Zhegalkin Zebra Motives Digital Recordings of Mirror Symmetry

Zhegalkin zebra motives are tilings of the plane by black and white polygons representing certain ${\mathbb F}_2$-valued functions on ${\mathbb R}^2$. They exhibit a rich geometric structure and provide easy to draw insightful visualizations of many topics in the physics and mathematics literature. The present paper gives some pieces of a general theory and a few explicit examples. Many more examples will be shown in the forthcoming article"Zhegalkin zebra motives: algebra and geometry in black and white".


Introduction
The constructions of motives in algebraic geometry heavily depend on the intersection theory of algebraic cycles and, hence, on the fairly delicate choice of an adequate equivalence relation on algebraic cycles. Chow motives, for instance, are based on rational equivalence, which is the finest equivalence relation on algebraic cycles yielding a good intersection theory [15]. On the contrary, the "motives" in the present paper are built with the usual set theoretical operations from simple subsets of the plane, which we call zebras. In 1927 Zhegalkin pointed out that functions with values in the field F 2 = Z/2Z with the usual addition and multiplication can replace the standard Boolean formalism.
The zebra with frequency v ∈ R 2 , v = 0, is the function on R 2 given by Here is the dot product on R 2 and for a real number r the integer r is such that 0 ≤ r − r < 1. It is sometimes convenient to identify the Euclidean plane R 2 and the complex plane C. In this paper we only use zebras for which the frequencies are positive integer multiples of the complex numbers with i = √ −1 and ε = e πi/6 ; see Figure 1.

Definition 1
We denote the zebra with frequency kv j by Z jk . The elements of the ring of F 2 -valued functions on R 2 generated by the zebras Z jk are called Zhegalkin zebra functions. Such a function F gives a tiling of the plane by white (F = 0) and black (F = 1) polygons. A Zhegalkin zebra function is convex (czzf) if all polygons in the tiling are bounded and convex. These tilings are the Zhegalkin Zebra Motives in the title.
Goal: Understand the deformation theory of these tilings.
Every Zhegalkin zebra function can be written as a Zhegalkin zebra polynomial, i.e. a polynomial in the variables Z jk in which all monomials have coefficients 1 and the variables in each monomial have exponent 1. Section 4.1 describes an efficient way for evaluating a Zhegalkin zebra polynomial and drawing the black-white picture. Convexity can easily be checked by visual inspection. See Figures 2,3,5,12 for examples.
A Zhegalkin zebra function F has an automorphism group consisting of translations leaving the tiling invariant: This is a lattice in R 2 if F is convex. For every sublattice Λ ⊂ Aut(F) the function F descends to a function on the torus R 2 /Λ and gives a tiling of this torus by black and white polygons. This brings us to the setting of dimer models (a.k.a. brane tilings), quivers with superpotential and discrete differential geometry. There is an extensive literature on these topics written from very different view-points, with very different terminologies, for very different applications. Our view-point will be that the pictures are realizations of an underlying combinatorial structure. Our Goal is: Understand the deformations of these realizations.
The combinatorial structure consists of the set E of edges in the picture, two permutations σ 0 , σ 1 of E and an injective homomorphism p : Z 2 → Perm(E) into the permutation group of E. The cycles (=orbits) of σ 0 and σ 1 correspond to the oriented boundaries of the white and black polygons, while the cycles of the permutation σ 2 = σ −1 1 σ 0 correspond to the vertices in the tiling. The orientation of the edges is such that the boundaries of the black (resp. white) polygons are oriented clockwise (resp. counter-clockwise). The homomorphism p comprises the action of Aut(F) and an isomorphism Aut(F) Z 2 . The permutations σ 0 and σ 1 commute with this action. Associated with a sublattice Λ of Z 2 is then the finite set E Λ = E/Λ equipped with the permutations σ 0 , σ 1 , σ 2 and an action of the finite group Z 2 /Λ. Since the torus R 2 /Λ has genus 1 the numbers of cycles of the permutations satisfy |σ 0 | + |σ 1 | + |σ 2 | = |E Λ |.
Here P Λ , P • Λ , P • Λ denote the respective sets of vertices, black and white polygons in the tiling of the torus R 2 /Λ and s, t, b, w are the respective maps which assign to an edge its source, target, adjacent black and white polygons.
In addition to the superpotential the actual pictures also contain a map ω : E → R 2 \ {0} which specifies for every edge the corresponding vector in R 2 . For a sublattice Λ ⊂ Z 2 we want this specification to be Λ-invariant; i.e. it should be a map ω : E Λ → R 2 \ {0}. We call such a map ω a realization of [F] Λ . It also gives a realization Λ ω of the lattice Λ by translations in the plane which leave the tiling specified by ω invariant. We denote the corresponding torus by The Zhegalkin zebra function F provides a tiling of R 2 with automorphism group Aut(F) and hence a realization ω F of [F] Λ for every sublattice Λ of Z 2 . It identifies Λ with a sublattice Λ ω F of Aut(F). A realization ω of the superpotential gives a tiling of R 2 which modulo Λ ω gives an embedding of the quiver (= graph with oriented edges) Γ Λ into the torus T ω as the 0-cells and 1-cells in the tiling. One can subsequently embed the graphs Γ ∨ Λ and D F ,Λ into this torus by means of a function θ : E Λ → R >0 for which the sum over each cycle of σ 0 and each cycle of σ 1 is equal to 1. This function is used to mark in each black/white polygon a point by taking a convex combination of the midpoints of its edges. This will be discussed in detail in Section 4.3.3. In [10] such a function θ is called a (positive) fractional matching. The existence of a fractional matching for Definition 3 An integer weight function for the superpotential [F] Λ is a map ν : E Λ → Z ≥0 for which the sum over each cycle of σ 0 and each cycle of σ 1 is equal to an integer deg ν (the degree of ν). The set of integer weight functions  with the operation + is a graded semi-group W Λ . An integer weight function of degree 1 is called a perfect matching, dimer covering or dimer configuration. The set of perfect matchings is denoted by M Λ . An integer weight function ν is said to be positive if ν(e) > 0 for all e ∈ E Λ . Perfect matchings play a crucial role all over the literature on dimer models. From the permutations σ 0 and σ 1 one can easily check whether perfect matchings exist and determine them all. Subsequently one can check whether the sum of all perfect matchings is a positive weight function, which then divided by its degree |M Λ | yields a positive fractional matching.

Definition 4
We say that the superpotential [F] Λ is dimer complete if the sum of all perfect matchings is a positive weight function.
If [F] Λ is dimer complete, the semi-group W Λ is generated by the perfect matchings (see Proposition 1): Rescaling the axes in the picture of the realization ω F by means of the diagonal matrix diag 1 2 , yields a realization of the superpotential [F] Λ with edge vectors ω F (e) diag 1 2 , √ 3 2 . It follows from (1) (see also §4.2.1) that the vertices in the rescaled tiling have coordinates in Q. By further rescaling with a positive integer factor one can clear the denominators and obtain a realization ω of [F] Λ with edge vectors in Z 2 , say ω(e) = (ω 1 (e), ω 2 (e)) with ω 1 , ω 2 : E Λ → Z. Now assume that the superpotential [F] Λ is dimer complete and let ν be a positive integer weight function. Then, for a sufficiently large integer N the maps ν 1 = ω 1 + N ν, ν 2 = ω 2 + N ν and ν 3 = N ν are positive weight functions with deg ν 1 = deg ν 2 = deg ν 3 . Then ω = (ν 1 − ν 3 , ν 2 − ν 3 ) and θ = 1 deg ν3 ν 3 yield for every edge e in the tiling of R 2 four points s( e), t( e), b( e), w( e), namely the endpoints of that edge and the marked (by θ) points in the polygons adjacent to that edge; see Figures 4,5,6,8. These quadrangles (for e ∈ E Λ ) constitute a tiling of R 2 . Taken modulo Λ ω the vertices and edges of the induced quadrangle-tiling give an embedding of the graph D F ,Λ into the torus T ω . So D F ,Λ is an S-quad-graph in the sense of [4] Definitions 3.1, 4.3. Since it is easy to draw pictures (see §4.3) the conditions in Definition 5 can easily be checked by visual inspection. In many examples one can find weight realizations by staring at the picture of the tiling for the Zhegalkin zebra function F drawn with the method of §4.1.
For purposes of further processing the following matrix A (u ν1 1 u ν2 2 u ν3 3 ) gives a very useful presentation of the maps s, t : E Λ → P Λ and ν 1 , ν 2 , ν 3 : E Λ → Z >0 . The rows and columns of A (u ν1 1 u ν2 2 u ν3 3 ) correspond with the elements of P Λ and its entries lie in the polynomial ring Z[u 1 , u 2 , u 3 ]; the entry in row s and column t is Figure 4: Quadrangles for the matrices A (u ν1 1 u ν2 2 u ν3 3 ) in Example 1.
Example 1 For F 2 and F 3 as in Figures 2 and 3 one has the weight realizations The 2-cells in these realizations are squares, resp. triangles with angles π 4 , π 4 , π 2 . The marked points in the 2-cells are their barycenters. The quadrangles for these weight realizations are shown in Figure 4.
The quadrangles in a weight realization constitute a tiling of the plane R 2 . When taken modulo Λ ω the s( e)t( e)-diagonals and the w( e)b( e)-diagonals show the graphs Γ Λ and Γ ∨ Λ embedded in the torus T ω and the duality between them.  Figure 2. Both ways of putting diagonals in the right-hand picture in Figure 4 lead to triangulations equivalent with the left-hand picture in Figure 3.

Example 3
The Zhegalkin zebra function shown on the left in Figure 5 has no fractional matchings because |P • Λ | = |P • Λ |. Nonetheless if one takes the baricentres of the polygons, one finds the tiling by quadrangles as shown on the right in Figure 5. The two ways of putting diagonals lead to respectively the left-hand picture in Figure 5 and the right-hand picture in Figure 3.  Each of its two diagonals divides a quadrangle into two triangles which we color black/white as indicated in Figure 6. When the quadrangles are put together to make a tiling of the plane the colored triangles for the s( e)t( e)diagonals fuse so as to form the black and white polygons in a tiling which we want to think of as the deformation, determined by (ν 1 , ν 2 , ν 3 ), of the tiling given by the Zhegalkin zebra function F. The weight realization (ν 1 , ν 2 , ν 3 ) itself can be deformed by where ν 1 , ν 2 , ν 3 , ν 1 , ν 2 , ν 3 ∈ W Λ are such that deg ν j = deg ν j for j = 1, 2, 3 and N ∈ Z ≥0 is so large that the positivity and strict convexity conditions are satisfied for the deformed triple.
w( e) The colored triangles for the w( e)b( e)-diagonals, on the other hand, make up a tiling of the plane R 2 by black and white triangles such that each triangle has one -vertex, one •-vertex and one •-vertex. It is a well-known [14] that from such a triangulation one can construct a branched covering B : T ω → CP 1 with precisely three branch points 0, 1, ∞: This is where Zhegalkin Zebra Motives meet Dessins d'Enfants. In the works on Dessins d'Enfants on Riemann surfaces of genus 1 one wants to find on the torus a structure of elliptic curve over a number field such that the branched covering map is a morphism of varieties, called a Belyi map. We will not elaborate on Dessins d'Enfants, but refer instead to [14,17,18]. The map B induces unramified coverings of C \ {0, 1} = CP 1 \ {0, 1, ∞}: One can normalize the formulas describing B such that the s( e)t( e)-diagonals of the quadrangles are mapped to the line z = 1 2 in C while the midpoints of these diagonals are mapped to the point 1 2 . Every path in C \ {0, 1} starting at the point 1 2 can be lifted uniquely to a collection of paths in R 2 \ P • , P • , P starting at the midpoints of the s( e)t( e)-diagonals.
Lifting the figure-∞-loop which starts at 1 2 in direction NW yields a collection of paths known as zigzags. It is evident from Figure 7 that these correspond to the orbits of the permutation σ 1 σ 0 . Zigzag paths play an important role in the literature on dimer models and are used to formulate consistency conditions [5,6,7,11,12]. Because not all dimer models which come from Zhegalkin zebra functions do satisfy these consistency conditions we will not say more about zigzag paths.
Instead we focus on the lifts of the arrows 0 ← 1 2 and 1 2 → 1 and 1 2 ↑ ∞ shown in the right-hand picture in Figure 8 as the vectors q w (e), q b (e) and q t (e), respectively. Obviously, q t (e) = 1 2 ω(e). Proposition 7 and Formula (81) explicitly give the vectors q w (e) and q b (e).
Section 4 describes how one can solve some practical matters (by computer). Formula (126) in §4.1 is basically computer code for evaluating the Zhegalkin zebra function F and drawing the picture of the tiling. In §4.2 I describe how one can compute the superpotential [F] Λ and the realization ω F . From the superpotential one can easily determine all perfect matchings. It is described in the text between Definitions 4 and 5 how to obtain from this a weight realization, which subsequently can be deformed with Formula (9). In §4.3 I describe how one can construct and draw the quadrangle tiling of R 2 for a given weight realization. In the remainder of the Introduction we summarize some results of Sections 2 and 3 and put them into perspective.

Symptoms of Mirror Symmetry:
There are evidently two sides to the story with the graphs Γ ∨ Λ and Γ Λ on different sides and the S-quad-graph D F ,Λ providing a "mirror correspondence". The appearance of the semi-group ring Z[W Λ ] of W Λ on one side and the semi-group (20)) on the other side is reminiscent of mirror symmetry as in the work of Batyrev and Borisov [2,3].
The semi-group ring Z[W Λ ] is a graded ring. A standard construction from algebraic geometry (see [13] Ch II) associates with this ring the projective scheme Proj(Z[W Λ ]). Alternatively, one can construct Proj(Z[W Λ ]) with the methods of toric geometry as follows. A perfect matching m defines a sub-semi-group of the group H 1 (Γ ∨ Λ , Z): The dual semi- Remark 2 Notice the analogy between Formula (9) for the deformations of weight realizations and the actions of This equivalence relation corresponds to the subgroup of H 1 (Γ ∨ Λ , Z) generated by the maps α v : E Λ → Z defined in (26). The set of equivalence classes is a graded semi-group of rank 3. The scheme A weight realization (ν 1 , ν 2 , ν 3 ) gives rise to a tiling of R 2 by quadrangles and, hence, for every e ∈ E Λ vectors q b (e) and q w (e) as in Figure 8. Using these vectors we define maps From Figure 8 and (14) one then sees that for m, Thus we obtain an injective map where A Λ = M Λ / ∼ denotes the set of ∼-equivalence classes of perfect matchings. The convex hull conv(A Λ ) of the set A Λ is called the Newton polygon of [F] Λ and through the map Q b − Q w it becomes a (concrete) polygon in R 2 . See Figure 11 for examples.
The Γ Λ -side: The counterpart of W Λ on the Γ Λ -side is the Jacobi algebra Jac([F] Λ ) of the superpotential [F] Λ . This is the quotient of the path algebra Z[Path(Γ Λ )] of the quiver Γ Λ by a two-sided ideal provided by the permutations σ 0 and σ 1 ; see (96) for a precise definition based on [5,6]. The Jacobi algebra comes with an injective algebra homomorphism, which we call the tautological representation, into the algebra of |P Λ | × |P Λ |-matrices over the semi-group ring Z[W ∨ Λ ] of the semi-group dual to W Λ : The tautological representation Φ induces an injective algebra homomorphism from the center of the Jacobi algebra into the semi-group-ring of the semi-group dual to W Λ ; see (15) and Proposition 6. For a weight realization (ν 1 , ν 2 , ν 3 ) "evaluation at ν 1 , ν 2 , ν 3 " gives a ring homomorphism In combination with the tautological representation this gives an injective algebra homomorphism It seems likely that this condition is satisfied in the case of Zhegalkin zebra motives.
The D F ,Λ -correspondence: Write σ 0 and σ 1 as permutation matrices; i.e. matrices with rows and columns indexed by the elements of E Λ and in column e only one non-zero entry, namely 1 in row σ 0 (e) (resp. σ 1 (e)). Fix a perfect matching m and multiply column e of matrix σ 0 (resp. σ 1 ) by 0 if m(e) = 1. This yields the nilpotent matrices ς m,0 and ς m,1 . We set Then τ m,0 and τ m,1 are unipotent matrices of size |E Λ | × |E Λ | with entries in Z ≥0 . They define injective homomorphisms of semi-groups Here ν, ν ∈ W Λ are viewed as column vectors. (82) and (83) then show that the maps (16) can be expressed as linear combinations of Conclusion: The maps T m,0 , T m,1 : W Λ −→ W ∨ Λ give the duality between the graphs Γ Λ and Γ ∨ Λ .
Every perfect matching m yields a pair of matrices τ m,0 and τ m,1 with entries in Z ≥0 and determinant 1. Products and transposes of such matrices also have entries in Z ≥0 and determinant 1. In this way one obtains lots of maps from W Λ to W ∨ Λ . We leave further analysis of this structure for future research on Zhegalkin Zebra motives.

The algebraic geometry of weights
In this section F is a convex Zhegalkin zebra function and Λ is a sublattice of Aut(F), such that the superpotential [F] Λ is dimer complete and such that a weight realization of the superpotential exists; see Definitions 1, 2, 4, 5. So there is a tiling of R 2 by strictly convex quadrangles and the diagonals provide embeddings of the graphs Γ Λ and Γ ∨ Λ into the torus T = R 2 /Λ.

The (co)homology of
There is only one linear relation between the maps α v (v ∈ P Λ ) and there is only one linear relation between the maps 2.1.2. The first cohomology group H 1 (Γ Λ , Z) of the graph Γ Λ is the subgroup of the group Z EΛ = Maps(E Λ , Z) consisting of the maps η : The rank of this group is The maps (27), (28), (29). The embedding Γ Λ → T induces a homomorphism of homology groups This homomorphism is surjective and its kernel is generated by the elementš In case k = 0 these are the constant paths supported on the vertices of Γ Λ . The homology class of a closed path p = (e 1 , . . . , e k ) is is generated by the homology classes of closed paths on Γ Λ . Special closed paths on Γ Λ are given by the boundaries of the black and white polygons in the tiling. Their homology classes areβ b andβ w as in (33). They generate Since there is only one linear relation between the equations in the system (35) the rank of the cohomology group is The maps α v for v ∈ P Λ are elements of H 1 (Γ ∨ Λ , Z). They generate a subgroup of rank |P Λ | − 1; see (26), (29).
The embedding Γ ∨ Λ → T induces a homomorphism of homology groups This homomorphism is surjective and its kernel is generated by the elementš

The geometry of Proj(Z[W Λ ])
In this section we investigate the geometry of the projective scheme Proj(Z[W Λ ]), which by general constructions in algebraic geometry is associated with the graded semi-group W Λ of integer weight functions for [F] Λ ; see [13] Ch.II.

By Definition 3 an integer weight function for the superpotential [F]
From (35) and (39) one sees that the difference ν − ν of two weight functions with the same degree is an element of H 1 (Γ ∨ Λ , Z). Conversely, if ν is a positive weight function (i.e. ν(e) > 0 for all e ∈ E Λ ) and θ is an element of H 1 (Γ ∨ Λ , Z), then for all sufficiently large integers N the function N ν − θ is a positive weight function. Thus we find that

2.2.2.
On the semi-group W Λ we define an equivalence relation ∼ by with α v as in (26). We denote the set of equivalence classes by W Λ : This is a graded semi-group of rank 3. The natural surjective homomorphism of semi-groups W Λ → W Λ is the analogue of the surjective homomorphism of groups H 1 (Γ ∨ Λ , Z) → H 1 (T, Z) induced by the embedding Γ ∨ Λ → T; cf. (37).

2.2.3.
Recall from Definition 3 that the integer weight functions of degree 1 are called perfect matchings and that M Λ is the set of perfect matchings. We denote the set of equivalence classes for the relation ∼ on M Λ by A Λ : The set of its vertices is precisely the set of perfect matchings M Λ .
ii. The semi-group W Λ is generated by the perfect matchings and the semigroup W Λ is generated by the set A Λ : iii. The matching polytope conv(M Λ ) has dimension |P Λ | + 1 and the Newton polygon conv(A Λ ) has dimension 2.
ii. Let ν ∈ W Λ , ν = 0, be given. By i. there are non-negative real numbers r ν,m (m ∈ M Λ ) such that Then ν(e) ≥ r ν,m m(e) for all m and e. Now take m such that r ν,m > 0. Then ν(e) − m(e) ≥ 0 for all e ∈ E Λ . This means that ν − m ∈ W Λ . Note that deg(ν − m) = deg ν − 1. If ν − m = 0 we repeat the preceding step with ν − m instead of ν. After finitely many steps we arrive at the situation that ν minus some linear combination of perfect matchings with positive integer coefficients is 0. This result passes well to ∼-equivalence classes.
iii. This follows from (40) 2.2.5. As a consequence of (45) we have a surjective homomorphism of rings where M denotes the lattice of Z-linear relations between the perfect matchings: The  Figure 38, of the local pairing at a white node w of the graph Γ ∨ Λ can be phrased as follows. Let e, e and e in E Λ be such that w(e) = w(e ) = w(e ) = w. Let E, E , E be the edges of the graph Γ ∨ Λ dual to e, e , e , respectively, and pointing away from the vertex w. Write the cycle of σ 0 which corresponds to w as (e 1 , . . . , e q ) with e 1 = e and let e = e j and e = e h . Then Formula (65) in [10] can be stated as A similar formula holds for the local pairing at a black node b of Γ ∨ Λ , but since in our convention the boundaries of the black polygons are oriented clockwise, there is an extra −-sign: Definition 8.2 in [10] builds the skew symmetric bilinear form ε on H 1 (Γ ∨ Λ , Z) from these local pairings. For reasons that will become clear in (63) we denote this form as ε + . The defining formula in [10] can then be stated as where X L denotes the element of Z[H 1 (Γ ∨ Λ , Z)] which corresponds to the (homology class of) the loop L on Γ ∨ Λ .

2.2.8.
We are now going to give a simple description of the form ε + (56) in terms of the permutations σ 0 and σ 1 . Fix a perfect matching m. Write σ 0 and σ 1 as permutation matrices; i.e. matrices with rows and columns indexed by the elements of E Λ and in column e only one non-zero entry, namely 1 in row σ 0 (e) (resp. σ 1 (e)). By multiplying for each e ∈ E Λ the corresponding column by 1 − m(e) we obtain two new matrices ς m,0 and ς m,1 , respectively. These are nilpotent matrices. We set The meaning of these matrices is as follows. Write Since H 1 (Γ ∨ Λ , Z) is the subgroup of Z EΛ which is generated by the differences of pairs of perfect matchings we conclude from (60)-(62): is the restriction of the bilinear form on Z EΛ associated with the matrix ρ m,0 + ρ m,1 − I : This holds for every perfect matching m.

2.2.9.
The difference of the matrices − 1 2 I+ρ m,0 and − 1 2 I+ρ m,1 induces another anti-symmetric bilinear form ε − on H 1 (Γ ∨ Λ , Z): The form ε − can also be defined with the method of §2.2.7, i.e. (cf. (56)) As a consequence the right-hand side of (64) is independent of the choice of the perfect matching m.  • the pair of permutations σ 0 and σ 1 plus the perfect matching m.

Definition 7
A zigzag for the superpotential [F] Λ = (E Λ , σ 0 , σ 1 ) is a cycle of the permutation σ 1 σ 0 . The set of zigzags is denoted by P z Λ . Every zigzag z defines a map α z : The matrix ρ m,0 + ρ m,1 − I is not invertible either, because, as one easily checks, for every zigzag z. Compare Formula (74) with [10] Lemma 1.1.

2.2.12.
The matrices −I+2ρ m,0 and −I+2ρ m,1 have entries in Z ≥0 and are of the form I + nilpotent. So, they give injective (but not surjective) homomorphisms of semi-groups T m,0 , T m,1 : where W ∨ Λ is the semi-group dual to W Λ ; see (20). The maps T m,0 and T m,1 do depend on the choice of the perfect matching m.

The geometry of
where M denotes the lattice of Z-linear relations between the perfect matchings: The following commutative diagram helps to locate these points Proof : Equations (142) and (143) yield The results (82)   From (41) and (72) we see that if m ∼ m , then This means that the map induces an embedding of A Λ = M Λ / ∼ and the Newton polygon conv(A Λ ) into R 2 .

Example 4
The above method yields for the superpotentials on the Zhegalkin zebra functions F 2 , F 3 , F 4 , F 6 and F in Figures 2, 3, 12 and Example 5 the Newton polygons in Figure 11, where we have also indicated the sizes of the Figure 11: Newton polygons for the superpotentials on the Zhegalkin zebra functions F 2 , F 3 , F 4 , F 6 and F in Figures 2, 3, 12 and Example 5.

2.4.3.
For a map ψ : E Λ → C and z ∈ C we can construct the maps With the constructions in §2.3.1 we thus obtain two (parametrized) curves in CSpec(Z[W Λ ]) through the point given by ψ:  (91) and (92) are non-zero, these formulas also define two curves

Since the coordinates in
The symplectic forms ε • and ε • on H 1 (Γ ∨ Λ , C * ) and the realization ω yield two vector fields ε • (-, ω) and ε • (-, ω). The Formulas (93), resp. (94), show that the curves S •ψ , resp. S •ψ , are integral curves for these vector fields. ii. The Jacobi algebra of the superpotential [F] Λ is the algebra where D • (e) | e ∈ E Λ is the two sided ideal generated by the elements Note: In [1,8] the master space is denoted as F .
So, D (e) | e ∈ E Λ is the Jacobi ideal of the polynomial F. The monomials in F correspond 1-1 with the cycles of the permutations σ 0 and σ 1 with neglect of the cyclic structure. Compare this with Remark 1 and the analogies (96)/(98) and (97)/(99).

The semi-group dual to
Let Mat P Λ denote the ring of matrices with rows and columns indexed by the elements of P Λ and let Z[W ∨ Λ ] denote the semi-group ring of W ∨ Λ . The map (100) can then be upgraded to an algebra homomorphism such that Φ(p) is the matrix with all entries 0 except for the (s(p), t(p))-entry, which is p viewed as an element of W ∨ Λ through (100); i.e.
It is clear from (39) and (97) that Φ induces an algebra homomorphism Definition 9 We call the above homomorphism Φ the tautological representation of the Jacobi algebra.
3.1.3. It follows from (39) and (99) that the ring homomorphism induces a ring homomorphism and, hence, a morphism of schemes Proposition 5 The image of the morphism (106) is an irreducible closed subscheme of the Master Space Spec(R ([F] Λ )). Note: in [1,8] the master space is denoted by F and the irreducible component by Irr F .
Proof : Since W Λ = Z ≥0 M Λ by (45) an element α of W ∨ Λ is completely determined by its values α(m) for m ∈ M Λ . So there is an injective ring homomorphism It follows that the ring Z[W ∨ Λ ] has no zero-divisors and that the kernel of the ring homomorphism (105) is a prime ideal.
The situation described by (109) is in an obvious sense dual to the situation described in (48)-(51).

Remark 6
The story in (98), (99), (50), (109) is well-known. It differs from the discussion of the Master Space and its irreducible component in [8,1] only in terminology and style and in that we have highlighted the role of the weight functions.
3.1.5. By definition the center of the Jacobi algebra is Applying Φ (103) to an element π in Z(Jac([F] Λ )) yields the matrix equations For ν ∈ W Λ "evaluation at ν " defines a homomorphism of semi-groups W ∨ Λ → Z ≥0 and, hence, a homomorphism of rings . By combining this homomorphism with Φ we obtain an algebra homomorphism such that for every e ∈ E Λ the only non-zero entry of the matrix Φ ν (e) is u ν(e) in position (s(e), t(e)). The matrix equations then imply that there is an element Now let ν, ν ∈ W Λ be such that ν ∼ ν (see (41)), say Then we have for all e ∈ E Λ : Consequently we have for π in Z(Jac([F] Λ )): and, hence, c π (ν ) = c π (ν). This means that c π is actually an element of the semi-group ring Z[W This proves: Proposition 6 There is an algebra homomorphism induced by the tautological representation Φ (103).

The matrix
3 ) contains the complete information on the edge vectors with which one can draw the quiver Γ embedded in R 2 . More precisely, consider a path p = (e 1 , . . . , e k ) and its subpaths p j = (e 1 , . . . , e j ) for j = 1, . . . , k. The path p j corresponds to the monomial in the (s(e 1 ), t(e j ))-entry of the matrix A (u ν1 1 u ν2 2 u ν3 3 ) j . In this way one sees the actual path p as it runs through the end points of the subpaths p j (j = 1, . . . , k). Thus one obtains from A (u ν1 1 u ν2 2 u ν3 3 ) the paths on Γ Λ , the boundary cycles of the polygons and the period lattice. Since this is all one needs for (96)-(97) we conclude: in (119), (103), (118) are injective.
Remark 7 Theorem 2 is closely related to Theorem 3.17 and Definition 3.10 in [5]. So it seems that the quivers with potential coming from Zhegalkin zebra functions are cancellative in the sense of [5] Def. 3.10. On the other hand, we do have examples of Zhegalkin zebra functions for which the quiver with potential is not consistent in the sense of [5] Thms. 1.37 and 3.11.

Remark 8
The above method of generating paths corresponds to the series expansion Since every entry of the matrix A (u ν1 1 u ν2 2 u ν3 3 ) is divisible by u 1 u 2 u 3 the series on the right-hand side converges in the topology provided by the powers of the principal ideal u 1 u 2 u 3 Z[u 1 , u 2 , u 3 ].

Practical matters
In this section I describe some methods for using a computer to draw the tiling associated with a Zhegalkin zebra polynomial F, compute the superpotential [F] Λ and check some conditions. Although the ideas work quite generally the exposition here is strongly influenced by my habit of using matlab.

How to draw the picture of the tiling of F
The defining formula for a czzf F can be rewritten as Formula (126) (below) with which one can easily draw the picture of the tiling. For the description of Formula (126) we define the function ¬ : R → {0, 1} by ¬(r) = 1 if r = 0 and ¬(r) = 0 if r = 0, we identify 0, 1 ∈ F 2 with 0, 1 ∈ R and we interpret in the matrix operations the matrix entries as elements of R.
Extract from the defining formula for F the 2 × n-matrix V of which the columns are the used frequency vectors. Put the coordinates of the points at which the function should be evaluated as rows in a k × 2-matrix X. Compute the matrix 2X · V and apply the function mod 2 to its entries. In short hand notation this can be summarized as 2X · V mod 2.
Extract from the defining formula for F the n×m-matrix M with entries 0, 1 of which the columns correspond to the monomials in the formula. Note that a monomial evaluates to 1 if and only if all variables it involves have value 1. This leads to the formula ¬ (¬( 2X · V mod 2)) · M for evaluating the monomials. In this formula the function ¬ is applied to the entries of the matrices.
The next and final step is to take the sum of the columns (or, equivalently, multiply on the right by the column vector 1 consisting of m 1's) and reduce the result modulo 2. The result is a vector of 0's and 1's which gives the value of F at the points listed in X. Thus the whole evaluation process reads: with which we reinterpret the frequency vectors in the defining formula for F. This clearly does not change the combinatorial structure of the picture, but it allows to do most computations with integer arithmetic. As in §4.1 we let the matrix V be such that its columns are the frequency vectors of F. The edges in the picture lie on lines with equation 2x · v = m with m ∈ Z and v a column of V . The vertices are intersection points of two such lines 2x · v = m and 2x · v = m with linearly independent v and v . The coordinates of the intersection point are then rational numbers with denominators dividing the number 2| det(v, v )|/|gcd(entries of v, v )|. Let K denote the least common multiple of these numbers for (v, v ) running over all pairs of linearly independent frequency vectors of F.
Fix a sufficiently large 1 positive integer N . Let X be the N 2 × 2 -matrix with set of rows {(n, m) ∈ Z 2 | 0 ≤ n, m < N } such that row (n, m) is above row (n , m ) if n + m √ 2 < n + m √ 2. We want to find among the rows of 1 K X those which are intersection points of two lines 2x · v = m and 2x · v = m with linearly independent v and v . First we determine which entries of the matrix 2X · V are divisible by K; with the notation of §4.1 this means the entries 1 in the matrix ¬((2X · V ) mod K). The intersection points correspond to the rows with a 1 for at least two linearly independent frequency vectors. Let X * denote the submatrix of X given by this selection of rows. Correspondingly we have the two matrices Looking at these two matrices column by column one easily determines what are the relevant lines and how the points in X * divide these lines into closed intervals with non-overlapping interiors. For this the initial ordering of the elements of X is very useful. Since it can happen that the same interval is produced from two different columns we remove the duplicates retaining for each interval exactly one copy. We list the intervals thus found by giving for each the two endpoints. The above calculations were done with integer arithmetic. In the next steps we have to work in R 2 and must therefore divide for all intervals in our list the coordinates of the endpoints by K. For each interval in the list, say I, take on both sides of the interval a point close to the midpoint and evaluate F at these two points using Formula (126). Remove I from the list if F has at these two points the same value. What is left is a list of intervals separating black and white regions. For each of these intervals we call one endpoint the source and the other the target, so that going along the interval from source to target the black region is on the right.
We make a new list with for each interval I besides the endpoints s(I) and t(I) also the midpoint m(I) = 1 N should at least be so large that the periodicity lattice which we want to implement later has two basis vectors in R 2 with non-negative coordinates ≤ 1 3 N

4.2.3.
Having a basis for Aut(F) we can specify the desired periodicity lattice Λ by an integer 2 × 2-matrix with non-zero determinant. In order to make the reduction modulo Λ we fix a basis for Λ and write m(I), s(I), t(I) and The vectors in (130) are given by their coordinates with respect to the chosen basis of Λ. Converting this back to the original coordinates on R 2 and multiplying by 2K turns (130) into a list of quadruples of elements of Z 2 . The first three elements in these quadruples have non-negative coordinates < 2K and can be made into integers using the injective map {0, . . . , 2K−1}×{0, . . . , 2K−1} → N, (a, b) → a + 2bK. These integers can be used as labels to identify the midpoint, source and target of the edge. The list which thus results from (130) contains many duplicates, which we remove. What remains is a list of labeled edges e with labeled source s(e) and target t(e) and the edge vector vec(e). It can still happen that this list contains edges e for which there is only one edge e with s(e ) = t(e), which case e and e must be fused. This will be taken care of in the final part of the next step.  Finally we define the edges σ 1 (e) and σ 0 (e) by s(σ 1 (e)) = s(σ 0 (e)) = t(e) and C(σ 1 (e)) ≤ C(e ) ≤ C(σ 0 (e)) for all e with s(e ) = t(e).

4.2.4.
After all this we have obtained the list of elements of E Λ and for every e ∈ E Λ the edge vector vec(e) as well as σ 0 (e) and σ 1 (e). The superpotential is then [F] Λ = (E Λ , σ 0 , σ 1 ) and ω F (e) = vec(e) gives the realization ω F of [F] Λ .

4.3.3.
For a realization ω and a positive fractional matching θ we now compute for every polygon in the tiling the point which is the convex combination specified by θ of the midpoints of the edges of that polygon. In order to do this in an efficient way we fix a perfect matching m. Consider a black polygon b ∈ P • Λ . Let (e 1 , . . . , e q ) be the corresponding cycle of the permutation σ 1 written such that e 1 ∈ m. Then the vertices of the polygon b are located at where v = s(e 1 ) and z v is as in 4.3.2. The midpoints of its sides are The convex combination of these midpoints specified by θ is therefore For a white polygon w one can construct in the same way a point W w,m,θ,ω . Using translations from the lattice Λ ω one subsequently obtains a marked point in every polygon in the tiling of R 2 and vectors connecting this point to the vertices of the polygon.
where on the right-hand side we view θ t as a row vector and ω as a |E Λ |×2-matrix with real entries. The source point of edge e k of b is located at ω(e j ) = z v +   e k -th row of matrix (−I + ρ m,1 ) · ω   .
The vector from the source point of edge e k of b to the marked point B b,m,θ,ω in b is obtained by subtracting (141) from (140). The term z v cancels out. Noticing that b = b(e k ) we are led to introduce the |E Λ | × |E Λ |-matrix B m,θ by e-th row of B m,θ = θ t · diag(β b(e) ) · (− 1 2 I + ρ m,1 ) .
The vector from the source point of edge e to the marked point in the polygon b(e) is then the the e-th row of the matrix B m,θ + I − ρ m,1 · ω. Proceeding in the same way for the white polygons we define the |E Λ | × |E Λ |matrix W m,θ by e-th row of W m,θ = θ t · diag(β w(e) ) · (− 1 2 I + ρ m,0 ) . For e ∈ E Λ take z s(e) as in as in 4.3.2 and draw for every z ∈ z s(e) +Λ ω the quadrangle as in Proposition 7 with vertex s( e) located at z and with the vectors from s( e) to b( e), w( e)and t( e) as specified in Proposition 7.