Landau Singularity Analysis Overview

• Landau singularity analysis is the study of conditions under which Feynman integrals become singular due to on-shell conditions and contour pinching.
• It employs classical Landau equations alongside modern methods like graph polynomial techniques and the principal Landau determinant to characterize singular structures.
• Contemporary approaches, including recursive unitarity cuts, Euler characteristic methods, and momentum-twistor geometry, allow precise identification of physical and higher-order singularities.

Searching arXiv for recent and foundational papers on Landau singularity analysis. {"query":"Landau singularity analysis Feynman integrals Landau equations Euler characteristic principal Landau determinant recursive Landau analysis", "max_results": 10} {"query":"(Correia et al., 2021, Fevola et al., 2023, Caron-Huot et al., 2024, He et al., 2024, Hollering et al., 26 Mar 2026, Chestnov et al., 28 Jun 2026) Landau singularities", "max_results": 10} Landau singularity analysis is the study of the kinematic loci at which a Feynman integral, or an amplitude built from such integrals, becomes singular because the integration contour is pinched by singular hypersurfaces of the integrand. In its classical form, the subject is organized by the Landau equations; in contemporary work, it is also formulated through graph polynomials, incidence varieties in momentum-twistor space, recursive cut constructions, and topological criteria based on Euler characteristics. Across these formulations, the common objective is to determine the branch-point structure of amplitudes as functions of external kinematics, to distinguish genuine singular loci from algebraic artifacts, and to relate those loci to thresholds, symbol alphabets, positivity, and nonperturbative analyticity domains (Fevola et al., 2023, Caron-Huot et al., 2024, He et al., 2024, Chestnov et al., 28 Jun 2026).

1. Classical formulation

For a graph with propagators labeled by $i$, internal momenta $k_i$, and Feynman parameters $\alpha_i$, the leading Landau singularity is determined by the simultaneous requirements that all propagators go on shell, momentum is conserved at each vertex, and the weighted loop equations hold on every loop, with $\alpha_i\neq 0$ and typically $\sum_i\alpha_i=1$ as a normalization (Correia et al., 2021). In one standard notation, the on-shell and pinch conditions are

$$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$

while in graph-polynomial language one may write bulk conditions as

$\mathcal G=\partial_\alpha \mathcal G=0,$

with $\mathcal G=\mathcal U+\mathcal F$ the graph polynomial built from the Symanzik polynomials (Fevola et al., 2023).

This framework distinguishes several notions that remain standard in the literature. A leading Landau singularity is the singularity of the full graph itself, with all propagators on shell; subleading singularities arise when some Feynman parameters vanish and the singularity is inherited from a contracted subgraph (Correia et al., 2021, He et al., 2024). A further distinction is between first-type singularities, associated with ordinary finite on-shell pinch configurations, and second-type singularities, which in one-loop analysis are tied to vanishing external Gram determinants (Flieger et al., 2022). The textbook reduced-diagram picture captures part of this structure, but modern algebraic treatments emphasize that more general correlated scalings

$\alpha_i\to \varepsilon^{w_i}\alpha_i$

must also be included; these are encoded by faces of the Newton polytope of $\mathcal G$, not only by simple edge contractions (Fevola et al., 2023).

A central physical refinement is the question of which Landau solutions lie on the physical sheet. In the analysis of $k_i$0 scattering of identical massive particles, $k_i$1-positive solutions,

$k_i$2

are used as the practical criterion for physical-sheet singularities; for planar graphs this correspondence is known, while for more general nonplanar graphs it is used as a working assumption rather than a general theorem (Correia et al., 2021).

2. Local singularity types and their limitations

Landau’s local analysis implies a restricted set of singular behaviors near a smooth codimension-one Landau locus with a unique pinch point and nonvanishing Hessian factor: poles, logarithmic divergences, or square-root-type branching. This is the origin of the long-standing expectation that generic codimension-one singularities are locally no worse than quadratic-root branch points (Bourjaily et al., 2022).

Modern examples show, however, that Feynman integrals and on-shell functions can involve irreducible cubic, sextic, octic, or still higher algebraic roots. The apparent contradiction is resolved by the observation that Landau’s local classification applies only at smooth codimension-one points of the Landau locus. Higher-order roots arise when several ordinary branch points collide, which requires kinematic limits of codimension greater than one; equivalently, the discriminant hypersurface itself becomes singular there (Bourjaily et al., 2022). In this sense, higher-order algebraic dependence is compatible with Landau’s argument, but only outside its generic codimension-one domain of validity.

A complementary one-loop result makes the relation between Landau discriminants and maximal residues explicit. Using the decomposition of loop momentum into parallel and perpendicular subspaces, it was shown that for one-loop $k_i$3-point scalar integrals the leading singularity is the inverse square root of the first-type Landau singularity when $k_i$4, while for $k_i$5 it is the inverse square root of the second-type Landau singularity, identified with the external Gram determinant (Flieger et al., 2022). This gives a direct analytic link between pinch loci and leading singularities and explains why the same discriminants also supply natural normalization factors for canonical differential equations (Flieger et al., 2022).

3. Computational frameworks

Several contemporary approaches recast Landau singularity analysis as a concrete elimination or recursion problem.

Approach	Core construction	Representative paper
Principal Landau determinant	Face-by-face analysis of $k_i$6 via Newton polytopes and codimension-one projections	(Fevola et al., 2023)
Recursive Landau analysis	Two-particle cuts, restricted Gram determinants, and inherited subgraph singularities	(Caron-Huot et al., 2024)
Landau-based Schubert analysis	Cut geometries in momentum twistor space uplifted to symbol letters by cross-ratios	(He et al., 2024)
Euler-characteristic method	Singular locus detected by drops in the Euler characteristic of a critical-point ideal	(Chestnov et al., 28 Jun 2026)

The principal Landau determinant framework was developed to handle dimensional regularization, massless propagators, and UV/IR-divergent configurations in a uniform way. For each face $k_i$7 of the Newton polytope $k_i$8, one studies the initial-form equations

$k_i$9

on $\alpha_i$0, decomposes the resulting incidence variety into irreducible components, projects those components to kinematic space, and retains only codimension-one images. The union of those images is the principal Landau determinant. The method was illustrated on 114 diagrams and implemented in the open-source Julia package PLD.jl; for a nonplanar two-loop five-point multiscale example the analysis produced 71 distinct kinematic singularities with polynomial degrees from 1 to 12 (Fevola et al., 2023).

The recursive approach starts from unitarity and rewrites the cut integral in a Gram-determinant representation in kinematic space. For a two-particle cut splitting a graph into subgraphs $\alpha_i$1 and $\alpha_i$2, one chooses a subset $\alpha_i$3, imposes the associated subgraph singularities together with the cut conditions, and defines the candidate singularity of the full graph by the vanishing of the restricted Gram determinant and its stationarity conditions. This bypasses Schwinger-parameter elimination and naturally incorporates inherited threshold, anomalous-threshold, and second-type components. The method reproduces known singularities and gives new predictions for multiloop diagrams relevant to Standard Model processes involving massive quarks and electroweak particles (Caron-Huot et al., 2024).

The Euler-characteristic method replaces direct solution of Landau equations by a topological criterion. In the Lee–Pomeransky representation, one forms a critical-point ideal

$\alpha_i$4

whose degree equals an Euler characteristic $\alpha_i$5, interpreted as the number of critical points. Landau singularities are exactly the loci where

$\alpha_i$6

Algorithmically, one projects $\alpha_i$7 onto each variable, computes a univariate elimination polynomial, and reads singular factors from the vanishing of its leading coefficient. Performed over finite fields, this makes elimination tractable for multi-scale, multi-loop examples. Subject to four diagnostics—zero-dimensionality sector by sector, absence of hidden degeneracy, agreement of sector Euler characteristics with the regulated total, and regulated verification of every candidate—the method returns what the paper calls the “genuine and complete set” of singularities (Chestnov et al., 28 Jun 2026).

4. Momentum twistors, Schubert problems, and cluster structure

In planar four-dimensional problems, Landau analysis often becomes most transparent in momentum-twistor space. External data are lines in $\alpha_i$8, loop variables are lines $\alpha_i$9, and cut conditions are incidence constraints of the form $\alpha_i\neq 0$0. This turns the Landau problem into a problem about varieties of lines in projective space, their projections to external kinematics, and the discriminants of those projections (He et al., 2024, Hollering et al., 26 Mar 2026).

A precise projective-geometric formulation defines, for a Landau diagram $\alpha_i\neq 0$1, an on-shell incidence variety $\alpha_i\neq 0$2 and a Landau map $\alpha_i\neq 0$3. When $\alpha_i\neq 0$4 has generically finite fibers, the points in a generic fiber are the leading singularities, their number is the LS degree, and the branch locus of $\alpha_i\neq 0$5 is the LS discriminant (Hollering et al., 26 Mar 2026). For planar tree diagrams, recursive factorization through four-mass-box substitutions was used to prove positivity on the positive Grassmannian and factorization of rational LS discriminants into cluster variables (Hollering et al., 26 Mar 2026).

In this same setting, Landau-based Schubert analysis explains how Landau loci are lifted to actual symbol letters. Landau analysis determines where singularities can occur, but a singularity locus such as $\alpha_i\neq 0$6 does not by itself distinguish among the possible rational or algebraic letters whose product vanishes on that locus. The Schubert refinement solves the cut geometry in momentum twistor space, identifies the relevant one-dimensional families of solutions, and constructs cross-ratios of the colliding solutions. In the examples analyzed, this reproduced known symbol alphabets and gave an explicit representation of alphabets as unions of type-$\alpha_i\neq 0$7 cluster algebras (He et al., 2024).

The seven-point case shows both the power and the limitations of graph-theoretic reduction. For the four-loop graph $\alpha_i\neq 0$8, the leading first-type Landau singularities match exactly the 42 heptagon letters in classes $\alpha_i\neq 0$9; the remaining class $\sum_i\alpha_i=1$0 appears only in subleading singularities (Lippstreu et al., 2022). A subsequent analysis of three-loop relaxations showed that two graphs have singularities outside the heptagon alphabet and established that $\sum_i\alpha_i=1$1 equivalence fails for certain branches of the Landau solution space. The new singularities are not cluster variables of $\sum_i\alpha_i=1$2, yet explicit comparison with $\sum_i\alpha_i=1$3 super-Yang–Mills leading singularities indicates that these extra scalar-graph singularities cancel from the full amplitudes (Lippstreu et al., 2023).

5. Physical-sheet structure and nonperturbative analyticity

Landau singularity analysis is not only a perturbative classification tool; it also constrains the nonperturbative analytic structure of amplitudes. A notable example is the exact $\sum_i\alpha_i=1$4 scattering amplitude $\sum_i\alpha_i=1$5 of identical massive particles in the region

$\sum_i\alpha_i=1$6

By systematically generating and filtering graphs, and solving the associated Landau equations for $\sum_i\alpha_i=1$7-positive leading singularities, it was shown that there is an infinite sequence of physical-sheet Landau curves accumulating at finite $\sum_i\alpha_i=1$8 on the curve

$\sum_i\alpha_i=1$9

This behavior differs sharply from the elastic region $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$0, where only finitely many Landau curves occur in a bounded region, and it suggests that for $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$1 the analytic structure implied by multi-particle unitarity is much richer than a finite collection of cuts and poles (Correia et al., 2021).

The same analysis identifies an “extended elastic unitarity region” below the first multi-particle Landau curve and argues that the new singularities are not anomalous thresholds of the usual heavy-particle type, but rather the branch loci of double discontinuities such as $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$2, i.e. the shadow of multi-particle unitarity in the two-body amplitude (Correia et al., 2021). This has direct consequences for any program, such as the S-matrix bootstrap, that assumes simple analyticity domains (Correia et al., 2021).

A different all-loop application occurs in four-point one-cycle negative geometries in planar $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$3 super-Yang–Mills theory. There, geometric Landau analysis of the relevant Landau diagrams shows that the integrated one-cycle negative geometry has branch points only at

$$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$4

to all loop orders (Paranjape et al., 24 Apr 2026). The result is weaker than a full symbol-alphabet classification, but it strongly constrains any nonperturbative resummation of one-cycle contributions to the logarithm of the four-point amplitude or the normalized quadrangular Wilson loop with a Lagrangian insertion (Paranjape et al., 24 Apr 2026).

6. Scope, controversies, and limitations

A persistent misconception is that solving the textbook Landau equations for reduced graphs exhausts the singularity problem. Modern work has shown that this is not generally correct. In dimensional regularization, especially with massless lines, some solutions exist for all external kinematics and correspond to UV or IR “permanent pinches” rather than genuine codimension-one kinematic singularities. The face-by-face Newton-polytope analysis behind the principal Landau determinant was designed precisely to separate those dominant components from true hypersurface singularities (Fevola et al., 2023).

A second misconception is that Landau loci alone determine symbol letters. They do not. A locus $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$5 may correspond to several rational or algebraic letters, for example letters involving square roots whose product vanishes on the same divisor. Landau-based Schubert analysis addresses this gap by constructing the cross-ratios of colliding Schubert solutions, but the distinction between singular loci and letters remains essential (He et al., 2024).

A third subtlety concerns completeness and specialization. The finite-field Euler-characteristic algorithm can return the “genuine and complete set” of singularities only under explicitly testable conditions, and the authors note that specializing kinematics and computing singularities do not commute in general (Chestnov et al., 28 Jun 2026). Likewise, the recursive unitarity-based method applies naturally to two-particle-reducible graphs; two-particle-irreducible topologies must serve as seeds and require other methods (Caron-Huot et al., 2024).

Finally, physical-sheet interpretation is still not uniform across all graph classes. In the $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$6 massive analysis, $$\alpha_i(q_i^2-m_i^2)=0, \qquad \sum_{i\in \text{loop}}\alpha_i q_i^\mu=0,$$7-positivity is used as the practical criterion for physical-sheet relevance, but a full proof is available only for planar diagrams; for more general nonplanar cases it remains a plausible but unproved assumption (Correia et al., 2021). Higher-order algebraic roots likewise show that the standard folklore about poles, logarithms, and square roots is a statement about smooth codimension-one points, not about the full global algebraic structure of cut solutions (Bourjaily et al., 2022).

Taken together, these developments have turned Landau singularity analysis from a largely formal pinch-condition calculus into a broad research program linking graph polynomials, elimination theory, finite-field algorithms, momentum-twistor geometry, cluster structures, and nonperturbative analyticity. The subject now spans exact amplitude singular sets, symbol-alphabet generation, positivity, and multi-particle physical-sheet structure, while retaining the Landau equations as its common core (Fevola et al., 2023, He et al., 2024, Hollering et al., 26 Mar 2026, Chestnov et al., 28 Jun 2026).