betaFeedback

WorksPhysics and Astronomy

Beyond hidden zeros and uniqueness in Tr(ϕ3)(ϕ^3): Block and rod zeros in any Tr(ϕm)(ϕ^m)

Abstract

Hidden zeros have proven powerful tools for reconstructing amplitudes in Tr(ϕ3)(ϕ^3), a theory of colored scalars with cubic interactions. We show that these structures extend to any Tr(ϕm)(ϕ^m) theory. For odd mm, we construct block zeros from orthogonal sums of consecutive momenta, generalizing the original cubic hidden zeros. For every m≥4m\geq4, including even mm, we describe a second family of nonlinear rod zeros through a rod packing problem. In Tr(ϕ4)(ϕ^4) this construction reduces to a simple rank-one condition on a matrix of momentum contractions. We connect this new zero to the classic threshold nullification zeros of Voloshin, where certain amplitudes in massive ϕ4ϕ^4 theory vanish when all produced particles are at rest. Finally, we demonstrate the reconstruction power of these zeros. Assuming locality, we prove that the rod zeros uniquely fix all Tr(ϕm)(ϕ^m) tree amplitudes for m≥4m\geq4 up to normalization. We also conjecture that the block zeros enforce both locality and unitarity for any Tr(ϕm)(ϕ^m) theory with odd mm.

Introduction

Hidden zeros are kinematic loci on which scattering amplitudes vanish while every propagator remains finite. In Tr⁡(ϕ3)\operatorname{Tr}(\phi^3), the nonlinear sigma model (NLSM), and Yang–Mills theory [1], they arise from nontrivial cancellations between diagrams that are not apparent from the usual Feynman expansion. Requiring an amplitude to vanish on such a locus relates terms with different poles, and can therefore constrain both the allowed pole combinations and their residues.

This gives hidden zeros a natural role in the uniqueness program, which asks how much of an amplitude can be determined from a small set of principles. In a conventional tree construction, locality determines which poles may occur together, while unitarity fixes their residues through products of lower-point amplitudes. Reconstruction from gauge invariance, soft behavior, and scaling at large complex momenta has shown that these properties need not always be assumed independently [2, 3, 4, 5, 6, 7, 8, 9].

Hidden zeros realize this possibility in tree-level Tr⁡(ϕ3)\operatorname{Tr}(\phi^3), where they enforce locality [10] and unitarity [11] at every multiplicity within a prescribed planar pole ansatz. Their constraining power extends to loop integrands, where, assuming locality, tree-boundary zeros together with zeros from Kontsevich’s matrix Airy function [12] determine relative factorization and cut data in the loop cases studied [13, 14]. Reconstruction from zeros has also been developed for pion amplitudes, cosmological wavefunctions or inverse Kawai–Lewellen–Tye (KLT) amplitudes [15, 16, 17], with further examples of hidden zeros found in particle and string theories and form factors [18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35]. The relation to lower-point functions becomes explicit in splitting relations motivated by zeros [36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51]. These cancellations also have a classical explanation based on the equation of motion of the scalar theory [17].

Despite this range of examples, many of the structures found in other theories remain connected to Tr⁡(ϕ3)\operatorname{Tr}(\phi^3) through kinematic shifts, scalar scaffolding, and double copy [52, 53, 54, 55, 56]. It is therefore important to determine whether zeros and their reconstruction power reflect a more general principle or depend on this cubic origin. Pure Tr⁡(ϕm)\operatorname{Tr}(\phi^m) theories provide a direct way to test this question without introducing spin or derivative interactions.

Extending the rectangle zeros has nevertheless proved nontrivial, as the next natural target, Tr⁡(ϕ4)\operatorname{Tr}(\phi^4), does not seem to admit a comparable general pattern of simple linear zeros.1

In this Letter we extend the zero structures and their reconstructive power to all Tr⁡(ϕm)\operatorname{Tr}(\phi^m). We find in fact that the simplest extension starts with Tr⁡(ϕ5)\operatorname{Tr}(\phi^5) and continues to every odd mm. The resulting block zeros replace orthogonality between individual momenta by orthogonality between consecutive momentum blocks,

Bi=pi+⋯+pi+m−3,pi⋅pj=0⟹Bi⋅Bj=0.(1)B_i = p_i+\cdots+p_{i+m-3}, \qquad p_i\cdot p_j=0 \quad\Longrightarrow\quad B_i\cdot B_j=0. \tag*{(1)}

A selection rule specifies which blocks must be orthogonal, recovering the usual rectangle zeros when m=3m=3. We also find that these conditions recover locality and relative factorization residues in finite explicit checks, even when the ansatz allows crossing and repeated poles.

Returning to Tr⁡(ϕ4)\operatorname{Tr}(\phi^4), we show that zeros can instead be specified by a nonlinear rank-one condition on a matrix of adjacent-pair contractions. Surprisingly, this condition connects hidden zeros to threshold nullification, the classic cancellation of certain production amplitudes when all produced particles are at rest [57, 58, 59], and extends that zero to continuous real scattering configurations away from threshold. To our knowledge, threshold nullification has not been yet connected to hidden zero developments.

We then extend the matrix condition to a second family, the rod zeros, described by a rod packing problem for every m≥4m \ge4. For this family, we prove that locality and the zero conditions together fix all relative diagram coefficients at every multiplicity.

Block zeros for odd interactions

We first introduce the block zeros, which generalize rectangular hidden zeros, which can be easily recovered when m=3m = 3.

We take massless outgoing momenta and define planar channels Xab=(pa+⋯+pb−1)2X_{ab} = (p_a + \cdots+ p_{b-1})^2. A tree with VV interaction vertices has n=(m−2)V+2n = (m - 2)V + 2 external legs and V−1V - 1 propagators. Its diagrams correspond to divisions of an nn-gon into mm-gons by noncrossing diagonals, called mm-angulations [60, 61, 62, 63].

For odd m≥3m \ge3 and V≥2V \ge2, the rule uses a set of nonadjacent external momenta. Schematically, the zero conditions have the form

Bi⋅Bj=0,Bi=∑r=0m−3pi+r,(2)B_i \cdot B_j = 0,\qquad B_i = \sum_{r=0}^{m-3} p_{i+r}, \tag*{(2)}

where BiB_i is a block momentum. The rule below determines which blocks must be orthogonal.

We give the construction in four steps. To aid visualization, we display the momenta pip_i as the vertices of a polygon in cyclic order, as in Figure 1.

Complete block zero in $\operatorname{Tr}(\phi^5)$

Figure 1. The complete block zero in Tr⁡(ϕ5)\operatorname{Tr}(\phi^5) with V=3V = 3 and n=11n = 11, defined by the M=5M = 5 picked momenta p2,p4,p7,p9,p11p_2,p_4,p_7,p_9,p_{11}. Vertices are momenta, with the initial selection in blue and the chosen separators in orange. The brackets identify the blocks BiB_i. Purple and green mark arcs AA and BB. Their picked-momentum densities (rA/NA,rB/NB)(r_A/N_A,r_B/N_B) are (2/6,1/3)(2/6,1/3) in (a) and (b), (1/3,2/6)(1/3,2/6) in (c), and (1/4,2/5)(1/4,2/5) in (d), always excluding the separators. Of the ten candidate separator pairs, exactly the three in (a)–(c) have equal density. Impose every displayed equation from these panels together. The contraction B1⋅B8B_1 \cdot B_8 occurs twice, leaving five distinct conditions. Panel (d) has unequal densities and contributes no conditions. The other six discarded pairs are not drawn.

Step 1. Pick the momenta. Pick

M=m−32V+2(3)M = \frac{m-3}{2}V + 2 \tag*{(3)}

nonadjacent external momenta. This choice defines the zero. The next steps determine all conditions defining this particular locus.

Step 2. Select the separator pairs. For each picked pair pi,pjp_i,p_j, let arc AA run from pi+1p_{i+1} to pj−1p_{j-1} and arc BB from pj+1p_{j+1} to pi−1p_{i-1}, cyclically. Let NA,NBN_A,N_B count all momenta in the arcs and rA,rBr_A,r_B count the picked momenta among them. Retain the pair exactly when

rANA=rBNB.(4)\frac{r_A}{N_A} = \frac{r_B}{N_B}. \tag*{(4)}

The integer counts restrict the possible arc lengths. The common density is (m−3)/[2(m−2)](m - 3)/[2(m - 2)], so the arc lengths and picked counts must take the form

NA=(m−2)h,rA=m−32h,N_A = (m - 2)h,\qquad r_A = \frac{m-3}{2}h,
NB=(m−2)(V−h),rB=m−32(V−h),(5)N_B = (m - 2)(V-h),\qquad r_B = \frac{m-3}{2}(V-h), \tag*{(5)}

with 1≤h<V1 \le h < V. Checking either arc suffices.2

Step 3. Impose the block rectangle. Starting immediately after each separator, divide the arcs into consecutive blocks of m−2m - 2 momenta. Require every block on one arc to be orthogonal to every block on the other,

Bi+1+α(m−2)⋅Bj+1+β(m−2)=0,0≤α<h,0≤β<V−h.(6)\begin{aligned} B_{i+1+\alpha(m-2)} \cdot B_{j+1+\beta(m-2)} &= 0,\\ 0 &\le\alpha< h,\qquad0 \le\beta< V-h. \tag*{(6)} \end{aligned}

This is an h×(V−h)h \times(V-h) rectangle, of width min⁡(h,V−h)\min(h,V-h).

Step 4. Combine the conditions. For one fixed picked set, impose every retained rectangle simultaneously. Each picked set specifies one locus, and these loci are tested separately.3

Every block equation is linear in physical channels, with the four-corner form in Eq. (30). We prove

An(m)=0(7)A_n^{(m)}=0 \tag*{(7)}

on a nonempty regular locus for every picked set. Section S2 proves regularity and establishes vanishing by factorization. Section S1 gives higher-point examples in Tr⁡(ϕ6)\operatorname{Tr}(\phi^6) and Tr⁡(ϕ7)\operatorname{Tr}(\phi^7).

The Tr(ϕ3\phi^3) limit and a Tr(ϕ5\phi^5) example

For m=3m=3, M=2M=2 and exactly two nonadjacent momenta are picked. Neither arc contains another picked momentum, so rA=rB=0r_A=r_B=0 and Eq. (4) is automatic. Each block is simply one momentum, Bi=piB_i=p_i, and the rule gives precisely

pa⋅pb=0,a∈A,b∈B,(8)p_a\cdot p_b=0,\qquad a\in A,\quad b\in B, \tag*{(8)}

the usual rectangle zeros [1], with no additional conditions. At six points, picking p2,p6p_2,p_6 gives the width-one zero

p1⋅p3=p1⋅p4=p1⋅p5=0,(9)p_1\cdot p_3=p_1\cdot p_4=p_1\cdot p_5=0, \tag*{(9)}

while picking p3,p6p_3,p_6 gives the width-two zero

pa⋅pb=0,a=1,2,b=4,5.(10)p_a\cdot p_b=0,\qquad a=1,2,\quad b=4,5. \tag*{(10)}

Thus the cubic specialization recovers the rectangular family at every multiplicity.

For Tr⁡(ϕ5)\operatorname{Tr}(\phi^5) at eight points, pick for example p2,p4,p6,p8p_2,p_4,p_6,p_8. The only retained separator pairs are (p4,p8)(p_4,p_8) and (p2,p6)(p_2,p_6). Each intervening arc has three momenta, one of them picked. With Bi=pi+pi+1+pi+2B_i=p_i+p_{i+1}+p_{i+2}, their conditions are

B1⋅B5=0⟺X15+X48=0,B3⋅B7=0⟺X26+X37=0.(11)\begin{aligned} B_1\cdot B_5&=0\quad\Longleftrightarrow\quad X_{15}+X_{48}=0,\\ B_3\cdot B_7&=0\quad\Longleftrightarrow\quad X_{26}+X_{37}=0. \tag*{(11)} \end{aligned}

They cancel the complete amplitude in pairs,

A8(5)=(1X15+1X48)+(1X26+1X37)=0.(12)A_8^{(5)}=\left(\frac{1}{X_{15}}+\frac{1}{X_{48}}\right)+\left(\frac{1}{X_{26}}+\frac{1}{X_{37}}\right)=0. \tag*{(12)}

Figure 1 shows all three retained separator pairs for one picked set at eleven points, together with a discarded pair. The retained rectangles give five distinct block conditions, all displayed below the polygons. The eleven- and fourteen-point channel conditions are given in § 2.1 A and S1 B.

A rank condition in Tr(ϕ4\phi^4)

We now consider Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) theory. In this case, we only find generic and natural zero loci that are nonlinear. At six points, the three diagrams give

A6(4)=1X14+1X25+1X36=X14X25+X14X36+X25X36X14X25X36.(13)A_6^{(4)}=\frac{1}{X_{14}}+\frac{1}{X_{25}}+\frac{1}{X_{36}} =\frac{X_{14}X_{25}+X_{14}X_{36}+X_{25}X_{36}}{X_{14}X_{25}X_{36}}. \tag*{(13)}

Remarkably, the numerator has a simple matrix interpretation. For any even n≥6n\ge6, form the matrix of contractions between adjacent momentum pairs,

Cij=2(p2i+1+p2i+2)⋅(p2j+2+p2j+3),(14)C_{ij}=2(p_{2i+1}+p_{2i+2})\cdot(p_{2j+2}+p_{2j+3}), \tag*{(14)}

with i,j=0,…,n/2−1i,j=0,\ldots,n/2-1 and cyclic labels. At six points it is

C=(X14−X14−X36X36X25X36−X25−X36−X14−X25X14X25).(15)C=\begin{pmatrix} X_{14} & -X_{14}-X_{36} & X_{36}\\ X_{25} & X_{36} & -X_{25}-X_{36}\\ -X_{14}-X_{25} & X_{14} & X_{25} \end{pmatrix}. \tag*{(15)}

Every 2×22\times2 minor is the numerator of Eq. (13), up to sign. Thus the amplitude at six points vanishes precisely when rank C=1C=1, provided all physical channels are nonzero. We show that the same condition gives zeros at every multiplicity,

rank⁡C=1⟹An(4)=0.(16)\operatorname{rank}C=1\quad\Longrightarrow\quad A_n^{(4)}=0. \tag*{(16)}

All 2×22\times2 minors must vanish. The determinant alone is uninformative because momentum conservation already makes every row and column sum zero.

For V=3V=3, the eight-point amplitude uses a 4×44\times4 matrix. Its upper-left minor gives

det⁡(X14X16−X14−X36X25X36)=0.(17)\det\begin{pmatrix} X_{14} & X_{16}-X_{14}-X_{36}\\ X_{25} & X_{36} \end{pmatrix}=0. \tag*{(17)}

All the other minors are imposed as well. The full matrix appears in § 2.1 D, Eq. (37). This minor already explains the cancellation mechanism. It gives

1X16(1X14+1X25+1X36)=1X14X25X36.(18)\frac{1}{X_{16}}\left(\frac{1}{X_{14}}+\frac{1}{X_{25}}+\frac{1}{X_{36}}\right)=\frac{1}{X_{14}X_{25}X_{36}}. \tag*{(18)}

The rank condition simplifies every proper interval in this way, and the final tree sum vanishes. The rod packing construction in the next section extends this mechanism to general mm.

Real scattering and threshold nullification

We connect the Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) rank condition to threshold nullification. In massive unbroken ϕ4\phi^4 theory, the full tree amplitude for 2→r2 \to r scattering vanishes for every even r>4r > 4 when all outgoing particles are at rest, through cancellations among finite diagrams [57, 58, 59]. We show that the corresponding ordered zero belongs to a continuous family of real zeros extending above threshold.

Real channel values need not come from on-shell Lorentzian momenta satisfying momentum conservation and the Gram constraints. On the rank-one locus, however, we show that every regular real point admits such a realization in 1+21+2 dimensions with a common mass. Section S4 A gives the construction, with alternating incoming and outgoing legs describing n/2→n/2n/2 \to n/2 scattering.

To reach threshold production, consider instead two incoming particles. Let n=2V+2n=2V+2 and give all external and exchanged scalars mass M≥0M \ge0, with inverse propagators X^ab=(pa+⋯+pb−1)2−M2\hat{X}_{ab}=(p_a+\cdots+p_{b-1})^2-M^2. In mostly-minus signature, take adjacent back-to-back pairs,

p2i−1=(bi/2,qi),p2i=(bi/2,−qi),p_{2i-1}=(b_i/2,\mathbf{q}_i),\qquad p_{2i}=(b_i/2,-\mathbf{q}_i),
∣qi∣2=bi2/4−M2,∑i=1V+1bi=0.(19)|\mathbf{q}_i|^2=b_i^2/4-M^2,\qquad\sum_{i=1}^{V+1}b_i=0. \tag*{(19)}

with ∣bi∣≥2M|b_i| \ge2M. Each pair carries only energy, so the matrix in Eq. (14) factorizes as

Ci−1,j−1=bi(bj+bj+1).(20)C_{i-1,j-1}=b_i(b_j+b_{j+1}). \tag*{(20)}

For regular configurations this has rank one, independently of the pair directions.

An allowed interval contains complete pairs with total energy BB and one constituent of a pair with energy bb. Its inverse propagator is

X^=(B+b/2)2−(b2/4−M2)−M2=B(B+b).(21)\hat{X}=(B+b/2)^2-(b^2/4-M^2)-M^2 =B(B+b). \tag*{(21)}

At fixed pair energies, the channels are independent of the pair directions and mass. The rank-one proof applies to X^\hat{X} because its adjacent boundary entries vanish and its cyclic mixed differences still give CC.

Taking b1<0b_1<0 and all other bi>0b_i>0 gives 2→n−22\to n-2 scattering with legs 1, 2 incoming. Each factor BB or B+bB+b is a nonempty proper subset sum of pair energies. It is positive if it excludes b1b_1, and otherwise equals minus the positive complementary sum. Thus no physical propagator is singular.

For M>0M>0, the threshold configuration is

b1=−2VM,b2=⋯=bV+1=2M.(22)b_1=-2VM,\qquad b_2=\cdots=b_{V+1}=2M. \tag*{(22)}

All 2V2V outgoing particles are then at rest. Increasing their pair energies and adjusting b1b_1 to conserve momentum moves above threshold while preserving the regular zero. The ordered amplitude with adjacent incoming legs already vanishes for 2→42\to4, whereas the full single-scalar amplitude first vanishes for 2→62\to6. Away from threshold, the ordered zero need not survive summing over orderings.

We thus connect threshold nullification to hidden zeros. The ordered threshold zero belongs to a continuous family of real 2→n−22\to n-2 scattering zeros on the rank-one locus, extending above threshold with every propagator finite.

Rod packing zeros for any Tr(m)

The eight-point cancellation in Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) suggests interpreting products of inverse channels as weights of nonoverlapping rods. Equation (18) then gives X16X_{16} as the sum of single-rod weights divided by the complete-tiling weight. We extend this packing rule to every m≥4m\ge4 and allowed multiplicity.

For n=(m−2)V+2n=(m-2)V+2 with V≥2V\ge2, choose a root leg and relabel it nn, opening the remaining ordering into a line of n−2n-2 gap cells between legs 1,…,n−11,\ldots,n-1. A rod now covers m−2m-2 consecutive cells and carries weight 1/Xi,i+m−11/X_{i,i+m-1} when it starts at cell ii. Rods may touch but cannot overlap, and multiplying their weights gives the weight of a packing.

Let Sk(a,b)S_k(a,b) be the sum of the weights of all packings of kk rods whose starting cells lie between aa and bb,

Sk(a,b)=∑a≤i1<⋯<ik≤bir+1−ir≥m−2∏r=1k1Xir,ir+m−1.(23)S_k(a,b)=\sum_{\substack{a\le i_1<\cdots<i_k\le b\\ i_{r+1}-i_r\ge m-2}}\prod_{r=1}^{k}\frac{1}{X_{i_r,i_r+m-1}}. \tag*{(23)}

The empty packing has weight S0(a,b)=1S_0(a,b)=1, while the sum is zero if no packing with k>0k>0 rods fits.

We define the rod-zero family G\mathcal{G} by two channel assignments. For every interval with 1≤a<b≤n1\le a<b\le n, b−a=(m−2)ℓ+1b-a=(m-2)\ell+1, and 2≤ℓ<V2\le\ell<V, set

Xab=Sℓ−1(a,b−m+1)Sℓ(a,b−m+1).(24)X_{ab}=\frac{S_{\ell-1}(a,b-m+1)}{S_{\ell}(a,b-m+1)}. \tag*{(24)}

For the rod covering the last m−2m-2 cells, set

Xn−m+1,n=−SV−2(1,n−2m+3)SV−1(1,n−m).(25)X_{n-m+1,n}=-\frac{S_{V-2}(1,n-2m+3)}{S_{V-1}(1,n-m)}. \tag*{(25)}

At V=2V=2, only the second assignment is needed. All physical channels and displayed denominators must be nonzero.

The family has n−mn-m free parameters in the space of n(V−1)/2n(V-1)/2 physical channels, and hence

# independent conditions=n(V−1)2−(n−m).(26)\#\text{ independent conditions}=\frac{n(V-1)}{2}-(n-m). \tag*{(26)}

The nn choices of root give distinct loci for m≥5m\geq5, V≥3V\geq3, but the same locus when m=4m=4 or V=2V=2 (Sec. S3 C). Section S3 proves vanishing and regularity at every allowed multiplicity.

Eight-point example. For Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) at n=8n=8, the interval assignment gives

X16=X14X36(1X14+1X25+1X36).(27)X_{16}=X_{14}X_{36}\left(\frac{1}{X_{14}}+\frac{1}{X_{25}}+\frac{1}{X_{36}}\right). \tag*{(27)}

Shifting the labels by one and two gives X27X_{27} and X38X_{38}. The remaining assignment is

X58=−1X14+1X25+1X361X14X36+1X14X47+1X25X47.(28)X_{58}=-\frac{\frac{1}{X_{14}}+\frac{1}{X_{25}}+\frac{1}{X_{36}}}{\frac{1}{X_{14}X_{36}}+\frac{1}{X_{14}X_{47}}+\frac{1}{X_{25}X_{47}}}. \tag*{(28)}

These four relations reduce the twelve diagrams to the six packing weights in Figure 2, whose sum vanishes. The X16X_{16} relation is the vanishing minor in Eq. (17), and all four relations together are equivalent to rank C=1C=1 on the regular locus.

Packing sum $S_2(1,5)$

Figure 2. The packing sum S2(1,5)S_2(1,5) for Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) at eight points. Two rods of length two occupy six cells, with allowed starting positions 1,…,51,\ldots,5. Each rod starting at ii has weight 1/Xi,i+31/X_{i,i+3}. Multiply the two rod weights within each packing and add the six resulting terms. In general, kk counts the rods and a,ba,b bound their starting positions in Sk(a,b)S_k(a,b).

§ 2.1 E gives explicit Tr⁡(ϕ6)\operatorname{Tr}(\phi^6) solutions at ten and fourteen points.

Locality and unitarity from zeros

We now discuss the reconstruction of amplitudes from trial expressions by imposing the two types of zero conditions.

Locality and unitarity from block zeros

To test whether zeros determine locality, take A~n\widetilde{A}_n to be a sum of every product of V−1V-1 inverse physical channels with independent constant coefficients, including crossing and repeated poles. No cyclic symmetry is assumed. We impose every rectangle selected by each tested picked set, treating its locus separately. In every case in Table I, only A~n=cAn(m)\widetilde{A}_n=cA_n^{(m)} survives. These finite results recover pole compatibility and relative factorization residues from the zeros and ansatz assumptions. We conjecture that the general rule does so at every odd mm and allowed multiplicity.

mmVVnnZero loci usedUnrestricted termsLocality and unitarity emerge
572346185 250 786✓
76323230 872 016✓
964444146 803 272✓
215979760 862 165✓
101439839835 641 099✓

Table I. Largest completed checks for each reported interaction order. Each check mark denotes uniqueness up to normalization without locality assumed. No counterexample was found in these checks.

Unitarity from rod zeros and locality

With locality assumed, we take an ansatz with constant coefficients

A~n=∑D∈Dn,mcD∏e∈D1Xe,(29)\widetilde{A}_n=\sum_{D\in\mathcal{D}_{n,m}}c_D\prod_{e\in D}\frac{1}{X_e}, \tag*{(29)}

where DD runs over compatible planar trees with vertices of valence mm.

For rod zeros, we prove uniqueness with locality for every m≥4m\geq4 and every allowed nn. Limits within G\mathcal{G} isolate an end vertex and leave a freely varying rod zero at lower multiplicity. Induction fixes the coefficients of diagrams containing that vertex. Diagrams sharing end vertices connect these equalities, forcing all cDc_D to agree. Section S3 D gives the proof.

The zero conditions fix A~n=cVAn(m)\widetilde{A}_n=c_VA_n^{(m)} but leave one normalization cVc_V for each number of vertices. Factorization into separately normalized lower amplitudes requires cV=cVLcVRc_V=c_{V_L}c_{V_R} whenever VL+VR=VV_L+V_R=V. Choosing cV=gVc_V=g^V, with elementary coupling gg, satisfies this condition. The homogeneous zero tests fix relative residues but do not relate these overall factors across multiplicities.

Discussion

Regular kinematic cancellations constrain diagram weights and allowed pole combinations beyond Tr⁡(ϕ3)\operatorname{Tr}(\phi^3). We establish block zeros through the equal-density rule at every odd valence and allowed multiplicity. Their reconstruction power is supported by the exact checks. For every m≥4m\geq4, we prove that rod zeros determine all relative diagram coefficients when locality is assumed. The Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) rank condition also connects these algebraic cancellations to real scattering and threshold nullification.

The main open question is to derive both locality and factorization from zeros at every interaction order and multiplicity. The optional sums of block contractions in Sec. S2 G may be evidence of further structure. An equations-of-motion derivation based on the methods of Ref. [17] could explain these zeros or reveal new ones. The structure of these summed conditions seems particularly natural in that approach. Finally, threshold nullification also occurs in ϕ3\phi^3 theory, where it appears to be connected to the continuous Tr⁡(ϕ3)\operatorname{Tr}(\phi^3) zeros [51].

Acknowledgments. LR is supported by the National Natural Science Foundation of China General Program No. 12475070 and the Beijing Natural Science Foundation International Scientist Project No. IS24014. AI tools were used to assist with checks and editing.

References

Supplemental Material

Beyond hidden zeros and uniqueness in Tr⁡(ϕ3)\operatorname{Tr}(\phi^3): Block and rod zeros in any Tr⁡(ϕm)\operatorname{Tr}(\phi^m)

We first give explicit examples, then prove the block and rod zeros and the uniqueness result for rod zeros with locality. The last section gives the real-momentum construction and further threshold connections. Throughout the algebraic arguments, the physical channels are independent, n=(m−2)V+2n=(m-2)V+2, and the vertices are normalized to one. Regularity means that every physical channel is nonzero. Only the last section imposes a fixed-dimensional momentum realization.

Explicit examples of zero loci

All examples describe loci in physical channels. Every physical channel and displayed denominator must be nonzero. An elementary mesh cell carries cij=−2pi⋅pjc_{ij}=-2p_i\cdot p_j. Summing over two blocks of m−2m-2 momenta cancels the interior channels and leaves four corners,

−2Bi⋅Bj=∑r,s=0m−3ci+r,j+s=Xij+Xi+m−2,j+m−2−Xi,j+m−2−Xi+m−2,j.(30)-2B_i\cdot B_j=\sum_{r,s=0}^{m-3}c_{i+r,j+s} =X_{ij}+X_{i+m-2,j+m-2}-X_{i,j+m-2}-X_{i+m-2,j}. \tag*{(30)}

A retained separator pair selects an h×(V−h)h\times(V-h) rectangle of these tiles. Each tile total vanishes separately. The elementary contractions inside one tile need not vanish.

The eleven-point example in Tr⁡(ϕ5)\operatorname{Tr}(\phi^5)

Pick p2,p4,p7,p9,p11p_2,p_4,p_7,p_9,p_{11}. The retained separator pairs are (2,9)(2,9), (4,11)(4,11), and (7,11)(7,11), each with a 1×21\times2 rectangle. Their six drawn tiles include one repeated contraction and give

B1⋅B5=B1⋅B8=B4⋅B8=0,B_1\cdot B_5=B_1\cdot B_8=B_4\cdot B_8=0,
B10⋅B3=B10⋅B6=0.(31)B_{10}\cdot B_3=B_{10}\cdot B_6=0. \tag*{(31)}

The five independent channel relations are

X18=X15+X48,X4,11=−X15,X_{18}=X_{15}+X_{48},\qquad X_{4,11}=-X_{15},
X7,11=−X18,X3,10=−X29,X_{7,11}=-X_{18},\qquad X_{3,10}=-X_{29},
X6,10=X26−X29.(32)X_{6,10}=X_{26}-X_{29}. \tag*{(32)}

They leave six physical channels unfixed and make the complete amplitude vanish throughout the regular locus.

The fourteen-point example in Tr⁡(ϕ5)\operatorname{Tr}(\phi^5)

At fourteen points, pick p2,p4,p7,p10,p12,p14p_2,p_4,p_7,p_{10},p_{12},p_{14}. The retained separator pairs are (2,12)(2,12), (4,14)(4,14), (7,14)(7,14), and (10,14)(10,14). These four rectangles give nine distinct block conditions. The pair (7,14)(7,14) contributes the 2×22\times2 rectangle with rows B1,B4B_1,B_4 and columns B8,B11B_8,B_{11}, while the other three have width one. With Bi=pi+pi+1+pi+2B_i=p_i+p_{i+1}+p_{i+2}, impose

B1⋅B5=B1⋅B8=B1⋅B11=0,B_1\cdot B_5=B_1\cdot B_8=B_1\cdot B_{11}=0,
B4⋅B8=B4⋅B11=B7⋅B11=0,B_4\cdot B_8=B_4\cdot B_{11}=B_7\cdot B_{11}=0,
B13⋅B3=B13⋅B6=B13⋅B9=0.(33)B_{13}\cdot B_3=B_{13}\cdot B_6=B_{13}\cdot B_9=0. \tag*{(33)}

The equivalent channel system is

X18=X15+X48,X1,11=X18+X7,11,X4,11=X48+X7,11,X_{18}=X_{15}+X_{48},\qquad X_{1,11}=X_{18}+X_{7,11},\qquad X_{4,11}=X_{48}+X_{7,11},
X4,14=−X15,X7,14=−X18,X10,14=−X1,11,X_{4,14}=-X_{15},\qquad X_{7,14}=-X_{18},\qquad X_{10,14}=-X_{1,11},
X3,13=−X2,12,X6,13=X26−X2,12,X9,13=X29−X2,12.(34)X_{3,13}=-X_{2,12},\qquad X_{6,13}=X_{26}-X_{2,12},\qquad X_{9,13}=X_{29}-X_{2,12}. \tag*{(34)}

The pairs (4,14)(4,14), (7,14)(7,14), and (10,14)(10,14) give the first two lines, while (2,12)(2,12) gives the last. Nine independent equations in 21 physical channels leave twelve free parameters.

The seventeen-point example in Tr⁡(ϕ7)\operatorname{Tr}(\phi^7)

Here V=3V=3 and every block contains five momenta. Pick

\{p_2,p_4,p_6,p_8,p_{11},p_{13},p_{15},p_{17}\}. \] (S6)

The retained separator pairs are (2,8)(2,8), (2,13)(2,13), (4,15)(4,15), (6,17)(6,17), and (11,17)(11,17). Their 1×21 \times2 rectangles give eight distinct conditions,

B1⋅B7=B1⋅B12=B6⋅B12=0,B3⋅B9=B3⋅B14=B8⋅B14=0,B16⋅B5=B16⋅B10=0.(35)\begin{aligned} B_1\cdot B_7&=B_1\cdot B_{12}=B_6\cdot B_{12}=0,\\ B_3\cdot B_9&=B_3\cdot B_{14}=B_8\cdot B_{14}=0,\\ B_{16}\cdot B_5&=B_{16}\cdot B_{10}=0. \tag*{(35)} \end{aligned}

In physical channels they are

X1,12=X1,7+X6,12,X6,17=−X1,7,X11,17=−X1,12,X3,14=X3,9+X8,14,X2,8=−X3,9,X2,13=−X3,14,X5,16=−X4,15,X10,16=X4,10−X4,15.(36)\begin{aligned} X_{1,12}&=X_{1,7}+X_{6,12},&X_{6,17}&=-X_{1,7},&X_{11,17}&=-X_{1,12},\\ X_{3,14}&=X_{3,9}+X_{8,14},&X_{2,8}&=-X_{3,9},&X_{2,13}&=-X_{3,14},\\ X_{5,16}&=-X_{4,15},&X_{10,16}&=X_{4,10}-X_{4,15}. \tag*{(36)} \end{aligned}

Nine physical channels remain unfixed. These relations make all 51 diagrams cancel on the regular locus.

The full matrix in Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) at eight points

All eight physical channels enter the matrix of adjacent-pair contractions,

C=(X14X16−X14−X36X36−X16−X38X38X25X36X38−X36−X58X58−X38−X25X27−X25−X47X47X58X25−X58−X27X47−X27−X14X14−X47−X16X16X27).(37)C=\begin{pmatrix} X_{14} & X_{16}-X_{14}-X_{36} & X_{36}-X_{16}-X_{38} & X_{38}\\ X_{25} & X_{36} & X_{38}-X_{36}-X_{58} & X_{58}-X_{38}-X_{25}\\ X_{27}-X_{25}-X_{47} & X_{47} & X_{58} & X_{25}-X_{58}-X_{27}\\ X_{47}-X_{27}-X_{14} & X_{14}-X_{47}-X_{16} & X_{16} & X_{27} \end{pmatrix}. \tag*{(37)}

The zero requires every 2×22 \times2 minor to vanish, with every physical channel nonzero. For example, the upper-left minor gives X25(X16−X14−X36)=X14X36X_{25}(X_{16}-X_{14}-X_{36})=X_{14}X_{36}, precisely the rod relation for X16X_{16}. Its shifts and the closing equation give the full rank-one condition, as in the main text.

Explicit rod solutions in Tr⁡(ϕ6)\operatorname{Tr}(\phi^6)

At ten points, V=2V=2 and one rod of length four has five positions on eight cells. Leave X1,6X_{1,6}, X2,7X_{2,7}, X3,8X_{3,8}, X4,9X_{4,9} unfixed. The complete condition is

X5,10=−(1X1,6+1X2,7+1X3,8+1X4,9)−1.(38)X_{5,10}=-\left(\frac{1}{X_{1,6}}+\frac{1}{X_{2,7}}+\frac{1}{X_{3,8}}+\frac{1}{X_{4,9}}\right)^{-1}. \tag*{(38)}

There are no longer proper channels. The five single-rod weights sum to zero.

At fourteen points, V=3V=3. Leave the eight shortest channels Xi,i+5X_{i,i+5}, i=1,…,8i=1,\ldots,8, unfixed. Two rods fit on twelve cells. Ten packings omit the last rod and five contain it, giving

X9,14=−∑i=151Xi,i+5∑i=14∑j=i+481Xi,i+5Xj,j+5.(39)X_{9,14}=-\frac{\sum_{i=1}^{5}\frac{1}{X_{i,i+5}}}{\sum_{i=1}^{4}\sum_{j=i+4}^{8}\frac{1}{X_{i,i+5}X_{j,j+5}}}. \tag*{(39)}

The other five channels are determined by the one-rod packings on eight cells,

Xa,a+9=Xa,a+5Xa+4,a+9(1Xa,a+5+1Xa+1,a+6+1Xa+2,a+7+1Xa+3,a+8+1Xa+4,a+9),a=1,…,5.(40)X_{a,a+9}=X_{a,a+5}X_{a+4,a+9}\left(\frac{1}{X_{a,a+5}}+\frac{1}{X_{a+1,a+6}}+\frac{1}{X_{a+2,a+7}}+\frac{1}{X_{a+3,a+8}}+\frac{1}{X_{a+4,a+9}}\right),\qquad a=1,\ldots,5. \tag*{(40)}

Together these equations give all fourteen physical channels in terms of the eight left unfixed. On the five longer-channel relations, the 35 diagrams reduce to fifteen packing weights,

A14(6)=∑i=15∑j=i+491Xi,i+5Xj,j+5,(41)A_{14}^{(6)}=\sum_{i=1}^{5}\sum_{j=i+4}^{9}\frac{1}{X_{i,i+5}X_{j,j+5}}, \tag*{(41)}

whose sum vanishes by Eq. (39). All physical channels and displayed denominators must remain nonzero in both examples.

Block zeros for odd interactions

We prove that every picked set in the equal-density prescription gives a regular zero at every allowed multiplicity. For m=3m=3, the prescription is exactly the usual rectangle zero [1]. We therefore take odd m≥5m \ge5 below and write q=(m−3)/2q=(m-3)/2. A zero at VV interaction vertices requires qV+2qV+2 nonadjacent picked momenta.

The proof uses induction on VV. We first show that the locus has regular points. We then examine each possible pole of the amplitude restricted to that locus. Usually, one of its two factorization amplitudes inherits a lower-point zero. The remaining poles occur in crossing pairs whose residues cancel. Once all poles have canceled, homogeneity forces the amplitude to vanish. The pole limits are used only in this proof. The zero itself is defined with every propagator nonzero.

Regularity and possible coincident poles

For this argument, place pip_i on edge (i,i+1)(i,i+1) of the channel polygon and mark that edge when pip_i is picked. This is the polygon whose diagonals carry the XabX_{ab}. Nonadjacent picked momenta give marked edges with disjoint endpoints.

To construct regular points, give the endpoints of each marked edge opposite values of an independent parameter. Give each remaining vertex its own parameter. Denote the resulting values by ziz_i and assign

Xab=za+zb,zi+1=−ziif pi is picked.(42)X_{ab}=z_a+z_b,\qquad z_{i+1}=-z_i\quad\text{if }p_i\text{ is picked}. \tag*{(42)}

The four terms in every block equation (Eq. (30)) cancel. Any boundary corner is a separator edge, where zi+zi+1=0z_i+z_{i+1}=0 already holds. Every physical channel is a nonzero linear form in two independent parameters, since no physical diagonal joins the endpoints of a marked edge. Generic parameter values therefore make all channels nonzero. These assignments establish regularity and are not additional conditions on the zero. They also establish regularity if extra complete-block rectangles between picked separators are imposed.

The same assignments restrict which poles can coincide after imposing the zero conditions. If two channel forms are proportional on the full locus, they must be proportional on Eq. (42). They must therefore involve the same two independent endpoint parameters. Replacing just one endpoint by its marked partner, or moving two endpoints in opposite directions, violates the physical length condition modulo m−2≥3m-2\ge3. The only possible distinct proportional channels are

Xab,Xa+1,b+1=−Xab,(43)X_{ab},\qquad X_{a+1,b+1}=-X_{ab}, \tag*{(43)}

with pa,pbp_a,p_b picked, or the same pair read in reverse. These two diagonals cross. No compatible tree contains both, so all restricted poles are simple. Moreover, a channel internal to either cut polygon cannot be proportional to its attaching channel. Thus the lower amplitudes have nonzero denominators at a generic point on the cut.

What must survive a factorization cut

Cut a physical diagonal (a,b)(a,b) and call the two sides LL and RR, with VL+VR=VV_L+V_R=V. Their original boundary paths contain (m−2)VL+1(m-2)V_L+1 and (m−2)VR+1(m-2)V_R+1 momenta. Each lower amplitude acquires one additional external momentum from the cut. On the left it is

kL=−∑r=ab−1pr=∑r=ba−1pr,kR=−kL,(44)k_L=-\sum_{r=a}^{b-1}p_r=\sum_{r=b}^{a-1}p_r,\qquad k_R=-k_L, \tag*{(44)}

where the second sum runs cyclically along the right path. On Xab=0X_{ab}=0, both cut momenta are massless. Before imposing the zero conditions, ordinary tree factorization gives

ResXab=0An(m)=AL(m)AR(m).(45)\mathop{\mathrm{Res}}_{X_{ab}=0} A_n^{(m)} = A_L^{(m)} A_R^{(m)}. \tag*{(45)}

To make this residue vanish, it is enough to find an admissible picked set on one side and show that every block condition of its lower zero follows from the parent conditions.

Let KL,KRK_L,K_R count picked momenta on the two original paths, excluding the new cut momenta. They obey

KL+KR=qV+2.(46)K_L+K_R=qV+2. \tag*{(46)}

There are only two possibilities. One side already has the qVi+2qV_i+2 picks required for a lower zero, or each side is short by exactly one. We treat them in turn.

One side already has enough picks

Suppose KL≥qVL+2K_L\ge qV_L+2. Keep a consecutive group of exactly qVL+2qV_L+2 picks in their order along the left path and leave kLk_L unpicked. The group is consecutive among the picks, not among all momenta. This gives an admissible lower picked set. It requires VL≥2V_L\ge2, since a one-vertex path has room for only q+1q+1 nonadjacent picks.

Consider any retained separator pair of this lower zero. Its arc avoiding the cut contains no discarded pick, because the kept group was consecutive. The arc therefore has exactly the same length and picked count in the parent. The separator pair passes the parent density test as well, so its entire parent rectangle is imposed.

The only change in its block partition is at the block containing kLk_L. Replacing kLk_L by the right path in Eq. (44) replaces one momentum by (m−2)VR+1(m-2)V_R+1 momenta. That lower block now contains

(m−3)+[(m−2)VR+1]=(m−2)(VR+1)(47)(m-3)+\bigl[(m-2)V_R+1\bigr]=(m-2)(V_R+1) \tag*{(47)}

parent momenta. Since the replacement adds a multiple of m−2m-2 entries, all other block boundaries remain aligned. The changed block is the sum of VR+1V_R+1 consecutive parent blocks. For any block BB on the other arc, its required contraction is consequently

B~⋅B=(∑ν=0VRB(ν))⋅B=∑ν=0VRB(ν)⋅B=0.(48)\widetilde{B}\cdot B=\left(\sum_{\nu=0}^{V_R}B^{(\nu)}\right)\cdot B=\sum_{\nu=0}^{V_R}B^{(\nu)}\cdot B=0. \tag*{(48)}

Every term is one of the imposed parent conditions. Blocks not containing the cut are unchanged. Thus every condition of the lower zero holds, and induction gives AL(m)=0A_L^{(m)}=0. The argument is identical with LL and RR exchanged.

Each side needs one more pick

If neither side has enough picks, Eq. (46) forces

KL=qVL+1,KR=qVR+1.(49)K_L=qV_L+1,\qquad K_R=qV_R+1. \tag*{(49)}

We try to supply the missing pick by selecting the cut momentum. On LL, this is allowed when neither pap_a nor pb−1p_{b-1} is picked, since these are its two neighbours. On RR, the corresponding neighbours are pbp_b and pa−1p_{a-1}. Nonadjacency of the original picks implies that both choices fail only when pa,pbp_a,p_b are picked, or when pa−1,pb−1p_{a-1},p_{b-1} are picked. These are the paired poles treated in the next subsection.

Otherwise select the cut on an allowed side, say LL, and retain all old picks there. This supplies exactly qVL+2qV_L+2 picks. Again VL≥2V_L\ge2, because a single mm-gon cannot hold this many nonadjacent picks. A lower rectangle between two old separators is inherited exactly as in the preceding subsection. It remains to consider a rectangle whose separators are the new cut and an old picked momentum pcp_c.

Finding an original separator to replace the cut. The right path contains (m−2)VR+1(m-2)V_R+1 momenta and qVR+1qV_R+1 picks. We need one of these picks to divide that path into two complete-block runs with the required density. The following elementary count guarantees it. On a path of (m−2)W+1(m-2)W+1 momenta with K=qW+1K=qW+1 picks, number the picked positions from zero as t1<⋯<tKt_1<\cdots<t_K and set

Hj=(m−2)(j−1)−qtj.(50)H_j=(m-2)(j-1)-qt_j. \tag*{(50)}

The first value is H1=−qt1≤0H_1=-qt_1\leq0, and the last is HK=q[(m−2)W−tK]≥0H_K=q[(m-2)W-t_K]\geq0. Nonadjacency gives

Hj+1−Hj=(m−2)−q(tj+1−tj)≤1.(51)H_{j+1}-H_j=(m-2)-q(t_{j+1}-t_j)\leq1. \tag*{(51)}

An integer sequence starting nonpositive and ending nonnegative, with no upward step larger than one, must pass through zero. Since qq and m−2=2q+1m-2=2q+1 are coprime, a zero gives

tj=(m−2)t,j−1=qt,0≤t≤W.(52)t_j=(m-2)t,\qquad j-1=qt,\qquad0\leq t\leq W. \tag*{(52)}

Removing this picked momentum leaves runs of lengths (m−2)t(m-2)t and (m−2)(W−t)(m-2)(W-t), with qtqt and q(W−t)q(W-t) picks. An empty run at an endpoint is allowed.

Apply this fact with W=VRW=V_R and call the resulting picked momentum pdp_d. The two arcs of the lower rectangle between pcp_c and kLk_L have lengths (m−2)h(m-2)h and (m−2)(VL−h)(m-2)(V_L-h) and picked counts qhqh and q(VL−h)q(V_L-h). Restoring the right path appends one of the runs ending at pdp_d to each arc. With tt chosen for the first run, the parent arcs between pc,pdp_c,p_d therefore have lengths and picked counts

lengthpicked momenta
first arc(m−2)(h+t)(m-2)(h+t)q(h+t)q(h+t)
second arc(m−2)(V−h−t)(m-2)(V-h-t)q(V−h−t)q(V-h-t)

Table 1.

Thus (c,d)(c,d) is retained in the parent. Each appended run contains a whole number of blocks, so every block in the lower rectangle is an unchanged block in this parent rectangle. All its contractions vanish. This completes the inheritance argument also when the cut is picked.

Induction needs only that the inherited lower kinematics satisfy the lower zero conditions and have generically nonzero denominators. They need not cover the entire lower zero locus, because the lower amplitude vanishes at every regular point of that locus.

Cancellation of the paired poles

The only remaining case has the counts (S21) and picked momenta pa,pbp_a,p_b, or equivalently the reverse shifted pair. Removing pap_a from the left path leaves (m−2)VL(m-2)V_L momenta and qVLqV_L picks. Thus (a,b)(a,b) passes the density test. Its rectangle relates the two crossing channels XabX_{ab} and Xa+1,b+1X_{a+1,b+1}.

To see that it relates their residues as well, label the block boundaries on the two arcs by

ur=a+1+r(m−2),0≤r≤VL,vs=b+1+s(m−2),0≤s≤VR.(53)u_r=a+1+r(m-2),\qquad0\leq r\leq V_L,\qquad v_s=b+1+s(m-2),\qquad0\leq s\leq V_R. \tag*{(53)}

Labels are cyclic, so uVL=bu_{V_L}=b and vVR=av_{V_R}=a. Each tile sets one mixed difference of Xur,vsX_{u_r,v_s} to zero. Summing the tile equations from any corner to the endpoints gives

Xur,vs=Xa,ur+Xb,vs−Xab.(54)X_{u_r,v_s}=X_{a,u_r}+X_{b,v_s}-X_{ab}. \tag*{(54)}

Here adjacent boundary channels vanish. Setting r=s=0r=s=0, or just one of them to zero, yields

Xa+1,b+1=−Xab,X_{a+1,b+1}=-X_{ab},
Xb+1,ur=Xa,ur−Xab,Xa+1,vs=Xb,vs−Xab.(55)X_{b+1,u_r}=X_{a,u_r}-X_{ab},\qquad X_{a+1,v_s}=X_{b,v_s}-X_{ab}. \tag*{(55)}

The interior values of uru_r and vsv_s exhaust the allowed internal diagonals incident on aa and bb in their respective cut polygons. At Xab=0X_{ab}=0, these equations identify their values with those of the shifted cut, while all other internal diagonals are unchanged. The two products of complete lower amplitudes are therefore equal. Their attaching propagators have opposite signs, so their residues cancel. The earlier classification of coincident poles ensures that no lower denominator vanishes identically in this comparison.

Completion of the induction

At V=2V=2, every polygon vertex belongs to a marked edge, so the picks alternate. The retained rectangles pair the m−1m-1 exchange channels with their opposite values, and the amplitude vanishes. This starts the induction.

For larger VV, every pole is covered by one of the preceding cases. Either one lower amplitude vanishes by induction, or two equal factorization numerators cancel against opposite propagators. If a proportional partner also occurs in the first case, each residue already vanishes separately. Thus every finite pole of the restricted amplitude cancels.

The result is a polynomial in linear coordinates on the zero locus. However, every tree term has homogeneous degree −(V−1)<0-(V-1)<0 in those coordinates. Such a polynomial must be zero. Together with regularity, this proves the claim at every allowed multiplicity and every odd m≥5m\geq5. The m=3m=3 restriction is the established cubic rectangle theorem. This argument proves vanishing of the physical amplitude, not uniqueness of a trial ansatz.

An optional relaxation of width-one strips

A particular picked set admits larger zeros obtained by combining block conditions. Pick every other position along a consecutive stretch of the cyclic ordering,

{pn,p2,p4,…,pn−V}.(56)\{p_n,p_2,p_4,\ldots,p_{n-V}\}. \tag*{(56)}

Write m−2=2q+1m-2=2q+1, with q=(m−3)/2q=(m-3)/2. For this picked set, the equal-density rule selects width-one rectangles with rows

c=1+2s,s=0,…,q(V−1),(57)c=1+2s,\qquad s=0,\ldots,q(V-1), \tag*{(57)}

and columns j=1,…,V−1j=1,\ldots,V-1. The entry in row cc, column jj is Bc⋅Bc+1+j(m−2)B_c\cdot B_{c+1+j(m-2)}.

To specify the relaxation, place boundaries before column 1 and after column V−1V-1. Retain an interior boundary after column jj, with 1≤j≤V−21\leq j\leq V-2, precisely when

[j=1 and s≤q(V−2)]ors+qj=q(V−1).(58)[j=1\text{ and }s\leq q(V-2)]\quad\text{or}\quad s+qj=q(V-1). \tag*{(58)}

Consecutive retained boundaries divide the columns into groups. If two consecutive boundary positions are u<vu<v, set I={u+1,…,v}I=\{u+1,\ldots,v\} and impose only

Bc⋅(∑j∈IBc+1+j(m−2))=0.(59)B_c\cdot\left(\sum_{j\in I}B_{c+1+j(m-2)}\right)=0. \tag*{(59)}

Impose this equation for every group in every row. Thus a group with one column gives an individual block condition, while a larger group allows cancellations among its contractions.

These grouped conditions give zeros at every allowed multiplicity for odd mm. The grouping is equivalent to a full-row sum together with a prefix sum at each retained interior boundary. Summing Eq. (30) over a prefix telescopes to the corresponding channel relation. Regularity follows from the regular points of the stronger pairwise locus. At m=3m=3, q=0q=0, every interior boundary is retained. No blocks are combined, and the usual cubic width-one zero is recovered.

The cancellation uses the ordinary current recursion (S36), with the row starts treated from largest to smallest. Opposite shortest channels remove child currents in pairs. Each omitted prefix condition multiplies a lower current already forced to vanish by the later rows, so it need not be imposed separately. Since m=3m=3 is even, this pair removal terminates. Before restoring the full-row condition at c=1c=1, the root sum reduces to

An(m)=(1X1,m+1Xm−1,n)An−m+2(m).(60)A_n^{(m)}=\left(\frac{1}{X_{1,m}}+\frac{1}{X_{m-1,n}}\right)A_{n-m+2}^{(m)}. \tag*{(60)}

where An−m+2(m)A_{n-m+2}^{(m)} is the complete lower tree with (pn,p1,…,pm−2)(p_n,p_1,\ldots,p_{m-2}) merged into one generally off-shell leg and all internal channels inherited. The remaining full-row sum gives Xm−1,n=−X1,mX_{m-1,n}=-X_{1,m} and cancels the prefactor. This current argument is separate from the factorization proof of the general pairwise prescription above.

For Tr⁡(ϕ5)\operatorname{Tr}(\phi^5) at fourteen points, q=1q=1 and V=4V=4. The four rows give

B1⋅B5=B1⋅(B8+B11)=0,B3⋅B7=B3⋅B10=B3⋅B13=0,B5⋅B9=B5⋅(B12+B1)=0,B7⋅(B11+B14+B3)=0.(61)\begin{aligned} B_1\cdot B_5&=B_1\cdot(B_8+B_{11})=0,\\ B_3\cdot B_7&=B_3\cdot B_{10}=B_3\cdot B_{13}=0,\\ B_5\cdot B_9&=B_5\cdot(B_{12}+B_1)=0,\\ B_7\cdot(B_{11}+B_{14}+B_3)&=0. \tag*{(61)} \end{aligned}

The cyclic labels have been reduced modulo fourteen. For example, B1⋅B8B_1\cdot B_8 and B1⋅B11B_1\cdot B_{11} can be nonzero and opposite. This is an additional relaxation of the displayed picked set. The general equal-density prescription and the reconstruction checks in the main text retain every selected pairwise block condition.

Rod zeros, regularity, and uniqueness with locality

Why the packing rules give a zero

The interval rule makes every proper current equal the weight of the complete packing on its interval. The root amplitude then becomes the sum over packings with one fewer rod, which the closing rule sets to zero. We establish these two statements to prove

An(m)∣G=0,m≥4,V≥2.(62)A_n^{(m)}\big|_{\mathcal{G}}=0,\qquad m\ge4,\quad V\ge2. \tag*{(62)}

Let J[a,b)J_{[a,b)} be the current including its attaching propagator. It obeys the ordinary tree recursion [64],

XabJ[a,b)=∑[a,b)=I1⊔⋯⊔Im−1∏rJIr,J[a,a+1)=1.(63)X_{ab}J_{[a,b)}=\sum_{[a,b)=I_1\sqcup\cdots\sqcup I_{m-1}}\prod_r J_{I_r},\qquad J_{[a,a+1)}=1. \tag*{(63)}

Each child has 1 mod (m−2)(m-2) leaves. For m≥4m\ge4, define trial currents by

b[a,b)=∏r=0ℓ−1Xa+(m−2)r,a+(m−2)(r+1)+1−1,b−a=(m−2)ℓ+1,b[a,a+1)=1.(64)b_{[a,b)}=\prod_{r=0}^{\ell-1}X^{-1}_{a+(m-2)r,a+(m-2)(r+1)+1},\qquad b-a=(m-2)\ell+1,\qquad b_{[a,a+1)}=1. \tag*{(64)}

The trial current b[a,b)b_{[a,b)} is precisely the complete packing weight Sℓ(a,b−m+1)S_\ell(a,b-m+1) in Eq. (24). For a single rod, the ratio in that equation is 1/(1/X)=X1/(1/X)=X. For several rods, division by the complete-packing weight cancels unchanged factors and gives the dimension of a channel.

A partition into m−1m-1 children contains a total of ℓ−1\ell-1 vertices below its first vertex. Each product of trial currents corresponds to a unique packing of ℓ−1\ell-1 rods, and conversely. To see this, write the rod starts as

ir=a+(m−2)(r−1)+dr,0≤d1≤⋯≤dℓ−1≤m−2.(65)i_r=a+(m-2)(r-1)+d_r,\qquad0\le d_1\le\cdots\le d_{\ell-1}\le m-2. \tag*{(65)}

The offset drd_r counts the cuts preceding rod rr and therefore specifies its child interval. The number of rods with each offset 0,…,m−20,\ldots,m-2 gives the number of vertices in that child. This explains the correspondence in both directions. The sum of child products is consequently Sℓ−1(a,b−m+1)S_{\ell-1}(a,b-m+1). Dividing this sum by the complete-packing weight b[a,b)b_{[a,b)} gives exactly the interval channel in Eq. (24). Thus Eq. (63) proves JI=bIJ_I=b_I by induction. At the amputated root, the same counting gives

An(m)∣longer-channel relations=SV−1(1,n−m+1).(66)\left.A_n^{(m)}\right|_{\text{longer-channel relations}}=S_{V-1}(1,n-m+1). \tag*{(66)}

Separating these packings according to whether they contain the final rod gives

SV−1(1,n−m+1)=SV−1(1,n−m)+SV−2(1,n−2m+3)Xn−m+1,n=0on G.(67)\begin{aligned} S_{V-1}(1,n-m+1)&=S_{V-1}(1,n-m)+\frac{S_{V-2}(1,n-2m+3)}{X_{n-m+1,n}}\\ &=0\qquad\text{on }\mathcal{G}. \tag*{(67)} \end{aligned}

The two terms omit or include that rod. Their cancellation follows from the assignment (25), proving that the amplitude vanishes. Singling out the last rod is only a parametrization choice.

Regular points at every multiplicity

To prove that the regular part is nonempty, it suffices to exhibit one point. For this purpose only, set every unfixed shortest channel to one in a common unit of momentum squared. Counting packings with and without the last rod gives Xn−m+1,n=−(V−1)/(m−2)X_{n-m+1,n}=-(V-1)/(m-2). For any proper interval b−a=(m−2)ℓ+1b-a=(m-2)\ell+1 the resulting channel is

Xab={ℓ+m−3m−2,b<n,−V−2m−2ℓ+m−4m−3,b=n.(68)X_{ab}= \begin{cases} \dfrac{\ell+m-3}{m-2},& b<n,\\ -\dfrac{V-2}{m-2}\dfrac{\ell+m-4}{m-3},& b=n. \end{cases} \tag*{(68)}

Both are nonzero for 1≤ℓ≤V1\le\ell\le V. The second expression follows by separating placements that contain the last rod, while the first counts all placements on the interval. Thus every allowed m,nm,n has a regular rod zero, and the n−mn-m unfixed shortest channels vary in a nonempty open set.

Root choices and the number of conditions

At a fixed root, the n−mn-m unfixed shortest channels are coordinates on the regular rod locus. Each remaining physical channel is determined by one equation. Thus the n(V−1)/2n(V-1)/2 physical channels give the independent-condition count in Eq. (26). A different choice of channel for solving the closing equation only changes the parametrization. Reversing the rooted line also gives the same packing rules.

For m=4m=4, cyclic invariance of the rank-one condition makes all roots equivalent. At V=2V=2, every root gives the same condition ∑eXe−1=0\sum_e X_e^{-1}=0. For m≥5m\geq5 and V≥3V\geq3, we can distinguish all nn roots using the regular point (S41) rooted at nn. Define, with cyclic labels,

Da=Xa,a+2m−3−Xa,a+m−1Xa+m−2,a+2m−3∑j=0m−21Xa+j,a+j+m−1.(69)D_a=X_{a,a+2m-3}-X_{a,a+m-1}X_{a+m-2,a+2m-3}\sum_{j=0}^{m-2}\frac{1}{X_{a+j,a+j+m-1}}. \tag*{(69)}

This is the first longer-channel rod relation. A rooted locus requires Da=0D_a=0 whenever its root is outside the consecutive run of 2m−32m-3 momenta starting at aa. Substituting Eq. (68) gives

Dn−2m+4=(V+m−5m−2)−1−(V+m−4m−2)(m−3−m−2V−1)<0,D_{n-2m+4}=\binom{V+m-5}{m-2}-1-\binom{V+m-4}{m-2}\left(m-3-\frac{m-2}{V-1}\right)<0,
Dn−m+1=V−1−(V+m−5m−3)<0,D_{n-m+1}=V-1-\binom{V+m-5}{m-3}<0,
Dn=(V+m−5m−3)−1−2m−2(V+m−6m−3)>0.(70)D_n=\binom{V+m-5}{m-3}-1-\frac{2}{m-2}\binom{V+m-6}{m-3}>0. \tag*{(70)}

The signs hold for all m≥5m\geq5, V≥3V\geq3. For the first, the ratio of its two binomial coefficients is (V−2)/(V+m−4)<(V−2)/(V−1)≤m−3−(m−2)/(V−1)(V-2)/(V+m-4)<(V-2)/(V-1)\leq m-3-(m-2)/(V-1). For the second, the binomial coefficient is at least V(V−1)/2V(V-1)/2. The third expression increases with VV by Pascal’s identity and is already (m−4)(m−1)/(m−2)>0(m-4)(m-1)/(m-2)>0 at V=3V=3.

The three corresponding momentum runs intersect only at leg nn. Every other root therefore requires at least one of these nonzero DaD_a to vanish. The regular point belongs only to its original rooted locus, and cyclic relabelling proves that all nn root choices give distinct loci. This counts the cyclic rod construction, without asserting completeness among all amplitude zeros.

Uniqueness when locality is assumed

Take the tree ansatz (29) with constant coefficients. An end vertex is attached to the rest of the tree by one internal line and carries m−1m-1 consecutive external legs. We first use limits within the rod locus to equate the coefficients of all diagrams containing a specified end vertex. Diagrams containing two such vertices then connect these groups of coefficients.

Starting the induction. At V=2V=2, the rod family covers ∑eXe−1=0\sum_e X_e^{-1}=0. In the variables 1/Xe1/X_e, this is a hyperplane. A linear combination ∑ece/Xe\sum_e c_e/X_e vanishes on it only when all cec_e agree.

For the induction, express the shortest channels as Xi,i+m−1=wi⋯wi+m−3X_{i,i+m-1}=w_i\cdots w_{i+m-3}. This introduces no restriction on them. One can leave the first m−3m-3 nonzero weights arbitrary and solve the consecutive products successively for the remaining weights. The rod assignments become

Xab=Kab(w)=∑0≤r1≤⋯≤rm−2<ℓ∏s=1m−2wa+s−1+(m−2)rs,b−a=(m−2)ℓ+1,K1n=0.(71)X_{ab}=K_{ab}(w)=\sum_{0\leq r_1\leq\cdots\leq r_{m-2}<\ell}\prod_{s=1}^{m-2}w_{a+s-1+(m-2)r_s},\qquad b-a=(m-2)\ell+1,\qquad K_{1n}=0. \tag*{(71)}

Indeed, in the recursion for trial currents, multiplying by all gap weights cancels every factor except the m−2m-2 cuts between children, yielding this sum.

An end vertex away from the root. For a start 2≤a≤n−m2\leq a\leq n-m, scale the m−2m-2 gap weights of its channel as εα1,…,εαm−2\varepsilon\alpha_1,\ldots,\varepsilon\alpha_{m-2}, with every αi≠0\alpha_i\ne0. Deleting these gaps at ε=0\varepsilon=0 leaves arbitrary lower rod data. The closing equation can be maintained by adjusting a noninserted weight, since it is linear in that weight and has a nonzero derivative at generic regular lower data. Every longer channel has a nonzero limit given by a channel of the smaller polygon. The chosen end propagator has order ε−(m−2)\varepsilon^{-(m-2)}. Compatible shortest propagators occupy disjoint weight segments separated by an unused gap, so diagrams omitting the chosen one cover at most m−3m-3 scaled gaps. Thus

0=lim⁡ϵ→0ϵm−2A~n=A~n−m+2(a)∣Gn−m+2α1⋯αm−2.(72)0=\lim_{\epsilon\to0}\epsilon^{m-2}\widetilde{A}_n=\frac{\widetilde{A}^{(a)}_{n-m+2}\lvert G_{n-m+2}}{\alpha_1\cdots\alpha_{m-2}}. \tag*{(72)}

Here A~n−m+2(a)\widetilde{A}^{(a)}_{n-m+2} is obtained by deleting the selected end vertex from every diagram that contains it, retaining its original coefficient. The remaining diagrams are precisely the lower compatible tree ansatz. The displayed limit says that this lower ansatz vanishes on the lower rod family. Induction therefore equates all its coefficients, without assuming their factorization values.

End vertices next to the root. The boundary requires a separate limit because two crossing channels vanish together. Write the final weights as (t1,…,tm−3,z)(t_1,\ldots,t_{m-3},z) and P=∏jtjP=\prod_j t_j. Writing FV=K1nF_V=K_{1n} and FV−1F_{V-1} for the root polynomial of the first V−1V-1 blocks gives FV=FV−1+zRF_V=F_{V-1}+zR, where

RP=1+∑k=1m−3Qkt1⋯tk,Qk=∑1≤i1≤⋯≤ik≤V−1∏j=1kw(m−2)(ij−1)+j.(73)\frac{R}{P}=1+\sum_{k=1}^{m-3}\frac{Q_k}{t_1\cdots t_k},\qquad Q_k=\sum_{1\le i_1\le\cdots\le i_k\le V-1}\prod_{j=1}^{k}w_{(m-2)(i_j-1)+j}. \tag*{(73)}

Along FV=0F_V=0, the crossing channels Xn−m+1,n=zPX_{n-m+1,n}=zP and X1,n−m+2=−zRX_{1,n-m+2}=-zR vanish together. A compatible diagram contains at most one of them. At z=0z=0, deleting either end vertex leaves the same lower rod data, built from the first V−1V-1 blocks of weights. Thus neither inherited lower function depends on the tjt_j. Denoting them by A~tail\widetilde{A}_{\mathrm{tail}} and A~root\widetilde{A}_{\mathrm{root}}, the 1/z1/z coefficient gives

A~root=RPA~tail.(74)\widetilde{A}_{\mathrm{root}}=\frac{R}{P}\widetilde{A}_{\mathrm{tail}}. \tag*{(74)}

For m≥4m\ge4, R/PR/P varies with the tjt_j at generic lower data. For example, setting the weights with residues 1,…,m−31,\ldots,m-3 modulo m−2m-2 to one makes Qm−3Q_{m-3} a positive integer. Thus both lower functions must vanish separately. Induction therefore fixes both boundary groups, and reversing the ordering fixes the opposite boundary.

Connecting the groups. A diagram with two compatible end vertices belongs to both groups, so their common coefficients must agree. For V≥4V\ge4, starts differing by at least m−1m-1 connect every nonroot end vertex to one of the two extreme starts, which are themselves compatible. At V=3V=3, the end vertex containing the root joins the otherwise isolated middle start to the others. Every tree has at least two end vertices and at most one contains the root. Hence all diagram coefficients agree. The proof requires m≥4m\ge4 and compatible simple poles. An overall normalization remains free.

Real kinematics and threshold connections

The classic threshold cancellation also has an interpretation in terms of reflectionless potentials [58, 59]. The threshold configurations and the distinction between ordered and full amplitudes are discussed in Sec. III.

There is also a related connection in Tr⁡(ϕ3)\operatorname{Tr}(\phi^3). The massive 2→32\to3 threshold cancellation [65] lies on the λ=4\lambda=4 member of the five-point zero family in Ref. [51], expressed in inverse propagators, and admits a regular real continuation away from threshold. The following construction concerns the full Tr⁡(ϕ4)\operatorname{Tr}(\phi^4) rank-one locus.

Real momenta on the rank-one locus

We now construct real momenta for any prescribed regular real rank-one channels. Write X^ab=Pab2−M2\widehat{X}_{ab}=P_{ab}^2-M^2 with common mass M≥0M\ge0. Factor the matrix as Cij=uivjC_{ij}=u_i v_j. Its zero row and column sums allow periodic sequences satisfying zi+1−zi=uiz_{i+1}-z_i=u_i and sj+1−sj=vjs_{j+1}-s_j=v_j. Integrating the channel mixed difference, with adjacent boundary channels set to zero, gives

X^2i+1,2j+2=yj−yi−zi(sj−si),yi−yi−1=zi(si−si−1).(75)\widehat{X}_{2i+1,2j+2}=y_j-y_i-z_i(s_j-s_i),\qquad y_i-y_{i-1}=z_i(s_i-s_{i-1}). \tag*{(75)}

Here yiy_i is also periodic, and the indices run cyclically from zero to VV, where n=2V+2n=2V+2. For a forward interval, Eq. (75) equals ∑i≤r<k<jurvk\sum_{i\le r<k<j}u_rv_k. Interleaving w2n+1=urw_{2n+1}=u_r and w2n+2=vrw_{2n+2}=v_r identifies this with the m=4m=4 rod formula (S44); the other endpoint parity gives the corresponding interchanged sum, and periodicity supplies the closing equation.

Conversely, cyclic mixed differences of these rod sums give Cij=uivjC_{ij}=u_i v_j. This explains the matrix and rod equivalence at all multiplicities.

The shortest channels are X^2i+1,2i+4=(zi+1−zi)(si+1−si)\widehat{X}_{2i+1,2i+4}=(z_{i+1}-z_i)(s_{i+1}-s_i). Regularity therefore makes every consecutive difference of sis_i nonzero. Set

zi=yi−yi−1si−si−1,ρ>max⁡j(yj+sj2),rj=ρ−yj−sj2.(76)z_i=\frac{y_i-y_{i-1}}{s_i-s_{i-1}},\qquad\rho>\max_j(y_j+s_j^2),\qquad r_j=\sqrt{\rho-y_j-s_j^2}. \tag*{(76)}

In signature (+,−,−)(+,-,-), define region momenta

Ej=(0,sj,rj),Oi=(ti,−zi/2,0),ti=(si+zi/2)2+ri2+M2.(77)E_j=(0,s_j,r_j),\qquad O_i=(t_i,-z_i/2,0),\qquad t_i=\sqrt{(s_i+z_i/2)^2+r_i^2+M^2}. \tag*{(77)}

Direct subtraction gives

(Oi−Ej)2−M2=yj−yi−zi(sj−si)=X^2i+1,2j+2.(78)(O_i-E_j)^2-M^2=y_j-y_i-z_i(s_j-s_i)=\widehat{X}_{2i+1,2j+2}. \tag*{(78)}

Set the external momenta to

p2i+1=Oi−Ei,p2i+2=Ei−Oi+1.(79)p_{2i+1}=O_i-E_i,\qquad p_{2i+2}=E_i-O_{i+1}. \tag*{(79)}

These momenta satisfy the three required checks. First, their sum telescopes to zero. Second, the adjacent channel values X^2i+1,2i+2=X^2i+3,2i+2=0\widehat{X}_{2i+1,2i+2}=\widehat{X}_{2i+3,2i+2}=0 give pa2=M2p_a^2=M^2. Third, every physical propagator joins an odd region to an even region, so Eq. (78) reproduces every prescribed channel. All remain nonzero by the assumed regularity. Positive tit_i makes odd legs outgoing and even legs incoming. Adding a zero spatial component embeds the same configuration in 1+31+3 dimensions.

References

  1. [1]N. Arkani-Hamed, Q. Cao, J. Dong, C. Figueiredo, and S. He, Hidden zeros for particle/string amplitudes and the unity of colored scalars, pions and gluons, J. High Energy Phys. 2024 (10), 231, arXiv:2312.16282.
  2. [2]N. Arkani-Hamed, L. Rodina, and J. Trnka, Locality and Unitarity of Scattering Amplitudes from Singularities and Gauge Invariance, Phys. Rev. Lett. 120, 231602 (2018), arXiv:1612.02797.
  3. [3]L. Rodina, Uniqueness from gauge invariance and the Adler zero, J. High Energy Phys. 2019 (09), 084, arXiv:1612.06342.DOI
  4. [4]L. Rodina, Uniqueness from locality and BCFW shifts, J. High Energy Phys. 2019 (09), 078, arXiv:1612.03885.DOI
  5. [5]D. A. McGady and L. Rodina, Recursion relations for graviton scattering amplitudes from Bose symmetry and bonus scaling laws, Phys. Rev. D 91, 105010 (2015), arXiv:1408.5125.
  6. [6]C. Cheung, K. Kampf, J. Novotný, and J. Trnka, Effective Field Theories from Soft Limits, Phys. Rev. Lett. 114, 221602 (2015), arXiv:1412.4095.DOI
  7. [7]L. Rodina, Scattering Amplitudes from Soft Theorems and Infrared Behavior, Phys. Rev. Lett. 122, 071601 (2019), arXiv:1807.09738.DOI
  8. [8]J. J. M. Carrasco and L. Rodina, UV considerations on scattering amplitudes in a web of theories, Phys. Rev. D 100, 125007 (2019), arXiv:1908.08033.DOI
  9. [9]L. Rodina, UV consistency conditions for Cachazo-He-Yuan integrands, Phys. Rev. D 102, 045012 (2020), arXiv:2005.06446.
  10. [10]L. Rodina, G. Cellaroli, and J. Valenzuela, A Combinatorial Origin of Locality (2026), arXiv:2608.23008.arxiv.org/abs/2608.23008
  11. [11]L. Rodina, Hidden zeros are equivalent to enhanced ultraviolet scaling and lead to unique amplitudes in Tr(ϕ³) theory, Phys. Rev. Lett. 134, 031601 (2025), arXiv:2406.04234.DOI
  12. [12]M. Kontsevich, Intersection theory on the moduli space of curves and the matrix Airy function, Commun. Math. Phys. 147, 1 (1992).DOI
  13. [13]J. V. Backus and L. Rodina, Emergence of Unitarity and Locality from Hidden Zeros at One-Loop Order, Phys. Rev. Lett. 135, 131601 (2025), arXiv:2503.03805.DOI
  14. [14]L. Rodina, Multiloop Unitarity from Airy Zeros (2026), arXiv:2601.114359.
  15. [15]X. Li and K. Zhou, A new recursion relation for tree-level NLSM amplitudes based on hidden zeros, J. High Energy Phys. 2026 (01), 010, arXiv:2508.12894.arxiv.org/abs/2508.12894
  16. [16]Y. Li and L. Rodina, Cosmological Wavefunctions as Amplitudes: Dual Shuffle Factorization and Uniquenessarxiv.org/abs/2604.01133
  17. [17]Y. Li, L. Rodina, and Z.-Z. Wang, Exact Splitting of Amplitudes and One-Loop Integrands: A Classical Origin of Zeros (2026), arXiv:2611.06810.arxiv.org/abs/2611.06810
  18. [18]H. Huang, Y. Yang, and K. Zhou, Note on hidden zeros and expansions of tree-level amplitudes, Eur. Phys. J. C 85, 685 (2025), arXiv:2502.07173.arxiv.org/abs/2502.07173
  19. [19]C. Bartsch, K. Kampf, J. Novotný, and J. Trnka, Universality of colored scalars from the stringy Kawai–Lewellen–Tye kernel, Phys. Rev. D 112, 085002 (2025), arXiv:2505.01501.arxiv.org/abs/2505.01501
  20. [20]S. Paranjape, M. Skowronek, M. Spradlin, and A. Volovich, Large deformations of Tr(ϕ³) and the world at infinity, J. High Energy Phys. 2026 (01), 076, arXiv:2504.11253.arxiv.org/abs/2504.11253
  21. [21]M. Carrillo González and F. Ward, Hidden Zeros in Massive Theories (2026), arXiv:2601.16860.arxiv.org/abs/2601.16860
  22. [22]Y. Zhang and P. Zhu, Fermionic hidden zeros (2026), arXiv:2609.06552.arxiv.org/abs/2609.06552
  23. [23]Q. Cao, Form factors from string amplitudes, Phys. Rev. Lett. 135, 021603 (2025), arXiv:2504.15702.DOI
  24. [24]L. Zhang, 2-Split of Form Factors via BCFW Recursion Relation (2025), arXiv:2509.12564.arxiv.org/abs/2509.12564
  25. [25]Y.-C. Chang, H.-Y. Chen, and Y.-t. Huang, All zeros of (super)String Theory (2025), arXiv:2506.15299.arxiv.org/abs/2506.15299
  26. [26]S. He, Y. Zhang, F. Zhu, and P. Zhu, String Amplitudes on Integer Rectangles (2026), arXiv:2610.31197.arxiv.org/abs/2610.31197
  27. [27]N. Arkani-Hamed, C. Figueiredo, and G. N. Remmen, Open String Amplitudes: Singularities, Asymptotics, and New Representations, J. High Energy Phys. 2025 (04), 039, arXiv:2412.20639.
  28. [28]B. Giménez Umbert and B. Stur̈mfels, Splitting CEGM Amplitudes (2025), arXiv:2501.10805.
  29. [29]N. Early, The CEGM NLSM (2025), arXiv:2502.08016.arxiv.org/abs/2502.08016
  30. [30]K. Zhou, Hidden zeros for higher-derivative YM and GR amplitudes at tree-level (2025), arXiv:2510.11070.arxiv.org/abs/2510.11070
  31. [31]S. De, S. Paranjape, A. Pokraka, M. Spradlin, and A. Volovich, Hidden zeros of the cosmological wave-function, J. High Energy Phys. 2025 (07), 174, arXiv:2503.23579.arxiv.org/abs/2503.23579
  32. [32]S. Mizera, Inverse of the String Theory KLT Kernel, J. High Energy Phys. 2017 (06), 084, arXiv:1610.04230.DOI
  33. [33]S. Mizera, Combinatorics and Topology of Kawai–Lewellen–Tye Relations, J. High Energy Phys. 2017 (08), 097, arXiv:1706.08527.DOI
  34. [34]C. Bartsch, K. Kampf, D. Podivín, and J. Stalknecht, Positive Geometry for Stringy Scalar Amplitudes, Phys. Rev. Lett. 136, 011601 (2026), arXiv:2508.20161.DOI
  35. [35]C. Bartsch, K. Kampf, J. Novotný, and J. Trnka, A surface integrand for the inverse KLT kernel, J. High Energy Phys. 2026 (10), 099, arXiv:2602.15102.arxiv.org/abs/2602.15102
  36. [36]F. Cachazo, N. Early, and B. Giménez Umbert, Smoothly splitting amplitudes and semi-locality, J. High Energy Phys. 2022 (08), 252, arXiv:2112.14191.DOI
  37. [37]Q. Cao, J. Dong, S. He, and C. Shi, A universal splitting of tree-level string and particle scattering amplitudes, Phys. Lett. B 856, 138934 (2024), arXiv:2403.08855.arxiv.org/abs/2403.08855
  38. [38]Q. Cao, J. Dong, S. He, C. Shi, and F. Zhu, On universal splittings of tree-level particle and string scattering amplitudes, J. High Energy Phys. 2024 (09), 049, erratum: J. High Energy Phys. 01, 126 (2025), arXiv:2406.03838.arxiv.org/abs/2406.03838
  39. [39]N. Arkani-Hamed and C. Figueiredo, All-order splits and multi-soft limits for particle and string amplitudes (2024), arXiv:2405.09608.
  40. [40]A. Guevara and Y. Zhang, New Factorizations of Yang-Mills Amplitudes (2024), arXiv:2406.08969.arxiv.org/abs/2406.08969
  41. [41]Y. Zhang, On the New Factorizations of Yang-Mills Amplitudes, J. High Energy Phys. 2025 (02), 074, arXiv:2412.15198.DOI
  42. [42]C. R. T. Jones and S. Paranjape, Smooth Splitting and Zeros from On-Shell Recursion (2025), arXiv:2505.02520.arxiv.org/abs/2505.02520
  43. [43]B. Feng, L. Zhang, and K. Zhou, Hidden Zeros and 2-split via BCFW Recursion Relation, J. High Energy Phys. 2025 (08), 205, arXiv:2504.14215.arxiv.org/abs/2504.14215
  44. [44]K. Zhou, Understanding zeros and splittings of ordered tree amplitudes via Feynman diagrams, J. High Energy Phys. 2025 (03), 154, arXiv:2411.07944.arxiv.org/abs/2411.07944
  45. [45]Y. Li, T. Wang, T. Brauner, and D. Roest, Diagrammatic Derivation of Hidden Zeros and Exact Factorization of Pion Scattering Amplitudes, Phys. Rev. Lett. 135, 181601 (2025), arXiv:2412.14858.arxiv.org/abs/2412.14858
  46. [46]B. Feng, L. Zhang, and K. Zhou, 2-split from Feynman diagrams and Expansions, J. High Energy Phys. 2026 (02), 204, arXiv:2508.21345.arxiv.org/abs/2508.21345
  47. [47]K. Zhou, Universal Interpretation of Hidden Zero and 2-Split of Tree-Level Amplitudes Using Feynman Diagrams, Part I: Tr(ϕ3), NLSM and YM (2026), arXiv:2604.23680.arxiv.org/abs/2604.23680
  48. [48]T. Azevedo, H. Gomez, and R. Lipinski Jusinskas, On-shell representation and further instances of the 2-split behavior of amplitudes, J. High Energy Phys. 2026 (06), 001, arXiv:2512.20790.DOI
  49. [49]L. Zhang and K. Zhou, Generalized 2-split for higher-derivative YM and GR amplitudes at tree-level (2026), arXiv:2601.00297.arxiv.org/abs/2601.00297
  50. [50]K. Zhou, Towards New Hidden Zero and 2-Split of Loop-Level Feynman Integrands in Tr(ϕ3) Model (2026), arXiv:2604.13810.arxiv.org/abs/2604.13810
  51. [51]Y. Li and L. Rodina, From Hidden Zeros to Exact Splittings and New Zeros (2026), arXiv:2610.01775.arxiv.org/abs/2610.01775
  52. [52]N. Arkani-Hamed, Q. Cao, J. Dong, C. Figueiredo, and S. He, NLSM ⊂ Tr(ϕ3), Phys. Rev. D 110, 065018 (2024), arXiv:2401.05483.
  53. [53]N. Arkani-Hamed and C. Figueiredo, Circles and Triangles, the NLSM and Tr(ϕ3) (2024), arXiv:2403.04826.
  54. [54]N. Arkani-Hamed, Q. Cao, J. Dong, C. Figueiredo, and S. He, Scalar-Scaffolded Gluons and the Combinatorial Origins of Yang-Mills Theory, J. High Energy Phys. 2025 (04), 078, arXiv:2401.00041.
  55. [55]C. Bartsch, T. V. Brown, K. Kampf, U. Oktem, S. Paranjape, and J. Trnka, Hidden amplitude zeros from the double-copy map, Phys. Rev. D 111, 045019 (2025), arXiv:2403.10594.arxiv.org/abs/2403.10594
  56. [56]Y. Li, D. Roest, and T. ter Veldhuis, Hidden zeros in exceptional field theories from double copy, J. High Energy Phys. 2025 (04), 121, arXiv:2403.12939.arxiv.org/abs/2403.12939
  57. [57]M. B. Voloshin, Summing one-loop graphs at multi-particle threshold, Phys. Rev. D 47, R357 (1993), arXiv:hep-ph/9209240.DOI
  58. [58]M. B. Voloshin, Zeros of tree-level amplitudes at multi-boson thresholds, Phys. Rev. D 47, 2573 (1993), arXiv:hep-ph/9210244.DOI
  59. [59]E. N. Argyres, C. G. Papadopoulos, and R. Kleiss, On amplitude zeros at threshold, Phys. Lett. B 302, 70 (1993), also Phys. Lett. B 319, 544 (1993), arXiv:hep-ph/9212280.DOI
  60. [60]N. Arkani-Hamed, Y. Bai, S. He, and G. Yan, Scattering Forms and the Positive Geometry of Kinematics, Color and the Worldsheet, J. High Energy Phys. 2018 (05), 096, arXiv:1711.09102.
  61. [61]P. Banerjee, A. Laddha, and P. Raman, Stokes polytopes: The positive geometry for ϕ4 interactions, J. High Energy Phys. 2019 (08), 067, arXiv:1811.05904.DOI
  62. [62]P. Raman, The positive geometry for ϕp interactions, J. High Energy Phys. 2019 (10), 271, arXiv:1906.02985.DOI
  63. [63]N. Arkani-Hamed, C. Figueiredo, H. Frost, and G. Salvatori, Tropical amplitudes for colored Lagrangians, J. High Energy Phys. 2025 (05), 051, arXiv:2402.06719.
  64. [64]F. A. Berends and W. T. Giele, Recursive calculations for processes with n gluons, Nucl. Phys. B 306, 759 (1988).DOI
  65. [65]Y. Makeenko, Exact multiparticle amplitudes at threshold in large-N component ϕ4 theory, Phys. Rev. D 50, 4137 (1994), arXiv:hep-ph/9404312.

Paper details

vlnc:W_cm4hac2p
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

AI tools

View statement
Looking for related papers…
Contents