WorksPhysics and Astronomy
Beyond hidden zeros and uniqueness in Tr: Block and rod zeros in any Tr
Abstract
Hidden zeros have proven powerful tools for reconstructing amplitudes in Tr, a theory of colored scalars with cubic interactions. We show that these structures extend to any Tr theory. For odd , we construct block zeros from orthogonal sums of consecutive momenta, generalizing the original cubic hidden zeros. For every , including even , we describe a second family of nonlinear rod zeros through a rod packing problem. In Tr 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 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 tree amplitudes for up to normalization. We also conjecture that the block zeros enforce both locality and unitarity for any Tr theory with odd .
Introduction
Hidden zeros are kinematic loci on which scattering amplitudes vanish while every propagator remains finite. In , 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 , 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 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 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, , 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 . We find in fact that the simplest extension starts with and continues to every odd . The resulting block zeros replace orthogonality between individual momenta by orthogonality between consecutive momentum blocks,
A selection rule specifies which blocks must be orthogonal, recovering the usual rectangle zeros when . 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 , 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 . 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 .
We take massless outgoing momenta and define planar channels . A tree with interaction vertices has external legs and propagators. Its diagrams correspond to divisions of an -gon into -gons by noncrossing diagonals, called -angulations [60, 61, 62, 63].
For odd and , the rule uses a set of nonadjacent external momenta. Schematically, the zero conditions have the form
where 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 as the vertices of a polygon in cyclic order, as in Figure 1.

Figure 1. The complete block zero in with and , defined by the picked momenta . Vertices are momenta, with the initial selection in blue and the chosen separators in orange. The brackets identify the blocks . Purple and green mark arcs and . Their picked-momentum densities are in (a) and (b), in (c), and 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 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
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 , let arc run from to and arc from to , cyclically. Let count all momenta in the arcs and count the picked momenta among them. Retain the pair exactly when
The integer counts restrict the possible arc lengths. The common density is , so the arc lengths and picked counts must take the form
with . Checking either arc suffices.2
Step 3. Impose the block rectangle. Starting immediately after each separator, divide the arcs into consecutive blocks of momenta. Require every block on one arc to be orthogonal to every block on the other,
This is an rectangle, of width .
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
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 and .
The Tr() limit and a Tr() example
For , and exactly two nonadjacent momenta are picked. Neither arc contains another picked momentum, so and Eq. (4) is automatic. Each block is simply one momentum, , and the rule gives precisely
the usual rectangle zeros [1], with no additional conditions. At six points, picking gives the width-one zero
while picking gives the width-two zero
Thus the cubic specialization recovers the rectangular family at every multiplicity.
For at eight points, pick for example . The only retained separator pairs are and . Each intervening arc has three momenta, one of them picked. With , their conditions are
They cancel the complete amplitude in pairs,
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()
We now consider theory. In this case, we only find generic and natural zero loci that are nonlinear. At six points, the three diagrams give
Remarkably, the numerator has a simple matrix interpretation. For any even , form the matrix of contractions between adjacent momentum pairs,
with and cyclic labels. At six points it is
Every minor is the numerator of Eq. (13), up to sign. Thus the amplitude at six points vanishes precisely when rank , provided all physical channels are nonzero. We show that the same condition gives zeros at every multiplicity,
All minors must vanish. The determinant alone is uninformative because momentum conservation already makes every row and column sum zero.
For , the eight-point amplitude uses a matrix. Its upper-left minor gives
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
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 .
Real scattering and threshold nullification
We connect the rank condition to threshold nullification. In massive unbroken theory, the full tree amplitude for scattering vanishes for every even 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 dimensions with a common mass. Section S4 A gives the construction, with alternating incoming and outgoing legs describing scattering.
To reach threshold production, consider instead two incoming particles. Let and give all external and exchanged scalars mass , with inverse propagators . In mostly-minus signature, take adjacent back-to-back pairs,
with . Each pair carries only energy, so the matrix in Eq. (14) factorizes as
For regular configurations this has rank one, independently of the pair directions.
An allowed interval contains complete pairs with total energy and one constituent of a pair with energy . Its inverse propagator is
At fixed pair energies, the channels are independent of the pair directions and mass. The rank-one proof applies to because its adjacent boundary entries vanish and its cyclic mixed differences still give .
Taking and all other gives scattering with legs 1, 2 incoming. Each factor or is a nonempty proper subset sum of pair energies. It is positive if it excludes , and otherwise equals minus the positive complementary sum. Thus no physical propagator is singular.
For , the threshold configuration is
All outgoing particles are then at rest. Increasing their pair energies and adjusting to conserve momentum moves above threshold while preserving the regular zero. The ordered amplitude with adjacent incoming legs already vanishes for , whereas the full single-scalar amplitude first vanishes for . 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 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 suggests interpreting products of inverse channels as weights of nonoverlapping rods. Equation (18) then gives as the sum of single-rod weights divided by the complete-tiling weight. We extend this packing rule to every and allowed multiplicity.
For with , choose a root leg and relabel it , opening the remaining ordering into a line of gap cells between legs . A rod now covers consecutive cells and carries weight when it starts at cell . Rods may touch but cannot overlap, and multiplying their weights gives the weight of a packing.
Let be the sum of the weights of all packings of rods whose starting cells lie between and ,
The empty packing has weight , while the sum is zero if no packing with rods fits.
We define the rod-zero family by two channel assignments. For every interval with , , and , set
For the rod covering the last cells, set
At , only the second assignment is needed. All physical channels and displayed denominators must be nonzero.
The family has free parameters in the space of physical channels, and hence
The choices of root give distinct loci for , , but the same locus when or (Sec. S3 C). Section S3 proves vanishing and regularity at every allowed multiplicity.
Eight-point example. For at , the interval assignment gives
Shifting the labels by one and two gives and . The remaining assignment is
These four relations reduce the twelve diagrams to the six packing weights in Figure 2, whose sum vanishes. The relation is the vanishing minor in Eq. (17), and all four relations together are equivalent to rank on the regular locus.

Figure 2. The packing sum for at eight points. Two rods of length two occupy six cells, with allowed starting positions . Each rod starting at has weight . Multiply the two rod weights within each packing and add the six resulting terms. In general, counts the rods and bound their starting positions in .
§ 2.1 E gives explicit 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 to be a sum of every product of 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 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 and allowed multiplicity.
| Zero loci used | Unrestricted terms | Locality and unitarity emerge | |||
| 5 | 7 | 23 | 46 | 185 250 786 | ✓ |
| 7 | 6 | 32 | 32 | 30 872 016 | ✓ |
| 9 | 6 | 44 | 44 | 146 803 272 | ✓ |
| 21 | 5 | 97 | 97 | 60 862 165 | ✓ |
| 101 | 4 | 398 | 398 | 35 641 099 | ✓ |
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
where runs over compatible planar trees with vertices of valence .
For rod zeros, we prove uniqueness with locality for every and every allowed . Limits within 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 to agree. Section S3 D gives the proof.
The zero conditions fix but leave one normalization for each number of vertices. Factorization into separately normalized lower amplitudes requires whenever . Choosing , with elementary coupling , 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 . 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 , we prove that rod zeros determine all relative diagram coefficients when locality is assumed. The 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 theory, where it appears to be connected to the continuous 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 : Block and rod zeros in any
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, , 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 . Summing over two blocks of momenta cancels the interior channels and leaves four corners,
A retained separator pair selects an rectangle of these tiles. Each tile total vanishes separately. The elementary contractions inside one tile need not vanish.
The eleven-point example in
Pick . The retained separator pairs are , , and , each with a rectangle. Their six drawn tiles include one repeated contraction and give
The five independent channel relations are
They leave six physical channels unfixed and make the complete amplitude vanish throughout the regular locus.
The fourteen-point example in
At fourteen points, pick . The retained separator pairs are , , , and . These four rectangles give nine distinct block conditions. The pair contributes the rectangle with rows and columns , while the other three have width one. With , impose
The equivalent channel system is
The pairs , , and give the first two lines, while gives the last. Nine independent equations in 21 physical channels leave twelve free parameters.
The seventeen-point example in
Here and every block contains five momenta. Pick
The retained separator pairs are , , , , and . Their rectangles give eight distinct conditions,
In physical channels they are
Nine physical channels remain unfixed. These relations make all 51 diagrams cancel on the regular locus.
The full matrix in at eight points
All eight physical channels enter the matrix of adjacent-pair contractions,
The zero requires every minor to vanish, with every physical channel nonzero. For example, the upper-left minor gives , precisely the rod relation for . Its shifts and the closing equation give the full rank-one condition, as in the main text.
Explicit rod solutions in
At ten points, and one rod of length four has five positions on eight cells. Leave , , , unfixed. The complete condition is
There are no longer proper channels. The five single-rod weights sum to zero.
At fourteen points, . Leave the eight shortest channels , , unfixed. Two rods fit on twelve cells. Ten packings omit the last rod and five contain it, giving
The other five channels are determined by the one-rod packings on eight cells,
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,
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 , the prescription is exactly the usual rectangle zero [1]. We therefore take odd below and write . A zero at interaction vertices requires nonadjacent picked momenta.
The proof uses induction on . 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 on edge of the channel polygon and mark that edge when is picked. This is the polygon whose diagonals carry the . 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 and assign
The four terms in every block equation (Eq. (30)) cancel. Any boundary corner is a separator edge, where 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 . The only possible distinct proportional channels are
with 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 and call the two sides and , with . Their original boundary paths contain and momenta. Each lower amplitude acquires one additional external momentum from the cut. On the left it is
where the second sum runs cyclically along the right path. On , both cut momenta are massless. Before imposing the zero conditions, ordinary tree factorization gives
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 count picked momenta on the two original paths, excluding the new cut momenta. They obey
There are only two possibilities. One side already has the 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 . Keep a consecutive group of exactly picks in their order along the left path and leave unpicked. The group is consecutive among the picks, not among all momenta. This gives an admissible lower picked set. It requires , since a one-vertex path has room for only 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 . Replacing by the right path in Eq. (44) replaces one momentum by momenta. That lower block now contains
parent momenta. Since the replacement adds a multiple of entries, all other block boundaries remain aligned. The changed block is the sum of consecutive parent blocks. For any block on the other arc, its required contraction is consequently
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 . The argument is identical with and exchanged.
Each side needs one more pick
If neither side has enough picks, Eq. (46) forces
We try to supply the missing pick by selecting the cut momentum. On , this is allowed when neither nor is picked, since these are its two neighbours. On , the corresponding neighbours are and . Nonadjacency of the original picks implies that both choices fail only when are picked, or when are picked. These are the paired poles treated in the next subsection.
Otherwise select the cut on an allowed side, say , and retain all old picks there. This supplies exactly picks. Again , because a single -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 .
Finding an original separator to replace the cut. The right path contains momenta and 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 momenta with picks, number the picked positions from zero as and set
The first value is , and the last is . Nonadjacency gives
An integer sequence starting nonpositive and ending nonnegative, with no upward step larger than one, must pass through zero. Since and are coprime, a zero gives
Removing this picked momentum leaves runs of lengths and , with and picks. An empty run at an endpoint is allowed.
Apply this fact with and call the resulting picked momentum . The two arcs of the lower rectangle between and have lengths and and picked counts and . Restoring the right path appends one of the runs ending at to each arc. With chosen for the first run, the parent arcs between therefore have lengths and picked counts
| length | picked momenta | |
| first arc | ||
| second arc |
Table 1.
Thus 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 , or equivalently the reverse shifted pair. Removing from the left path leaves momenta and picks. Thus passes the density test. Its rectangle relates the two crossing channels and .
To see that it relates their residues as well, label the block boundaries on the two arcs by
Labels are cyclic, so and . Each tile sets one mixed difference of to zero. Summing the tile equations from any corner to the endpoints gives
Here adjacent boundary channels vanish. Setting , or just one of them to zero, yields
The interior values of and exhaust the allowed internal diagonals incident on and in their respective cut polygons. At , 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 , every polygon vertex belongs to a marked edge, so the picks alternate. The retained rectangles pair the exchange channels with their opposite values, and the amplitude vanishes. This starts the induction.
For larger , 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 in those coordinates. Such a polynomial must be zero. Together with regularity, this proves the claim at every allowed multiplicity and every odd . The 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,
Write , with . For this picked set, the equal-density rule selects width-one rectangles with rows
and columns . The entry in row , column is .
To specify the relaxation, place boundaries before column 1 and after column . Retain an interior boundary after column , with , precisely when
Consecutive retained boundaries divide the columns into groups. If two consecutive boundary positions are , set and impose only
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 . 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 , , 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 is even, this pair removal terminates. Before restoring the full-row condition at , the root sum reduces to
where is the complete lower tree with merged into one generally off-shell leg and all internal channels inherited. The remaining full-row sum gives and cancels the prefactor. This current argument is separate from the factorization proof of the general pairwise prescription above.
For at fourteen points, and . The four rows give
The cyclic labels have been reduced modulo fourteen. For example, and 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
Let be the current including its attaching propagator. It obeys the ordinary tree recursion [64],
Each child has 1 mod leaves. For , define trial currents by
The trial current is precisely the complete packing weight in Eq. (24). For a single rod, the ratio in that equation is . For several rods, division by the complete-packing weight cancels unchanged factors and gives the dimension of a channel.
A partition into children contains a total of vertices below its first vertex. Each product of trial currents corresponds to a unique packing of rods, and conversely. To see this, write the rod starts as
The offset counts the cuts preceding rod and therefore specifies its child interval. The number of rods with each offset gives the number of vertices in that child. This explains the correspondence in both directions. The sum of child products is consequently . Dividing this sum by the complete-packing weight gives exactly the interval channel in Eq. (24). Thus Eq. (63) proves by induction. At the amputated root, the same counting gives
Separating these packings according to whether they contain the final rod gives
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 . For any proper interval the resulting channel is
Both are nonzero for . The second expression follows by separating placements that contain the last rod, while the first counts all placements on the interval. Thus every allowed has a regular rod zero, and the unfixed shortest channels vary in a nonempty open set.
Root choices and the number of conditions
At a fixed root, the unfixed shortest channels are coordinates on the regular rod locus. Each remaining physical channel is determined by one equation. Thus the 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 , cyclic invariance of the rank-one condition makes all roots equivalent. At , every root gives the same condition . For and , we can distinguish all roots using the regular point (S41) rooted at . Define, with cyclic labels,
This is the first longer-channel rod relation. A rooted locus requires whenever its root is outside the consecutive run of momenta starting at . Substituting Eq. (68) gives
The signs hold for all , . For the first, the ratio of its two binomial coefficients is . For the second, the binomial coefficient is at least . The third expression increases with by Pascal’s identity and is already at .
The three corresponding momentum runs intersect only at leg . Every other root therefore requires at least one of these nonzero to vanish. The regular point belongs only to its original rooted locus, and cyclic relabelling proves that all 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 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 , the rod family covers . In the variables , this is a hyperplane. A linear combination vanishes on it only when all agree.
For the induction, express the shortest channels as . This introduces no restriction on them. One can leave the first nonzero weights arbitrary and solve the consecutive products successively for the remaining weights. The rod assignments become
Indeed, in the recursion for trial currents, multiplying by all gap weights cancels every factor except the cuts between children, yielding this sum.
An end vertex away from the root. For a start , scale the gap weights of its channel as , with every . Deleting these gaps at 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 . Compatible shortest propagators occupy disjoint weight segments separated by an unused gap, so diagrams omitting the chosen one cover at most scaled gaps. Thus
Here 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 and . Writing and for the root polynomial of the first blocks gives , where
Along , the crossing channels and vanish together. A compatible diagram contains at most one of them. At , deleting either end vertex leaves the same lower rod data, built from the first blocks of weights. Thus neither inherited lower function depends on the . Denoting them by and , the coefficient gives
For , varies with the at generic lower data. For example, setting the weights with residues modulo to one makes 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 , starts differing by at least connect every nonroot end vertex to one of the two extreme starts, which are themselves compatible. At , 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 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 . The massive threshold cancellation [65] lies on the 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 rank-one locus.
Real momenta on the rank-one locus
We now construct real momenta for any prescribed regular real rank-one channels. Write with common mass . Factor the matrix as . Its zero row and column sums allow periodic sequences satisfying and . Integrating the channel mixed difference, with adjacent boundary channels set to zero, gives
Here is also periodic, and the indices run cyclically from zero to , where . For a forward interval, Eq. (75) equals . Interleaving and identifies this with the 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 . This explains the matrix and rod equivalence at all multiplicities.
The shortest channels are . Regularity therefore makes every consecutive difference of nonzero. Set
In signature , define region momenta
Direct subtraction gives
Set the external momenta to
These momenta satisfy the three required checks. First, their sum telescopes to zero. Second, the adjacent channel values give . 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 makes odd legs outgoing and even legs incoming. Adding a zero spatial component embeds the same configuration in dimensions.
References
- [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]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]L. Rodina, Uniqueness from gauge invariance and the Adler zero, J. High Energy Phys. 2019 (09), 084, arXiv:1612.06342.DOI
- [4]L. Rodina, Uniqueness from locality and BCFW shifts, J. High Energy Phys. 2019 (09), 078, arXiv:1612.03885.DOI
- [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]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]L. Rodina, Scattering Amplitudes from Soft Theorems and Infrared Behavior, Phys. Rev. Lett. 122, 071601 (2019), arXiv:1807.09738.DOI
- [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]L. Rodina, UV consistency conditions for Cachazo-He-Yuan integrands, Phys. Rev. D 102, 045012 (2020), arXiv:2005.06446.
- [10]L. Rodina, G. Cellaroli, and J. Valenzuela, A Combinatorial Origin of Locality (2026), arXiv:2608.23008.arxiv.org/abs/2608.23008
- [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]M. Kontsevich, Intersection theory on the moduli space of curves and the matrix Airy function, Commun. Math. Phys. 147, 1 (1992).DOI
- [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]L. Rodina, Multiloop Unitarity from Airy Zeros (2026), arXiv:2601.114359.
- [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]Y. Li and L. Rodina, Cosmological Wavefunctions as Amplitudes: Dual Shuffle Factorization and Uniquenessarxiv.org/abs/2604.01133
- [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]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]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]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]M. Carrillo González and F. Ward, Hidden Zeros in Massive Theories (2026), arXiv:2601.16860.arxiv.org/abs/2601.16860
- [22]Y. Zhang and P. Zhu, Fermionic hidden zeros (2026), arXiv:2609.06552.arxiv.org/abs/2609.06552
- [23]Q. Cao, Form factors from string amplitudes, Phys. Rev. Lett. 135, 021603 (2025), arXiv:2504.15702.DOI
- [24]L. Zhang, 2-Split of Form Factors via BCFW Recursion Relation (2025), arXiv:2509.12564.arxiv.org/abs/2509.12564
- [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]S. He, Y. Zhang, F. Zhu, and P. Zhu, String Amplitudes on Integer Rectangles (2026), arXiv:2610.31197.arxiv.org/abs/2610.31197
- [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]B. Giménez Umbert and B. Stur̈mfels, Splitting CEGM Amplitudes (2025), arXiv:2501.10805.
- [29]N. Early, The CEGM NLSM (2025), arXiv:2502.08016.arxiv.org/abs/2502.08016
- [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]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]S. Mizera, Inverse of the String Theory KLT Kernel, J. High Energy Phys. 2017 (06), 084, arXiv:1610.04230.DOI
- [33]S. Mizera, Combinatorics and Topology of Kawai–Lewellen–Tye Relations, J. High Energy Phys. 2017 (08), 097, arXiv:1706.08527.DOI
- [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]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]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]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]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]N. Arkani-Hamed and C. Figueiredo, All-order splits and multi-soft limits for particle and string amplitudes (2024), arXiv:2405.09608.
- [40]A. Guevara and Y. Zhang, New Factorizations of Yang-Mills Amplitudes (2024), arXiv:2406.08969.arxiv.org/abs/2406.08969
- [41]Y. Zhang, On the New Factorizations of Yang-Mills Amplitudes, J. High Energy Phys. 2025 (02), 074, arXiv:2412.15198.DOI
- [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]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]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]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]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]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]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]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]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]Y. Li and L. Rodina, From Hidden Zeros to Exact Splittings and New Zeros (2026), arXiv:2610.01775.arxiv.org/abs/2610.01775
- [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]N. Arkani-Hamed and C. Figueiredo, Circles and Triangles, the NLSM and Tr(ϕ3) (2024), arXiv:2403.04826.
- [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]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]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]M. B. Voloshin, Summing one-loop graphs at multi-particle threshold, Phys. Rev. D 47, R357 (1993), arXiv:hep-ph/9209240.DOI
- [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]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]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]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]P. Raman, The positive geometry for ϕp interactions, J. High Energy Phys. 2019 (10), 271, arXiv:1906.02985.DOI
- [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]F. A. Berends and W. T. Giele, Recursive calculations for processes with n gluons, Nucl. Phys. B 306, 759 (1988).DOI
- [65]Y. Makeenko, Exact multiparticle amplitudes at threshold in large-N component ϕ4 theory, Phys. Rev. D 50, 4137 (1994), arXiv:hep-ph/9404312.