) u n − 1 / ( 1 + u ) 3 s + 1 , the claim is
B s ( z ) : = ∑ n = 1 s ( − 1 ) n − 1 T ( s , n ) z n − 1 ≡ c o n j . ∑ τ ∈ T s z ν ( τ ) − 2 , R s ( u ) = u B s ( − u ) ( 1 + u ) 3 s + 1 . (70) B_s(z):=\sum_{n=1}^{s}(-1)^{n-1}T(s,n)z^{n-1}\mathrel{\overset{\mathrm{conj.}}{\equiv}}\sum_{\tau\in\mathcal{T}_s}z^{\nu(\tau)-2},
\qquad
R_s(u)=\frac{uB_s(-u)}{(1+u)^{3s+1}}. \tag*{(70)} B s ( z ) := n = 1 ∑ s ( − 1 ) n − 1 T ( s , n ) z n − 1 ≡ conj. τ ∈ T s ∑ z ν ( τ ) − 2 , R s ( u ) = ( 1 + u ) 3 s + 1 u B s ( − u ) . (70)
Independent enumeration agrees with the differential recursion for 1 ≤ s ≤ 9 1\le s\le9 1 ≤ s ≤ 9 ; the first rows are B 1 = 1 B_1=1 B 1 = 1 , B 2 = 2 + 2 z B_2=2+2z B 2 = 2 + 2 z , B 3 = 5 + 14 z + 5 z 2 B_3=5+14z+5z^2 B 3 = 5 + 14 z + 5 z 2 , and B 4 = 14 + 74 z + 74 z 2 + 14 z 3 B_4=14+74z+74z^2+14z^3 B 4 = 14 + 74 z + 74 z 2 + 14 z 3 . Thus the alternating signs of T ( s , n ) T(s,n) T ( s , n ) would record a signed statistic rather than negative cardinalities. The conjecture refines the row-sum relation B s ( 1 ) = 2 s ( 3 s ) ! / [ ( s + 1 ) ! ( 2 s + 1 ) ! ] = A 000309 ( s ) B_s(1)=2^s(3s)!/[(s+1)!(2s+1)!]=\mathrm{A000309}(s) B s ( 1 ) = 2 s ( 3 s )! / [( s + 1 )! ( 2 s + 1 )!] = A000309 ( s ) in this indexing. The same unrefined numbers count rooted loopless triangulations, for which Bernardi gave a bijective enumeration using Kreweras walks [16 ] . This gives a walk connection for the row sum; it does not identify the statistic ν \nu ν on those walks. Both extreme statistic classes have C a t s = ( 2 s ) ! / [ s ! ( s + 1 ) ! ] \mathrm{Cat}_s=(2s)!/[s!(s+1)!] Cat s = ( 2 s )! / [ s ! ( s + 1 )!] members: ν = s + 1 \nu=s+1 ν = s + 1 assigns the unique maximal labeling to every rooted plane tree shape, whereas ν = 2 \nu=2 ν = 2 forces a chain whose nonmaximal labels encode a Dyck path (Figure 1 and Figure 2 ); for the classical Catalan families of plane trees and Dyck paths, see Ref. [17 ] . These endpoint counts follow directly within the tree model; the refinement of all intermediate columns remains conjectural. Finally, if B ( q , z ) = ∑ s ≥ 1 q s B s ( z ) B(q,z)=\sum_{s\geq1}q^sB_s(z) B ( q , z ) = ∑ s ≥ 1 q s B s ( z ) and t = h 2 / 4 t=h^2/4 t = h 2 /4 , the GWW series (35) gives the exact formal identity
Figure 1. All four description trees with two edges. Blue leaves and orange maximal internal vertices each contribute one to ν \nu ν ; white internal vertices do not contribute. The root is always maximal. The distribution ( 2 , 2 ) (2,2) ( 2 , 2 ) gives B 2 ( z ) = 2 + 2 z B_2(z)=2+2z B 2 ( z ) = 2 + 2 z and the signed GWW row ( 2 , − 2 ) (2,-2) ( 2 , − 2 ) .
Figure 2. The five three-edge chains in the class ν = 2 \nu=2 ν = 2 and their Dyck paths from ( 0 , 0 ) (0,0) ( 0 , 0 ) to ( 3 , 3 ) (3,3) ( 3 , 3 ) , using north and east steps above the diagonal. Read the labels from the leaf upward, excluding the root, as ℓ 0 , … , ℓ s − 1 \ell_0,\ldots,\ell_{s-1} ℓ 0 , … , ℓ s − 1 and set a i = ℓ i − 2 a_i=\ell_i-2 a i = ℓ i − 2 . The nonmaximality conditions are precisely a 0 = 0 a_0=0 a 0 = 0 and 0 ≤ a i ≤ a i − 1 + 1 0\leq a_i\leq a_{i-1}+1 0 ≤ a i ≤ a i − 1 + 1 ; the ( i + 1 ) (i+1) ( i + 1 ) st north step has i − a i i-a_i i − a i preceding east steps. This is an explicit bijection for the endpoint class, not a claimed path bijection for the intermediate statistic classes.
g ( − z , 2 t ) + z 2 = t 1 − z + z t 1 − z B ( t ( 1 − z ) 3 , z ) . (71) g(-z,2\sqrt{t})+\frac{z}{2}=\frac{t}{1-z}+\frac{zt}{1-z}\mathcal{B}\left(\frac{t}{(1-z)^3},z\right). \tag*{(71)} g ( − z , 2 t ) + 2 z = 1 − z t + 1 − z z t B ( ( 1 − z ) 3 t , z ) . (71)
Under the conjecture, q q q marks edges and z z z marks ν − 2 \nu-2 ν − 2 in B \mathcal{B} B . This identifies the tree refinement inside the algebraic GWW generating function while leaving a direct statistic-preserving bijection with the matrix-integral expansion as an additional problem.
An algebraic check of the finite tree comparison. For q = 2 q=2 q = 2 , the uniform covering of Ref. [12 Theorem C.1] provides a second way to compute the GWW numerator polynomials. In the present normalization, twice its reduced function, expanded in h 2 / 4 h^2/4 h 2 /4 , reproduces R s R_s R s through s = 9 s=9 s = 9 . The positive numerator at s = 9 s=9 s = 9 has coefficients
4862 , 126008 , 964264 , 3027080 , 4385332 , 3027080 , 964264 , 126008 , 4862. 4862,\ 126008,\ 964264,\ 3027080,\ 4385332,\ 3027080,\ 964264,\ 126008,\ 4862. 4862 , 126008 , 964264 , 3027080 , 4385332 , 3027080 , 964264 , 126008 , 4862.
These agree with the description-tree counts in (70) . The covering verifies the algebraic side of the comparison independently of the differential recursion; the general equality with the tree statistic remains conjectural.
The algebraic curve and functional transformations of one-dimensional QCD
One-cut loop equation and the origin of the sextic
The QCD curve can be derived directly from the planar loop equation. For the exact determinant formulation and its ’t Hooft–Veneziano limit we refer to Ref. [6 ] , published in 2025. The following calculation uses the same one-cut resolvent and discriminant construction as (3.5)–(3.8), with the exponential GWW weight replaced by the two fermion determinants. Since D q Q = D − q Q D_q^{\mathrm{Q}}=D_{-q}^{\mathrm{Q}} D q Q = D − q Q , use the equivalent insertion ( det U ) − q (\det U)^{-q} ( det U ) − q and write v = q / N = k u v=q/N=ku v = q / N = k u . Converting the unitary Vandermonde to a holomorphic eigenvalue integral gives the potential
V ( λ ) = ( 1 + k + v ) log λ − k log ( 1 + h λ ) − k log ( λ + h ) , V ′ = 1 + k + v λ − k λ + h − k λ + h − 1 . (72) V(\lambda)=(1+k+v)\log\lambda-k\log(1+h\lambda)-k\log(\lambda+h),\qquad V'=\frac{1+k+v}{\lambda}-\frac{k}{\lambda+h}-\frac{k}{\lambda+h^{-1}}. \tag*{(72)} V ( λ ) = ( 1 + k + v ) log λ − k log ( 1 + hλ ) − k log ( λ + h ) , V ′ = λ 1 + k + v − λ + h k − λ + h − 1 k . (72)
The first logarithm includes the unitary measure. Let ω ( λ ) = ∫ ρ ( ξ ) / ( λ − ξ ) d ξ \omega(\lambda)=\int\rho(\xi)/(\lambda-\xi)\,\mathrm{d}\xi ω ( λ ) = ∫ ρ ( ξ ) / ( λ − ξ ) d ξ be normalized by ω ∼ 1 / λ \omega\sim1/\lambda ω ∼ 1/ λ at infinity. The planar loop equation is ω 2 − V ′ ω + f = 0 \omega^2-V'\omega+f=0 ω 2 − V ′ ω + f = 0 , where f = ∫ [ V ′ ( λ ) − V ′ ( ξ ) ] ρ ( ξ ) / ( λ − ξ ) d ξ f=\int[V'(\lambda)-V'(\xi)]\rho(\xi)/(\lambda-\xi)\,\mathrm{d}\xi f = ∫ [ V ′ ( λ ) − V ′ ( ξ )] ρ ( ξ ) / ( λ − ξ ) d ξ . Put a = h ω ( − h ) a=h\omega(-h) a = hω ( − h ) , b = h − 1 ω ( − h − 1 ) b=h^{-1}\omega(-h^{-1}) b = h − 1 ω ( − h − 1 ) and G = x ∂ x Ψ Q G=x\partial_x\Psi_{\mathrm{Q}} G = x ∂ x Ψ Q . Expansion at infinity and direct differentiation of the determinant give
a + b = u − 1 , G = k 2 ( 1 + b − a ) , a = u 2 − G k , b = u 2 − 1 + G k . (73) a+b=u-1,\qquad G=\frac{k}{2}(1+b-a),\qquad a=\frac{u}{2}-\frac{G}{k},\qquad b=\frac{u}{2}-1+\frac{G}{k}. \tag*{(73)} a + b = u − 1 , G = 2 k ( 1 + b − a ) , a = 2 u − k G , b = 2 u − 1 + k G . (73)
The vanishing 1 / λ 1/\lambda 1/ λ coefficient of f f f eliminates ω ( 0 ) \omega(0) ω ( 0 ) , leaving
f ( λ ) = k a λ ( λ + h ) + k b λ ( λ + h − 1 ) . (74) f(\lambda)=\frac{ka}{\lambda(\lambda+h)}+\frac{kb}{\lambda(\lambda+h^{-1})}. \tag*{(74)} f ( λ ) = λ ( λ + h ) k a + λ ( λ + h − 1 ) k b . (74)
Consequently, with
D h ( λ ) = h λ 3 + ( 1 + h 2 ) λ 2 + h λ , (75) D_h(\lambda)=h\lambda^3+(1+h^2)\lambda^2+h\lambda, \tag*{(75)} D h ( λ ) = h λ 3 + ( 1 + h 2 ) λ 2 + hλ , (75)
N h ( λ ) = ( 1 − k + v ) h λ 2 + ( 1 + v ) ( 1 + h 2 ) λ + ( 1 + k + v ) h , (76) N_h(\lambda)=(1-k+v)h\lambda^2+(1+v)(1+h^2)\lambda+(1+k+v)h, \tag*{(76)} N h ( λ ) = ( 1 − k + v ) h λ 2 + ( 1 + v ) ( 1 + h 2 ) λ + ( 1 + k + v ) h , (76)
the squared spectral function is
( V ′ − 2 ω ) 2 = Q h ( λ ) D h ( λ ) 2 , (77) (V'-2\omega)^2=\frac{Q_h(\lambda)}{D_h(\lambda)^2}, \tag*{(77)} ( V ′ − 2 ω ) 2 = D h ( λ ) 2 Q h ( λ ) , (77)
Q h ( λ ) = N h ( λ ) 2 − 4 D h ( λ ) [ ( v − k ) h λ + v 2 ( 1 + h 2 ) − k h 2 + G ( h 2 − 1 ) ] . Q_h(\lambda)=N_h(\lambda)^2-4D_h(\lambda)\left[(v-k)h\lambda+\frac{v}{2}(1+h^2)-kh^2+G(h^2-1)\right]. Q h ( λ ) = N h ( λ ) 2 − 4 D h ( λ ) [ ( v − k ) hλ + 2 v ( 1 + h 2 ) − k h 2 + G ( h 2 − 1 ) ] .
On a regular one-cut branch the spectral numerator has the form ( A λ + B ) 2 ( λ − λ 1 ) ( λ − λ 2 ) (A\lambda+B)^2(\lambda-\lambda_1)(\lambda-\lambda_2) ( A λ + B ) 2 ( λ − λ 1 ) ( λ − λ 2 ) . Thus Disc λ Q h = 0 \operatorname{Disc}_{\lambda}Q_h=0 Disc λ Q h = 0 . For E 0 = k 2 + ( k + 1 ) 2 + v 2 E_0=k^2+(k+1)^2+v^2 E 0 = k 2 + ( k + 1 ) 2 + v 2 , z = ( 1 − h 2 ) / ( 1 + h 2 ) z=(1-h^2)/(1+h^2) z = ( 1 − h 2 ) / ( 1 + h 2 ) and
W = z G − k z / 2 − k 2 / 2 E 0 , (78) W=\frac{zG-kz/2-k^2/2}{E_0}, \tag*{(78)} W = E 0 z G − k z /2 − k 2 /2 , (78)
this discriminant is exactly
Disc λ Q h = 2 12 h 4 E 0 6 ( 1 − h 2 ) 2 ( 1 + h 2 ) 6 P 6 ( W ) . (79) \operatorname{Disc}_{\lambda}Q_h=2^{12}h^4E_0^6(1-h^2)^2(1+h^2)^6P_6(W). \tag*{(79)} Disc λ Q h = 2 12 h 4 E 0 6 ( 1 − h 2 ) 2 ( 1 + h 2 ) 6 P 6 ( W ) . (79)
Here P 6 P_6 P 6 is the monic sextic below; its invariant arguments are r = [ k 2 ( k + 1 ) 2 + ( k 2 + ( k + 1 ) 2 ) v 2 ] / E 0 2 r=[k^2(k+1)^2+(k^2+(k+1)^2)v^2]/E_0^2 r = [ k 2 ( k + 1 ) 2 + ( k 2 + ( k + 1 ) 2 ) v 2 ] / E 0 2 , s = k 2 ( k + 1 ) 2 v 2 / E 0 3 s=k^2(k+1)^2v^2/E_0^3 s = k 2 ( k + 1 ) 2 v 2 / E 0 3 and t = z 2 t=z^2 t = z 2 . The identity was checked as a full polynomial identity in k k k , v v v , h h h , W W W , not through a truncated source series. It supplies the missing spectral derivation under the stated one-cut and planar-factorization assumptions. The discriminant condition includes other algebraic sheets; the small-h h h boundary condition and contour continuation are still required to select a physical branch. Colliding-pole limits such as h = 1 h=1 h = 1 are taken by continuation, rather than by cancelling vanishing factors at that point.
The sextic in projective variables
For the QCD triple in (10) , set
E 0 = k 2 + ( k + 1 ) 2 + k 2 u 2 , z = 1 − h 2 1 + h 2 , t = z 2 , (80) E_0 = k^2 + (k+1)^2 + k^2u^2,\qquad z = \frac{1-h^2}{1+h^2},\qquad t = z^2, \tag*{(80)} E 0 = k 2 + ( k + 1 ) 2 + k 2 u 2 , z = 1 + h 2 1 − h 2 , t = z 2 , (80)
and form r r r , s s s from the same symmetric expressions (13) . The useful projective generating function is
W = k z R − k 2 / 2 E 0 = k 2 E 0 ( z h S h − k ) . (81) W = \frac{kzR-k^2/2}{E_0} = \frac{k}{2E_0}(zhS_h-k). \tag*{(81)} W = E 0 k z R − k 2 /2 = 2 E 0 k ( z h S h − k ) . (81)
This definition contains both a multiplicative normalization and an additive shift. Neither can be omitted when comparing the transformed mass derivative with an invariant algebraic variable. The resulting algebraic curve is the sextic
P 6 ( W ) = W 3 ( W 3 + 1 2 W 2 + r 4 W + s 8 ) + t P 1 ( W ) + t 2 P 2 ( W ) + t 3 Δ 256 = 0 , (82) P_6(W) = W^3\left(W^3+\frac{1}{2}W^2+\frac{r}{4}W+\frac{s}{8}\right)+\frac{tP_1(W)+t^2P_2(W)+t^3\Delta}{256}=0, \tag*{(82)} P 6 ( W ) = W 3 ( W 3 + 2 1 W 2 + 4 r W + 8 s ) + 256 t P 1 ( W ) + t 2 P 2 ( W ) + t 3 Δ = 0 , (82)
where
P 1 ( W ) = ( 16 − 192 r ) W 4 − ( 64 r + 288 s ) W 3 − ( 32 r 2 + 120 s ) W 2 − 72 r s W − 27 s 2 , (83) P_1(W) = (16-192r)W^4-(64r+288s)W^3-(32r^2+120s)W^2-72rsW-27s^2, \tag*{(83)} P 1 ( W ) = ( 16 − 192 r ) W 4 − ( 64 r + 288 s ) W 3 − ( 32 r 2 + 120 s ) W 2 − 72 r s W − 27 s 2 , (83)
P 2 ( W ) = ( − 8 r + 48 r 2 − 72 s ) W 2 + ( 8 r 2 − 48 s + 72 r s ) W + 4 r 3 − 18 r s + 54 s 2 . (84) P_2(W) = (-8r+48r^2-72s)W^2+(8r^2-48s+72rs)W+4r^3-18rs+54s^2. \tag*{(84)} P 2 ( W ) = ( − 8 r + 48 r 2 − 72 s ) W 2 + ( 8 r 2 − 48 s + 72 r s ) W + 4 r 3 − 18 r s + 54 s 2 . (84)
The discriminant Δ \Delta Δ in this formula is the symmetric invariant in (15) . Unlike the oriented GWW quartic, the sextic does not involve a choice of sign for d d d . This difference affects the normalization of the invariant generating function, although the projective parameter group is the same in both models.
For the positive fixed-sector germ, the small-h h h boundary condition is
R ( k , u , 0 ) = u − 1 2 , W ( k , u , 0 ) = k u − k − k 2 2 E 0 . (85) R(k,u,0)=\frac{u-1}{2},\qquad W(k,u,0)=\frac{ku-k-k^2}{2E_0}. \tag*{(85)} R ( k , u , 0 ) = 2 u − 1 , W ( k , u , 0 ) = 2 E 0 k u − k − k 2 . (85)
Equivalently, in the chart [ 1 : M : u ] [1:M:u] [ 1 : M : u ] with M = 1 + 1 / k M=1+1/k M = 1 + 1/ k ,
W ( 0 ) = ( M − 1 ) u − M 2 ( 1 + M 2 + u 2 ) . (86) W(0)=\frac{(M-1)u-M}{2(1+M^2+u^2)}. \tag*{(86)} W ( 0 ) = 2 ( 1 + M 2 + u 2 ) ( M − 1 ) u − M . (86)
The endpoint h = 0 h=0 h = 0 corresponds to t = 1 t=1 t = 1 , not t = 0 t=0 t = 0 . The distinction matters because the polynomial degenerates differently on the two mass slices:
P 6 ∣ t = 0 = W 3 ( W 3 + 1 2 W 2 + r 4 W + s 8 ) , (87) P_6\big|_{t=0}=W^3\left(W^3+\frac{1}{2}W^2+\frac{r}{4}W+\frac{s}{8}\right), \tag*{(87)} P 6 t = 0 = W 3 ( W 3 + 2 1 W 2 + 4 r W + 8 s ) , (87)
P 6 ∣ t = 1 = ( 4 W + 1 ) 2 256 ( 16 W 4 − 8 r W 2 − 16 s W + r 2 − 4 s ) . (88) P_6\big|_{t=1}=\frac{(4W+1)^2}{256}\left(16W^4-8rW^2-16sW+r^2-4s\right). \tag*{(88)} P 6 t = 1 = 256 ( 4 W + 1 ) 2 ( 16 W 4 − 8 r W 2 − 16 s W + r 2 − 4 s ) . (88)
A repeated polynomial root on one of these slices does not specify which regular branch approaches it. Boundary data and continuation remain part of the fixed-sector definition.
For the expansions and limits below, set
K = k ( k + 1 ) , E = 2 K + 1 + v 2 , (89) K=k(k+1),\qquad E=2K+1+v^2, \tag*{(89)} K = k ( k + 1 ) , E = 2 K + 1 + v 2 , (89)
and define a polynomial by clearing the projective denominators of the sextic:
Q ( T ; K , v , x ) = 16 ( 1 + x ) 6 E 6 P 6 ( w ; r , s , t ) . (90) Q(T;K,v,x)=16(1+x)^6E^6P_6(w;r,s,t). \tag*{(90)} Q ( T ; K , v , x ) = 16 ( 1 + x ) 6 E 6 P 6 ( w ; r , s , t ) . (90)
where
w = 2 T + v ( 1 − x ) − K ( 1 + x ) 2 ( 1 + x ) E , r = K 2 + ( 2 K + 1 ) v 2 E 2 , (91) w=\frac{2T+v(1-x)-K(1+x)}{2(1+x)E},\qquad r=\frac{K^2+(2K+1)v^2}{E^2}, \tag*{(91)} w = 2 ( 1 + x ) E 2 T + v ( 1 − x ) − K ( 1 + x ) , r = E 2 K 2 + ( 2 K + 1 ) v 2 , (91)
s = K 2 v 2 E 3 , t = ( 1 − x ) 2 ( 1 + x ) 2 . (92) s=\frac{K^2v^2}{E^3},\qquad t=\frac{(1-x)^2}{(1+x)^2}. \tag*{(92)} s = E 3 K 2 v 2 , t = ( 1 + x ) 2 ( 1 − x ) 2 . (92)
The relation to the original QCD derivative is
T = ( 1 − x ) ( G Q − v 2 ) + k x . (93) T=(1-x)\left(G_Q-\frac{v}{2}\right)+kx. \tag*{(93)} T = ( 1 − x ) ( G Q − 2 v ) + k x . (93)
Independent finite-N N N Toeplitz check
For integer N f N_f N f the exact determinant representation of one-dimensional QCD [6 ] gives, for 0 ≤ ℓ ≤ N f 0\leq\ell\leq N_f 0 ≤ ℓ ≤ N f ,
c ℓ ( h ) = ∑ j = 0 N f − ℓ ( N f j + ℓ ) ( N f j ) h 2 j + ℓ , c − ℓ = c ℓ . (94) c_\ell(h)=\sum_{j=0}^{N_f-\ell}\binom{N_f}{j+\ell}\binom{N_f}{j}h^{2j+\ell},\qquad c_{-\ell}=c_\ell. \tag*{(94)} c ℓ ( h ) = j = 0 ∑ N f − ℓ ( j + ℓ N f ) ( j N f ) h 2 j + ℓ , c − ℓ = c ℓ . (94)
Set T i j = c q + i − j T_{ij}=c_{q+i-j} T ij = c q + i − j and A i j = x ∂ x c q + i − j A_{ij}=x\partial_xc_{q+i-j} A ij = x ∂ x c q + i − j . The independent observable is
G N = 1 N 2 Tr ( T − 1 A ) , D Q q = det T . (95) G_N=\frac{1}{N^2}\operatorname{Tr}(T^{-1}A),\qquad D_Q^q=\det T. \tag*{(95)} G N = N 2 1 Tr ( T − 1 A ) , D Q q = det T . (95)
Calculations at 150 decimal digits give the following comparison. The extrapolation fits G N = G ∞ + a / N 2 + b / N 4 G_N=G_\infty+a/N^2+b/N^4 G N = G ∞ + a / N 2 + b / N 4 using N = 32 , 48 , 64 N=32,48,64 N = 32 , 48 , 64 in the first two rows and N = 24 , 36 , 48 N=24,36,48 N = 24 , 36 , 48 in the last.
Table 1.
The finite matrices are constructed without using the sextic. For example, at ( 1 , 1 / 2 , 3 / 5 ) (1,1/2,3/5) ( 1 , 1/2 , 3/5 ) the unextrapolated N = 64 N=64 N = 64 value is 0.368634327577591. These regular-point checks support the selected physical branch in addition to the spectral derivation. The quoted differences are observed extrapolation differences, not rigorous error bounds or a global saddle-dominance proof.
Returning to the physical affine chart
The physical QCD coordinates satisfy b − a = 1 b-a=1 b − a = 1 . After a permutation or sign change, let the raw coordinates be ( a ~ , b ~ , c ~ ) (\tilde{a},\tilde{b},\tilde{c}) ( a ~ , b ~ , c ~ ) . Restoring the physical chart requires
λ = 1 b ~ − a ~ , k ′ = λ a ~ , k ′ u ′ = λ c ~ . (96) \lambda=\frac{1}{\tilde{b}-\tilde{a}},\qquad k'=\lambda\tilde{a},\qquad k'u'=\lambda\tilde{c}. \tag*{(96)} λ = b ~ − a ~ 1 , k ′ = λ a ~ , k ′ u ′ = λ c ~ . (96)
Thus E 0 ′ = λ 2 E 0 E'_0=\lambda^2E_0 E 0 ′ = λ 2 E 0 , whereas r r r , s s s , t t t , and the transported root W W W are unchanged. For pure transpositions the affine maps are
T a b : ( k , u ) ⟼ ( − k − 1 , k u k + 1 ) , (97) T_{ab}:(k,u)\longmapsto\left(-k-1,\frac{ku}{k+1}\right), \tag*{(97)} T ab : ( k , u ) ⟼ ( − k − 1 , k + 1 k u ) , (97)
T b c : ( k , u ) ⟼ ( 1 u − 1 , k + 1 k ) , (98) T_{bc}:(k,u)\longmapsto\left(\frac{1}{u-1},\frac{k+1}{k}\right), \tag*{(98)} T b c : ( k , u ) ⟼ ( u − 1 1 , k k + 1 ) , (98)
T a c : ( k , u ) ⟼ ( k u k + 1 − k u , 1 u ) . (99) T_{ac}:(k,u)\longmapsto\left(\frac{ku}{k+1-ku},\frac{1}{u}\right). \tag*{(99)} T a c : ( k , u ) ⟼ ( k + 1 − k u k u , u 1 ) . (99)
The frequently used variant u ′ = − k u / ( k + 1 ) u'=-ku/(k+1) u ′ = − k u / ( k + 1 ) in the first map corresponds to the signed operation ( a , b , c ) ⟼ ( b , a , − c ) (a,b,c)\longmapsto(b,a,-c) ( a , b , c ) ⟼ ( b , a , − c ) . Both versions exchange A 1 A_1 A 1 and A 2 A_2 A 2 . Their difference is a charge sign change, which belongs to the kernel of the action on squared coordinate labels. Keeping the signed lift explicit prevents a sign convention from being mistaken for a different projective symmetry.
The defining relation (81) determines the transformed mass derivative. Integrating that derivative at fixed external parameters gives
S ( k ′ , u ′ ; h ) = λ 2 k k ′ S ( k , u ; h ) + k ′ 2 − λ 2 k 2 k ′ ℓ ( h ) + C T ( k , u ) , ℓ ( h ) = log h 1 − h 2 . (100) S(k',u';h)=\frac{\lambda^2 k}{k'}S(k,u;h)+\frac{k'^2-\lambda^2 k^2}{k'}\ell(h)+C_T(k,u),\qquad\ell(h)=\log\frac{h}{1-h^2}. \tag*{(100)} S ( k ′ , u ′ ; h ) = k ′ λ 2 k S ( k , u ; h ) + k ′ k ′2 − λ 2 k 2 ℓ ( h ) + C T ( k , u ) , ℓ ( h ) = log 1 − h 2 h . (100)
The logarithm follows from h ℓ h = ( 1 + h 2 ) / ( 1 − h 2 ) = z − 1 h\ell_h=(1+h^2)/(1-h^2)=z^{-1} h ℓ h = ( 1 + h 2 ) / ( 1 − h 2 ) = z − 1 . The function C T C_T C T is independent of h h h and carries the integration freedom left by the mass derivative. Its value depends on the normalization and the continued logarithmic sheets. The group relations impose corresponding consistency conditions on these additive functions; the differential equation in h h h alone does not determine them.
Writing T ∗ S = α T S + β T ℓ + C T T^*S=\alpha_T S+\beta_T\ell+C_T T ∗ S = α T S + β T ℓ + C T , the coefficients are
T α T β T a b − k / ( k + 1 ) − ( 2 k + 1 ) / ( k + 1 ) b c 1 / [ k ( u − 1 ) ] 0 a c 1 / [ u ( k + 1 − k u ) ] k ( u 2 − 1 ) / [ u ( k + 1 − k u ) ] (101) \begin{array}{c|c|c}
T & \alpha_T & \beta_T \\
\hline
ab & -k/(k+1) & -(2k+1)/(k+1) \\
bc & 1/[k(u-1)] & 0 \\
ac & 1/[u(k+1-ku)] & k(u^2-1)/[u(k+1-ku)] \tag*{(101)}
\end{array} T ab b c a c α T − k / ( k + 1 ) 1/ [ k ( u − 1 )] 1/ [ u ( k + 1 − k u )] β T − ( 2 k + 1 ) / ( k + 1 ) 0 k ( u 2 − 1 ) / [ u ( k + 1 − k u )] (101)
The signed a b ab ab variant has the same displayed coefficients when the same algebraic root is transported. These affine relations concern the analytic free energy reconstructed from W W W , not an independent selection of a physical saddle at each transformed parameter value.
The sign operations can also be written directly in the physical chart. For example, b ↦ − b b\mapsto-b b ↦ − b gives k ′ = − k / ( 2 k + 1 ) k'=-k/(2k+1) k ′ = − k / ( 2 k + 1 ) and u ′ = u u'=u u ′ = u , while c ↦ − c c\mapsto-c c ↦ − c gives u ′ = − u u'=-u u ′ = − u . Together with (97) , these generate the full projective signed-permutation group.
There is a further mass inversion. At finite N N N ,
D q Q ( h − 1 ) = h − 2 N N f D q Q ( h ) , (102) D_q^{\mathrm Q}(h^{-1})=h^{-2NN_f}D_q^{\mathrm Q}(h), \tag*{(102)} D q Q ( h − 1 ) = h − 2 N N f D q Q ( h ) , (102)
which follows by factoring h − N det U h^{-N}\det U h − N det U and h − N det U − 1 h^{-N}\det U^{-1} h − N det U − 1 from the two determinants. In the normalization (6) , this yields S ( h − 1 ) = S ( h ) S(h^{-1})=S(h) S ( h − 1 ) = S ( h ) with consistent logarithms. It sends G Q ↦ k − G Q G_{\mathrm Q}\mapsto k-G_{\mathrm Q} G Q ↦ k − G Q and R ↦ − R R\mapsto-R R ↦ − R , while z ↦ − z z\mapsto-z z ↦ − z ; hence t t t and W W W are fixed. This inversion commutes with the parameter permutations and has a different origin from the affine chart transformations.
A homogeneous potential for the free energy
The affine law (100) suggests subtracting its specific logarithmic term before discussing the free-energy operators. Define
Φ ( k , u ; h ) = k S − k 2 ℓ ( h ) , S ^ = Φ E 0 . (103) \Phi(k,u;h)=kS-k^2\ell(h),\qquad\widehat{S}=\frac{\Phi}{E_0}. \tag*{(103)} Φ ( k , u ; h ) = k S − k 2 ℓ ( h ) , S = E 0 Φ . (103)
Equation (100) becomes Φ ′ = λ 2 Φ + k ′ C T \Phi'=\lambda^2\Phi+k'C_T Φ ′ = λ 2 Φ + k ′ C T . Thus S ^ \widehat{S} S is invariant up to an h h h -independent function. This is exactly the ambiguity to which a differential equation containing a mass derivative is insensitive. Moreover,
W = z h 2 ∂ h S ^ = t ( t − 1 ) ∂ t S ^ . (104) W=\frac{zh}{2}\partial_h\widehat{S}=t(t-1)\partial_t\widehat{S}. \tag*{(104)} W = 2 z h ∂ h S = t ( t − 1 ) ∂ t S . (104)
The logarithmic subtraction is therefore determined by the desired projective derivative; it is not an arbitrary simplification of the free energy.
To give Φ \Phi Φ a homogeneous meaning away from the physical chart, let d 0 = b − a d_0=b-a d 0 = b − a and set
Φ ( a , b , c ; h ) = a ( b − a ) S ( a b − a , c b − a ; h ) − a 2 ℓ ( h ) . (105) \Phi(a,b,c;h)=a(b-a)S\left(\frac{a}{b-a},\frac{c}{b-a};h\right)-a^2\ell(h). \tag*{(105)} Φ ( a , b , c ; h ) = a ( b − a ) S ( b − a a , b − a c ; h ) − a 2 ℓ ( h ) . (105)
This expression is homogeneous of degree two in ( a , b , c ) (a,b,c) ( a , b , c ) , or degree one in ( A 1 , A 2 , A 3 ) (A_1,A_2,A_3) ( A 1 , A 2 , A 3 ) on a chosen local square-root chart. The homogeneity relation supplies the redundant scale equation when three Euler derivatives are used to describe a system with only two independent external parameter ratios.
Differential recursions and the QCD coefficient pyramid
The QCD recurrence is most transparent after resumming the small-charge series, rather than comparing successive powers of the bare mass parameter. We first give the row and column forms of the recurrence and describe the boundary data. The projective derivation then relates the resulting PDE to the sextic. Appendix § 11 retains the alternative recursions for the original mass and free-energy coefficients.
The small-charge chart and the row recursion
We now expand the selected root of the QCD sextic (82) in the charge per color v = k u v=ku v = k u . The following normalization organizes that expansion into a coefficient pyramid:
K = k ( k + 1 ) , v = k u , x = h 2 , τ = K x 1 − x , t p = 4 τ , ζ = − v 1 − x . (106) K=k(k+1),\qquad v=ku,\qquad x=h^2,\qquad\tau=\frac{Kx}{1-x},\qquad t_p=4\tau,\qquad\zeta=-\frac{v}{1-x}. \tag*{(106)} K = k ( k + 1 ) , v = k u , x = h 2 , τ = 1 − x K x , t p = 4 τ , ζ = − 1 − x v . (106)
The resummed variable t p t_p t p is distinct from the invariant mass coordinate t t t in (80) . Define
G Q = v 2 + k 2 x 1 − x + ( 1 − x ) A p ( ζ , t p , x ) = v 2 + k 2 x 1 − x + ∑ m ≥ 0 ( − v ) m + 1 ( 1 − x ) m B m ( t p , x ) , \mathcal{G}_Q=\frac{v}{2}+\frac{k^2x}{1-x}+(1-x)A_p(\zeta,t_p,x)=\frac{v}{2}+\frac{k^2x}{1-x}+\sum_{m\geq0}\frac{(-v)^{m+1}}{(1-x)^m}B_m(t_p,x), G Q = 2 v + 1 − x k 2 x + ( 1 − x ) A p ( ζ , t p , x ) = 2 v + 1 − x k 2 x + m ≥ 0 ∑ ( 1 − x ) m ( − v ) m + 1 B m ( t p , x ) ,
A p = ∑ m ≥ 0 ζ m + 1 B m ( t p , x ) , B m ( t p , x ) = ∑ j ≥ 0 A m , j ( t p ) x j . (107) A_p=\sum_{m\geq0}\zeta^{m+1}B_m(t_p,x),\qquad B_m(t_p,x)=\sum_{j\geq0}A_{m,j}(t_p)x^j. \tag*{(107)} A p = m ≥ 0 ∑ ζ m + 1 B m ( t p , x ) , B m ( t p , x ) = j ≥ 0 ∑ A m , j ( t p ) x j . (107)
Here ζ \zeta ζ , t p t_p t p , x x x are treated as independent in formal expansions and differentiations. Expanding each A m , j A_{m,j} A m , j in t p t_p t p gives the third index of the array.
This GF is a root of the same sextic in a different chart. In the cleared polynomial Q Q Q of (90) , substitute
K = τ ( 1 − x ) x , v = − ζ ( 1 − x ) , T = ( 1 − x ) τ + ( 1 − x ) 2 a . (108) K=\frac{\tau(1-x)}{x},\qquad v=-\zeta(1-x),\qquad T=(1-x)\tau+(1-x)^2a. \tag*{(108)} K = x τ ( 1 − x ) , v = − ζ ( 1 − x ) , T = ( 1 − x ) τ + ( 1 − x ) 2 a . (108)
The root is a = A p ( ζ , 4 τ , x ) a=A_p(\zeta,4\tau,x) a = A p ( ζ , 4 τ , x ) . Let P ~ ( a ; ζ , t p , x ) \widetilde{P}(a;\zeta,t_p,x) P ( a ; ζ , t p , x ) be the polynomial obtained after clearing denominators and removing parameter-only content, with constant normalization fixed in (5.11) below. Thus the pyramid is an expansion of the QCD curve, with the fixed-sector germ selecting its branch.
The main differential recurrence for this expansion is
L m + 2 B m + 2 = 2 E m M m B m , m ≥ 0 , (109) \mathcal{L}_{m+2}B_{m+2}=2\mathcal{E}_m\mathcal{M}_mB_m,\qquad m\geq0, \tag*{(109)} L m + 2 B m + 2 = 2 E m M m B m , m ≥ 0 , (109)
with the two initial equations L 0 B 0 = L 1 B 1 = 0 \mathcal{L}_0B_0=\mathcal{L}_1B_1=0 L 0 B 0 = L 1 B 1 = 0 . Writing ϑ = t p ∂ t p \vartheta=t_p\partial_{t_p} ϑ = t p ∂ t p , the operators are
E m = m − 1 + 2 ϑ , H m = m − ϑ , L m = [ t p ( 1 − x ) E m − 4 x ( H m + 1 ) ] ∂ x + t p E m H m + 2 H m ( H m + 1 ) , M m = 2 x ( 1 − x ) 2 ∂ x + E m + 1 + 2 x H m − 3 m x 2 . (110) \begin{aligned}
\mathcal{E}_m&=m-1+2\vartheta,\qquad H_m=m-\vartheta,\\
\mathcal{L}_m&=\left[t_p(1-x)\mathcal{E}_m-4x(H_m+1)\right]\partial_x+t_p\mathcal{E}_mH_m+2H_m(H_m+1),\\
\mathcal{M}_m&=2x(1-x)^2\partial_x+\mathcal{E}_m+1+2xH_m-3mx^2. \tag*{(110)}
\end{aligned} E m L m M m = m − 1 + 2 ϑ , H m = m − ϑ , = [ t p ( 1 − x ) E m − 4 x ( H m + 1 ) ] ∂ x + t p E m H m + 2 H m ( H m + 1 ) , = 2 x ( 1 − x ) 2 ∂ x + E m + 1 + 2 x H m − 3 m x 2 . (110)
Products denote composition; the rightmost operator acts first. In particular, the equation differentiates both rows. Its derivation is the coefficient extraction of the exact operator identity (150) , which follows from the sextic PDE (136) . It is not a recurrence inferred from the displayed low-order coefficients.
The generating PDE and reconstruction by columns
Set Q = ζ ∂ ζ Q=\zeta\partial_\zeta Q = ζ ∂ ζ , D p = Q + 2 ϑ D_p=Q+2\vartheta D p = Q + 2 ϑ and J b = Q − ϑ − 1 J_b=Q-\vartheta-1 J b = Q − ϑ − 1 . On a term ζ m + 1 B m \zeta^{m+1}B_m ζ m + 1 B m , E Q − 1 \mathcal{E}_{Q-1} E Q − 1 and H Q − 1 H_{Q-1} H Q − 1 act as E m \mathcal{E}_m E m and H m H_m H m . Summing (109) , including the two initial equations, gives
R p A p = 0 , R p = − L Q − 1 + 2 ζ 2 E Q − 1 M Q − 1 . (111) \mathcal{R}_pA_p=0,\qquad\mathcal{R}_p=-\mathcal{L}_{Q-1}+2\zeta^2\mathcal{E}_{Q-1}\mathcal{M}_{Q-1}. \tag*{(111)} R p A p = 0 , R p = − L Q − 1 + 2 ζ 2 E Q − 1 M Q − 1 . (111)
Conversely, the coefficient of ζ m + 3 \zeta^{m+3} ζ m + 3 gives (109) . In particular it is the factor ζ 2 \zeta^2 ζ 2 , placed to the left of the Euler operators, that shifts the row index by two. The recurrence and the PDE therefore encode the same constraint.
For computation it is convenient to expand the same operator in x x x :
R p = ∑ i = 0 3 x i M i ∂ x + ∑ i = 0 2 x i L i , \mathcal{R}_p=\sum_{i=0}^{3}x^iM_i\partial_x+\sum_{i=0}^{2}x^iL_i, R p = i = 0 ∑ 3 x i M i ∂ x + i = 0 ∑ 2 x i L i ,
M 0 = − t p ( D p − 2 ) , M 1 = ( t p + 4 ζ 2 ) ( D p − 2 ) + 4 ( J b + 1 ) , M 2 = − 8 ζ 2 ( D p − 2 ) , M 3 = 4 ζ 2 ( D p − 2 ) , L 0 = 2 ζ 2 ( D p − 1 ) ( D p − 2 ) − t p ( D p − 2 ) J b − 2 J b ( J b + 1 ) , L 1 = 4 ζ 2 ( D p − 2 ) J b , L 2 = − 6 ζ 2 ( Q − 1 ) ( D p − 2 ) . (112) \begin{aligned}
M_0&=-t_p(D_p-2), & M_1&=(t_p+4\zeta^2)(D_p-2)+4(J_b+1),\\
M_2&=-8\zeta^2(D_p-2), & M_3&=4\zeta^2(D_p-2),\\
L_0&=2\zeta^2(D_p-1)(D_p-2)-t_p(D_p-2)J_b-2J_b(J_b+1),\\
L_1&=4\zeta^2(D_p-2)J_b, & L_2&=-6\zeta^2(Q-1)(D_p-2). \tag*{(112)}
\end{aligned} M 0 M 2 L 0 L 1 = − t p ( D p − 2 ) , = − 8 ζ 2 ( D p − 2 ) , = 2 ζ 2 ( D p − 1 ) ( D p − 2 ) − t p ( D p − 2 ) J b − 2 J b ( J b + 1 ) , = 4 ζ 2 ( D p − 2 ) J b , M 1 M 3 L 2 = ( t p + 4 ζ 2 ) ( D p − 2 ) + 4 ( J b + 1 ) , = 4 ζ 2 ( D p − 2 ) , = − 6 ζ 2 ( Q − 1 ) ( D p − 2 ) . (112)
These L i , M i L_i,M_i L i , M i are column operators, distinct from the row operators L m , M m \mathcal{L}_m,\mathcal{M}_m L m , M m . With A j = [ x j ] A p = ∑ m ζ m + 1 A m , j A_j=[x^j]\mathcal{A}_p=\sum_m\zeta^{m+1}A_{m,j} A j = [ x j ] A p = ∑ m ζ m + 1 A m , j , coefficient extraction now gives
( j + 1 ) t p ( D p − 2 ) A j + 1 = ( L 0 + j M 1 ) A j + [ L 1 + ( j − 1 ) M 2 ] A j − 1 + [ L 2 + ( j − 2 ) M 3 ] A j − 2 , j ≥ 0. (113) \begin{aligned}
(j+1)t_p(D_p-2)A_{j+1}&=(L_0+jM_1)A_j\\
&\quad+[L_1+(j-1)M_2]A_{j-1}\\
&\quad+[L_2+(j-2)M_3]A_{j-2},\qquad j\ge0. \tag*{(113)}
\end{aligned} ( j + 1 ) t p ( D p − 2 ) A j + 1 = ( L 0 + j M 1 ) A j + [ L 1 + ( j − 1 ) M 2 ] A j − 1 + [ L 2 + ( j − 2 ) M 3 ] A j − 2 , j ≥ 0. (113)
Columns of negative index vanish. In the physical formal class ζ t p Q [ [ ζ , t p ] ] \zeta t_pQ[[\zeta,t_p]] ζ t p Q [[ ζ , t p ]] the Euler factor is invertible term by term:
( D p − 2 ) − 1 ( ζ a t p b ) = ζ a t p b a + 2 b − 2 , a , b ≥ 1. (114) (D_p-2)^{-1}(\zeta^a t_p^b)=\frac{\zeta^a t_p^b}{a+2b-2},\qquad a,b\ge1. \tag*{(114)} ( D p − 2 ) − 1 ( ζ a t p b ) = a + 2 b − 2 ζ a t p b , a , b ≥ 1. (114)
The right-hand side of (113) is first divided by t p t_p t p , then acted on by this inverse and divided by j + 1 j+1 j + 1 . Existence of the regular algebraic expansion supplies the required divisibility for the chosen branch. No such divisibility is assumed for arbitrary data in place of the boundary column. It follows that the first column determines all other columns in this formal class. The fixed-row version for computing individual entries is given in Appendix § 12 .
The GWW and Narayana boundary triangles
The recurrence (113) builds a three-index array. The two boundaries discussed here are different specializations of that array: the GWW column at x = 0 x=0 x = 0 and the Narayana triangle at leading order in t p t_p t p . The formal slice x = 1 x=1 x = 1 will give a separate exact tree enumeration.
The first rows are
B 0 = 1 − 1 − t p 2 , B 1 = t p 4 ( 1 − t p ) , B_0=\frac{1-\sqrt{1-t_p}}{2},\qquad B_1=\frac{t_p}{4(1-t_p)}, B 0 = 2 1 − 1 − t p , B 1 = 4 ( 1 − t p ) t p ,
B 2 = t p 4 ( 1 − t p ) 5 / 2 + x t p 4 ( 1 − t p ) 3 / 2 , B_2=\frac{t_p}{4(1-t_p)^{5/2}}+\frac{xt_p}{4(1-t_p)^{3/2}}, B 2 = 4 ( 1 − t p ) 5/2 t p + 4 ( 1 − t p ) 3/2 x t p ,
B 3 = t p ( 1 + t p ) 4 ( 1 − t p ) 4 + 3 x t p 4 ( 1 − t p ) 3 + x 2 t p 4 ( 1 − t p ) 2 . (115) B_3=\frac{t_p(1+t_p)}{4(1-t_p)^4}+\frac{3xt_p}{4(1-t_p)^3}+\frac{x^2t_p}{4(1-t_p)^2}. \tag*{(115)} B 3 = 4 ( 1 − t p ) 4 t p ( 1 + t p ) + 4 ( 1 − t p ) 3 3 x t p + 4 ( 1 − t p ) 2 x 2 t p . (115)
On the slice k = 1 k=1 k = 1 , the first row also identifies a known integer sequence. With the physical substitution t p = 8 x / ( 1 − x ) t_p=8x/(1-x) t p = 8 x / ( 1 − x ) ,
B 0 ( 8 x 1 − x , x ) = 1 − ( 1 − 9 x ) / ( 1 − x ) 2 = 2 ( G M ( x ) − 1 ) , G M ( x ) = 5 − ( 1 − 9 x ) / ( 1 − x ) 4 . B_0\left(\frac{8x}{1-x},x\right)=\frac{1-\sqrt{(1-9x)/(1-x)}}{2}=2(G_M(x)-1),\qquad G_M(x)=\frac{5-\sqrt{(1-9x)/(1-x)}}{4}. B 0 ( 1 − x 8 x , x ) = 2 1 − ( 1 − 9 x ) / ( 1 − x ) = 2 ( G M ( x ) − 1 ) , G M ( x ) = 4 5 − ( 1 − 9 x ) / ( 1 − x ) .
Muntingh obtained G M G_M G M as the generating function for the number of terms in implicit divided differences (OEIS A162326) [18 Prop. 1] . The identity concerns this row and this specialization.
The first column is precisely the GWW boundary, with elementary terms removed. In the normalization of (108) , the exact specialization of the transformed sextic is
P ~ ( a ; ζ , t p , 0 ) = 256 t p 2 P 4 ( a − ζ 2 + t p 4 ; − ζ , t p ) . (116) \widetilde{P}(a;\zeta,t_p,0)=256t_p^2P_4\left(a-\frac{\zeta}{2}+\frac{t_p}{4};-\zeta,\sqrt{t_p}\right). \tag*{(116)} P ( a ; ζ , t p , 0 ) = 256 t p 2 P 4 ( a − 2 ζ + 4 t p ; − ζ , t p ) . (116)
Selecting the same analytic germ as in (35) gives
A p ( ζ , t p , 0 ) = g ( − ζ , t p ) + ζ 2 − t p 4 . (117) A_p(\zeta,t_p,0)=g(-\zeta,\sqrt{t_p})+\frac{\zeta}{2}-\frac{t_p}{4}. \tag*{(117)} A p ( ζ , t p , 0 ) = g ( − ζ , t p ) + 2 ζ − 4 t p . (117)
Inserting (35) identifies this column with − ∑ n ≥ 0 R n ( − ζ ) ( t p / 4 ) n + 1 -\sum_{n\geq0}R_n(-\zeta)(t_p/4)^{n+1} − ∑ n ≥ 0 R n ( − ζ ) ( t p /4 ) n + 1 , so the GWW coefficient triangle, in the rational normalization of § 3 , supplies the data for (113) . This is also the many-flavor limit of the two matrix models, as explained in § 6 . Another exact boundary relation is
A m , 1 ( t p ) = 6 t p m + 1 d 2 d t p 2 A m − 2 , 0 ( t p ) , m ≥ 2. (118) A_{m,1}(t_p)=\frac{6t_p}{m+1}\frac{\mathrm{d}^2}{\mathrm{d}t_p^2}A_{m-2,0}(t_p),\qquad m\geq2. \tag*{(118)} A m , 1 ( t p ) = m + 1 6 t p d t p 2 d 2 A m − 2 , 0 ( t p ) , m ≥ 2. (118)
In generating-function form this says Q A 1 = 6 t p ζ 2 ∂ t p 2 A 0 QA_1=6t_p\zeta^2\partial_{t_p}^2A_0 Q A 1 = 6 t p ζ 2 ∂ t p 2 A 0 . It follows by implicit differentiation of the transformed sextic at x = 0 x=0 x = 0 and reduction modulo the boundary quartic (Appendix § 12 ). The derivative suggests a pair of marked components; a bijection would also have to account for the prefactor and the charge-index shift.
For m ≥ 1 m\geq1 m ≥ 1 a useful normalization is
B m ( t p , x ) = t p 4 ( 1 − t p ) ( 3 m − 1 ) / 2 N m ( x ( 1 − t p ) ; t p ) . (119) B_m(t_p,x)=\frac{t_p}{4(1-t_p)^{(3m-1)/2}}N_m\left(x(1-t_p);t_p\right). \tag*{(119)} B m ( t p , x ) = 4 ( 1 − t p ) ( 3 m − 1 ) /2 t p N m ( x ( 1 − t p ) ; t p ) . (119)
Its boundary is the Narayana polynomial,
N m ( w ; 0 ) = ∑ j = 0 m − 1 1 m ( m j ) ( m j + 1 ) w j . (120) N_m(w;0)=\sum_{j=0}^{m-1}\frac{1}{m}\binom{m}{j}\binom{m}{j+1}w^j. \tag*{(120)} N m ( w ; 0 ) = j = 0 ∑ m − 1 m 1 ( j m ) ( j + 1 m ) w j . (120)
In the classical lattice-path interpretation the coefficient of w j w^j w j counts Dyck paths of semilength m m m with j + 1 j+1 j + 1 peaks [19 pp. 100–101] . This boundary follows at all orders from the transformed sextic. Write A p = t p ζ Z / 4 + O ( t p 2 ) A_p=t_p\zeta Z/4+O(t_p^2) A p = t p ζ Z /4 + O ( t p 2 ) . Inserting a = t p ζ Z / 4 a=t_p\zeta Z/4 a = t p ζ Z /4 in P ~ \widetilde{P} P gives
[ t p 2 ] P ~ ( t p ζ Z / 4 ; ζ , t p , x ) = 16 ζ 3 x [ ζ x Z 2 + ζ ( 1 − x ) Z − Z + 1 ] × [ ζ 2 ( 1 − x ) 2 − 2 ζ ( 1 + x ) + 1 ] 2 . (121) [t_p^2]\widetilde{P}(t_p\zeta Z/4;\zeta,t_p,x)=16\zeta^3x\left[\zeta xZ^2+\zeta(1-x)Z-Z+1\right]
\times\left[\zeta^2(1-x)^2-2\zeta(1+x)+1\right]^2. \tag*{(121)} [ t p 2 ] P ( t p ζ Z /4 ; ζ , t p , x ) = 16 ζ 3 x [ ζ x Z 2 + ζ ( 1 − x ) Z − Z + 1 ] × [ ζ 2 ( 1 − x ) 2 − 2 ζ ( 1 + x ) + 1 ] 2 . (121)
Lower powers vanish in this substitution. The squared factor is a formal unit near ζ = 0 \zeta=0 ζ = 0 , and the regular branch obeys Z = 1 + ζ Z ( 1 − x + x Z ) Z=1+\zeta Z(1-x+xZ) Z = 1 + ζ Z ( 1 − x + x Z ) . With U = Z − 1 U=Z-1 U = Z − 1 this is U = ζ ( 1 + U ) ( 1 + x U ) U=\zeta(1+U)(1+xU) U = ζ ( 1 + U ) ( 1 + xU ) . Lagrange inversion yields, for m ≥ 1 m\geq1 m ≥ 1 ,
[ ζ m ] U = 1 m [ y m − 1 ] ( 1 + y ) m ( 1 + x y ) m = ∑ j = 0 m − 1 1 m ( m j ) ( m j + 1 ) x j . (122) [\zeta^m]U=\frac{1}{m}[y^{m-1}](1+y)^m(1+xy)^m=\sum_{j=0}^{m-1}\frac{1}{m}\binom{m}{j}\binom{m}{j+1}x^j. \tag*{(122)} [ ζ m ] U = m 1 [ y m − 1 ] ( 1 + y ) m ( 1 + x y ) m = j = 0 ∑ m − 1 m 1 ( j m ) ( j + 1 m ) x j . (122)
Since w = x w=x w = x at t p = 0 t_p=0 t p = 0 , this is (120) . The leading resummed coefficient is meant here; setting h = 0 h=0 h = 0 at fixed k k k in the original integral is a different substitution.
For instance the fourth normalized row is
N 4 ( w ; 4 τ ) = 1 + 13 τ + 16 τ 2 + ( 6 + 18 τ ) w + ( 6 + τ ) w 2 + w 3 . (123) N_4(w;4\tau)=1+13\tau+16\tau^2+(6+18\tau)w+(6+\tau)w^2+w^3. \tag*{(123)} N 4 ( w ; 4 τ ) = 1 + 13 τ + 16 τ 2 + ( 6 + 18 τ ) w + ( 6 + τ ) w 2 + w 3 . (123)
At τ = 0 \tau=0 τ = 0 its coefficients are ( 1 , 6 , 6 , 1 ) (1,6,6,1) ( 1 , 6 , 6 , 1 ) , the fourth Narayana row. The τ \tau τ terms display how the interior of the pyramid deforms this row.
Expansion about the Narayana boundary. Let s = 1 − t p s=\sqrt{1-t_{\mathrm{p}}} s = 1 − t p , w = x s 2 w=xs^2 w = x s 2 , and ξ = ζ / s 3 \xi=\zeta/s^3 ξ = ζ / s 3 . Set
H ( ξ ; w , t p ) = ∑ m ≥ 1 N m ( w ; t p ) ξ m , A p = ζ [ 1 − s 2 + t p s 4 H ] . (124) H(\xi;w,t_{\mathrm{p}})=\sum_{m\ge1}N_m(w;t_{\mathrm{p}})\xi^m,\qquad\mathcal{A}_{\mathrm{p}}=\zeta\left[\frac{1-s}{2}+\frac{t_{\mathrm{p}}s}{4}H\right]. \tag*{(124)} H ( ξ ; w , t p ) = m ≥ 1 ∑ N m ( w ; t p ) ξ m , A p = ζ [ 2 1 − s + 4 t p s H ] . (124)
The substitution in the sextic of (108) gives
P ( H ; ξ , w , t p ) = P ~ ( ξ s 3 [ ( 1 − s ) / 2 + t p s H / 4 ] ; ξ s 3 , t p , w / s 2 ) ξ 2 t p 2 s 12 = 0. (125) \mathcal{P}(H;\xi,w,t_{\mathrm{p}})=\frac{\widetilde{P}\left(\xi s^3\left[(1-s)/2+t_{\mathrm{p}}sH/4\right];\xi s^3,t_{\mathrm{p}},w/s^2\right)}{\xi^2t_{\mathrm{p}}^2s^{12}}=0. \tag*{(125)} P ( H ; ξ , w , t p ) = ξ 2 t p 2 s 12 P ( ξ s 3 [ ( 1 − s ) /2 + t p sH /4 ] ; ξ s 3 , t p , w / s 2 ) = 0. (125)
The quotient is a polynomial in H , ξ , w , t p H,\xi,w,t_{\mathrm{p}} H , ξ , w , t p ; after cancellation it contains only even powers of s s s . It is generically irreducible of degree six in H H H . For example, the primitive specialization at ( t p , ξ , w ) = ( 2 , 1 , 2 ) (t_{\mathrm{p}},\xi,w)=(2,1,2) ( t p , ξ , w ) = ( 2 , 1 , 2 ) reduces, up to a unit, to the irreducible polynomial − H 6 − H 5 + 2 H 4 − 2 H 3 − 2 H + 1 -H^6-H^5+2H^4-2H^3-2H+1 − H 6 − H 5 + 2 H 4 − 2 H 3 − 2 H + 1 over F 5 \mathbb{F}_5 F 5 . At t p = 0 t_{\mathrm{p}}=0 t p = 0 one has
P ( H ; ξ , w , 0 ) = 16 ξ w ( ξ ( 1 + H ) ( 1 + w H ) − H ) ( 1 − 2 ξ ( 1 + w ) + ξ 2 ( 1 − w ) 2 ) 2 . (126) \mathcal{P}(H;\xi,w,0)=16\xi w\bigl(\xi(1+H)(1+wH)-H\bigr)\bigl(1-2\xi(1+w)+\xi^2(1-w)^2\bigr)^2. \tag*{(126)} P ( H ; ξ , w , 0 ) = 16 ξ w ( ξ ( 1 + H ) ( 1 + w H ) − H ) ( 1 − 2 ξ ( 1 + w ) + ξ 2 ( 1 − w ) 2 ) 2 . (126)
Thus the chosen branch begins with the Narayana series y = ξ ( 1 + y ) ( 1 + w y ) y=\xi(1+y)(1+wy) y = ξ ( 1 + y ) ( 1 + w y ) . In what follows, set z = w y z=wy z = w y , A = ( 1 + y ) ( 1 + z ) A=(1+y)(1+z) A = ( 1 + y ) ( 1 + z ) and D = 1 − y z D=1-yz D = 1 − y z , so ξ = y / A \xi=y/A ξ = y / A .
The substitution H = y + D V H=y+DV H = y + D V , τ = D 3 λ \tau=D^3\lambda τ = D 3 λ removes the singular factors of the perturbation equation. In fact, put
R ( V , λ ; y , z ) = A 6 16 D 6 P ( y + D V ; y / A , z / y , 4 D 3 λ ) . (127) \mathcal{R}(V,\lambda;y,z)=\frac{A^6}{16D^6}\mathcal{P}(y+DV;y/A,z/y,4D^3\lambda). \tag*{(127)} R ( V , λ ; y , z ) = 16 D 6 A 6 P ( y + D V ; y / A , z / y , 4 D 3 λ ) . (127)
Direct expansion in (125) gives R ∈ Z [ y , z , V , λ ] \mathcal{R}\in\mathbb{Z}[y,z,V,\lambda] R ∈ Z [ y , z , V , λ ] and R ( V , 0 ; y , z ) = z V ( z V − 1 ) \mathcal{R}(V,0;y,z)=zV(zV-1) R ( V , 0 ; y , z ) = z V ( z V − 1 ) . For the divisibility check, write A 6 P ( y + η ; y / A , z / y , 4 τ ) / 16 A^6\mathcal{P}(y+\eta;y/A,z/y,4\tau)/16 A 6 P ( y + η ; y / A , z / y , 4 τ ) /16 as a polynomial in τ , η \tau,\eta τ , η over Z [ y , z ] \mathbb{Z}[y,z] Z [ y , z ] . The coefficient of τ 0 η j \tau^0\eta^j τ 0 η j is divisible by D 5 D^5 D 5 for j = 1 j=1 j = 1 and by D 4 D^4 D 4 for $j=2; the other coefficients at τ 0 \tau^0 τ 0 vanish. For τ 1 \tau^1 τ 1 the required powers of D D D at j = 0 , 1 , 2 , 3 j=0,1,2,3 j = 0 , 1 , 2 , 3 are $3,2,1,0$, respectively, and the remaining coefficients at τ 1 \tau^1 τ 1 vanish. All terms with τ \tau τ -degree at least two already carry D 6 D^6 D 6 after scaling.
The coefficient recursion for R = 0 \mathcal{R}=0 R = 0 with V ( 0 ) = 0 V(0)=0 V ( 0 ) = 0 divides only by − z -z − z . Consequently
H = y + ∑ c ≥ 1 t p c H c ( y , w ) , 4 c H c ( y , w ) = C c ( y , w y ) ( 1 − w y 2 ) 3 c − 1 , C c ∈ Z [ y , z ] . (128) H=y+\sum_{c\ge1}t_{\mathrm{p}}^cH_c(y,w),\qquad4^cH_c(y,w)=\frac{C_c(y,wy)}{(1-wy^2)^{3c-1}},\qquad C_c\in\mathbb{Z}[y,z]. \tag*{(128)} H = y + c ≥ 1 ∑ t p c H c ( y , w ) , 4 c H c ( y , w ) = ( 1 − w y 2 ) 3 c − 1 C c ( y , w y ) , C c ∈ Z [ y , z ] . (128)
Indeed, initially C c = [ λ c ] V C_c=[\lambda^c]V C c = [ λ c ] V belongs to Z [ y , z , z − 1 ] \mathbb{Z}[y,z,z^{-1}] Z [ y , z , z − 1 ] . The original formal branch, expressed at ξ = y / [ ( 1 + y ) ( 1 + w y ) ] \xi=y/[(1+y)(1+wy)] ξ = y / [( 1 + y ) ( 1 + w y )] , is regular in w w w . Since D D D is a formal unit at w = 0 w=0 w = 0 , this rules out negative powers of z z z in each C c C_c C c . The factor 4 c 4^c 4 c is essential. For instance,
4 H 1 ( y , w ) = y 3 ( 4 + ( 1 + 6 w + w 2 ) y + 4 w y 2 ) ( 1 − w y 2 ) 2 . (129) 4H_1(y,w)=\frac{y^3\bigl(4+(1+6w+w^2)y+4wy^2\bigr)}{(1-wy^2)^2}. \tag*{(129)} 4 H 1 ( y , w ) = ( 1 − w y 2 ) 2 y 3 ( 4 + ( 1 + 6 w + w 2 ) y + 4 w y 2 ) . (129)
Degree bounds and finite-order positivity. At ξ = 0 \xi=0 ξ = 0 the polynomial in (125) is H t p ( H t p − 4 ) H t_{\mathrm{p}}(H t_{\mathrm{p}}-4) H t p ( H t p − 4 ) . Solving successively in ξ \xi ξ yields coefficients polynomial in w w w and Laurent polynomial in t p t_{\mathrm{p}} t p ; regularity at t p = 0 t_{\mathrm{p}}=0 t p = 0 makes these polynomials in both variables. To see the degree bounds, use a new series variable ρ \rho ρ and a small parameter ϵ \epsilon ϵ . The two rescalings
( ξ , w , H ) = ( ϵ ρ , ϵ − 1 , ϵ U ) , ( ξ , t p , H ) = ( ϵ ρ , ϵ − 1 , ϵ ρ + ϵ 2 U ) (\xi,w,H)=(\epsilon\rho,\epsilon^{-1},\epsilon U),\qquad(\xi,t_{\mathrm{p}},H)=(\epsilon\rho,\epsilon^{-1},\epsilon\rho+\epsilon^2U) ( ξ , w , H ) = ( ϵ ρ , ϵ − 1 , ϵ U ) , ( ξ , t p , H ) = ( ϵ ρ , ϵ − 1 , ϵ ρ + ϵ 2 U )
give, respectively, the following lowest-order terms of ϵ − 1 P \epsilon^{-1}\mathcal{P} ϵ − 1 P :
− 4 ( ρ − 1 ) 3 ( ( ρ − 1 ) U + ρ ) [ t p ( 2 ρ − 1 ) 2 + 4 ρ ( 1 − ρ ) ] , -4(\rho-1)^3\bigl((\rho-1)U+\rho\bigr)\bigl[t_{\mathrm{p}}(2\rho-1)^2+4\rho(1-\rho)\bigr], − 4 ( ρ − 1 ) 3 ( ( ρ − 1 ) U + ρ ) [ t p ( 2 ρ − 1 ) 2 + 4 ρ ( 1 − ρ ) ] ,
4 ( ρ − 1 ) 3 [ ( 1 − ρ ) ( U − w ρ 2 ) − ρ 2 ] . 4(\rho-1)^3\bigl[(1-\rho)(U-w\rho^2)-\rho^2\bigr]. 4 ( ρ − 1 ) 3 [ ( 1 − ρ ) ( U − w ρ 2 ) − ρ 2 ] .
Both limiting equations have a simple branch at ρ = 0 \rho= 0 ρ = 0 over the corresponding parameter field. They give U = ρ / ( 1 − ρ ) U = \rho/(1 - \rho) U = ρ / ( 1 − ρ ) and U = w ρ 2 + ρ 2 / ( 1 − ρ ) U = w\rho^2 + \rho^2/(1 - \rho) U = w ρ 2 + ρ 2 / ( 1 − ρ ) , respectively. Formal uniqueness therefore gives deg w N m = m − 1 \deg_w N_m = m - 1 deg w N m = m − 1 , with leading coefficient one, and deg t p N m = m − 2 \deg_{t_p} N_m = m - 2 deg t p N m = m − 2 , with leading coefficient one, for m ≥ 3 m \ge3 m ≥ 3 . The small rows are N 1 = 1 N_1 = 1 N 1 = 1 and N 2 = 1 + w N_2 = 1 + w N 2 = 1 + w .
For 1 ≤ c ≤ 6 1 \le c \le6 1 ≤ c ≤ 6 , there are nonnegative integer polynomials P c , j P_{c,j} P c , j such that
C c ( y , z ) = ∑ j = 1 3 c − 1 P c , j ( y , z ) D 3 c − 1 − j . (130) C_c(y,z) = \sum_{j=1}^{3c-1} P_{c,j}(y,z)D^{3c-1-j}. \tag*{(130)} C c ( y , z ) = j = 1 ∑ 3 c − 1 P c , j ( y , z ) D 3 c − 1 − j . (130)
The certificate for c = 1 c = 1 c = 1 is P 1 , 2 = y 2 ( y 2 + 4 y z 2 + 6 y z + 4 y + z 2 ) P_{1,2} = y^2(y^2 + 4yz^2 + 6yz + 4y + z^2) P 1 , 2 = y 2 ( y 2 + 4 y z 2 + 6 y z + 4 y + z 2 ) . The ancillary file positivity_certificates_c1_c6.txt lists the explicit polynomials P c , j P_{c,j} P c , j for 1 ≤ c ≤ 6 1 \le c \le6 1 ≤ c ≤ 6 . These decompositions make coefficientwise nonnegativity manifest: each term P c , j ( y , z ) / ( 1 − y z ) j P_{c,j}(y,z)/(1-yz)^j P c , j ( y , z ) / ( 1 − y z ) j has a nonnegative integer power-series expansion in y , z y,z y , z .
The companion script verify_narayana_certificates.py reconstructs the scaled sextic and substitutes the right-hand sides of (130) as coefficients of λ c \lambda^c λ c . The residual vanishes through λ 6 \lambda^6 λ 6 in Z [ y , z ] \mathbb{Z}[y,z] Z [ y , z ] , so formal uniqueness verifies all six identities. Substitution of the Narayana series, which has nonnegative integer coefficients, proves
[ τ c ] N m ( w ; 4 τ ) ∈ Z ≥ 0 [ w ] , m ≥ 1 , 0 ≤ c ≤ 6. (131) [\tau^c]N_m(w;4\tau) \in\mathbb{Z}_{\ge0}[w], \qquad m \ge1,\qquad0 \le c \le6. \tag*{(131)} [ τ c ] N m ( w ; 4 τ ) ∈ Z ≥ 0 [ w ] , m ≥ 1 , 0 ≤ c ≤ 6. (131)
The degree bound implies positivity of the full row for m ≤ 8 m \le8 m ≤ 8 . Direct exact extraction from (125) verifies the full rows for m ≤ 14 m \le14 m ≤ 14 .
For the original rows B m B_m B m , a monomial τ c w j \tau^c w^j τ c w j of N m ( w ; 4 τ ) N_m(w;4\tau) N m ( w ; 4 τ ) contributes a multiple of
τ c + 1 x j ( 1 − 4 τ ) − 3 m − 1 − 2 j 2 , 0 ≤ j ≤ m − 1. \tau^{c+1}x^j(1-4\tau)^{-\frac{3m-1-2j}{2}}, \qquad0 \le j \le m-1. τ c + 1 x j ( 1 − 4 τ ) − 2 3 m − 1 − 2 j , 0 ≤ j ≤ m − 1.
The integer 3 m − 1 − 2 j 3m-1-2j 3 m − 1 − 2 j is positive, and ( 1 − 4 τ ) − 1 / 2 = ∑ a ≥ 0 ( 2 a a ) τ a (1-4\tau)^{-1/2} = \sum_{a\ge0} \binom{2a}{a}\tau^a ( 1 − 4 τ ) − 1/2 = ∑ a ≥ 0 ( a 2 a ) τ a . Thus B m B_m B m has nonnegative integer coefficients through τ 7 \tau^7 τ 7 for every m ≥ 1 m \ge1 m ≥ 1 , and to all orders in each fully positive row. The coefficient of τ \tau τ is the Narayana polynomial. The physical substitution τ = K x / ( 1 − x ) \tau= Kx/(1-x) τ = K x / ( 1 − x ) preserves these sign and integrality statements in the corresponding powers of K K K . The exceptional row B 0 = ( 1 − 1 − 4 τ ) / 2 B_0 = (1-\sqrt{1-4\tau})/2 B 0 = ( 1 − 1 − 4 τ ) /2 has nonnegative integer coefficients by the Catalan formula.
The polar coefficient and the critical GWW column. Define the critical GWW series
G ( X ) = H ( X ; 0 , 1 ) = ∑ m ≥ 1 N m ( 0 ; 1 ) X m . G(X) = H(X;0,1) = \sum_{m\ge1} N_m(0;1)X^m. G ( X ) = H ( X ; 0 , 1 ) = m ≥ 1 ∑ N m ( 0 ; 1 ) X m .
The specialization of (125) is
J ( U , X ) = 4 X U 3 + ( 1 − 24 X ) U 2 + ( 30 X − 4 ) U + 4 X − 27 X 2 , J ( G ( X ) , X ) = 0 , G ( 0 ) = 0. (132) J(U,X) = 4XU^3 + (1-24X)U^2 + (30X-4)U + 4X - 27X^2,\qquad J(G(X),X)=0,\qquad G(0)=0. \tag*{(132)} J ( U , X ) = 4 X U 3 + ( 1 − 24 X ) U 2 + ( 30 X − 4 ) U + 4 X − 27 X 2 , J ( G ( X ) , X ) = 0 , G ( 0 ) = 0. (132)
There is an exact identity at the polar divisor:
R ( V , λ ; y , 1 / y ) = 1 16 J ( 4 V y , 4 ( 1 + y ) 4 y λ ) . (133) \mathcal{R}(V,\lambda;y,1/y) = \frac{1}{16}J\left(\frac{4V}{y},\frac{4(1+y)^4}{y}\lambda\right). \tag*{(133)} R ( V , λ ; y , 1/ y ) = 16 1 J ( y 4 V , y 4 ( 1 + y ) 4 λ ) . (133)
This follows by retaining terms of weight six in (127) , with D , τ , η D,\tau,\eta D , τ , η of weights 1 , 3 , 1 1,3,1 1 , 3 , 1 . It identifies the regular branch of the scaled equation with a rescaled copy of the critical GWW series. In particular, for every c ≥ 1 c \ge1 c ≥ 1 ,
4 c H c = 4 c − 1 y 1 − c ( 1 + y ) 4 c N c ( 0 ; 1 ) ( 1 − w y 2 ) 3 c − 1 + O ( ( 1 − w y 2 ) − ( 3 c − 2 ) ) , (134) 4^c H_c = \frac{4^{c-1}y^{1-c}(1+y)^{4c}N_c(0;1)}{(1-wy^2)^{3c-1}} + O\left((1-wy^2)^{-(3c-2)}\right), \tag*{(134)} 4 c H c = ( 1 − w y 2 ) 3 c − 1 4 c − 1 y 1 − c ( 1 + y ) 4 c N c ( 0 ; 1 ) + O ( ( 1 − w y 2 ) − ( 3 c − 2 ) ) , (134)
where the Laurent expansion is taken at fixed generic y y y .
To evaluate the coefficient, parametrize (132) by X = ( r 2 − 1 ) / ( 2 r 3 ) X=(r^2-1)/(2r^3) X = ( r 2 − 1 ) / ( 2 r 3 ) and G = 2 − ( r + 3 / r ) / 2 G=2-(r+3/r)/2 G = 2 − ( r + 3/ r ) /2 with r = 1 r=1 r = 1 at X = 0 X=0 X = 0 . Writing R = ( r 2 − 1 ) / 2 R=(r^2-1)/2 R = ( r 2 − 1 ) /2 gives R = X ( 1 + 2 R ) 3 / 2 R=X(1+2R)^{3/2} R = X ( 1 + 2 R ) 3/2 and d G / d R = ( 1 − R ) ( 1 + 2 R ) − 3 / 2 \mathrm{d}G/\mathrm{d}R=(1-R)(1+2R)^{-3/2} d G / d R = ( 1 − R ) ( 1 + 2 R ) − 3/2 . Lagrange inversion then gives
N m ( 0 ; 1 ) = 1 m [ ρ m − 1 ] ( 1 − ρ ) ( 1 + 2 ρ ) 3 ( m − 1 ) / 2 = 4 j Γ ( 3 j + 1 ) Γ ( j + 2 ) Γ ( 2 j + 2 ) , j = m − 1 2 . (135) N_m(0;1)=\frac{1}{m}[\rho^{m-1}](1-\rho)(1+2\rho)^{3(m-1)/2}=\frac{4^j\Gamma(3j+1)}{\Gamma(j+2)\Gamma(2j+2)},\qquad j=\frac{m-1}{2}. \tag*{(135)} N m ( 0 ; 1 ) = m 1 [ ρ m − 1 ] ( 1 − ρ ) ( 1 + 2 ρ ) 3 ( m − 1 ) /2 = Γ ( j + 2 ) Γ ( 2 j + 2 ) 4 j Γ ( 3 j + 1 ) , j = 2 m − 1 . (135)
Thus (128) has pole order exactly 3 c − 1 3c-1 3 c − 1 at generic points of the divisor. For m = 2 j + 1 m=2j+1 m = 2 j + 1 these coefficients are the Kreweras excursion numbers (OEIS A006335). The entire odd GWW column has a related description: iterating (36) on monomials in (117) gives
N 2 j + 1 ( 0 ; t p ) = ( 1 − t p ) 3 j + 1 ∑ n ≥ 0 ( n + j + 1 ) ! ( 2 n + 2 j + 1 ) ! ( n + 1 ) ! ( 2 n + 1 ) ! ( j + 1 ) ! ( 2 j + 1 ) ! t p n . N_{2j+1}(0;t_p)=(1-t_p)^{3j+1}\sum_{n\geq0}\frac{(n+j+1)!(2n+2j+1)!}{(n+1)!(2n+1)!(j+1)!(2j+1)!}t_p^n. N 2 j + 1 ( 0 ; t p ) = ( 1 − t p ) 3 j + 1 n ≥ 0 ∑ ( n + 1 )! ( 2 n + 1 )! ( j + 1 )! ( 2 j + 1 )! ( n + j + 1 )! ( 2 n + 2 j + 1 )! t p n .
By the numerator definition of OEIS A140136, this is its row j j j . The identity (133) relates the deformation order to the index of the critical GWW column. It does not require an exchange symmetry of the full coefficient array.
An invariant second-order PDE
The projective generating function obeys a compact equation in the independent invariant coordinates ( r , s , t ) (r,s,t) ( r , s , t ) :
0 = D i n v W : = 2 r ( 4 r − 1 ) W r r + 4 s ( 6 r − 1 ) W r s + 18 s 2 W s s + 2 t ( t − 1 ) W r t + ( 4 r − t ) W r + 12 s W s . (136) 0=\mathcal{D}_{\mathrm{inv}}W:=2r(4r-1)W_{rr}+4s(6r-1)W_{rs}+18s^2W_{ss}+2t(t-1)W_{rt}+(4r-t)W_r+12sW_s. \tag*{(136)} 0 = D inv W := 2 r ( 4 r − 1 ) W r r + 4 s ( 6 r − 1 ) W r s + 18 s 2 W ss + 2 t ( t − 1 ) W r t + ( 4 r − t ) W r + 12 s W s . (136)
The equation has total order two and contains no zeroth-order term. Therefore the centered variable W + 1 / 12 W+1/12 W + 1/12 satisfies the same PDE. Its validity for all regular roots of the sextic is established by the polynomial-remainder calculation given in Appendix § 9 . This is a full parameter identity rather than an inference from a finite coefficient expansion.
Substituting W = t ( t − 1 ) S ^ t W=t(t-1)\widehat{S}_t W = t ( t − 1 ) S t gives a third-order equation for the normalized potential itself:
0 = 2 r ( 4 r − 1 ) S ^ r r t + 4 s ( 6 r − 1 ) S ^ r s t + 18 s 2 S ^ s s t + 2 t ( t − 1 ) S ^ r t t + ( 4 r + 3 t − 2 ) S ^ r t + 12 s S ^ s t . (137) 0=2r(4r-1)\widehat{S}_{rrt}+4s(6r-1)\widehat{S}_{rst}+18s^2\widehat{S}_{sst}+2t(t-1)\widehat{S}_{rtt}+(4r+3t-2)\widehat{S}_{rt}+12s\widehat{S}_{st}. \tag*{(137)} 0 = 2 r ( 4 r − 1 ) S r r t + 4 s ( 6 r − 1 ) S r s t + 18 s 2 S ss t + 2 t ( t − 1 ) S r tt + ( 4 r + 3 t − 2 ) S r t + 12 s S s t . (137)
The coefficient 4 r + 3 t − 2 4r+3t-2 4 r + 3 t − 2 includes the derivative of the normalizing factor t ( t − 1 ) t(t-1) t ( t − 1 ) . This term is needed when the equation is expressed for the potential rather than for its derivative. In particular, projective normalization of the free energy preserves the third order of its equation; choosing its mass derivative as a new dependent function is the separate operation that yields order two.
The homogeneous operator triple
Let D i = A i ∂ A i D_i=A_i\partial_{A_i} D i = A i ∂ A i and η = t ( t − 1 ) \eta=t(t-1) η = t ( t − 1 ) . The lift (105) has degree one in A i A_i A i , so
( D 1 + D 2 + D 3 ) Φ = Φ , Y Q = E 0 W = η Φ t . (138) (D_1+D_2+D_3)\Phi=\Phi,\qquad Y_Q=E_0W=\eta\Phi_t. \tag*{(138)} ( D 1 + D 2 + D 3 ) Φ = Φ , Y Q = E 0 W = η Φ t . (138)
The parameter equation for Y Q Y_Q Y Q admits three operators of identical form,
K i = D i ( D i − 1 − η ∂ t + t 2 ) , i = 1 , 2 , 3. (139) \mathcal{K}_i=D_i\left(D_i-1-\eta\partial_t+\frac{t}{2}\right),\qquad i=1,2,3. \tag*{(139)} K i = D i ( D i − 1 − η ∂ t + 2 t ) , i = 1 , 2 , 3. (139)
They commute: the Euler derivatives commute with each other and with the common mass operator − 1 − η ∂ t + t / 2 -1-\eta\partial_t+t/2 − 1 − η ∂ t + t /2 . The complete equation is the weighted cyclic relation
( A 2 − A 3 ) K 1 Y Q + ( A 3 − A 1 ) K 2 Y Q + ( A 1 − A 2 ) K 3 Y Q = 0. (140) (A_2-A_3)\mathcal{K}_1Y_Q+(A_3-A_1)\mathcal{K}_2Y_Q+(A_1-A_2)\mathcal{K}_3Y_Q=0. \tag*{(140)} ( A 2 − A 3 ) K 1 Y Q + ( A 3 − A 1 ) K 2 Y Q + ( A 1 − A 2 ) K 3 Y Q = 0. (140)
A coordinate permutation exchanges the three K i \mathcal{K}_i K i and the corresponding weights. The entire cyclic operator is alternating: an odd permutation changes its sign while preserving its equation.
This decomposition can be checked before selecting an algebraic root. On the chart A 1 = 1 A_1=1 A 1 = 1 , A 2 = B A_2=B A 2 = B , A 3 = C A_3=C A 3 = C , a homogeneous degree-one function is acted on by
D 1 = 1 − B ∂ B − C ∂ C , D 2 = B ∂ B , D 3 = C ∂ C . (141) D_1=1-B\partial_B-C\partial_C,\qquad D_2=B\partial_B,\qquad D_3=C\partial_C. \tag*{(141)} D 1 = 1 − B ∂ B − C ∂ C , D 2 = B ∂ B , D 3 = C ∂ C . (141)
Let E = 1 + B + C E=1+B+C E = 1 + B + C and δ ∗ = ( 1 − B ) ( B − C ) ( C − 1 ) \delta_*=(1-B)(B-C)(C-1) δ ∗ = ( 1 − B ) ( B − C ) ( C − 1 ) . With r = ( B + C + B C ) / E 2 r=(B+C+BC)/E^2 r = ( B + C + B C ) / E 2 and s = B C / E 3 s=BC/E^3 s = B C / E 3 , direct chain-rule substitution gives
∑ c y c ( A j − A k ) K i Y = − δ ∗ 2 E D i n v [ Y E ] , (142) \sum_{\mathrm{cyc}}(A_j-A_k)\mathcal{K}_iY=-\frac{\delta_*}{2E}\mathcal{D}_{\mathrm{inv}}\left[\frac{Y}{E}\right], \tag*{(142)} cyc ∑ ( A j − A k ) K i Y = − 2 E δ ∗ D inv [ E Y ] , (142)
where the invariant operator on the right is pulled back to the ( B , C , t ) (B,C,t) ( B , C , t ) chart. Equation (142) holds for an arbitrary function Y Y Y , with the invariant derivatives interpreted on a regular local chart. It establishes that the cyclic triple is an exact decomposition of the previously verified equation. It has not been introduced by multiplying that equation by three unrelated functions.
The corresponding operators for the free energy follow from
K i [ η Φ t ] = η T i ( t ) Φ , T i ( t ) = D i ( D i − η ∂ t − 3 t 2 ) ∂ t . (143) \mathcal{K}_i[\eta\Phi_t]=\eta\mathcal{T}^{(t)}_i\Phi,\qquad\mathcal{T}^{(t)}_i=D_i\left(D_i-\eta\partial_t-\frac{3t}{2}\right)\partial_t. \tag*{(143)} K i [ η Φ t ] = η T i ( t ) Φ , T i ( t ) = D i ( D i − η ∂ t − 2 3 t ) ∂ t . (143)
Each T i ( t ) \mathcal{T}^{(t)}_i T i ( t ) has total order three. To return to x = h 2 x=h^2 x = h 2 , observe that
η ∂ t = χ ( x ) ∂ x , χ ( x ) = x ( 1 − x ) 1 + x , Y Q = χ ( x ) Φ x . (144) \eta\partial_t=\chi(x)\partial_x,\qquad\chi(x)=\frac{x(1-x)}{1+x},\qquad Y_Q=\chi(x)\Phi_x. \tag*{(144)} η ∂ t = χ ( x ) ∂ x , χ ( x ) = 1 + x x ( 1 − x ) , Y Q = χ ( x ) Φ x . (144)
Differentiating the mass-dependent prefactor gives, after a common scalar normalization,
T i ( x ) = D i [ 2 ( 1 + x ) D i + x − 3 − 2 x ( 1 − x ) ∂ x ] ∂ x . (145) \mathcal{T}^{(x)}_i=D_i[2(1+x)D_i+x-3-2x(1-x)\partial_x]\partial_x. \tag*{(145)} T i ( x ) = D i [ 2 ( 1 + x ) D i + x − 3 − 2 x ( 1 − x ) ∂ x ] ∂ x . (145)
More precisely, K i [ χ ( x ) Φ x ] = χ ( x ) T i ( x ) Φ / [ 2 ( 1 + x ) ] \mathcal{K}_i[\chi(x)\Phi_x]=\chi(x)\mathcal{T}^{(x)}_i\Phi/[2(1+x)] K i [ χ ( x ) Φ x ] = χ ( x ) T i ( x ) Φ/ [ 2 ( 1 + x )] . The shifted free energy therefore satisfies
∑ c y c ( A j − A k ) T i ( x ) Φ = 0. (146) \sum_{\mathrm{cyc}}(A_j-A_k)\mathcal{T}^{(x)}_i\Phi=0. \tag*{(146)} cyc ∑ ( A j − A k ) T i ( x ) Φ = 0. (146)
This is the desired permuted triple of third-order operators. The literal permutation applies in homogeneous coordinates; restoring the physical chart introduces conjugation by the degree-two scale factor, consistent with the transformation of Φ \Phi Φ . Independent coordinate sign changes leave the A i A_i A i and their labels fixed, so the action of the full S 4 S_4 S 4 on the triple factors through S 3 S_3 S 3 .
The projective action in pyramid coordinates
The connection between the pyramid recurrence and the invariant sextic PDE can be established at operator level. Let ( a 1 , a 2 , a 3 ) = ( a , b , c ) (a_1,a_2,a_3)=(a,b,c) ( a 1 , a 2 , a 3 ) = ( a , b , c ) and d 0 = b − a d_0=b-a d 0 = b − a . In homogeneous coordinates put
E h = a 2 + b 2 + c 2 , z = 1 − x 1 + x , E_h=a^2+b^2+c^2,\qquad z=\frac{1-x}{1+x}, E h = a 2 + b 2 + c 2 , z = 1 + x 1 − x ,
and
H = E h W z = f 0 A p + ℓ 0 , f 0 = d 0 2 ( 1 − x ) , ℓ 0 = 1 2 ( c d 0 − a b ) . (147) \mathcal{H}=\frac{E_hW}{z}=f_0A_p+\ell_0,\qquad f_0=d_0^2(1-x),\qquad\ell_0=\frac{1}{2}(cd_0-ab). \tag*{(147)} H = z E h W = f 0 A p + ℓ 0 , f 0 = d 0 2 ( 1 − x ) , ℓ 0 = 2 1 ( c d 0 − ab ) . (147)
The physical chart is d 0 = 1 d_0=1 d 0 = 1 ; before restoring this chart the pyramid arguments are ζ = − c / [ d 0 ( 1 − x ) ] \zeta=-c/[d_0(1-x)] ζ = − c / [ d 0 ( 1 − x )] and t p = 4 a b x / [ d 0 2 ( 1 − x ) ] t_p=4abx/[d_0^2(1-x)] t p = 4 ab x / [ d 0 2 ( 1 − x )] . Conjugating (139) by z z z gives
J ‾ i = D i ( D i − 1 2 − χ ∂ x ) , J i = f 0 − 1 J ‾ i f 0 . (148) \overline{\mathcal{J}}_i=D_i\left(D_i-\frac{1}{2}-\chi\partial_x\right),\qquad\mathcal{J}_i=f_0^{-1}\overline{\mathcal{J}}_if_0. \tag*{(148)} J i = D i ( D i − 2 1 − χ ∂ x ) , J i = f 0 − 1 J i f 0 . (148)
The second equation is operator conjugation. The mass derivative in the homogeneous expression holds the three coordinates fixed. Each J i \mathcal{J}_i J i annihilates a b ab ab , a c ac a c , b c bc b c separately: on these monomials the squared-coordinate Euler derivative D i D_i D i has eigenvalue zero or 1 / 2 1/2 1/2 . In particular the affine term ℓ 0 \ell_0 ℓ 0 does not introduce a source into the equation for A p A_p A p .
For explicit comparison with (111) , put δ = 2 k + 1 = 1 + t p ( 1 − x ) / x \delta= 2k + 1 = \sqrt{1 + t_p(1-x)/x} δ = 2 k + 1 = 1 + t p ( 1 − x ) / x on the physical chart, retaining its sheet under parameter transformations. The operators in pyramid variables are
∂ 1 = δ 4 ( D p − 2 ) − Q − 2 4 , ∂ 2 = − δ 4 ( D p − 2 ) − Q − 2 4 , ∂ 3 = Q 2 , \partial_1 = \frac{\delta}{4}(D_p - 2) - \frac{Q - 2}{4}, \qquad\partial_2 = -\frac{\delta}{4}(D_p - 2) - \frac{Q - 2}{4}, \qquad\partial_3 = \frac{Q}{2}, ∂ 1 = 4 δ ( D p − 2 ) − 4 Q − 2 , ∂ 2 = − 4 δ ( D p − 2 ) − 4 Q − 2 , ∂ 3 = 2 Q ,
X = χ ∂ x + ϑ + x Q 1 + x , J i = ∂ i ( ∂ i − 1 2 − X + x 1 + x ) . (149) X = \chi\partial_x + \frac{\vartheta+ xQ}{1+x}, \qquad\mathcal{J}_i = \partial_i\left(\partial_i - \frac{1}{2} - X + \frac{x}{1+x}\right). \tag*{(149)} X = χ ∂ x + 1 + x ϑ + x Q , J i = ∂ i ( ∂ i − 2 1 − X + 1 + x x ) . (149)
Here X X X represents χ ∂ x \chi\partial_x χ ∂ x at fixed ( k , v ) (k,v) ( k , v ) , whereas the displayed ∂ x \partial_x ∂ x holds ( ζ , t p ) (\zeta,t_p) ( ζ , t p ) fixed. The terms ϑ + x Q \vartheta+xQ ϑ + x Q are necessary for the chain rule. With weights w i w_i w i as in (C.25), expansion of (149) gives the exact identity
C p : = ∑ i w i J i = δ ( 1 − x ) 8 ( 1 + x ) R p . (150) \mathcal{C}_p := \sum_i w_i\mathcal{J}_i = \frac{\delta(1-x)}{8(1+x)}\mathcal{R}_p. \tag*{(150)} C p := i ∑ w i J i = 8 ( 1 + x ) δ ( 1 − x ) R p . (150)
The weights multiply externally. Both sides have the same action on an arbitrary test function, before choosing a root. Combining (142) , (147) and (150) derives (109) from the sextic. The row projection and exact verification are given in Appendix D D D .
The homogeneous operators in (148) are permuted by S 4 S_4 S 4 , with the label action through S 4 / V 4 ≃ S 3 S_4/V_4 \simeq S_3 S 4 / V 4 ≃ S 3 . For a signed permutation P P P , let ( a ′ , b ′ , c ′ ) = P ( a , b , c ) (a',b',c')=P(a,b,c) ( a ′ , b ′ , c ′ ) = P ( a , b , c ) , d 0 ′ = b ′ − a ′ d'_0=b'-a' d 0 ′ = b ′ − a ′ and ρ = ( d 0 / d 0 ′ ) 2 \rho=(d_0/d'_0)^2 ρ = ( d 0 / d 0 ′ ) 2 . On the transported sheet, the affine-chart laws are
P ∗ J i ( P ∗ ) − 1 = ρ J σ ( i ) ρ − 1 , P^*\mathcal{J}_i(P^*)^{-1}=\rho\mathcal{J}_{\sigma(i)}\rho^{-1}, P ∗ J i ( P ∗ ) − 1 = ρ J σ ( i ) ρ − 1 ,
P ∗ A p = ρ A p + c d 0 − a b − c ′ d 0 ′ + a ′ b ′ 2 ( d 0 ′ ) 2 ( 1 − x ) . (151) P^*A_p=\rho A_p+\frac{cd_0-ab-c'd'_0+a'b'}{2(d'_0)^2(1-x)}. \tag*{(151)} P ∗ A p = ρ A p + 2 ( d 0 ′ ) 2 ( 1 − x ) c d 0 − ab − c ′ d 0 ′ + a ′ b ′ . (151)
The affine part divided by ρ \rho ρ lies in the common kernel of the J i \mathcal{J}_i J i , by the bilinear-monomial observation above. Thus the operator actions transform covariantly even though the regular pyramid GF is not an invariant scalar. This is an action on the generating system and on its charts, not an asserted permutation of the individual rows of one physical series.
A useful alternative takes divided differences. On the physical chart put b i = a i 2 b_i=a_i^2 b i = a i 2 and, for cyclic ( i , j , k ) (i,j,k) ( i , j , k ) , set
Z i = J j − J k b j − b k . (152) \mathcal{Z}_i=\frac{\mathcal{J}_j-\mathcal{J}_k}{b_j-b_k}. \tag*{(152)} Z i = b j − b k J j − J k . (152)
The denominator acts by left multiplication. Away from coinciding squared coordinates, the equality of the three actions on A p A_p A p is equivalent to the single weighted equation C p A p = 0 \mathcal{C}_pA_p=0 C p A p = 0 . In fact,
Z 1 − Z 3 = − C p w 1 w 3 , Z 2 − Z 3 = C p w 2 w 3 . (153) \mathcal{Z}_1-\mathcal{Z}_3=-\frac{\mathcal{C}_p}{w_1w_3}, \qquad\mathcal{Z}_2-\mathcal{Z}_3=\frac{\mathcal{C}_p}{w_2w_3}. \tag*{(153)} Z 1 − Z 3 = − w 1 w 3 C p , Z 2 − Z 3 = w 2 w 3 C p . (153)
This reformulation does not imply pairwise equality of the original components J i \mathcal{J}_i J i . Its relation to the standard GWW third-order operators is established in § 6.5 .
Exact tree enumeration on the collapsed slice
A complete tree interpretation is available after summing over the explicit x x x statistic. Define
T Q ( ζ , τ ) = A p ( ζ , 4 τ , 1 ) = ∑ m , a ≥ 0 q ( m , a ) ζ m + 1 τ a + 1 . (154) T_Q(\zeta,\tau)=A_p(\zeta,4\tau,1)=\sum_{m,a\geq0}q(m,a)\zeta^{m+1}\tau^{a+1}. \tag*{(154)} T Q ( ζ , τ ) = A p ( ζ , 4 τ , 1 ) = m , a ≥ 0 ∑ q ( m , a ) ζ m + 1 τ a + 1 . (154)
In this formula x = 1 x=1 x = 1 is a formal specialization of the stripped rows B m B_m B m , with τ \tau τ held independent. It is not the physical massless limit b = 1 b=1 b = 1 at fixed k k k , for which (106) is singular. The exact coefficient formula is
q ( m , a ) = 4 m + a ( ( a + 1 ) / 2 ) m ( ( m + 1 ) / 2 ) a ( m + 1 ) ! ( a + 1 ) ! a , m , a ≥ 0. (155) q(m,a)=\frac{4^{m+a}\left((a+1)/2\right)_m\left((m+1)/2\right)_a}{(m+1)!(a+1)!}\,a,\qquad m,a\geq0. \tag*{(155)} q ( m , a ) = ( m + 1 )! ( a + 1 )! 4 m + a ( ( a + 1 ) /2 ) m ( ( m + 1 ) /2 ) a a , m , a ≥ 0. (155)
Here ( b ) n = b ( b + 1 ) ⋯ ( b + n − 1 ) (b)_n = b(b+1)\cdots(b+n-1) ( b ) n = b ( b + 1 ) ⋯ ( b + n − 1 ) and ( b ) 0 = 1 (b)_0 = 1 ( b ) 0 = 1 . The first entries are
( q ( m , a ) ) 0 ≤ m , a ≤ 3 = ( 1 1 2 5 1 4 16 64 2 16 100 560 5 64 560 4096 ) . (156) \left(q(m,a)\right)_{0\leq m,a\leq3} =
\begin{pmatrix}
1 & 1 & 2 & 5 \\
1 & 4 & 16 & 64 \\
2 & 16 & 100 & 560 \\
5 & 64 & 560 & 4096
\end{pmatrix}. \tag*{(156)} ( q ( m , a ) ) 0 ≤ m , a ≤ 3 = 1 1 2 5 1 4 16 64 2 16 100 560 5 64 560 4096 . (156)
In particular q ( m , 0 ) = Cat m q(m,0)=\operatorname{Cat}_m q ( m , 0 ) = Cat m and q ( m , 1 ) = 4 m q(m,1)=4^m q ( m , 1 ) = 4 m . The symmetry q ( m , a ) = q ( a , m ) q(m,a)=q(a,m) q ( m , a ) = q ( a , m ) is manifest in (155) ; the following construction explains both that symmetry and the integrality of the formula.
Consider unrooted bipartite plane trees with at least one edge. Each vertex of degree d d d carries a distinguished corner and a full plane binary tree with d d d leaves. The cyclically ordered incident edges, read from the distinguished corner, correspond in order to the binary tree’s leaves. There are Cat d − 1 \operatorname{Cat}_{d-1} Cat d − 1 binary decorations; a leaf of the original tree has a unique corner and decoration. Isomorphisms preserve the colors, the plane embedding, the corners, and the decorations. Then q ( m , a ) q(m,a) q ( m , a ) counts these objects with m + 1 m+1 m + 1 black vertices and a + 1 a+1 a + 1 white vertices. No global root is chosen. The corner at each vertex is a local mark, not an additional root of the whole tree.
For a planted tree the parent edge fixes a reference in the cyclic order. A vertex with n n n children has degree n + 1 n+1 n + 1 and therefore ( n + 1 ) Cat n = ( 2 n n ) (n+1)\operatorname{Cat}_n = \binom{2n}{n} ( n + 1 ) Cat n = ( n 2 n ) local choices. Consequently, the two planted generating functions satisfy
U = ζ 1 − 4 V , V = τ 1 − 4 U , U ( 0 , 0 ) = V ( 0 , 0 ) = 0. (157) U = \frac{\zeta}{\sqrt{1-4V}}, \qquad V = \frac{\tau}{\sqrt{1-4U}}, \qquad U(0,0)=V(0,0)=0. \tag*{(157)} U = 1 − 4 V ζ , V = 1 − 4 U τ , U ( 0 , 0 ) = V ( 0 , 0 ) = 0. (157)
The weights ζ \zeta ζ and τ \tau τ mark black and white vertices, respectively. Removing the global root by the dissymmetry identity for enriched trees [20 ] gives
T Q = ζ 2 ( 1 − 1 − 4 V ) + τ 2 ( 1 − 1 − 4 U ) − U V . (158) \mathcal{T}_{\mathrm{Q}} = \frac{\zeta}{2}\left(1-\sqrt{1-4V}\right) + \frac{\tau}{2}\left(1-\sqrt{1-4U}\right) - UV. \tag*{(158)} T Q = 2 ζ ( 1 − 1 − 4 V ) + 2 τ ( 1 − 1 − 4 U ) − U V . (158)
The two positive terms count vertex-rooted trees, separated by root color; U V UV U V counts trees with a distinguished edge. Color exchange interchanges U , V U,V U , V and ζ , τ \zeta,\tau ζ , τ , giving the coefficient symmetry directly. Figure 3 shows the local decoration, and Figure 4 gives the complete example q ( 1 , 1 ) = 4 q(1,1)=4 q ( 1 , 1 ) = 4 . The algebraic identification with the QCD curve and the coefficient extraction are proved in Appendix § 13 .
Figure 3. A vertex of degree three has two full plane binary-tree decorations. The orange mark distinguishes a corner and orders the incident edges, which are matched to the binary leaves. When a parent edge is fixed, there are three possible positions for this mark and two decorations, hence 3 Cat 2 = 6 = ( 4 2 ) 3\operatorname{Cat}_2=6=\binom{4}{2} 3 Cat 2 = 6 = ( 2 4 ) local choices. The binary trees are decorations of one vertex, not extra vertices counted by ζ \zeta ζ or τ \tau τ .
Figure 4. The four decorated bipartite plane trees counted by q ( 1 , 1 ) = 4 q(1,1)=4 q ( 1 , 1 ) = 4 : two black vertices and two white vertices. The underlying tree must be a path. Each of its two degree-two vertices has a unique binary decoration and two choices of distinguished corner, indicated above or below the path. Leaf corners and their trivial decorations are suppressed. Plane isomorphisms preserve cyclic order; reflections are not identified. The resulting contribution is 4 ζ − 2 4\zeta^{-2} 4 ζ − 2 .
This result counts a formal coefficient slice of the analytic QCD generating function. It does not identify these trees with the original gauge configurations, nor does it imply that the full refined pyramid has already been enumerated. The GWW description-tree refinement in (70) and the present decorated bipartite trees therefore have different logical status: the former statistic identity remains conjectural here, whereas (155) and (158) are exact for the collapsed QCD slice.
The many-flavor limit and the relation between the two models
We connect the quartic of § 3 to the QCD sextic of § 4 at the level of the matrix integral, the algebraic equation, and the pyramid recurrence. The coefficient limit and the operator limit must be taken with the same rescaled charge and source.
The matrix-integral limit and the free-energy normalization
The relation between the models is obtained by retaining the charge per color rather than the charge per flavor. Let
k ⟶ ∞ , u Q = v k , h Q = H 2 k , (159) k\longrightarrow\infty,\qquad u_{\mathrm{Q}}=\frac{v}{k},\qquad h_{\mathrm{Q}}=\frac{H}{2k}, \tag*{(159)} k ⟶ ∞ , u Q = k v , h Q = 2 k H , (159)
with v , H v,H v , H fixed. At the level of the potential,
k log ( 1 + h Q U ) = H 2 U + O ( k − 1 ) , (160) k\log(1+h_{\mathrm{Q}}U)=\frac{H}{2}U+O(k^{-1}), \tag*{(160)} k log ( 1 + h Q U ) = 2 H U + O ( k − 1 ) , (160)
and the same expansion holds for U − 1 U^{-1} U − 1 . The two exact determinant factors in the QCD integral therefore tend to the GWW exponential. The sector insertion remains fixed because q / N = k u Q = v q/N=ku_{\mathrm{Q}}=v q / N = k u Q = v . With the static mass factor removed and the corresponding branch continued consistently, this gives
Ψ Q ( k , v k , H 2 k ) ⟶ Ψ G ( v , H ) , G Q ⟶ G G . (161) \Psi_{\mathrm{Q}}\left(k,\frac{v}{k},\frac{H}{2k}\right)\longrightarrow\Psi_{\mathrm{G}}(v,H),\qquad\mathcal{G}_{\mathrm{Q}}\longrightarrow\mathcal{G}_{\mathrm{G}}. \tag*{(161)} Ψ Q ( k , k v , 2 k H ) ⟶ Ψ G ( v , H ) , G Q ⟶ G G . (161)
The normalization in this statement matters. Since Ψ Q = k ( S + log h Q ) \Psi_{\mathrm{Q}}=k(S+\log h_{\mathrm{Q}}) Ψ Q = k ( S + log h Q ) , the free energy S S S itself does not tend directly to Ψ G \Psi_{\mathrm{G}} Ψ G . Likewise, R → − 1 / 2 R\to-1/2 R → − 1/2 , and the finite information is carried by k ( R + 1 / 2 ) = G Q k(R+1/2)=\mathcal{G}_{\mathrm{Q}} k ( R + 1/2 ) = G Q . An unrescaled comparison of R R R would lose the GWW generating function.
The determinant expansion gives the matrix-integral origin of the limit. When it is used for a planar branch, one must retain the branch that continues the relevant finite-source germ. A theorem about uniform convergence across a change of dominant saddle would require additional control. The algebraic limit below identifies the curves and their local branches without imposing such a global assertion.
The degeneration of the projective triples
The naive projective limit of the QCD triple is
[ k : k + 1 : k u Q ] = [ k : k + 1 : v ] ⟶ [ 1 : 1 : 0 ] , t ⟶ 1. (162) [k:k+1:ku_{\mathrm{Q}}]=[k:k+1:v]\longrightarrow[1:1:0],\qquad t\longrightarrow1. \tag*{(162)} [ k : k + 1 : k u Q ] = [ k : k + 1 : v ] ⟶ [ 1 : 1 : 0 ] , t ⟶ 1. (162)
Thus simply taking a limit of r , s r,s r , s erases the finite charge v v v and the source H H H . The GWW variables are transverse data near this degenerate point. They are retained by the rescaled triple
[ i ( b − a ) : c : 2 a h Q ] = [ i : v : H ] , (163) [i(b-a):c:2ah_{\mathrm{Q}}]=[i:v:H], \tag*{(163)} [ i ( b − a ) : c : 2 a h Q ] = [ i : v : H ] , (163)
which becomes [ v : H : i ] [v:H:i] [ v : H : i ] after a permutation. Equation (163) explains both the similarity and the difference of the two projective descriptions. They are related through a degeneration with a retained transverse scale, rather than by a regular identification of two generic parameter triples. In geometric terms this is the local rescaling associated with a blow-up of the degenerate point. In particular, a GWW permutation exchanging source and charge cannot be identified with an ordinary permutation of the generic QCD triple while leaving its additional mass coordinate untouched. The source enters the transverse data of the limit.
An explicit degeneration of the sextic to the quartic
Use the cleared sextic (90) . The reduction in algebraic degree can be verified directly, without radical expressions. For a formal generating-function variable g g g , substitution of x = H 2 / ( 4 k 2 ) x=H^2/(4k^2) x = H 2 / ( 4 k 2 ) gives the exact coefficientwise limit
lim k → ∞ k − 4 Q ( ( 1 − x ) ( g − v / 2 ) + k x ; k ( k + 1 ) , v , x ) = 16 P 4 ( g ; v , H ) . (164) \lim_{k\to\infty} k^{-4}Q\left((1-x)(g-v/2)+kx;k(k+1),v,x\right)=16P_4(g;v,H). \tag*{(164)} k → ∞ lim k − 4 Q ( ( 1 − x ) ( g − v /2 ) + k x ; k ( k + 1 ) , v , x ) = 16 P 4 ( g ; v , H ) . (164)
The right-hand polynomial is exactly (25) with u = v u=v u = v and h = H h=H h = H . This identity was obtained by a full expansion of the cleared polynomial; it does not depend on a finite truncation of a root series. The leading polynomial in the scaled generating-function variable has degree four. The other two generic sextic roots do not remain finite roots in this scaling, so the degree reduction is consistent with the six-sheeted curve before degeneration.
The coefficient limit and the order of the GWW recursion
The auxiliary QCD coefficients recover the GWW sequence through
R n ( v ) = lim k → ∞ f n ( k , v / k ) k 2 n , n ≥ 0. (165) R_n(v)=\lim_{k\to\infty}\frac{f_n(k,v/k)}{k^{2n}},\qquad n\geq0. \tag*{(165)} R n ( v ) = k → ∞ lim k 2 n f n ( k , v / k ) , n ≥ 0. (165)
For the first two terms, the limit gives R 0 = v / ( 1 + v ) R_0=v/(1+v) R 0 = v / ( 1 + v ) and R 1 = v / ( 1 + v ) 4 R_1=v/(1+v)^4 R 1 = v / ( 1 + v ) 4 immediately from (C.5). Together with the source rescaling in (159) , the coefficient relation (C.3) also reproduces the regular GWW free-energy series. At order H 2 n H^{2n} H 2 n , the leading contribution for n ≥ 2 n\geq2 n ≥ 2 comes from f n − 1 f_{n-1} f n − 1 , while the lower auxiliary coefficients have lower powers of k k k . The first H 2 H^2 H 2 coefficient must be retained separately, since its elementary contribution combines with f 0 f_0 f 0 .
The differential-order comparison is more subtle than the coefficient limit alone suggests. In the variables K , v K,v K , v , write
f n ( K , v ) = K n R n ( v ) + K n − 1 S n ( v ) + ⋯ . (166) f_n(K,v)=K^nR_n(v)+K^{n-1}S_n(v)+\cdots. \tag*{(166)} f n ( K , v ) = K n R n ( v ) + K n − 1 S n ( v ) + ⋯ . (166)
Coefficients multiplying parameter derivatives can raise the power of K K K . Consequently, the leading equation obtained from the QCD second-order constraint can involve both R n R_n R n and S n S_n S n . The first nontrivial example is
f 1 ( K , v ) = K v ( 1 + v ) 4 − v 2 ( 1 + v ) 3 , (167) f_1(K,v)=K\frac{v}{(1+v)^4}-\frac{v^2}{(1+v)^3}, \tag*{(167)} f 1 ( K , v ) = K ( 1 + v ) 4 v − ( 1 + v ) 3 v 2 , (167)
so that S 1 = − v 2 / ( 1 + v ) 3 S_1=-v^2/(1+v)^3 S 1 = − v 2 / ( 1 + v ) 3 . The leading parameter relation contains
0 = v 2 ( 1 − v 2 ) R 1 ′ ′ − 2 v ( 3 v 2 + 2 ) R 1 ′ + 6 ( 1 − v 2 ) R 1 + 2 v S 1 ′ − 2 v R 0 ′ . (168) 0=v^2(1-v^2)R_1''-2v(3v^2+2)R_1'+6(1-v^2)R_1
+2vS_1'-2vR_0'. \tag*{(168)} 0 = v 2 ( 1 − v 2 ) R 1 ′′ − 2 v ( 3 v 2 + 2 ) R 1 ′ + 6 ( 1 − v 2 ) R 1 + 2 v S 1 ′ − 2 v R 0 ′ . (168)
The S 1 S_1 S 1 contribution does not vanish. Dropping it does not yield a closed second-order raising equation for R 1 R_1 R 1 . This example explains why the QCD second-order parameter constraint is compatible with the GWW third-derivative recursion. The resummed chart below makes the transverse contribution explicit at generating-function level.
How the generating systems meet at the GWW boundary
The many-flavor scaling has two complementary effects. At the algebraic level, (164) reduces the QCD sextic to the GWW quartic. In the small-charge chart the same scaling is x → 0 x \to0 x → 0 with ζ , t p \zeta,t_p ζ , t p fixed and H = t p H=\sqrt{t_p} H = t p . The limit of the stripped QCD derivative is therefore A 0 A_0 A 0 , the GWW column in (117) . Conversely, once this boundary column is specified, (113) generates the regular transverse coefficients of the QCD pyramid. This gives a coefficient-level connection, in addition to the limit of the matrix potentials. In particular, the first transverse column is given exactly by (118) .
The operator limit retains this distinction between boundary and transverse information. From (149) and (152) ,
L 3 = 1 4 ( D p − 2 ) ( Q − 1 + 2 X − 2 x 1 + x ) . (169) \mathcal{L}_3=\frac{1}{4}(D_p-2)\left(Q-1+2X-\frac{2x}{1+x}\right). \tag*{(169)} L 3 = 4 1 ( D p − 2 ) ( Q − 1 + 2 X − 1 + x 2 x ) . (169)
At x → 0 x \to0 x → 0 , δ → ∞ \delta\to\infty δ → ∞ , X → ϑ X \to\vartheta X → ϑ on regular jets, and w 1 ∼ δ 2 / 4 w_1 \sim\delta^2/4 w 1 ∼ δ 2 /4 , w 2 ∼ − δ 2 / 4 w_2 \sim-\delta^2/4 w 2 ∼ − δ 2 /4 , w 3 = − δ w_3=-\delta w 3 = − δ . Equations (150) and (153) therefore show that all three L i \mathcal{L}_i L i approach
L 0 = 1 4 ( D p − 2 ) ( D p − 1 ) . (170) \mathcal{L}_0=\frac{1}{4}(D_p-2)(D_p-1). \tag*{(170)} L 0 = 4 1 ( D p − 2 ) ( D p − 1 ) . (170)
For the undivided normalized actions U i = J i A p / E 0 \mathcal{U}_i=\mathcal{J}_iA_p/E_0 U i = J i A p / E 0 , the limit is ( L 0 A 0 / 2 , L 0 A 0 / 2 , 0 ) (\mathcal{L}_0A_0/2,\mathcal{L}_0A_0/2,0) ( L 0 A 0 /2 , L 0 A 0 /2 , 0 ) : J 3 \mathcal{J}_3 J 3 remains finite and b 1 / E 0 , b 2 / E 0 → 1 / 2 b_1/E_0,b_2/E_0\to1/2 b 1 / E 0 , b 2 / E 0 → 1/2 . The two differences multiplied by δ 2 \delta^2 δ 2 approach the same boundary operator, R p ∣ x = 0 / 2 \mathcal{R}_p|_{x=0}/2 R p ∣ x = 0 /2 , which acts as ( L 0 A 0 − t p ( D p − 2 ) A 1 ) / 2 (\mathcal{L}_0A_0-t_p(D_p-2)A_1)/2 ( L 0 A 0 − t p ( D p − 2 ) A 1 ) /2 . It is not an equation for A 0 A_0 A 0 alone. Together with Q A 1 = 6 t p ζ 2 ∂ t p 2 A 0 QA_1=6t_p\zeta^2\partial_{t_p}^2A_0 Q A 1 = 6 t p ζ 2 ∂ t p 2 A 0 it gives the closed third-order identity
[ Q L 0 − 6 t p ( D p − 2 ) ∘ ( t p ζ 2 ∂ t p 2 ) ] A 0 = 0. (171) \left[Q\mathcal{L}_0-6t_p(D_p-2)\circ\left(t_p\zeta^2\partial_{t_p}^2\right)\right]A_0=0. \tag*{(171)} [ Q L 0 − 6 t p ( D p − 2 ) ∘ ( t p ζ 2 ∂ t p 2 ) ] A 0 = 0. (171)
The extra differential order comes from eliminating the first transverse coefficient, not from a regular limit of second-order operators.
For v = − ζ v=-\zeta v = − ζ , the centered GWW variable is τ G = − 4 A 0 + 2 ζ − ( 1 + ζ 2 + t p ) / 2 \tau_G=-4A_0+2\zeta-(1+\zeta^2+t_p)/2 τ G = − 4 A 0 + 2 ζ − ( 1 + ζ 2 + t p ) /2 , in the normalization of (28) . Equation (170) then gives
− 16 D p ( L 0 A 0 ) = D p ( D p − 1 ) ( D p − 2 ) τ G . (172) -16D_p(\mathcal{L}_0A_0)=D_p(D_p-1)(D_p-2)\tau_G. \tag*{(172)} − 16 D p ( L 0 A 0 ) = D p ( D p − 1 ) ( D p − 2 ) τ G . (172)
This recovers the radial third-order GWW action after an additional Euler derivative. The full standard GWW system (51) –(53) follows on the limiting quartic, where both independent differences have zero polynomial remainder (Appendix § 9 ). It is not a componentwise limit of (152) . For example, the QCD exchange a ↔ b a\leftrightarrow b a ↔ b induces ( v , H ) ↦ ( − v , − H ) (v,H)\mapsto(-v,-H) ( v , H ) ↦ ( − v , − H ) in the chosen scaling, not an exchange of the GWW charge and source. This is consistent with the transverse rescaling in (163) and explains why the shared abstract S 4 S_4 S 4 does not identify the two triples in this degeneration.
Differential order and ordinary equations in a finite algebra
Algebraic degree, PDE order, and fixed-parameter ODE order
The quartic and sextic count algebraic branches at generic parameters. Their degrees do not specify the order of every differential equation satisfied by a particular generating function. A mixed PDE may use parameter derivatives that cannot be eliminated without increasing the order in the remaining mass variable. The QCD equation (136) , for example, has order two in ( r , s , t ) (r,s,t) ( r , s , t ) , whereas an ordinary equation at fixed r , s r,s r , s can have higher order.
There are two further distinctions. First, an invertible change of independent coordinates and an invertible affine change of dependent variable preserve the order of a specified differential operator. Finding an additional second-order equation for GWW does not turn its known third-order equation into a second-order one by a coordinate change. Second, replacing a potential by its derivative is not an invertible point transformation. It loses integration data, and the order of the corresponding equation can decrease by one. This is what occurs in passing from (137) for the potential to (136) for its normalized mass derivative.
The coefficient recursions illustrate the same distinction in discrete form. GWW has a third-order operator acting explicitly on the preceding coefficient. QCD has second-order parameter derivatives acting on both adjacent coefficients. The latter is an implicit relation and is not a lower-order explicit raising formula. On special parameter slices the algebraic curve and its differential equations can degenerate further; such reductions do not determine the generic order.
Differentiation in the quotient algebra
A uniform construction of ordinary equations is available directly from an algebraic polynomial. Let P ( t , w ) P(t,w) P ( t , w ) be monic and separable of degree d d d in w w w , and let K K K be the rational-function field in t t t and the fixed external parameters. In the finite algebra
A = K [ w ] / ( P ) , (173) \mathcal{A} = K[w]/(P), \tag*{(173)} A = K [ w ] / ( P ) , (173)
differentiation along a local root is represented by
∇ t f = [ ∂ t f − ∂ w f P t P w − 1 ] m o d P . (174) \nabla_t f = \left[\partial_t f - \partial_w f P_t P_w^{-1}\right] \bmod P. \tag*{(174)} ∇ t f = [ ∂ t f − ∂ w f P t P w − 1 ] mod P . (174)
The inverse of P w P_w P w is taken modulo P P P . It exists on the regular part of the finite covering, where the polynomial is separable. All successive derivatives of an algebraic function can therefore be represented in a fixed finite-dimensional space, using rational coefficients and polynomial reduction.
The trace in this algebra is computable without writing its roots:
Tr A / K ( f ) = [ w d − 1 ] rem P ( f P w ) . (175) \operatorname{Tr}_{\mathcal{A}/K}(f) = [w^{d-1}]\,\operatorname{rem}_P(fP_w). \tag*{(175)} Tr A / K ( f ) = [ w d − 1 ] rem P ( f P w ) . (175)
If a centered element V V V has zero trace, all its derivatives have zero trace as well. They lie in a space of dimension at most d − 1 d-1 d − 1 . Set v j = ∇ t j V v_j = \nabla_t^j V v j = ∇ t j V and form
H i j = Tr ( v i v j ) , b i = Tr ( v i v d − 1 ) , 0 ≤ i , j ≤ d − 2. (176) H_{ij} = \operatorname{Tr}(v_i v_j), \qquad b_i = \operatorname{Tr}(v_i v_{d-1}), \qquad0 \le i,j \le d-2. \tag*{(176)} H ij = Tr ( v i v j ) , b i = Tr ( v i v d − 1 ) , 0 ≤ i , j ≤ d − 2. (176)
On a chart where det H ≠ 0 \det H \ne0 det H = 0 , the coefficients of an ordinary annihilator are
[ ∂ t d − 1 − ∑ j = 0 d − 2 γ j ∂ t j ] V = 0 , γ j = det H [ j ← b ] det H . (177) \left[\partial_t^{d-1} - \sum_{j=0}^{d-2} \gamma_j \partial_t^j\right]V = 0, \qquad\gamma_j = \frac{\det H[j \leftarrow b]}{\det H}. \tag*{(177)} [ ∂ t d − 1 − j = 0 ∑ d − 2 γ j ∂ t j ] V = 0 , γ j = det H det H [ j ← b ] . (177)
This is a closed rational formula even when expanded coefficients are large. The determinant also identifies where the chosen cyclic basis fails; some resulting poles can be apparent rather than physical singularities. Establishing the minimal possible order requires a separate rank argument, not just the degree count.
For the depressed GWW quartic, τ \tau τ has zero trace, and the construction gives an equation of order at most three at fixed charge. For QCD the coefficient of W 5 W^5 W 5 is 1 / 2 1/2 1/2 , so the six roots sum to − 1 / 2 -1/2 − 1/2 . Hence
V Q = W + 1 12 , Tr V Q = 0 , (178) V_Q = W + \frac{1}{12}, \qquad\operatorname{Tr} V_Q = 0, \tag*{(178)} V Q = W + 12 1 , Tr V Q = 0 , (178)
and the centered mass-dependent algebraic function has an ordinary annihilator of order at most five. This order is compatible with the second-order mixed PDE because the latter also differentiates the external invariants.
The relation to the uncentered mass derivative can be stated explicitly. At fixed k , u k,u k , u , let
L 5 = ∂ t 5 − ∑ j = 0 4 γ j ∂ t j , D t = ∂ t + 1 2 t . (179) L_5=\partial_t^5-\sum_{j=0}^{4}\gamma_j\partial_t^j,\qquad D_t=\partial_t+\frac{1}{2t}. \tag*{(179)} L 5 = ∂ t 5 − j = 0 ∑ 4 γ j ∂ t j , D t = ∂ t + 2 t 1 . (179)
Using z 2 = t z^2=t z 2 = t and (81) ,
R = E 0 k z V Q + 6 k 2 − E 0 12 k z . (180) R=\frac{E_0}{kz}V_Q+\frac{6k^2-E_0}{12kz}. \tag*{(180)} R = k z E 0 V Q + 12 k z 6 k 2 − E 0 . (180)
The extra elementary term can be eliminated by an additional first-order factor. On a generic chart with γ 0 ≠ 0 \gamma_0\ne0 γ 0 = 0 , this gives
( D t − ∂ t γ 0 γ 0 ) ( D t 5 − ∑ j = 0 4 γ j D t j ) R = 0. (181) \left(D_t-\frac{\partial_t\gamma_0}{\gamma_0}\right)\left(D_t^5-\sum_{j=0}^{4}\gamma_jD_t^j\right)R=0. \tag*{(181)} ( D t − γ 0 ∂ t γ 0 ) ( D t 5 − j = 0 ∑ 4 γ j D t j ) R = 0. (181)
Products here mean composition of noncommuting differential operators: the outer derivative also differentiates inner coefficients. Equation (181) is an ordinary sixth-order representation at fixed parameters. It does not replace or contradict the third-order mixed equation for S S S .
Structural comparison and scope of the formulation
The two models share a projective parameter geometry, but their algebraic and differential realizations retain distinct features. The GWW invariant quartic requires an orientation covariant unless the algebraic variable is multiplied by the same orientation factor. The QCD sextic is symmetric in the squared homogeneous coordinates from the outset. In GWW the source is part of the projective triple; in QCD the mass appears as a separate invariant coordinate. These differences explain why the common signed-permutation group does not make the two operator systems identical at generic parameters.
The comparison can be summarized as follows.
Table 2.
The cyclic QCD equation explains how an alternating complete operator can coexist with permuted components. The counterexample (C.33) concerns the original actions on the shifted free energy; the divided differences (152) have equal actions precisely because they express one weighted equation. Their common GWW boundary operator is given by (170) . The full third-order GWW system comes from the limiting quartic, with the role of the first transverse coefficient made explicit by (171) .
The row recurrence (109) is a differential constraint on two rows, not a one-step raising operator. Nevertheless, (113) gives an explicit reconstruction in the formal class, using GWW data and a diagonal Euler inverse. This statement is separate from generation by the original small-h h h rational ansatz in Appendix § 11 , for which an all-order uniqueness theorem in that ansatz would be an additional result.
The combinatorial comparison likewise has a precise scope. Equation (117) embeds the GWW series as one boundary of the QCD pyramid, and (122) gives the Narayana triangle on the other. The exact identity (133) recovers the critical GWW series at the Narayana pole. The positivity certificates in (130) apply through the sixth deformation order in every row. Equation (155) counts a different, formally collapsed slice by two vertex colors. A tree statistic retaining the full x x x dependence, and a direct relation between the two tree families, remain to be established. Color exchange of the collapsed array is not by itself an identification with the projective S 4 S_4 S 4 action on the original physical parameters.
Finally, the projective symmetries describe analytic algebraic branches. They do not by themselves select the dominant physical saddle after a charge extremization or a sector sum. The real GWW parameterization provides a useful branch prescription within its one-cut family, while the QCD small-h h h boundary condition identifies a fixed-sector germ. Global continuation across physical phase boundaries, complex-source cancellations, and alternative operator decompositions with a more direct contraction to GWW remain separate questions. These questions can be addressed using the algebraic curves and differential systems without changing the normalization dictionary developed above.
Exact algebraic verification of the differential identities
The differential identities can be checked without selecting radical expressions for individual roots. Let P ( w , ξ ) = 0 P(w,\xi)=0 P ( w , ξ ) = 0 be a monic algebraic relation and assume P w ≠ 0 P_w\ne0 P w = 0 on a regular branch. Implicit differentiation gives
w i = − P i P w , w i j = J i j P w 3 . (182) w_i=-\frac{P_i}{P_w},\qquad w_{ij}=\frac{J_{ij}}{P_w^3}. \tag*{(182)} w i = − P w P i , w ij = P w 3 J ij . (182)
with
J i j = − P i j P w 2 + ( P i w P j + P j w P i ) P w − P w w P i P j . (183) J_{ij}=-P_{ij}P_w^2+(P_{iw}P_j+P_{jw}P_i)P_w-P_{ww}P_iP_j. \tag*{(183)} J ij = − P ij P w 2 + ( P i w P j + P j w P i ) P w − P w w P i P j . (183)
Multiplying a proposed second-order PDE by P w 3 P_w^3 P w 3 reduces its verification to a polynomial identity modulo P P P .
For the QCD equation (136) , the exact remainder is
0 = rem P 6 [ 2 r ( 4 r − 1 ) J r r + 4 s ( 6 r − 1 ) J r s + 18 s 2 J s s + 2 t ( t − 1 ) J r t − ( 4 r − t ) P r P w 2 − 12 s P s P w 2 ] . (184) 0=\operatorname{rem}_{P^6}\left[2r(4r-1)J_{rr}+4s(6r-1)J_{rs}+18s^2J_{ss}+2t(t-1)J_{rt}-(4r-t)P_rP_w^2-12sP_sP_w^2\right]. \tag*{(184)} 0 = rem P 6 [ 2 r ( 4 r − 1 ) J r r + 4 s ( 6 r − 1 ) J r s + 18 s 2 J ss + 2 t ( t − 1 ) J r t − ( 4 r − t ) P r P w 2 − 12 s P s P w 2 ] . (184)
Every coefficient vanishes in rational polynomial arithmetic. This proves the relation for every regular local root of the sextic. The same procedure verifies (59) against the canonical quartic. At a ramification point, the displayed implicit-derivative fractions require a local continuation analysis, even though the polynomial identities remain meaningful.
For a third derivative along a parameter derivation δ \delta δ , such as ∂ u \partial_u ∂ u , ∂ h \partial_h ∂ h , or u ∂ u + h ∂ h u\partial_u+h\partial_h u ∂ u + h ∂ h , define
J δ = − ( δ 2 P ) P w 2 + 2 ( δ P w ) ( δ P ) P w − P w w ( δ P ) 2 , (185) J_\delta=-(\delta^2P)P_w^2+2(\delta P_w)(\delta P)P_w-P_{ww}(\delta P)^2, \tag*{(185)} J δ = − ( δ 2 P ) P w 2 + 2 ( δ P w ) ( δ P ) P w − P w w ( δ P ) 2 , (185)
N δ = [ ( δ J δ ) w − ( J δ ) w δ P ] P w − 3 J δ [ ( δ P w ) P w − P w w δ P ] . (186) N_\delta=\left[(\delta J_\delta)_w-(J_\delta)_w\delta P\right]P_w-3J_\delta\left[(\delta P_w)P_w-P_{ww}\delta P\right]. \tag*{(186)} N δ = [ ( δ J δ ) w − ( J δ ) w δ P ] P w − 3 J δ [ ( δ P w ) P w − P w w δ P ] . (186)
Then δ 2 w = J δ / P w 3 \delta^{2}w=J_{\delta}/P_{w}^{3} δ 2 w = J δ / P w 3 and δ 3 w = N δ / P w 5 \delta^{3}w=N_{\delta}/P_{w}^{5} δ 3 w = N δ / P w 5 . For GWW, the two independent equations in (53) reduce to
rem P 4 ( h N ∂ h − u N ∂ u ) = 0 (187) \operatorname{rem}_{P_{4}}(hN_{\partial h}-uN_{\partial u})=0 \tag*{(187)} rem P 4 ( h N ∂ h − u N ∂ u ) = 0 (187)
and
rem P 4 [ N D − 3 J D P w 2 − 2 ( D P ) P w 4 − u N ∂ u ] = 0. (188) \operatorname{rem}_{P_{4}}\left[N_{D}-3J_{D}P_{w}^{2}-2(DP)P_{w}^{4}-uN_{\partial u}\right]=0. \tag*{(188)} rem P 4 [ N D − 3 J D P w 2 − 2 ( D P ) P w 4 − u N ∂ u ] = 0. (188)
Both remainders vanish exactly. The second formula accounts for the lower powers of D D D in Δ 0 = D 3 − 3 D 2 + 2 D \Delta_{0}=D^{3}-3D^{2}+2D Δ 0 = D 3 − 3 D 2 + 2 D .
The distinction between these all-parameter verifications and a series check is useful in practice. A finite list of coefficients can reveal or refute a proposed relation, but it does not prove the identity on the complete curve. Conversely, an exact remainder proves the algebraic identity but does not select its physical branch. These are complementary checks with different purposes.
Derivation of the cyclic QCD decomposition
The cyclic construction can be made explicit on the homogeneous chart used in (142) . For a degree-one function Y ( A 1 , A 2 , A 3 ; t ) Y(A_{1},A_{2},A_{3};t) Y ( A 1 , A 2 , A 3 ; t ) with A 1 = 1 A_{1}=1 A 1 = 1 , write B = A 2 B=A_{2} B = A 2 and C = A 3 C=A_{3} C = A 3 . Homogeneity implies
Y A 1 = Y − B Y B − C Y C , Y A 1 A 1 = B 2 Y B B + 2 B C Y B C + C 2 Y C C . (189) Y_{A_{1}}=Y-BY_{B}-CY_{C},\qquad Y_{A_{1}A_{1}}=B^{2}Y_{BB}+2BCY_{BC}+C^{2}Y_{CC}. \tag*{(189)} Y A 1 = Y − B Y B − C Y C , Y A 1 A 1 = B 2 Y B B + 2 B C Y B C + C 2 Y C C . (189)
Using these formulas in (140) gives the differential expression
C Y = B 2 ( B − 1 ) Y B B + 2 B C ( B − C ) Y B C − C 2 ( C − 1 ) Y C C + t 2 [ B ( B − 2 C + 1 ) Y B + C ( 2 B − C − 1 ) Y C − ( B − C ) Y ] − t 2 [ B ( B − 2 C + 1 ) Y B + C ( 2 B − C − 1 ) Y C − ( B − C ) Y ] . (190) \begin{aligned}
\mathcal{C}Y={}&B^{2}(B-1)Y_{BB}+2BC(B-C)Y_{BC}-C^{2}(C-1)Y_{CC}\\
&+\frac{t}{2}\left[B(B-2C+1)Y_{B}+C(2B-C-1)Y_{C}-(B-C)Y\right]\\
&-\frac{t}{2}\left[B(B-2C+1)Y_{B}+C(2B-C-1)Y_{C}-(B-C)Y\right]. \tag*{(190)}
\end{aligned} C Y = B 2 ( B − 1 ) Y B B + 2 B C ( B − C ) Y B C − C 2 ( C − 1 ) Y C C + 2 t [ B ( B − 2 C + 1 ) Y B + C ( 2 B − C − 1 ) Y C − ( B − C ) Y ] − 2 t [ B ( B − 2 C + 1 ) Y B + C ( 2 B − C − 1 ) Y C − ( B − C ) Y ] . (190)
The form of the coefficients follows from the same cyclic differences that multiply the three Euler operators. Substitution of r = ( B + C + B C ) / ( 1 + B + C ) 2 r=(B+C+BC)/(1+B+C)^{2} r = ( B + C + B C ) / ( 1 + B + C ) 2 , s = B C / ( 1 + B + C ) 3 s=BC/(1+B+C)^{3} s = B C / ( 1 + B + C ) 3 , and W = Y / ( 1 + B + C ) W=Y/(1+B+C) W = Y / ( 1 + B + C ) into (136) yields (142) term by term. This supplies a direct operator-level derivation of the triple from the invariant PDE.
The logarithmic leading term also admits a simple homogeneous check. In the physical chart it has coefficient L = k u − k ( k + 1 ) L=ku-k(k+1) L = k u − k ( k + 1 ) . Its homogeneous lift is
L ( a , b , c ) = b c − a c − a b . (191) L(a,b,c)=bc-ac-ab. \tag*{(191)} L ( a , b , c ) = b c − a c − ab . (191)
For each squared-coordinate Euler operator, D i ( D i − 1 / 2 ) L = 0 D_{i}(D_{i}-1/2)L=0 D i ( D i − 1/2 ) L = 0 , because every monomial contains each unsquared coordinate at most once. Applying (145) to ( L / 2 ) log x (L/2)\log x ( L /2 ) log x gives
J i ( x ) [ L 2 log x ] = 1 + x x D i ( D i − 1 / 2 ) L = 0. (192) \mathcal{J}_{i}^{(x)}\left[\frac{L}{2}\log x\right]=\frac{1+x}{x}D_{i}(D_{i}-1/2)L=0. \tag*{(192)} J i ( x ) [ 2 L log x ] = x 1 + x D i ( D i − 1/2 ) L = 0. (192)
Thus the logarithmic leading term is annihilated by each component separately. The coefficient recursion (C.29) follows from the regular part alone, without an omitted inhomogeneous contribution from the leading logarithm.
Finally, inserting ∑ n ≥ 1 ϕ n x n \sum_{n\geq1}\phi_{n}x^{n} ∑ n ≥ 1 ϕ n x n into (C.27) gives the coefficient of x n x^{n} x n as
( n + 1 ) [ 2 U − ( 2 n + 3 ) V ] ϕ n + 1 + n [ 2 U + ( 2 n − 1 ) V ] ϕ n . (193) (n+1)[2U-(2n+3)V]\phi_{n+1}+n[2U+(2n-1)V]\phi_{n}. \tag*{(193)} ( n + 1 ) [ 2 U − ( 2 n + 3 ) V ] ϕ n + 1 + n [ 2 U + ( 2 n − 1 ) V ] ϕ n . (193)
The x 0 x^{0} x 0 coefficient gives ( 2 U − 3 V ) ϕ 1 = 0 (2U-3V)\phi_{1}=0 ( 2 U − 3 V ) ϕ 1 = 0 . This derivation also explains why derivatives of the new coefficient are an intrinsic part of the recurrence, rather than an artifact of notation.
Recurrences in the original QCD parameters
These formulas use the expansion at fixed ( k , u ) (k,u) ( k , u ) in powers of x = h 2 x=h^2 x = h 2 . They give complementary differential constraints to the resummed small-charge recurrence in § 5.1 . The normalization dictionary is useful when computing the original free-energy coefficients or checking the action of projective maps.
The small-h h h series and an auxiliary generating function
For comparison with the pyramid expansion, use the physical variables k , u k,u k , u . Write
S = S l e a d i n g + ∑ n ≥ 1 C n ( k , u ) x n , ∂ h S l e a d i n g = u − 1 h . (194) S = S_{\mathrm{leading}} + \sum_{n \ge1} C_n(k,u)x^n,\qquad\partial_h S_{\mathrm{leading}} = \frac{u-1}{h}. \tag*{(194)} S = S leading + n ≥ 1 ∑ C n ( k , u ) x n , ∂ h S leading = h u − 1 . (194)
An auxiliary generating function F = ∑ n ≥ 0 f n x n F=\sum_{n\ge0}f_nx^n F = ∑ n ≥ 0 f n x n is defined by
G Q = k u 2 + k 2 x − k ( k + 1 ) x F 1 − x . (195) G_Q = \frac{ku}{2}+\frac{k^2x-k(k+1)xF}{1-x}. \tag*{(195)} G Q = 2 k u + 1 − x k 2 x − k ( k + 1 ) x F . (195)
It is related to the original action coefficients by
C n = k − ( k + 1 ) ∑ j = 0 n − 1 f j n . (196) C_n = \frac{k-(k+1)\sum_{j=0}^{n-1}f_j}{n}. \tag*{(196)} C n = n k − ( k + 1 ) ∑ j = 0 n − 1 f j . (196)
Conversely,
f 0 = k − C 1 k + 1 , f n = n C n − ( n + 1 ) C n + 1 k + 1 ( n ≥ 1 ) . (197) f_0 = \frac{k-C_1}{k+1},\qquad f_n = \frac{nC_n-(n+1)C_{n+1}}{k+1}\quad(n\ge1). \tag*{(197)} f 0 = k + 1 k − C 1 , f n = k + 1 n C n − ( n + 1 ) C n + 1 ( n ≥ 1 ) . (197)
The distinction between f n f_n f n and C n C_n C n is substantive: an operator acting on one of these sequences must be conjugated or combined with the coefficient relation before it is applied to the other. The first auxiliary coefficients are
f 0 = k u 1 + k u . (198) f_0 = \frac{ku}{1+ku}. \tag*{(198)} f 0 = 1 + k u k u . (198)
f 1 = k 2 u ( 1 − u ) [ 1 + k ( 1 + u ) ] ( 1 + k u ) 4 , (199) f_1 = \frac{k^2u(1-u)[1+k(1+u)]}{(1+ku)^4}, \tag*{(199)} f 1 = ( 1 + k u ) 4 k 2 u ( 1 − u ) [ 1 + k ( 1 + u )] , (199)
f 2 = k 2 u ( 1 − u ) [ 1 + k ( 1 + u ) ] ( 1 + k u ) 7 [ ( 1 + k u ) ( 1 + k 2 u 2 ) + 2 k ( k + 1 ) ( 1 − k u ) ] . (200) f_2 = \frac{k^2u(1-u)[1+k(1+u)]}{(1+ku)^7}
\left[(1+ku)(1+k^2u^2)+2k(k+1)(1-ku)\right]. \tag*{(200)} f 2 = ( 1 + k u ) 7 k 2 u ( 1 − u ) [ 1 + k ( 1 + u )] [ ( 1 + k u ) ( 1 + k 2 u 2 ) + 2 k ( k + 1 ) ( 1 − k u ) ] . (200)
In particular,
C 1 = k ( 1 − u ) 1 + k u . (201) C_1 = \frac{k(1-u)}{1+ku}. \tag*{(201)} C 1 = 1 + k u k ( 1 − u ) . (201)
As a normalization check, at k = 1 k=1 k = 1 , u = 1 / 2 u=1/2 u = 1/2 the first five action coefficients are
( C 1 , C 2 , C 3 , C 4 , C 5 ) = ( 1 3 , 7 162 , − 431 6561 , − 6739 78732 , − 40159 7971615 ) . (202) (C_1,C_2,C_3,C_4,C_5)=\left(\frac{1}{3},\frac{7}{162},-\frac{431}{6561},-\frac{6739}{78732},-\frac{40159}{7971615}\right). \tag*{(202)} ( C 1 , C 2 , C 3 , C 4 , C 5 ) = ( 3 1 , 162 7 , − 6561 431 , − 78732 6739 , − 7971615 40159 ) . (202)
At saturation u = 1 u=1 u = 1 , all these regular coefficients vanish. At u = 0 u=0 u = 0 , the weak branch has C n = k / n C_n=k/n C n = k / n , giving Ψ Q = − k 2 log ( 1 − h 2 ) \Psi_Q=-k^2\log(1-h^2) Ψ Q = − k 2 log ( 1 − h 2 ) with Ψ Q ( 0 ) = 0 \Psi_Q(0)=0 Ψ Q ( 0 ) = 0 . This latter expression describes a local weak small-h h h branch and should not be extrapolated as the physical answer at arbitrary h h h without a separate branch analysis.
The implicit differential recursion
Introduce the coefficient polynomials
a 0 = 2 k 2 ( 1 − u 2 ) + 2 k + 1 , b 0 = k 2 ( 1 + u 2 ) + k , c 0 = k 2 ( u 2 − 1 ) − 4 k − 2 (203) a_0 = 2k^2(1-u^2)+2k+1,\qquad b_0 = k^2(1+u^2)+k,\qquad c_0 = k^2(u^2-1)-4k-2 \tag*{(203)} a 0 = 2 k 2 ( 1 − u 2 ) + 2 k + 1 , b 0 = k 2 ( 1 + u 2 ) + k , c 0 = k 2 ( u 2 − 1 ) − 4 k − 2 (203)
and the second-order operator
L = k 2 u ( u 2 − 1 ) ∂ u 2 + 2 k ( k + 1 ) u [ ( k + 1 ) 2 − k 2 u 2 ] ∂ u ∂ k − k 2 ( k + 1 ) 2 ( 2 k + 1 ) ∂ k 2 . (204) L = k^2u(u^2-1)\partial_u^2+2k(k+1)u[(k+1)^2-k^2u^2]\partial_u\partial_k-k^2(k+1)^2(2k+1)\partial_k^2. \tag*{(204)} L = k 2 u ( u 2 − 1 ) ∂ u 2 + 2 k ( k + 1 ) u [( k + 1 ) 2 − k 2 u 2 ] ∂ u ∂ k − k 2 ( k + 1 ) 2 ( 2 k + 1 ) ∂ k 2 . (204)
The two operators needed for the recursion are
M = L − 2 k ( k + 1 ) b 0 ∂ k + 2 u c 0 ∂ u + 2 ( 2 k + 1 ) , (205) M=L-2k(k+1)b_0\partial_k+2uc_0\partial_u+2(2k+1), \tag*{(205)} M = L − 2 k ( k + 1 ) b 0 ∂ k + 2 u c 0 ∂ u + 2 ( 2 k + 1 ) , (205)
P = 2 k ( k + 1 ) a 0 ∂ k + 2 u c 0 ∂ u + 2 ( 2 k + 1 ) . (206) P=2k(k+1)a_0\partial_k+2uc_0\partial_u+2(2k+1). \tag*{(206)} P = 2 k ( k + 1 ) a 0 ∂ k + 2 u c 0 ∂ u + 2 ( 2 k + 1 ) . (206)
Terms without derivatives denote multiplication. Although M M M and P P P enter symmetrically in the notation, their differential orders differ: M M M is second order and P P P is first order.
The exact adjacent-coefficient relation is
( M + n P ) f n = ( n P − M ) f n − 1 , n ≥ 1. (207) (M+nP)f_n=(nP-M)f_{n-1},\qquad n\ge1. \tag*{(207)} ( M + n P ) f n = ( n P − M ) f n − 1 , n ≥ 1. (207)
This is a differential constraint on both the new and the previous coefficient. It can be used effectively once the rational class of f n f_n f n is specified, but it is not an explicit second-order replacement for the GWW raising operator. In particular, the left-hand derivatives cannot be discarded when testing the recurrence.
A finite ansatz compatible with the first coefficients is
f n = k 2 u ( 1 − u ) [ 1 + k ( 1 + u ) ] ( 1 + k u ) 3 n + 1 ∑ a = 0 n − 1 ∑ b = 0 3 ( n − 1 ) − 2 a c a b [ k ( k + 1 ) ] a ( k u ) b , n ≥ 1. (208) f_n=\frac{k^2u(1-u)[1+k(1+u)]}{(1+ku)^{3n+1}}\sum_{a=0}^{n-1}\sum_{b=0}^{3(n-1)-2a}c_{ab}[k(k+1)]^a(ku)^b,\qquad n\ge1. \tag*{(208)} f n = ( 1 + k u ) 3 n + 1 k 2 u ( 1 − u ) [ 1 + k ( 1 + u )] a = 0 ∑ n − 1 b = 0 ∑ 3 ( n − 1 ) − 2 a c ab [ k ( k + 1 ) ] a ( k u ) b , n ≥ 1. (208)
Substituting this ansatz into (C.14) gives a finite linear system for the coefficients c a b c_{ab} c ab . The system has a unique solution in this class for the first five nontrivial coefficients that have been checked. This supplies a concrete generation procedure; a proof of uniqueness for every n n n would be an additional result.
Using (C.3), the recurrence can also be written directly for the original free-energy coefficients:
( n + 1 ) ( M + n P ) [ C n + 1 k + 1 ] = n [ ( n + 1 ) P − M ] [ C n k + 1 ] . (209) (n+1)(M+nP)\left[\frac{C_{n+1}}{k+1}\right]=n[(n+1)P-M]\left[\frac{C_n}{k+1}\right]. \tag*{(209)} ( n + 1 ) ( M + n P ) [ k + 1 C n + 1 ] = n [( n + 1 ) P − M ] [ k + 1 C n ] . (209)
The denominator belongs inside the argument of the operator. Since the operators differentiate in k k k , moving 1 / ( k + 1 ) 1/(k+1) 1/ ( k + 1 ) outside them changes the recurrence. This is a useful practical distinction when checking individual coefficients in a symbolic calculation.
The equation for the unshifted free energy
The coefficient relation can be summed into a PDE for the complete free energy. Define
A = ( k + 1 ) M ( k + 1 ) − 1 , B = ( k + 1 ) P ( k + 1 ) − 1 , D 0 = ( k + 1 ) 2 − k 2 u 2 . (210) A=(k+1)M(k+1)^{-1},\qquad B=(k+1)P(k+1)^{-1},\qquad D_0=(k+1)^2-k^2u^2. \tag*{(210)} A = ( k + 1 ) M ( k + 1 ) − 1 , B = ( k + 1 ) P ( k + 1 ) − 1 , D 0 = ( k + 1 ) 2 − k 2 u 2 . (210)
Explicitly,
A = L + 2 k 3 ( k + 1 ) ( 1 − u 2 ) ∂ k + 2 u ( c 0 − k D 0 ) ∂ u + 2 [ k 3 ( u 2 − 1 ) + 2 k + 1 ] . (211) A=L+2k^3(k+1)(1-u^2)\partial_k+2u(c_0-kD_0)\partial_u+2[k^3(u^2-1)+2k+1]. \tag*{(211)} A = L + 2 k 3 ( k + 1 ) ( 1 − u 2 ) ∂ k + 2 u ( c 0 − k D 0 ) ∂ u + 2 [ k 3 ( u 2 − 1 ) + 2 k + 1 ] . (211)
B = 2 k ( k + 1 ) a 0 ∂ k + 2 u c 0 ∂ u + 2 ( 2 k + 1 − k a 0 ) . (212) B=2k(k+1)a_0\partial_k+2uc_0\partial_u+2(2k+1-ka_0). \tag*{(212)} B = 2 k ( k + 1 ) a 0 ∂ k + 2 u c 0 ∂ u + 2 ( 2 k + 1 − k a 0 ) . (212)
The full equation, including the logarithmic leading term, is
E S : = h ( 1 − h 2 ) B [ S h h ] + [ 2 ( 1 + h 2 ) A − ( 1 + 3 h 2 ) B ] [ S h ] = − 4 k ( 1 − h 2 ) h D 0 = : R . (213) \mathcal{E}S := h(1-h^2)B[S_{hh}] + [2(1+h^2)A-(1+3h^2)B][S_h] = -\frac{4k(1-h^2)}{h}D_0 =:\mathcal{R}. \tag*{(213)} E S := h ( 1 − h 2 ) B [ S hh ] + [ 2 ( 1 + h 2 ) A − ( 1 + 3 h 2 ) B ] [ S h ] = − h 4 k ( 1 − h 2 ) D 0 =: R . (213)
The first term contains one parameter derivative and two mass derivatives. The second contains up to two parameter derivatives and one mass derivative. The total order is therefore three, even though S h h S_{hh} S hh appears multiplied by a differential operator. Replacing the coefficient index by a source derivative does not produce a fourth order here because the coefficient of the higher source derivative is only first order in the external parameters.
The operators in (C.20) are not literally invariant scalar operators on S S S . If T ∗ T^* T ∗ denotes parameter pullback, the exact laws are
T ∗ A ( T ∗ ) − 1 = ρ T α T A α T − 1 , T ∗ B ( T ∗ ) − 1 = ρ T α T B α T − 1 , (214) T^*A(T^*)^{-1}=\rho_{T\alpha}TA_\alpha T^{-1}, \qquad T^*B(T^*)^{-1}=\rho_{T\alpha}TB_\alpha T^{-1}, \tag*{(214)} T ∗ A ( T ∗ ) − 1 = ρ T α T A α T − 1 , T ∗ B ( T ∗ ) − 1 = ρ T α T B α T − 1 , (214)
with
ρ a b = − 1 , ρ b c = − 1 k 2 ( u − 1 ) 2 , ρ a c = − 1 ( k + 1 − k u ) 2 . (215) \rho_{ab}=-1,\qquad\rho_{bc}=-\frac{1}{k^2(u-1)^2},\qquad\rho_{ac}=-\frac{1}{(k+1-ku)^2}. \tag*{(215)} ρ ab = − 1 , ρ b c = − k 2 ( u − 1 ) 2 1 , ρ a c = − ( k + 1 − k u ) 2 1 . (215)
Consequently
E T = ρ T α T E α T − 1 , T ∗ R = ρ T α T R + E T [ β T ℓ ] . (216) \mathcal{E}_T=\rho_{T\alpha}T\mathcal{E}_\alpha T^{-1},\qquad T^*\mathcal{R}=\rho_{T\alpha}T\mathcal{R}+\mathcal{E}_T[\beta_T\ell]. \tag*{(216)} E T = ρ T α T E α T − 1 , T ∗ R = ρ T α T R + E T [ β T ℓ ] . (216)
The source term and the logarithmic part of the free-energy transformation are thus tied to the operator covariance. For example, T a b ∗ B ( T a b ∗ ) − 1 = − B + 2 a 0 T^*_{ab}B(T^*_{ab})^{-1}=-B+2a_0 T ab ∗ B ( T ab ∗ ) − 1 = − B + 2 a 0 , which exhibits its literal noninvariance. After the weight of S S S and the affine logarithmic term are included, the transformed equation is equivalent to the original one.
This covariance concerns a specific complete operator and its chosen two-component representation. It does not rule out a different representation in which a triple of components is permuted. The homogeneous potential (105) makes precisely that representation possible.
Physical-coordinate realization and the original action coefficients
The homogeneous construction can be used without differentiating in three new independent parameters. To derive its physical realization, write d 0 = b − a d_0=b-a d 0 = b − a , k = a / d 0 k=a/d_0 k = a / d 0 , u = c / a u=c/a u = c / a , and Φ ( a , b , c ) = d 0 2 Φ ( k , u ) \Phi(a,b,c)=d_0^2\Phi(k,u) Φ ( a , b , c ) = d 0 2 Φ ( k , u ) . Applying D i = ( a i / 2 ) ∂ a i D_i=(a_i/2)\partial_{a_i} D i = ( a i /2 ) ∂ a i and then setting d 0 = 1 d_0=1 d 0 = 1 gives
D 1 = k ( k + 1 ) 2 ∂ k − u 2 ∂ u − k , D 2 = − k ( k + 1 ) 2 ∂ k + ( k + 1 ) , D 3 = u 2 ∂ u . (217) \begin{aligned}
D_1&=\frac{k(k+1)}{2}\partial_k-\frac{u}{2}\partial_u-k,\\
D_2&=-\frac{k(k+1)}{2}\partial_k+(k+1),\\
D_3&=\frac{u}{2}\partial_u. \tag*{(217)}
\end{aligned} D 1 D 2 D 3 = 2 k ( k + 1 ) ∂ k − 2 u ∂ u − k , = − 2 k ( k + 1 ) ∂ k + ( k + 1 ) , = 2 u ∂ u . (217)
The zeroth-order terms are the contributions of the homogeneous weight. They ensure D 1 + D 2 + D 3 = 1 D_1+D_2+D_3=1 D 1 + D 2 + D 3 = 1 and the commutativity of the physical-coordinate representation. Removing them would no longer represent the same Euler operators on the degree-one functions.
The cyclic weights become
w 1 = ( k + 1 ) 2 − k 2 u 2 = D 0 , w 2 = k 2 ( u 2 − 1 ) , w 3 = − ( 2 k + 1 ) . (218) w_1=(k+1)^2-k^2u^2=D_0,\qquad w_2=k^2(u^2-1),\qquad w_3=-(2k+1). \tag*{(218)} w 1 = ( k + 1 ) 2 − k 2 u 2 = D 0 , w 2 = k 2 ( u 2 − 1 ) , w 3 = − ( 2 k + 1 ) . (218)
They multiply from the left in (146) . The parameter derivatives inside each I i \mathcal{I}_i I i do not act on the external weight w i w_i w i . Define
U = ∑ i = 1 3 w i D i 2 , V = ∑ i = 1 3 w i D i . (219) U=\sum_{i=1}^{3}w_iD_i^2,\qquad V=\sum_{i=1}^{3}w_iD_i. \tag*{(219)} U = i = 1 ∑ 3 w i D i 2 , V = i = 1 ∑ 3 w i D i . (219)
Here D i 2 D_i^2 D i 2 denotes operator composition, including differentiation of its internal coefficients. The cyclic equation has the compact form
2 ( 1 + x ) U [ Φ x ] + ( x − 3 ) V [ Φ x ] − 2 x ( 1 − x ) V [ Φ x x ] = 0. (220) 2(1+x)U[\Phi_x]+(x-3)V[\Phi_x]-2x(1-x)V[\Phi_{xx}]=0. \tag*{(220)} 2 ( 1 + x ) U [ Φ x ] + ( x − 3 ) V [ Φ x ] − 2 x ( 1 − x ) V [ Φ xx ] = 0. (220)
Substitution of Φ = k S − k 2 ℓ \Phi=kS-k^2\ell Φ = k S − k 2 ℓ shows that its left-hand side is exactly − k / 2 -k/2 − k /2 times the residual of (C.20) expressed in x x x , namely the residual in h h h divided by 4 h 4h 4 h . This checks the source term and logarithmic shift together with the principal higher-derivative terms.
The relation to the original small-source coefficients is particularly simple:
Φ = k u − k ( k + 1 ) 2 log x + Φ 0 ( k , u ) + ∑ n ≥ 1 ϕ n ( k , u ) x n , ϕ n = k C n − k 2 n . (221) \Phi=\frac{ku-k(k+1)}{2}\log x+\Phi_0(k,u)+\sum_{n\geq1}\phi_n(k,u)x^n,\qquad\phi_n=kC_n-\frac{k^2}{n}. \tag*{(221)} Φ = 2 k u − k ( k + 1 ) log x + Φ 0 ( k , u ) + n ≥ 1 ∑ ϕ n ( k , u ) x n , ϕ n = k C n − n k 2 . (221)
The leading logarithm separately satisfies (C.27); the function Φ 0 \Phi_0 Φ 0 is annihilated by the mass derivative. Thus the regular coefficients obey the homogeneous differential recursion
( n + 1 ) [ 2 U − ( 2 n + 3 ) V ] ϕ n + 1 = − n [ 2 U + ( 2 n − 1 ) V ] ϕ n , n ≥ 1. (222) (n+1)[2U-(2n+3)V]\phi_{n+1}=-n[2U+(2n-1)V]\phi_n,\qquad n\geq1. \tag*{(222)} ( n + 1 ) [ 2 U − ( 2 n + 3 ) V ] ϕ n + 1 = − n [ 2 U + ( 2 n − 1 ) V ] ϕ n , n ≥ 1. (222)
The initial coefficient and its equation are
ϕ 1 = − k 2 ( k + 1 ) u 1 + k u , ( 2 U − 3 V ) ϕ 1 = 0. (223) \phi_1=-\frac{k^2(k+1)u}{1+ku},\qquad(2U-3V)\phi_1=0. \tag*{(223)} ϕ 1 = − 1 + k u k 2 ( k + 1 ) u , ( 2 U − 3 V ) ϕ 1 = 0. (223)
The next one is
ϕ 2 = − k 2 u ( k + 1 ) [ k 3 u 3 + 2 k 2 u 2 + k 2 + 2 k u + k + 1 ] 2 ( 1 + k u ) 4 . (224) \phi_2=-\frac{k^2u(k+1)[k^3u^3+2k^2u^2+k^2+2ku+k+1]}{2(1+ku)^4}. \tag*{(224)} ϕ 2 = − 2 ( 1 + k u ) 4 k 2 u ( k + 1 ) [ k 3 u 3 + 2 k 2 u 2 + k 2 + 2 k u + k + 1 ] . (224)
At k = 1 k=1 k = 1 , u = 1 / 2 u=1/2 u = 1/2 ,
( ϕ 1 , ϕ 2 , ϕ 3 ) = ( − 2 3 , − 37 81 , − 2618 6561 ) . (225) (\phi_1,\phi_2,\phi_3)=\left(-\frac{2}{3},-\frac{37}{81},-\frac{2618}{6561}\right). \tag*{(225)} ( ϕ 1 , ϕ 2 , ϕ 3 ) = ( − 3 2 , − 81 37 , − 6561 2618 ) . (225)
After the parameter derivatives are evaluated, the two sides of (C.29) equal 6 at n = 1 n=1 n = 1 and 250 / 9 250/9 250/9 at n = 2 n=2 n = 2 . Four consecutive transitions, through n = 4 n=4 n = 4 , have been checked exactly against the rational coefficients obtained independently from the auxiliary recurrence.
The cyclic structure is genuinely different from the pairwise equality in (51) . In the same example, the three actions at x → 0 x\to0 x → 0 are
( J 1 ( x ) Φ , J 2 ( x ) Φ , J 3 ( x ) Φ ) = ( 10 27 , − 14 27 , 16 27 ) . (226) \left(\mathcal{J}_1^{(x)}\Phi,\mathcal{J}_2^{(x)}\Phi,\mathcal{J}_3^{(x)}\Phi\right)=\left(\frac{10}{27},-\frac{14}{27},\frac{16}{27}\right). \tag*{(226)} ( J 1 ( x ) Φ , J 2 ( x ) Φ , J 3 ( x ) Φ ) = ( 27 10 , − 27 14 , 27 16 ) . (226)
while ( w 1 , w 2 , w 3 ) = ( 15 / 4 , − 3 / 4 , − 3 ) (w_1,w_2,w_3)=(15/4,-3/4,-3) ( w 1 , w 2 , w 3 ) = ( 15/4 , − 3/4 , − 3 ) gives a zero weighted sum. Replacing the cyclic equation by pairwise equality would therefore impose an incorrect additional condition. A different normalization or additional operators would be needed to realize that stronger analogy. The present formulation already supplies the permuted components and the exact third-order equation, without assuming the stronger property.
The counterexample concerns these original components on the shifted free energy. The divided differences in (152) act on the regular pyramid derivative and re-express one cyclic constraint; their equal actions do not contradict this test.
Coefficient extraction and exact checks for the pyramid
The coefficient form of (113) , useful when working at fixed row m m m , is
( j + 1 ) t p E m A m , j + 1 = [ t p E m ( j − H m ) − 2 ( H m − 2 j ) ( H m + 1 ) ] A m , j + 2 E m − 2 ( E m − 2 + 2 j + 1 ) A m − 2 , j + 4 E m − 2 ( H m − 2 − 2 j + 2 ) A m − 2 , j − 1 + ( 4 j − 6 m + 4 ) E m − 2 A m − 2 , j − 2 . (227) \begin{aligned}
(j+1)t_pE_mA_{m,j+1}={}&[t_pE_m(j-H_m)-2(H_m-2j)(H_m+1)]A_{m,j}\\
&+2E_{m-2}(E_{m-2}+2j+1)A_{m-2,j}\\
&+4E_{m-2}(H_{m-2}-2j+2)A_{m-2,j-1}\\
&+(4j-6m+4)E_{m-2}A_{m-2,j-2}. \tag*{(227)}
\end{aligned} ( j + 1 ) t p E m A m , j + 1 = [ t p E m ( j − H m ) − 2 ( H m − 2 j ) ( H m + 1 )] A m , j + 2 E m − 2 ( E m − 2 + 2 j + 1 ) A m − 2 , j + 4 E m − 2 ( H m − 2 − 2 j + 2 ) A m − 2 , j − 1 + ( 4 j − 6 m + 4 ) E m − 2 A m − 2 , j − 2 . (227)
Terms with a negative row or column index are zero. The diagonal inverse of E m E_m E m on t p b t_p^b t p b divides by m − 1 + 2 b m-1+2b m − 1 + 2 b . This is the coefficient form of (114) , so it does not require solving a differential boundary-value problem in t p t_p t p .
For a direct operator check of (150) , restrict (149) to the coefficient of ζ m + 1 \zeta^{m+1} ζ m + 1 and write J i , m \mathcal{J}_{i,m} J i , m for the resulting operator. Let f = δ ( 1 − x ) / [ 8 ( 1 + x ) ] f=\delta(1-x)/[8(1+x)] f = δ ( 1 − x ) / [ 8 ( 1 + x )] . Then
1 4 ( δ + 1 ) 2 J 1 , m − 1 4 ( δ − 1 ) 2 J 2 , m − δ J 3 , m = − f L m , ( 1 − x ) 2 ( J 2 , m − J 1 , m ) = 2 f E m M m . (228) \frac{1}{4}(\delta+1)^2\mathcal{J}_{1,m}-\frac{1}{4}(\delta-1)^2\mathcal{J}_{2,m}-\delta\mathcal{J}_{3,m}=-f\mathcal{L}_m,
\qquad
(1-x)^2(\mathcal{J}_{2,m}-\mathcal{J}_{1,m})=2fE_m\mathcal{M}_m. \tag*{(228)} 4 1 ( δ + 1 ) 2 J 1 , m − 4 1 ( δ − 1 ) 2 J 2 , m − δ J 3 , m = − f L m , ( 1 − x ) 2 ( J 2 , m − J 1 , m ) = 2 f E m M m . (228)
The first line collects the charge-independent part of the weights; the second multiplies the contribution shifted by ζ 2 \zeta^2 ζ 2 . Composition in these identities differentiates the internal coefficient δ \delta δ . The symbolic operator check uses an arbitrary test function; the additional checks on ten explicit rows are not needed for the identity.
To obtain the second-column relation independently of the coefficient table, put p 0 = P ~ ∣ x = 0 / t p 2 p_0=\left.\widetilde{P}\right|_{x=0}/t_p^2 p 0 = P x = 0 / t p 2 and p 1 = ∂ x P ~ ∣ x = 0 / t p 2 p_1=\left.\partial_x\widetilde{P}\right|_{x=0}/t_p^2 p 1 = ∂ x P x = 0 / t p 2 . At a = A 0 a=A_0 a = A 0 the first transverse coefficient is A 1 = − p 1 / p 0 , a A_1=-p_1/p_{0,a} A 1 = − p 1 / p 0 , a . Define the derivative along the quartic by ∇ i = ∂ i − ( p 0 , i / p 0 , a ) ∂ a \nabla_i=\partial_i-(p_{0,i}/p_{0,a})\partial_a ∇ i = ∂ i − ( p 0 , i / p 0 , a ) ∂ a for i = ζ , t p i=\zeta,t_p i = ζ , t p . Clearing denominators in
∇ ζ ( − p 1 p 0 , a ) − 6 t p ζ ∇ t p 2 a (229) \nabla_\zeta\left(-\frac{p_1}{p_{0,a}}\right)-6t_p\zeta\nabla_{t_p}^2a \tag*{(229)} ∇ ζ ( − p 0 , a p 1 ) − 6 t p ζ ∇ t p 2 a (229)
gives a numerator of zero polynomial remainder modulo p 0 p_0 p 0 . This proves the relation at every regular point of the algebraic expansion, and hence as a formal identity for the selected germ. The companion scripts construct the cleared curve, perform this reduction, and check the leading factor used in the Narayana derivation.
The collapsed QCD slice: derivation and enumeration
We give an algebraic proof connecting the tree construction to the QCD sextic, rather than infer that connection from a finite table. Use the substitution (108) in the cleared polynomial (90) , with a = A p ( ζ , 4 τ , x ) a=A_p(\zeta,4\tau,x) a = A p ( ζ , 4 τ , x ) . After this substitution Q Q Q has a factor ( 1 − x ) 10 (1-x)^{10} ( 1 − x ) 10 . Dividing by this factor before taking x → 1 x\to1 x → 1 gives exactly the following quintic for g = T Q g=T_Q g = T Q , with a = g a=g a = g . Put σ = ζ + τ \sigma=\zeta+\tau σ = ζ + τ and π = ζ τ \pi=\zeta\tau π = ζ τ . Then
F ( g ; ζ , τ ) = 16 g 5 + c 4 g 4 + c 3 g 3 + c 2 g 2 + c 1 g + c 0 = 0 , (230) F(g;\zeta,\tau)=16g^5+c_4g^4+c_3g^3+c_2g^2+c_1g+c_0=0, \tag*{(230)} F ( g ; ζ , τ ) = 16 g 5 + c 4 g 4 + c 3 g 3 + c 2 g 2 + c 1 g + c 0 = 0 , (230)
with
c 4 = − 32 π + 16 σ 2 − 40 σ + 1 , c 3 = 2 ( 8 π 2 + 32 π σ + 8 π − 16 σ 3 + 16 σ 2 − σ ) , c 2 = − 24 π 2 σ − 62 π 2 − 16 π σ 2 − 24 π σ + π + 16 σ 4 − 8 σ 3 + σ 2 , c 1 = π ( 72 π 2 − 24 π σ 2 + 62 π σ − 2 π − 16 σ 3 + 8 σ 2 − σ ) , c 0 = − π 2 ( 27 π 2 + 36 π σ − π − 16 σ 3 + 8 σ 2 − σ ) . (231) \begin{aligned}
c_4&=-32\pi+16\sigma^2-40\sigma+1,\\
c_3&=2(8\pi^2+32\pi\sigma+8\pi-16\sigma^3+16\sigma^2-\sigma),\\
c_2&=-24\pi^2\sigma-62\pi^2-16\pi\sigma^2-24\pi\sigma+\pi+16\sigma^4-8\sigma^3+\sigma^2,\\
c_1&=\pi(72\pi^2-24\pi\sigma^2+62\pi\sigma-2\pi-16\sigma^3+8\sigma^2-\sigma),\\
c_0&=-\pi^2(27\pi^2+36\pi\sigma-\pi-16\sigma^3+8\sigma^2-\sigma). \tag*{(231)}
\end{aligned} c 4 c 3 c 2 c 1 c 0 = − 32 π + 16 σ 2 − 40 σ + 1 , = 2 ( 8 π 2 + 32 π σ + 8 π − 16 σ 3 + 16 σ 2 − σ ) , = − 24 π 2 σ − 62 π 2 − 16 π σ 2 − 24 π σ + π + 16 σ 4 − 8 σ 3 + σ 2 , = π ( 72 π 2 − 24 π σ 2 + 62 π σ − 2 π − 16 σ 3 + 8 σ 2 − σ ) , = − π 2 ( 27 π 2 + 36 π σ − π − 16 σ 3 + 8 σ 2 − σ ) . (231)
The coefficient germ is g = ζ τ + O t o t ( 3 ) g=\zeta\tau+O_{\mathrm{tot}}(3) g = ζ τ + O tot ( 3 ) . The polynomial is symmetric in ζ , τ \zeta,\tau ζ , τ , but its derivative with respect to g g g is not a unit at the origin, so the ordinary implicit-function theorem does not establish branch uniqueness there. Instead assign weights 2 , 1 , 1 2,1,1 2 , 1 , 1 to g , ζ , τ g,\zeta,\tau g , ζ , τ . The lowest weighted part of F F F is
ζ τ ( ζ + τ ) ( ζ τ − g ) . (232) \zeta\tau(\zeta+\tau)(\zeta\tau-g). \tag*{(232)} ζ τ ( ζ + τ ) ( ζ τ − g ) . (232)
For two putative roots with the indicated germ, the divided difference of F F F has leading term − ζ τ ( ζ + τ ) ≠ 0 -\zeta\tau(\zeta+\tau)\ne0 − ζ τ ( ζ + τ ) = 0 . Since the formal power-series ring is an integral domain, the two roots coincide. Thus the germ, whenever constructed, is unique.
To check the tree expression exactly, put
U = 1 − r 0 2 4 , V = 1 − s 0 2 4 , ζ = U s 0 , τ = V r 0 , r 0 = s 0 = 1 at the origin . (233) U = \frac{1-r_0^2}{4}, \qquad V = \frac{1-s_0^2}{4}, \qquad\zeta= Us_0, \qquad\tau= Vr_0, \qquad r_0=s_0=1\text{ at the origin}. \tag*{(233)} U = 4 1 − r 0 2 , V = 4 1 − s 0 2 , ζ = U s 0 , τ = V r 0 , r 0 = s 0 = 1 at the origin . (233)
Equations (157) hold, and the right-hand side g t r e e g_{\mathrm{tree}} g tree of (158) becomes
16 g t r e e = ( r 0 − 1 ) ( s 0 − 1 ) ( 3 r 0 s 0 + r 0 + s 0 − 1 ) . (234) 16g_{\mathrm{tree}}=(r_0-1)(s_0-1)(3r_0s_0+r_0+s_0-1). \tag*{(234)} 16 g tree = ( r 0 − 1 ) ( s 0 − 1 ) ( 3 r 0 s 0 + r 0 + s 0 − 1 ) . (234)
Substitution into (E.1) gives the zero polynomial in r 0 r_0 r 0 , s 0 s_0 s 0 , with no series truncation. Its germ is ζ τ + O t o t ( 3 ) \zeta\tau+O_{\mathrm{tot}}(3) ζ τ + O tot ( 3 ) , establishing the identity with the collapsed QCD branch by uniqueness.
For completeness, coefficient extraction gives the product formula without eliminating U U U , V V V . Every tree counted by ζ M τ N \zeta^{M}\tau^{N} ζ M τ N has M + N − 1 M+N-1 M + N − 1 edges. Edge marking therefore yields
( ζ ∂ ζ + τ ∂ τ − 1 ) T Q = U V . (235) (\zeta\partial_{\zeta}+\tau\partial_{\tau}-1)T_Q=UV. \tag*{(235)} ( ζ ∂ ζ + τ ∂ τ − 1 ) T Q = U V . (235)
The multivariate Lagrange–Good formula applied to (157) reads
[ ζ M τ N ] U V = [ u 0 M − 1 v 0 N − 1 ] ( 1 − 4 v 0 ) − M / 2 ( 1 − 4 u 0 ) − N / 2 ( 1 − 4 u 0 v 0 ( 1 − 4 u 0 ) ( 1 − 4 v 0 ) ) = 4 M + N − 2 ( N / 2 ) M − 1 ( M / 2 ) N − 1 ( M − 1 ) ! ( N − 1 ) ! ( 1 − ( M − 1 ) ( N − 1 ) M N ) . (236) \begin{aligned}
[\zeta^M\tau^N]UV=[u_0^{M-1}v_0^{N-1}](1-4v_0)^{-M/2}(1-4u_0)^{-N/2}\left(1-\frac{4u_0v_0}{(1-4u_0)(1-4v_0)}\right) \\
&=\frac{4^{M+N-2}(N/2)_{M-1}(M/2)_{N-1}}{(M-1)!(N-1)!}\left(1-\frac{(M-1)(N-1)}{MN}\right). \tag*{(236)}
\end{aligned} [ ζ M τ N ] U V = [ u 0 M − 1 v 0 N − 1 ] ( 1 − 4 v 0 ) − M /2 ( 1 − 4 u 0 ) − N /2 ( 1 − ( 1 − 4 u 0 ) ( 1 − 4 v 0 ) 4 u 0 v 0 ) = ( M − 1 )! ( N − 1 )! 4 M + N − 2 ( N /2 ) M − 1 ( M /2 ) N − 1 ( 1 − M N ( M − 1 ) ( N − 1 ) ) . (236)
Combining (E.6) and (E.7), then setting M = m + 1 M=m+1 M = m + 1 , N = a + 1 N=a+1 N = a + 1 , proves (155) . The locally marked corners also rule out nontrivial automorphisms: a tree automorphism fixes its central vertex or its central edge; the colors prevent exchanging the endpoints of the latter, and a fixed vertex with a marked corner fixes all its incident edges. This justifies ordinary, rather than automorphism-weighted, counting in the dissymmetry identity.