Twisted Neumann–Zagier matrices

The Neumann–Zagier matrices of an ideal triangulation are integer matrices with symplectic properties whose entries encode the number of tetrahedra that wind around each edge of the triangulation. They can be used as input data for the construction of a number of quantum invariants that include the loop invariants, the 3D-index and state-integrals. We define a twisted version of Neumann–Zagier matrices, describe their symplectic properties, and show how to compute them from the combinatorics of an ideal triangulation. As a sample application, we use them to define a twisted version of the 1-loop invariant (a topological invariant) which determines the 1-loop invariant of the cyclic covers of a hyperbolic knot complement, and conjecturally is equal to the adjoint twisted Alexander polynomial.

Motivation.Ideal triangulations of 3-manifolds with torus boundary components were introduced by Thurston [Thu77] as a convenient way to describe and effectively compute [CDW] complete hyperbolic structures on 3-manifolds.To do so, one assigns a complex number different from 0 or 1 to each tetrahedron and a polynomial equation around each edge of the triangulation.These so-called gluing equations have special shape that can be described by two matrices A and B (with rows and columns indexed by the edges and by the tetrahedra, respectively) whose entries describe the number of times (but not the order by which) tetrahedra wind around an edge.One of the main discoveries of Neumann-Zagier is that the matrix (A|B) becomes, after some minor modifications, the upper part of a symplectic matrix with integer entries [NZ85].The symplectic property of the NZ matrices and of the corresponding gluing equations define a linear symplectic structure on a vector space whose quantization leads to a plethora of quantum invariants that include the loop invariants of Dimofte and the first author [DG13,DG18], the 3D-index in both the original formulation of Dimofte-Gaiotto-Gukov [DGG14,DGG13] as well as the state-integral formulation of Kashaev and the first author [GK19] and Kashaev-Luo-Vartanov state-integral [KLV16,AGK].All of those invariants are defined using the NZ matrices of a suitable ideal triangulation, and their topological invariance follows by proving that they are unchanged under Pachner 2-3 moves.
Our original motivation was to study the behavior of the loop invariants of [DG13] under cyclic covers.Since the latter are defined in terms of NZ data of an ideal triangulation, we were led to study the behavior of the NZ matrices under cyclic covers.By elementary topology, each tetrahedron of an ideal triangulation lifts to n tetrahedra in the n-fold cyclic cover, and lifting all the way to the infinite cyclic cover leads to the notion of NZ matrices which we can call equivariant or twisted (as is common in algebraic and geometric topology) or t-deformed (as is common in physics).We will use the term "twisted", and keep in mind that the variable t below encodes topological information of cyclic covers.
Said differently, as the triangulation of the cyclic cover unfolds, so do its edges and the tetrahedra that wind around them.This is the content of the twisted NZ matrices.How can this elementary idea be nontrivial or interesting?
It turns out that the twisted NZ matrices have twisted symplectic properties which come from topology and using them one can give twisted versions of the above mentioned invariants, i.e., of the loop invariants [DG13,DG18], the 3D-index in [DGG14,DGG13] and [GK19] and KLV state-integral [KLV16,AGK].A key property of such a twisted invariant is that it determines the corresponding (untwisted) invariant of all cyclic covers.For example, the twisted 1-loop invariant of a knot complement defined below satisfies this property.
Our goal is to define the twisted NZ matrices, describe their properties and show how to compute them in terms of the methods developed by SnapPy [CDW].Having done so, we can use the twisted NZ matrices to define a twisted version of the 1-loop invariant, prove its topological invariance and conjecture that it equals to the adjoint twisted Alexander polynomial.It is interesting to note that the twisted 1-loop invariant depends only on the combinatorics of the NZ matrices of the infinite cyclic cover (which is abelian information) whereas the adjoint twisted Alexander polynomial depends on the complete hyperbolic structure given as a representation of the fundamental group.
Further applications of twisted NZ matrices will be given in forthcoming work.
1.2.Torsion and its twisted version.Before discussing twisting matters further, let us recall a key motivating example.A basic invariant of a compact 3-manifold is the order of the torsion of its first homology.The behavior of this invariant for all cyclic covers of a knot complement is determined by a single Laurent polynomial, the Alexander polynomial.Explicitly, we have where ∆ K (t) ∈ Z[t ±1 ] is the Alexander polynomial of a knot K ⊂ S 3 , M (n) is the nfold cyclic cover of M = S 3 \ K, and the right-hand side is the order of the torsion part of H 1 (M (n) ; Z).This classical result connecting the torsion of the first homology to the Alexander polynomial (see e.g.[Fox56,(6.3),p.417]) is deeply rooted in the idea that the Alexander polynomial is the torsion of the infinite cyclic cover of M twisted by the abelianization map α : In other words, we have The twisted torsion determines the torsion of the cyclic covers (2) This idea has been extended in several directions.Among them, one can define the torsion of a 3-manifold using interesting representations of its fundamental group.For example, when M is a cusped hyperbolic 3-manifold, one can define a torsion using a (lifted) geometric SL 2 (C)-representation ρ of its complement, or a symmetric power thereof.There are some technical difficulties that one must overcome, stemming from the fact that sometimes the corresponding chain complexes are not acyclic, hence the torsion depends on a choice of peripheral curves, as well as normalization issues, since the torsion is usually well-defined up to a sign.These issues have been addressed in detail by [Por97,DY12,DFJ12].Among the several torsion invariants, we will be interested in the adjoint (Reidemeister) torsion τ γ (M ) of a one-cusped hyperbolic 3-manifold (such as a knot complement) using the adjoint representation Ad ρ = Sym 2 (ρ) : π 1 (M ) → SL 3 (C), where γ is a fixed peripheral curve.The adjoint torsion τ γ (M ) ∈ F × /± is a nonzero element of the trace field F of M , well-defined up to a sign [Por97].Just as in the case of the Alexander polynomial, there is a version of the adjoint twisted Alexander polynomial τ (M, α, t) ∈ F [t ±1 ]/(±t Z ) (abbreviated by τ (M, t) when α is clear) defined in [Wad94, DFJ12, DY12] using an epimorphism α : π 1 (M ) → Z.
Here, the ambiguity is given by multiplication by an element of ±t Z := {±t r | r ∈ Z}.
Two key properties of the adjoint twisted Alexander polynomial are the behavior under finite cyclic covers [DY12] τ and the relation with the adjoint torsion [Yam08], namely where λ is the canonical longitude, i.e., the peripheral curve satisfying α(λ) = 0.
1.3.Neumann-Zagier matrices and their twisted version.Having already discussed the torsion and twisted torsion, we now recall some basic facts from ideal triangulations of 3-manifolds and their gluing equations, following [Thu77,NZ85].Let M be an oriented hyperbolic 3-manifold with a torus boundary component (often called a one-cusped hyperbolic 3-manifold) equipped with an ideal triangulation T .An Euler characteristic argument shows that the number N of tetrahedra in T is equal to the number of edges.We order the edges e i and the tetrahedra ∆ j of T for 1 ≤ i, j ≤ N .A quad type of a tetrahedron is a pair of opposite edges; hence each tetrahedron has three quad types.We fix an orientation and a quad type of each tetrahedron ∆ j so that each edge of ∆ j admits a shape parameter among with opposite edges having same parameters (see Figure 1).We denote by the three pairs of opposite edges of a tetrahedron, so that the edges of are assigned the edge parameter z .A complete hyperbolic structure of M can be described by a special solution of the gluing equations.The latter describe the holonomy of the hyperbolic structure around each edge e i of T , and have the (logarithmic) form Here G, G and G are the gluing equation matrices whose rows and columns are indexed by the edges and by the tetrahedra of T , respectively, such that the (i, j)-entry of G is the number of edges of ∆ j with parameter z j is incident to the edge e i in T .It will be convenient to introduce ζ-variables following [Sie].Note that the three shape parameters in each tetrahedron satisfy the relation Note that where diag(ζ ) denotes the diagonal matrix with diagonal entries ζ 1 , . . ., ζ N .
The NZ matrices have remarkable properties discovered in [NZ85] which is a starting point for the quantization of the gluing equations and for passing from hyperbolic geometry to quantum topology.The symplectic property of the NZ matrices implies that There are two enhancements of the gluing equations and of their corresponding matrices.The first is obtained by looking at the cusp, i.e., the peripheral (also called boundary) torus of the 3-manifold.By a peripheral curve we mean an oriented, homotopically non-trivial, simple closed curve in the peripheral torus of M .A peripheral curve γ gives a triple (C γ , C γ , C γ ) of row vectors in Z N that describe the completeness equation as Fixing a peripheral curve γ, let G denote the matrix obtained from G by replacing the last row by C γ , and likewise for A and B.
The second enhancement is a combinatorial flattening of T , that is a triple (f, f , f ) of column vectors in Z N such that for any peripheral curve γ.This term was introduced in [DG13, Sec.4.4] as a necessary ingredient to define the 1-loop invariant (and there, it was called a combinatorial flattening compatible with any peripheral curve).Every ideal triangulation has combinatorial flattenings [Neu92,Thm.4.5].This concludes our discussion of the NZ matrices.We now introduce a twisted version of the NZ matrices of an ideal triangulation T of a 3-manifold M as above.We will fix an epimorphism α : π 1 (M ) → Z and a peripheral curve µ satisfying α(µ) = 1.For instance, if M is a knot complement in S 3 , then α is the abelianization map and µ is a meridian of the knot.Let M and M (n) denote the cyclic covers of M corresponding to α −1 (0) and α −1 (nZ), respectively.The ideal triangulation T of the infinite cyclic cover M induced from T is equipped with an action of the deck transformation group Z (generated by the µ-action), and induces an ideal triangulation T (n) of the n-fold cyclic cover M (n) .
We choose lifts e i and ∆ j of e i and ∆ j to T , respectively, and for k ∈ Z let G k be the N × N matrix whose (i, j)-entry is the number of edges of µ k • ∆ j with parameter z j is incident to the edge e i in T .Note that G k is a zero matrix for all but finitely many k, as there are finitely many tetrahedra around e i .We define the twisted gluing equation matrices and the twisted Neumann-Zagier matrices by (More generally, if X(t) is a matrix with entries in Z[t ±1 ], we denote by X k the coefficient of t k in X(t).)Note that the rows and columns of these matrices are indexed by the edges and tetrahedra of T , respectively, and that their entries are in the ring Z[t ±1 ].Since the above matrices are well-defined after fixing lifts of each edge of T , a different choice of lifts changes G (t), A(t) and B(t) by multiplication from the left by the same diagonal matrix diag(t c 1 , t c 2 , . . ., t c N ) for integers c 1 , c 2 , . . ., c N .This ambiguity propagates to any invariant constructed using these matrices.
A first property of the twisted NZ matrices (A(t) | B(t)) is that they determine the NZ matrices (A (n) | B (n) ) of the cyclic covers T (n) of T , and moreover they do so via block circulant matrices, i.e., square matrices such that each row is obtained by its predecessor by a cyclic shift.We refer to [Dav79] and [RCdSL88] for details on the properties of block circulant matrices.
Theorem 1.1.We have The twisted NZ matrices satisfy a twisted version of the symplectic property of Equation (9).Theorem 1.2.We have In other words, A(t)B(1/t) T is symmetric under tranposition followed by t → 1/t.In particular, A(ω)B(ω) * is Hermitian for ω ∈ C with |ω| = 1 and so is B(ω is non-singular.Here X * is the complex conjugate of the transpose of a matrix X.Note that given any ideal triangulation, we can choose quads so that B (hence also B(t), since We now come to a conjectural property of the twisted NZ matrices, which we have checked in numerous examples.
We next discuss the behavior of the twisted NZ matrices under a 2-3 Pachner move relating an ideal triangulation T with N tetrahedra to another one T with N + 1 tetrahedra.Recall that such a move is determined by two tetrahedra with a common face in T , which become three tetrahedra with a common edge in T as shown in Figure 2 below.Let us write the twisted NZ matrices A(t) and B(t) of the triangulation T schematically in columns as with a i meaning (a 3 , a 4 , ..., a N ) and similarly for b i .Let A(t) and B(t) denote the corresponding twisted NZ matrices of T .
Theorem 1.4.With the above notation, there exists a lower triangular matrix P with 1's on the diagonal such that This corrects the omission of P in [DG13, Eqn.(3.27)], which does not affect the validity of the proofs in [DG13].
1.4.The 1-loop invariant and its twisted version.The adjoint Reidemeister torsion has a conjectural description in terms of the 1-loop invariant of [DG13].The latter depends on the NZ matrices of an ideal triangulation, its shapes, their flattenings and a peripheral curve γ.With the notation of Section 1.3 the 1-loop invariant is defined by where the last equality follows from Equation (8).In [DG13] it is conjectured that the 1-loop invariant τ CS γ (T ) is equal to the adjoint torsion τ γ (M ) with respect to γ τ CS γ (T ) Given the above conjecture and the discussion of Section 1.2, it is natural to predict the existence of a twisted version of the 1-loop invariant, defined in terms of the twisted NZ matrices.With the notation of Section 1.3, we define the twisted 1-loop invariant by where the second equality follows from the fact (analogous to Equation (8)) An elementary observation is that τ CS (T , t) is well-defined up to multiplication by an element in ±t Z .Indeed, the computation in [DG13, Sec.3.5]shows that up to sign, the twisted 1-loop invariant is independent of the choice of a combinatorial flattening.It also manifestly independent of the choice of a quad type of ∆ j , as the definition ( 22) is symmetric with respect to ζ, ζ , and ζ .Finally, a different choice of lifts of the edges of T results to left multiplication of all of the matrices G (t) by the same diagonal matrix diag(t c 1 , t c 2 , . . ., t c N ) for integers c 1 , c 2 , . . ., c N , hence the observation follows.
By its very definition of τ CS (T , t), it leads to a twisted version of Conjecture (21), namely We now list some properties of the twisted 1-loop invariant.The first concerns the topological invariance.
Combining the above theorem with Proposition 1.7 of [DG13], we conclude that the twisted 1-loop invariant defines a topological invariant of one-cusped hyperbolic 3-manifolds.
The next two theorems concern the properties (3) and (4) of the adjoint twisted Alexander polynomial.
A corollary of the above theorems is a relation between the twisted 1-loop invariant and the 1-loop invariant of [DG13] for cyclic covers, namely The next result concerns the symmetries of the twisted 1-loop invariant.Conjecture 1.3 implies a symmetry of the twisted 1-loop invariant which is known to hold for the adjoint twisted Alexander polynomial [Kit96, KL99], namely τ (M, t) = τ (M, 1/t).
Corollary 1.8.(assuming Conjecture 1.3) We have We end our discussion on the twisted 1-loop invariant with a remark which suggests that the t-deformation variable is independent from the variable m which is an eigenvalue of the meridian of an SL 2 (C)-representation of π 1 (M ).
Remark 1.9.By varying the representation, the adjoint Reidemeister torsion can be extended to a rational function on the geometric component X M of the SL 2 (C)-character variety of a knot complement M [DG16], and consequently to a rational function of the geometric component of the A-polynomial curve.The same holds for the 1-loop invariant; see [DG13,Sec.4].Likewise, we can extend the twisted 1-loop invariant to a twisted rational function on the geometric component of the character variety, i.e., to an element of the ring C(X M )[t ±1 ]/(±t Z ), where C(X M ) is the field of rational functions on X M .Then, m ∈ C(X M ) is a variable independent of t.

Proofs: twisted NZ matrices
In this section we give proofs of the properties of the twisted NZ matrices, namely the relation to the NZ matrices of cyclic covers (Theorem 1.1), the symplectic properties (Theorem 1.2), and the behavior under 2-3 Pachner moves (Theorem 1.4).
Proof.(of Theorem 1.1)It is clear that T (n) has nN tetrahedra, as T has N tetrahedra ∆ j , j = 1, . . ., N .Fixing a lift ∆ j of each ∆ j to the infinite cyclic cover, we choose lifts of the tetrahedra of T (n) by µ k • ∆ j for k = 0, . . ., n − 1.Similarly, we choose lifts of the edges of T (n) .Then it is clear from the construction of the n-fold cyclic cover that for all k ∈ Z, which implies where ≡ means the equality of integers in modulo n.This completes the proof of Theorem 1.1.
Proof.(of Theorem 1.2) Following Neumann [Neu92], we consider the Z[t ±1 ]-module V generated by variables Z j , Z j , Z j subject to linear relations Z j + Z j + Z j = 0 for 1 ≤ j ≤ N .
We define a symplectic form on V as and ] where δ ij is the Kronecker delta and ¯: We associate an element E i ∈ V to each edge e i of T as In what follows, we use the same argument given in the proof of [Cho06, Thm.
Considering the linear expansion of E a , E b using the equation (31), we have a non-trivial term when the edges e a and e b appear in some ∆ j with different shape parameters.Without loss of generality, suppose that e a and e b have parameters Z j and Z j , respectively.It is clear that ∆ j contains µ α • e a and µ β • e b for some α and β ∈ Z. Equivalently, µ −α • ∆ j and µ −β • ∆ j are attached to e a and e b , respectively.We thus get a non-trivial term in the linear expansion from the triple ( ∆ j , µ α • e a , µ β • e b ).On the other hand, considering the face-pairing of the face of ∆ j containing e a and e b , there is another tetrahedron ∆ i (possibly i = j), which also contains both e a and e b .It follows that ∆ i contains µ α+γ • e a and µ β+γ • e b for some γ ∈ Z. From this triple ( ∆ i , µ α+γ • e a , µ β+γ • e b ), we get another non-trivial term −t (β+γ)−(α+γ) = −t β−α which we can pair with the term (33) to cancel out.Note that we have the minus sign here, since the face-pairing is orientation-reversing: the shape parameters of e a and e b for ∆ i are either (Z i , Z i ), (Z i , Z i ), or (Z i , Z i ).In this way, every non-trivial term that appears in the expansion of E a , E b is paired to cancel out.This proves the equation (32).We then directly obtain the desired equation ( 17), since we have Proof.(of Theorem 1.4) Recall that a 2-3 Pachner move on T is determined by two tetrahedra ∆ α and ∆ β with a common face in T .Dividing the bipyramid ∆ α ∪ ∆ β into three tetrahedra ∆ a , ∆ b , and ∆ c as in Figure 2, we obtain a new ideal triangulation T with one additional edge e 0 .We choose lifts of the five tetrahedra ∆ α , ∆ β , ∆ a , ∆ b , ∆ c and the edge e 0 (to the infinite cyclic cover) such that ∆ α ∪ ∆ β and ∆ a ∪ ∆ b ∪ ∆ c are the same bipyramid containing e 0 .Note that the (twisted) gluing equation of e 0 is log There are nine on the bipyramid whose gluing equations are affected by the Pachner move: three at the triangle of the base pyramid and six others on the bipyramid but not on its base.The six edges give and the three edges give The relations (36) are obtained from the relations (35) up to the gluing equation ( 34) of e 0 and the relation log Therefore, we can obtain the twisted gluing equation matrices G (t) for T from the twisted gluing equation matrices for T by substituting log z α and log z β in terms of the right hand side of equation ( 35) and the substitution is correct up to the gluing equation of e 0 and the relation log z + log z + log z = π √ −1.In matrix form, this implies that for some lower triangular matrix Q = 1 0 * I N and an integer matrix C, we have Here I N is the identity matrix of size N and X j is the j-th column of a matrix X. Multiplying on both sides of Equation (37), we obtain It that where P is a lower triangular matrix with 1's on the diagonal.

Examples
In this section we explain how to compute the twisted NZ matrices of an ideal triangulation using the methods of SnapPy.As is customary, we do so by the example of the simplest hyperbolic 4 1 knot, and later by the 6 3 knot.
3.1.The knot 4 1 .The knot complement M of the knot 4 1 decomposes into two ideal tetrahedra ∆ 1 and ∆ 2 .Using the default triangulation of SnapPy for 4 1 with isometry signature cPcbbbiht BaCB, the gluing equations are and the NZ matrices are We can read the above information by looking at the triangulated cusp shown in Figure 3.To describe the ideal triangulation T of the infinite cyclic cover M of M , we consider the triangulation of the peripheral torus induced from T = {∆ 1 , ∆ 2 } (see Figure 3) together with its lift to the universal cover, the Euclidean plane E 2 .Note that each triangle of E 2 with corner z j has a label j ∈ {1, 2} of the corresponding tetrahedron.
We now describe second labeling of each triangle of E 2 by an integer, obtained by choosing a cocycle representative of the abelianization map α : π 1 (M ) Z.This construction is standard and is appears in unpublished work of Goerner and also in Zickert [Zic16,Sec.3.5].
Let G denote the dual dual 1-skeleton of the triangulation of E 2 (see Figure 4).Each vertex and edge of G correspond to a tetrahedron and face-pairing of T , respectively, and that the edges of G are oriented from ∆ 0 to ∆ 1 .Since the face-pairings generate the fundamental group π 1 (M ), one can assign an integer α(p i ) to each face-pairing p i corresponding to the abelianization map α : π 1 (M ) Z. Since in our case there are two tetrahedra, we have four face-pairings, thus four integers α(p 1 ), . . ., α(p 4 ) that are required to satisfy Note that these equations are directly obtained from Figure 4 with the µ-action described in Figure 3.
Figure 5.The dual 1-complex of E 2 , where the labeling α(v) j of a vertex v indicates that v is labeled by the j-th tetrahedron.
Since E 2 is a contractible space, we can assign an integer α(v) Explicitly, the general solution of the linear equations ( 39) is given by It follows from the construction that if a vertex v corresponds to ∆ j in T , then it corresponds to µ α(v) • ∆ j in T .Choosing (α(p 1 ), α(p 2 ), α(p 3 ), α(p 4 )) = (0, 0, −1, −1) and α(v) and e i as in Figure 4, we obtain the table where the (i, k)-entry is the product of shape-parameters contributed from µ k • ∆ 1 and µ k • ∆ 2 to e i (the empty entries are 1).The gluing equation around e i is the product of the entries of the corresponding row in the above table, i.e., the product over the vertices of polygons around each edge e i in Figure 4. Hence, we have giving that which specializes to the matrix AB T when t = 1.
Choosing a lift e i of e i as in Figure 7, we have the table where the (i, k)-entry is the product of shape-parameters contributed from µ k • ∆ 1 , . . ., µ k • ∆ 6 to e i .Hence, we have e 3 e 2 e 4 e 5 e 1 e 6 0 6 giving that which specializes to matrix AB T when t = 1.

Proofs: twisted 1-loop invariant
In this section we give proofs of the properties of the twisted 1-loop invariant.
Proof.(of Theorem 1.5) We will use the notation of Theorem 1.4 which relates the NZ matrices of two triangulations T and T obtained by a 2-3 Pachner move shown in Figure 2.
Recall that a 2-3 Pachner move on T is determined by two tetrahedra ∆ α and ∆ β with a common face in T .Dividing the bipyramid ∆ α ∪ ∆ β into three tetrahedra ∆ a , ∆ b , and ∆ c as in Figure 2, we obtain a new ideal triangulation T with one additional edge e 0 .Note that one can read off relations of shape parameters between T and T directly from Figure 2: with the gluing equation z a z b z c = 1 for the edge e 0 .
We first consider the flattening part of the definition (22).A combinatorial flattening of T satisfies the equation f a + f b + f c = 2 (coming from the edge e 0 ) and determines that of T by Then a straightforward computation shows that Moreover, combining Equation ( 37) with ( 46), one checks that where the determinant of the 3 × 3 matrix in the left (resp., right) hand side is (1 Combining this fact with the equation ( 48) and the fact that det Q = 1, we conclude that τ CS (T , t) = ±τ CS (T , t).
Proof.(of Theorem 1.6) Using Equation (28) and the fact that a combinatorial flattening T (n) is simply given by n copies of that of T , it follows that where W r = k∈Z G nk+r diag(ζ ) t nk for 0 ≤ r ≤ n − 1.The block matrix in the equation ( 50) is called a factor block circulant matrix where the factor is diag(t n , . . ., t n ).It is known that a factor block circulant matrix admits a blockdiagonalization in terms of the representer where ω is a primitive n-th root of unity and V is a block Vandermonde matrix We refer to [RCdSL88] for details.In particular, we have det where, by definition, Combining the above calculations, we obtain Proof.(of Theorem 1.7) Let R(t) = G (t) diag(ζ ) and R i (t) denote its i-th row.Since the sum of all rows of G (1) = G is (2, . . ., 2) and ζ j +ζ j +ζ j = 0, the sum R 1 (1)+• • •+R N (1) is a zero vector.It follows that det R(1) = 0 and thus τ CS (T , 1) = 0.
Since the determinant is a linear map, we have For i = N we exchange the i-th row ∂ ∂t t=1 R i (t) with the N -th row R N (1) and replace R N (1) by −R 1 (1) − • • • − R N −1 (1).Then after some elementary row operations, we obtain Comparing the definitions of equation ( 20) and ( 22), it suffices to show that Let Σ be the pre-image of the peripheral torus Σ under the covering map M → M .Note that T induces a triangulation of Σ where vertices and triangles of Σ correspond to edges and tetrahedra of T , respectively.For each triangle ∆ of Σ corresponding to the tetrahedron µ k • ∆ j , we assign a corner parameter t k log z j to each corner of ∆ according to the quad type of ∆ j .Then the sum r i (t) of corner parameters around a vertex of Σ corresponding to the edge e i is It follows that (see Equation ( 6)) We choose any lift λ of λ to Σ. Since α(λ) = 0, λ is still a loop in Σ.Up to homotopy, we may assume that λ is an edge-path in Σ so that the surface S bounded by λ and µ • λ is triangulated.Here we view S as a closed cylinder whose boundary is consisted of λ and µ • λ.Let c λ (t) (resp., čλ (t)) be the sum of corner parameters in S that are adjacent (resp., not adjacent) to λ.By definition, the equation c λ (1) = 0 represents the completeness equation of λ.Hence, Also, since S consists of triangles and ζ j + ζ j + ζ j = 0 for all j, we have On the other hand, tc λ (t) is the sum of corner parameters adjacent to µ • λ and not in S. It follows that the sum of corner parameters around vertices in D \ λ is given by tc λ (t) + čλ (t).Since D \ λ is a fundamental domain of Σ, exactly two vertices in D \ λ correspond to a lift of e i for each 1 ≤ i ≤ N .It follows that for some a i and b i ∈ Z and thus Taking the partial derivative with respect to z j , we obtain where the second equation is followed from the equation (60) with the fact tc λ (t) + čλ (t) = (t − 1)c λ (t) + c λ (t) + čλ (t).Combining the equations ( 58), (59), and (64), we obtain the desired result that for some ε = ±1 and some integer r and some choice of quads of T .We make such a choice such that B is non-singular.This is always possible; see [ Note that Theorem 1.2 implies that B(t) −1 A(t) = (B(1/t) −1 A(1/t)) T which is used in the second equality above.
We end this section by giving a proof of Equation (26).Taking the derivative of Equation (25) at t = 1 and then simplifying the result by Theorem 1.7, we obtain which is equivalent to Equation (26) due to Theorem 4.1 of [Por97] (explicitly, the ratio τ CS λ (T )/τ CS µ (T ) is the cusp shape of M and τ CS λ (T (n) )/τ CS µ (T (n) ) is n times of that).

Examples, continued
In this section, which is a continuation of Section 3, we compute the twisted 1-loop invariant of an ideal triangulation using the methods of SnapPy.
5.1.The knot 4 1 .In Section 3.1 we already gave the twisted NZ matrices of the 4 1 knot.Using the same notation, we now give the remaining data, namely the shapes of the complete hyperbolic structure and flattenings, which are needed to compute the twisted 1-loop invariant.
up to multiplication by a monomial ±t r for some integer r.

Figure 6 .
Figure 6.The triangulation of the peripheral torus of 6 3 .