Positive Geometry and Amplituhedron

⚡ Key Takeaways

Positive geometry is a mathematical framework in which physical observables (notably scattering amplitudes) are encoded as the canonical form of a region carved out by positivity conditions. The amplituhedron is the prototypical example: a generalization of polytopes living in a Grassmannian whose canonical form computes the tree-level and loop integrand of planar maximally supersymmetric ( $\mathcal{N}=4$ ) Yang–Mills scattering amplitudes — without reference to Feynman diagrams, locality, or unitarity as inputs. Applies to: high-energy theory, combinatorial algebraic geometry, cosmology, and conformal field theory. Introduced by Nima Arkani-Hamed and Jaroslav Trnka (2013), with the general "positive geometry" abstraction by Arkani-Hamed, Bai, and Lam (2017).

💡 The Core Insight

A scattering amplitude is not fundamentally a sum over Feynman diagrams; it is the canonical differential form of a positive geometry. Singularities (poles) of the amplitude are dictated entirely by the facet structure of the region. The recursion "residue at a boundary = canonical form of that boundary" is the geometric avatar of unitarity factorization. Thus physics is read off from a shape's combinatorial boundary stratification.

DefinitionPositive geometry (Arkani-Hamed–Bai–Lam)

Let $X$ be a $d$ -dimensional irreducible complex projective variety defined over $\mathbb{R}$ , and let $X_{\geq 0} \subset X(\mathbb{R})$ be a closed semialgebraic subset of the real points, of full dimension $d$ . The pair $(X, X_{\geq 0})$ is a positive geometry if there exists a unique rational top-form $\Omega(X, X_{\geq 0})$ , the canonical form, satisfying the recursive adjointness property:

If $d = 0$ , then $X_{\geq 0}$ is a point and $\Omega = \pm 1$ .
If $d > 0$ , then $\Omega$ has poles only along the boundary components $C$ of $X_{\geq 0}$ (each $C$ being itself a positive geometry $(\,\overline{C}, C_{\geq 0})$ of dimension $d-1$ ), and along each such boundary the residue of $\Omega$ equals the canonical form of that boundary:

1

\quad \operatorname{Res}_{C}\,\Omega(X, X_{\geq 0}) \;=\; \Omega(\overline{C}, C_{\geq 0}).

Derivation hint: This is the defining axiom; uniqueness follows because any two candidate forms differ by a form with no poles on a projective variety, i.e. a global holomorphic top-form, which is forced to vanish by matching residues down the entire boundary stratification by induction on $d$ .

DefinitionCanonical form of a simplex and a polytope

For the standard projective simplex $\Delta \subset \mathbb{P}^d$ with vertices $Z_1,\dots,Z_{d+1}$ (columns regarded as points in $\mathbb{P}^d$ ), the canonical form in an affine chart with coordinates $x=(x_1,\dots,x_d)$ is

2

\quad \Omega(\Delta) \;=\; \frac{\langle Z_1 \cdots Z_{d+1}\rangle^{\,d}\; \langle Y\, d^d Y\rangle}{\prod_{i=1}^{d+1}\langle Y\, Z_{i_1}\cdots Z_{i_d}\rangle},

where $Y$ is the projective point being integrated over, $\langle \cdots \rangle$ denotes the $\mathrm{SL}(d+1)$ -invariant bracket (determinant), and each factor in the denominator vanishes on one facet hyperplane. For a convex polytope $P$ with facets $\{F_a\}$ , $\Omega(P)$ has simple poles exactly on the hyperplanes spanned by the facets, with residue $\Omega(F_a)$ .

TheoremExistence and uniqueness for convex polytopes

Every convex polytope $P \subset \mathbb{P}^d$ is a positive geometry, and its canonical form $\Omega(P)$ is unique. In an affine chart, $\Omega(P) = \omega(x)\, d^d x$ where the rational function $\omega(x)$ is the unique function with simple poles on the facet hyperplanes, correct residues, and overall degree $-(d+1)$ in homogeneous coordinates (i.e. it is a genuine projective top-form). Its numerator is fixed up to normalization by demanding no spurious poles in the interior.

Proof

Step 1 (Triangulation). Any polytope $P$ admits a triangulation $P = \bigcup_i \Delta_i$ into simplices with disjoint interiors. For simplices the canonical form is given explicitly by Eq. [2].

Step 2 (Additivity). The canonical form is additive under triangulation:

3

\quad \Omega(P) \;=\; \sum_i \Omega(\Delta_i).

Derivation hint: Internal facets of adjacent simplices coincide as hyperplanes but appear with opposite-sign residues (opposite orientations), so the would-be poles on internal walls cancel pairwise, leaving only poles on the genuine facets of $P$ .

Step 3 (Spurious-pole cancellation). Each $\Omega(\Delta_i)$ has poles on internal hyperplanes ("spurious poles"). Summing, residues on each internal hyperplane cancel because the two simplices sharing it induce opposite orientations. Hence $\sum_i \Omega(\Delta_i)$ has poles only on the true facets of $P$ .

Step 4 (Uniqueness). Suppose $\Omega'$ also satisfies the axioms. Then $\Omega - \Omega'$ has no poles anywhere on $\mathbb{P}^d$ (residues match on every facet by recursion). A global holomorphic top-form on $\mathbb{P}^d$ is zero, so $\Omega = \Omega'$ . $\blacksquare$

∎

Figure

Canonical form of a pentagon: poles on the five edges, no interior poles

DefinitionThe (tree) amplituhedron

\mathcal{A}_{n,k,m}

Fix integers $n > k+m$ , with $m$ even (physically $m=4$ ). Let $Z$ be an $n \times (k+m)$ matrix whose ordered maximal minors are all positive — a point in the positive part of a configuration space, $Z \in \mathrm{Mat}_{n,k+m}^{>0}$ . The amplituhedron is the image of the positive Grassmannian under the linear map induced by $Z$ :

4

\quad \mathcal{A}_{n,k,m} \;=\; \big\{\, Y = C\cdot Z \;:\; C \in \mathrm{Gr}_{\geq 0}(k,n)\,\big\} \;\subset\; \mathrm{Gr}(k, k+m),

where $\mathrm{Gr}_{\geq 0}(k,n)$ is the totally nonnegative Grassmannian (all $k\times k$ Plücker coordinates $\geq 0$ ), $C \in \mathrm{Mat}_{k,n}$ , and $Y$ is a $k$ -plane in $(k+m)$ -dimensional space. The case $k=1$ , $m=2$ reduces to a convex polygon; general $(n,k,m)$ is its "Grassmannian polytope" analogue.

TheoremAmplituhedron / amplitude correspondence (Arkani-Hamed–Trnka)

The tree-level scattering amplitude integrand of planar $\mathcal{N}=4$ SYM, for $n$ particles and N $^{k}$ MHV degree, is the canonical form of the $m=4$ amplituhedron:

5

\quad \Omega(\mathcal{A}_{n,k,4}) \;=\; \mathcal{A}^{\text{tree}}_{n,k}\big|_{\text{integrand in } Y} ,

read off via the "bosonized" momentum-twistor map $Z$ . The $L$ -loop integrand is the canonical form of the loop amplituhedron $\mathcal{A}_{n,k,4}^{(L)}$ , obtained by adjoining $L$ planes ("loop lines") with the appropriate mutual-positivity conditions. Derivation hint: Decompose $\mathcal{A}$ into images of positroid cells of $\mathrm{Gr}_{\geq 0}(k,n)$ (a "BCFW triangulation"); each cell's canonical form is a product of d log forms, and their sum reproduces the BCFW recursion result for the amplitude.

Proof

From positivity to $d\log$ forms.

Step 1 (Positroid cells). The positive Grassmannian $\mathrm{Gr}_{\geq 0}(k,n)$ decomposes into positroid cells, each parametrized by a network/plabic graph with edge or face variables $\alpha_i > 0$ . On a top cell, coordinates can be chosen so the canonical form is a product of logarithmic derivatives:

6

\quad \Omega\big(\mathrm{Gr}_{\geq 0}(k,n)\,\text{cell}\big) \;=\; \pm\,\frac{d\alpha_1}{\alpha_1}\wedge\frac{d\alpha_2}{\alpha_2}\wedge\cdots\wedge\frac{d\alpha_D}{\alpha_D},

where $D$ is the cell dimension. The boundaries $\alpha_i \to 0$ are exactly the poles, matching axiom [1]. Derivation hint: Positivity coordinates $\alpha_i>0$ make each boundary the locus $\alpha_i=0$ ; the $d\log$ form has a simple pole there with residue equal to the canonical form of the lower-dimensional cell, by direct computation of $\operatorname{Res}_{\alpha_i=0}$ .

Step 2 (Pushforward). The map $C \mapsto Y = C\cdot Z$ pushes the cell's $d\log$ form forward to a form on $\mathrm{Gr}(k,k+m)$ . For a triangulating collection of cells $\{S_\sigma\}$ of dimension $4k$ (the dimension of $\mathcal{A}_{n,k,4}$ ),

7

\quad \Omega(\mathcal{A}_{n,k,4}) \;=\; \sum_{\sigma}\, (\pi_Z)_{*}\,\Omega(S_\sigma).

Derivation hint: A triangulation tiles $\mathcal{A}$ with non-overlapping cell images; spurious boundaries (internal walls between cells) cancel by the same opposite-orientation argument as in the polytope proof (Step 3 there), leaving poles only on physical boundaries.

Step 3 (Locality and unitarity emerge). Physical boundaries of $\mathcal{A}$ correspond to consecutive Plücker brackets $\langle Y\, i\, i{+}1\, j\, j{+}1\rangle \to 0$ . In momentum-twistor variables these are precisely the propagator poles $1/P^2$ . The residue (boundary canonical form) factorizes into lower-point amplituhedra — this is unitarity factorization. Derivation hint: $\langle Y\, i\,i{+}1\,j\,j{+}1\rangle$ translates to a Mandelstam invariant $P_{i\cdots j}^2$ ; its vanishing is an on-shell channel, and the residue equals a product of two boundary canonical forms = product of two sub-amplitudes.

∎

NA

Nima Arkani-Hamed & Jaroslav Trnka

2013–present

Arkani-Hamed (b. 1972, IAS Princeton) and his then-student Trnka (b. ~1984, now UC Davis) proposed the amplituhedron in 2013 as the culmination of a decade-long program — Grassmannian formulations, on-shell diagrams, the positroid stratification — aimed at deriving scattering amplitudes from geometry rather than spacetime Lagrangians. Their guiding conviction: locality and unitarity, the twin pillars of quantum field theory, are emergent features of a more primitive combinatorial-geometric object, and spacetime itself may be reconstructed from positivity. The abstract notion of "positive geometry" was distilled with Yuntao Bai and Thomas Lam (2017), connecting the physics to total positivity, cluster algebras (Fomin–Zelevinsky), and the work of Postnikov on the nonnegative Grassmannian.

🧠 Intuition

Think first of a triangle in the plane. There is essentially one natural rational 2-form with simple poles exactly on its three edges and nothing inside — its canonical form. Now imagine an amplitude is "the canonical form of a generalized triangle." The amplituhedron is that generalized polytope, but it lives inside a Grassmannian (a space of $k$ -planes) rather than ordinary projective space, and "being inside the shape" means "all the relevant determinants are positive." The amplitude's poles — the places where it blows up, physically where intermediate particles go on-shell — are exactly the faces of this shape. So computing an amplitude becomes: find the shape, read its boundary structure, write down the unique form with poles on those boundaries. No diagrams, no integrals over loop momenta in the usual sense — just geometry. The "magic" is that the unique-form requirement (no spurious interior poles) does all the work that gauge invariance and Ward identities do in the textbook approach.

ExampleThe MHV

n=4

amplitude as a square's canonical form (

k=1, m=2

)

Given: Take $k=1$ , $m=2$ , $n=4$ . Then $Y$ is a point in $\mathbb{P}^2$ , $C \in \mathrm{Gr}_{\geq 0}(1,4)$ is a row $(c_1,c_2,c_3,c_4)$ with $c_i \geq 0$ , and $Z$ is $4\times 3$ with positive minors. The image $Y = C\cdot Z$ traces a convex quadrilateral with vertices $Z_1,Z_2,Z_3,Z_4 \in \mathbb{P}^2$ .

Steps: 1. The amplituhedron is the convex polygon $\langle Z_1 Z_2 Z_3 Z_4\rangle$ ; its facets are the four edges $\langle Y\, Z_i Z_{i+1}\rangle = 0$ . 2. The canonical form of a convex quadrilateral, by triangulating into two triangles $\{Z_1Z_2Z_3\}$ and $\{Z_1Z_3Z_4\}$ , is

8

\quad \Omega = \frac{\langle Z_1Z_2Z_3\rangle^2\,\langle Y d^2Y\rangle}{\langle Y12\rangle\langle Y23\rangle\langle Y31\rangle} + \frac{\langle Z_1Z_3Z_4\rangle^2\,\langle Y d^2Y\rangle}{\langle Y13\rangle\langle Y34\rangle\langle Y41\rangle}.

Derivation hint: Apply Eq. [2] to each triangle; the spurious pole $\langle Y13\rangle$ cancels between the two terms because its residues are equal and opposite. 3. After cancellation the diagonal pole $\langle Y13\rangle$ disappears.

Boxed answer:

\boxed{\;\Omega(\text{quad}) = \frac{\langle 1234\rangle^2\,\langle Y\,d^2Y\rangle}{\langle Y12\rangle\langle Y23\rangle\langle Y34\rangle\langle Y41\rangle}\;}

(units: a projective 2-form, dimensionless after stripping the measure). This is the geometric origin of the four-point MHV amplitude — four physical poles, no spurious interior pole.

ExampleCounting boundaries: the dimension and facet count of

\mathcal{A}_{n,k,4}

Given: The tree amplituhedron $\mathcal{A}_{n,k,4}$ , with $m=4$ .

Steps: 1. Dimension: $Y \in \mathrm{Gr}(k, k+4)$ has $\dim = k\cdot m = 4k$ . So

9

\quad \dim \mathcal{A}_{n,k,4} = 4k.

Derivation hint: $\dim \mathrm{Gr}(k,k{+}m) = k(k{+}m) - k^2 = km$ ; with $m=4$ gives $4k$ . 2. For the MHV sector $k=1$ : $\dim = 4$ . The physical poles are the $\binom{n}{2} - n$ "non-adjacent" plus adjacent channels; for $n$ particles the boundary facets correspond to $\langle Y\, i\, i{+}1\, j\, j{+}1\rangle = 0$ . 3. For $n=5,\ k=1$ : count of facets of the cyclic-polytope-like region equals the number of consecutive 4-brackets that can vanish on the boundary, which gives the $5$ physical propagator structures.

Boxed answer:

\boxed{\;\dim \mathcal{A}_{n,k,4} = 4k,\qquad \dim \mathcal{A}_{n,k,4}^{(L)} = 4k + 4L\;}

where each loop adds a 4-dimensional plane (a line in momentum-twistor space, contributing $4$ dimensions). For $L$ -loop MHV ( $k=0$ adjusted): the loop integrand lives in $4L$ dimensions.

⚠️ Common mistake

The amplituhedron is not a polytope in ordinary space. It lives in a Grassmannian, not projective space, and "positivity" refers to ordered Plücker minors, not coordinate positivity. Its boundaries are stratified by positroid combinatorics, not by simple hyperplane intersections. Treating it as a literal convex polytope only works in the degenerate $k=1,m=2$ case.

The canonical form gives the integrand, not the integrated amplitude. At loop level, $\Omega(\mathcal{A}^{(L)})$ is the loop integrand in dual/momentum-twistor variables; one must still perform the loop integration (which introduces IR divergences and transcendental functions) separately. The amplituhedron geometrizes the rational integrand, not the final polylogarithmic answer.

Positivity of $Z$ matters. If the external data $Z$ does not have all-positive ordered minors, the image need not be a positive geometry and triangulations may fail; the construction is genuinely about total positivity.

Counterexample

A non-convex region need not be a positive geometry with a sign-definite form. Consider a non-convex polygon (e.g. a "pac-man" shape) traced by a self-intersecting boundary. Its naive canonical form acquires the wrong residues or spurious poles, and the recursive residue property [1] fails on the reflex vertex. Equivalently, if one drops the total-positivity condition on $Z$ and allows minors of mixed sign, the "amplituhedron" image can fold over itself, and the canonical form develops spurious poles inside the would-be region — demonstrating that positivity is not an optional decoration but the defining structure. This is precisely why physical amplitudes (with their clean pole structure) require the positive, convex setting.

connection

Positive geometry sits at a crossroads of several fields:

Total Positivity and the Nonnegative Grassmannian — Postnikov's positroid stratification provides the combinatorial backbone; cells, plabic graphs, and the boundary poset directly give the amplituhedron's facet structure.
Cluster Algebras — The Fomin–Zelevinsky cluster structure on $\mathrm{Gr}(k,n)$ governs which singularities (symbol letters) appear; positivity is a cluster-positivity statement.
BCFW Recursion — The Britto–Cachazo–Feng–Witten on-shell recursion corresponds geometrically to a triangulation of the amplituhedron into positroid cells.
Generalized Polytopes and Canonical Forms — The polytope case [2]–[3] is the toy model; the amplituhedron is its Grassmannian generalization.
Cosmological Polytopes — The same canonical-form philosophy computes wavefunction coefficients in cosmology.

🔬 Where this appears

⏳ History

- 1994 (Postnikov, earlier roots; formalized 2006): Postnikov develops the theory of the totally nonnegative Grassmannian, its positroid cell decomposition and plabic graphs — the combinatorial foundation.

- 2003–2004 (Witten; BCFW): Twistor-string theory (Witten) and BCFW recursion (Britto, Cachazo, Feng, Witten) reveal hidden simplicity and on-shell structure of amplitudes.

- 2009–2012: Arkani-Hamed, Bourjaily, Cachazo, Goncharov, Postnikov, Trnka develop the Grassmannian integral, on-shell diagrams, and the positroid picture (culminating in the 2012 "Scattering Amplitudes and the Positive Grassmannian").

- 2013 (Arkani-Hamed & Trnka): The amplituhedron is proposed; "The Amplituhedron" (arXiv:1312.2007) and "Into the Amplituhedron" appear.

- 2014–2016: Loop amplituhedron, geometry of unitarity and the "sign-flip" characterization developed; explicit canonical forms and triangulations computed.

- 2017 (Arkani-Hamed, Bai, Lam): "Positive Geometries and Canonical Forms" abstracts the framework; the ABHY associahedron and cosmological polytope appear, extending positive geometry beyond

\mathcal{N}=4

SYM.

- 2017–present: Connections to cluster algebras (symbol alphabets), the momentum amplituhedron (He, Damgaard, Plefka, Spradlin, 2019) for amplitudes directly in spinor-helicity space, and tropical/stringy generalizations.

- 2018 (Galashin–Karp–Lam and others): Topology of the amplituhedron studied — it is shown to be homeomorphic to a ball in cases, addressing well-definedness of triangulations.

- 2020s: "Amplituhedron-prime," cluster adjacency proofs, and positivity in form factors and the energy correlators program.

open-problem

Major open questions remain: 1. Existence/uniqueness of triangulations. Proving that BCFW (and other) cell decompositions actually tile the amplituhedron — with no gaps or overlaps — was settled only partially; a fully general, intrinsic definition of the amplituhedron's boundary stratification and a proof that it is a genuine positive geometry for all $(n,k,L)$ is still being completed. 2. An intrinsic, triangulation-free formula for $\Omega(\mathcal{A}_{n,k,L})$ — a "master form" not requiring a choice of triangulation — is known only in special cases (the four-particle integrand, some MHV cases). 3. Beyond planar $\mathcal{N}=4$ . Non-planar amplitudes, theories with less supersymmetry, and gravity lack a known positive-geometry description; whether one exists is open. 4. Integration. The canonical form is the integrand; geometrizing the integrated amplitude (transcendental functions, the symbol/coaction structure) within positive geometry is an active, unfinished program (links to cluster algebras and motives). 5. The deeper meaning. If spacetime and unitarity are emergent from positivity, what selects the positive geometry? A first-principles derivation that does not presuppose the answer is the grand open question.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also