Foundations of Zeta Space: Geometry, Spectral States, and Fourier Mellin Duality

Each face center has four incident arms. Each of the twelve cube edges contributes one common midpoint, of degree two. Removing the subdivision leaves one vertex for each face and one edge for each pair of adjacent faces. A face is adjacent to every face except itself and its opposite. The three pairs of opposite faces are therefore the three parts of K_{2,2,2}. The graph is connected, so its first Betti number is 12-6+1=7. Every construction uses face adjacency and midpoints, both preserved by cube symmetries. Finally, those symmetries are the signed permutations of three centered coordinate axes, a group of order 2^3\cdot3!=48.

Each face cross cuts its face into four quadrants. The three quadrants incident to a given cube vertex meet across the original cube edges, forming one region around that vertex. Midpoints separate this region from the analogous regions at neighboring vertices. Hence the 24 quadrants form 8 regions of 3 quadrants each. If the faces F_1,F_2,F_3 meet at v, the boundary becomes the cycle through c_{F_1},c_{F_2},c_{F_3} in the reduced graph. A cube edge has two endpoints, giving the two triangles containing its dual seam edge. The antipodal correspondence follows from the cube’s central inversion.

Every triangle uses three distinct edges. The graph is simple and has no loops or double edges, so no cycle uses fewer than three edges.

The form norm is the direct-sum H^1 norm. Vertex evaluation is continuous on each H^1 interval, so the continuity constraints define a closed subspace. Functions compactly supported in edge interiors are dense in the direct-sum L^2 space, proving density. The representation theorem for closed nonnegative forms gives a unique nonnegative self-adjoint operator. Testing against interior functions gives -f_e'' and edgewise H^2 regularity. Integration by parts leaves at each vertex a boundary coefficient proportional to the sum in (3.1); the common test value at that vertex is arbitrary, so that coefficient vanishes. Conversely, those conditions remove every boundary term. Finally, a finite direct sum of H^1 intervals embeds compactly into the direct-sum L^2 space. The form-domain embedding, and hence the resolvent, is compact.

Signed permutations act transitively on oriented edges. There is also a symmetry that reverses a chosen edge: for the edge from e_1 to e_2, exchange the first two coordinates. Therefore an invariant function is represented on every edge by the same profile F, independent of orientation, and F(x)=F(\ell-x). Conversely, any such profile gives an invariant edgewise function.

At the form level, continuity holds automatically because F(0)=F(\ell). Restriction to [0,\ell/2] determines F by reflection, and

Thus f\mapsto\sqrt{24}\,F|_{[0,\ell/2]} is a unitary equivalence of forms with the unrestricted H^1 energy on that interval. Its associated operator is the Neumann Laplacian. Its eigenfunctions are \cos(\pi mx/(\ell/2)), giving (3.2). Equivalently, the Kirchhoff condition gives 4F'(0)=0, while reflection gives F'(\ell/2)=0.

The normalized eigenvalues are \pi m^2. The heat trace is their convergent exponential sum. Doubling it and subtracting one restores the two signs of each nonzero integer while counting the constant mode once.

Each edge carries F(x)=\cos(2\pi mx/\ell), and reversal does not alter it. Concatenating the three edge profiles therefore gives the same \ell-periodic function three times. Its circular frequency 2\pi m/\ell equals 2\pi(3m)/(3\ell).

At t=1, differentiate the Fourier integral in \xi and integrate by parts in x. The result satisfies F'(\xi)=-2\pi\xi F(\xi), while F(0)=\int e^{-\pi x^2}\,d x=1, as follows from the two-dimensional Gaussian integral in polar coordinates. Thus F(\xi)=e^{-\pi\xi^2}; scaling proves the formula for t>0. For Poisson summation, periodize f as \sum_m f(x+m). Schwartz decay permits termwise differentiation and integration; its Fourier coefficient at n is \widehat f(n). The Fourier series converges absolutely, and evaluation at x=0 gives the identity.

Apply Lemma 4.1 to e^{-\pi t x^2} and use Corollary 3.5.

For real s>1, Tonelli’s theorem gives

The same calculation bounds the absolute value for complex s by the expression at \Re s, proving absolute convergence and the identity on \Re s>1.

Using Z_\Gamma-1=(\theta_\Gamma-1)/2, split the integral at 1. On (0,1) insert (4.2) and substitute t\mapsto1/t in the decaying part. This yields

The integral is entire: on every compact s-set its integrand and all s-derivatives are bounded by a polynomial in t and \log t times Ce^{-\pi t}. The rational part supplies exactly the stated poles and residues. Both parts of (?) are invariant under s\mapsto1-s. Multiplication by s(s-1)/2 cancels the two poles and preserves the reflection.

The action on vertices is transitive. Invariance therefore makes all components equal. Averaging over the group gives each original component the same weight 1/8 in every output component.

The determinant in (u,v) coordinates is one. Let r be meridian arclength, so dr=du/\sqrt\phi, and write f(r)=\sqrt{\phi(u(r))}. Then g=dr^2+f(r)^2dv^2, with f'=\phi'/2 and f''=\phi''f/2. The curvature is -f''/f=-\phi''/2=b. Near u=0, set u=ar^2/4 at leading order. The metric becomes dr^2+(a^2/4)r^2dv^2 to leading order, giving cone angle aT/2. The other endpoint has slope \phi'(a/b)=-a and the same angle. Integrating the area form gives T(a/b).

For a complete finite-area hyperbolic surface, Gauss–Bonnet on cusp truncations gives -\operatorname{area}=2\pi\chi. The horocyclic boundary contributions tend to zero as the cusps are truncated farther out. A cylinder has Euler characteristic zero, contradicting positive area. For the second statement, write a cylinder as \mathbb R\times S^1 with metric dr^2+f(r)^2dx^2, where f>0 is smooth and equals a positive multiple of e^{-|r|} outside a compact interval. The ends are cusp metrics after shifting r, and the intervening curvature is -f''/f.

The norm \|u\|_{H_3}^2+q_3(u,u) is equivalent to the finite direct-sum H^1 norm, since all weights are fixed and positive. The trace identifications and external homogeneous conditions are kernels of bounded trace maps, so V_3 is closed in that norm. Interior compactly supported functions are dense in H_3, giving density of the form. The representation theorem gives a unique nonnegative self-adjoint operator.

Interior test functions identify its differential expression. For a sufficiently regular domain element, integration by parts against v\in V_3 produces the common edge test trace multiplied by the weighted sum of conormal derivatives. This sum must vanish in the dual trace space. Conversely, the stated differential expression and vanishing boundary terms satisfy the defining weak identity of the form operator. The weak domain supplied by the form also fixes behavior at seam vertices.

Finally, the group action preserves both the Hilbert inner product and the form. If u=(A_3+1)^{-1}f, its weak defining equation transforms to the same equation with right side U_gf. Uniqueness gives U_g(A_3+1)^{-1}=(A_3+1)^{-1}U_g, and the spectral functional calculus yields the remaining commutations.

The Reynolds projection commutes with the functional calculus by Theorem 6.1. Unitarity preserves the norm, and commutation with A_3^{1/2} preserves the energy on the form domain.

Otherwise choose u_j with L^2 norm one and energy tending to zero. The sequence is bounded in the finite piecewise H^1 space. Rellich compactness gives a subsequence converging strongly in L^2 and weakly in H^1 to u. Its gradient vanishes on every piece. The trace constraints pass to the weak limit, so these constants agree across each connected component. The external zero trace makes each constant zero, contradicting the unit L^2 norm.

Choose a lift w\in V(K) of b. Seek u_b=w+z with z\in V_0(K). The equation for z is q_K(z,v)=-q_K(w,v). Lemma 6.3 makes q_K coercive on V_0(K), so Lax–Milgram gives a unique z and an energy bound. Replacing w by another lift changes it by an element of V_0(K) and produces the same u_b. Taking the infimum over lifts gives continuity in the quotient trace norm. Orthogonality shows q_K(u_b+v,u_b+v)=q_K(u_b,u_b)+q_K(v,v), proving minimality and uniqueness. Interior and seam equations follow by testing as in Theorem 6.1. Equivariance follows by transforming the weak equation and using uniqueness.

Apply Green’s identity on each piece. The interior terms vanish by the local equation, and the identified test traces allow the conormal terms to be collected edge by edge. Their sum is (6.6). Vanishing is sufficient by the weak definition of A_3 and necessary by arbitrary local edge tests, interpreted in the compatible trace space.

The excluded Dirichlet spectra guarantee existence and uniqueness of each piecewise boundary solution by the resolvent of its Dirichlet realization. A global eigenfunction has a common trace and zero seam defect, hence lies over the kernel. Conversely, a kernel trace produces a unique family of solutions, and Proposition 6.6 glues them to a global state. A zero trace gives only the zero piecewise solution, so the correspondence is injective. Symmetry follows by applying the second Green identity to two boundary solutions at the same real \lambda and summing over pieces.

A quarter-turn may be written (u,w)\mapsto(1-w,u). It exchanges the two central lines and sends (u-1/2)(w-1/2) to its negative. Pullback transports a tensor intrinsically; retaining the orbit of the decorated chart records this action without declaring its coordinate formula fixed.

Substitute t=\sqrt{\sigma/a}\,e^u. The exponent becomes -2\sqrt{a\sigma}\cosh u, and the power and measure give (\sigma/a)^{\nu/2}e^{\nu u}\,d u. Split the integral over \mathbb R at zero and combine the two halves into 2\cosh(\nu u). This is the integral representation K_\nu(z)=\int_0^\infty e^{-z\cosh u}\cosh(\nu u)\,d u for z>0. Both endpoints are exponentially controlled, justifying the substitution for every fixed \nu; see also [DLMFK].

Conjugating the sum by the actions of an element h merely reindexes it by g\mapsto hg. Measure preservation makes the integral defining the constant mode invariant.

Write \Psi=\sum_na_n(y)e^{2\pi inx}. For n\ne0 the radial equation is

Set a_n(y)=\sqrt y\,b(2\pi|n|y). Direct differentiation gives z^2b''+zb'-(z^2+(ir)^2)b=0. The decaying solution of this modified Bessel equation is K_{ir}(z); the independent solution has exponential growth. For n=0, substituting a_0=y^\eta gives \eta(1-\eta)=1/4+r^2. The repeated-root case yields the logarithmic second solution.

For fixed r, the asymptotic K_{ir}(z)=\sqrt{\pi/(2z)}e^{-z}(1+O(z^{-1})) gives a positive nonzero value for all sufficiently large positive z. This permits a common Y. The ratio in (8.2) has exponential factor e^{-2\pi|n|(y-Y)}; its differentiated forms have at most additional polynomial factors in |n| on compact y-ranges. Smoothness of b gives rapid Fourier decay, and the exponential factor gives normal convergence away from the initial collar. At y=Y the ratio is one. Modewise uniqueness of the decaying solution proves uniqueness among zero-mode-free decaying solutions with that trace.

Let B_\xi(X)=\log((1-|X|^2)/|X-\xi|^2). In upper half-space coordinates with \xi=\infty, this is \log y up to an additive constant. The hyperbolic identities are |\nabla B_\xi|^2=1 and \Delta B_\xi=-2. Therefore -\Delta e^{\beta B_\xi}=\beta(2-\beta)e^{\beta B_\xi}. Isometries give the same identity for every ideal point. On a compact subset of the ball the kernel and all its derivatives are uniformly bounded in \xi, so differentiation under the L^1 integral is justified. Orthogonal transformations preserve its distance formula and the spherical measure, proving equivariance.

The normal-coordinate metric is dr^2+\cosh^2r\,g_N, so \Delta_{\mathbb H^3}=\partial_r^2+2\tanh r\,\partial_r+\cosh^{-2}r\,\Delta_N. Evaluate at r=0 and use the positive-Laplacian convention.

Use the integral representation K_\nu(z)=\tfrac12(z/2)^\nu\int_0^\infty a^{-\nu-1}e^{-a-z^2/(4a)}\,d a, which follows by substitution from (7.2). Integrating first in z gives

The remaining a integral is \Gamma((s-\nu)/2). The condition on \Re s makes both absolute integrals finite, justifying Fubini. Finally substitute z=cy. This is the standard identity recorded in [DLMFMellin].

For even parity, set I_0(s)=\int_0^\infty u(iy)y^{s-1/2}\,d y/y. Parity gives A(-n)=A(n), and Lemma 10.1 yields I_0(s)=\tfrac12\Lambda_u(s) initially in a right half-plane. Invariance under inversion gives u(i/y)=u(iy). Substituting y\mapsto1/y therefore gives I_0(s)=I_0(1-s).

For odd parity let v(y)=\partial_xu(x+iy)|_{x=0}. Then v(y)=4\pi i\sqrt y\sum_{n\ge1}nA(n)K_{ir}(2\pi ny). Define I_1(s)=\int_0^\infty v(y)y^{s+1/2}\,d y/y. The Bessel integral gives I_1(s)=i\pi\Lambda_u(s). Differentiating u(z)=u(-1/z) at z=iy gives v(y)=-y^{-2}v(1/y), so substitution yields I_1(s)=-I_1(1-s).

At infinity, cusp decay controls both u and its derivative exponentially. The inversion relations transfer this to decay faster than any power at zero. The two period integrals and all their s-derivatives therefore converge locally uniformly for every s\in\mathbb C, proving entire continuation and the stated functional equations.

Write L=B\mathbb Z^k with |\det B|=v. Periodizing a Schwartz function on \mathbb R^k/L as in Lemma 4.1 gives \sum_{\ell\in L}f(\ell)=v^{-1}\sum_{\ell^*\in L^*}\widehat f(\ell^*). The k-dimensional Gaussian transform is the product of the one-dimensional transforms, equal to t^{-k/2}e^{-\pi|\xi|^2/t}. This proves the first formula. Since L^* has covolume v^{-1}, multiplying by v^{1/2} proves the second.

At zero, reciprocity gives \theta_\pm(t)=t^{-k/2}(a_\mp+O(e^{-c/t})), so the initial integral converges absolutely for \Re s>k. Split it at 1 and use reciprocity in the small-t part. The terms involving the constants integrate to a_\mp/(s-k)-a_\pm/s, while the other term changes to the second integral over [1,\infty). Exponential decay makes that integral entire by locally uniform domination, including all derivatives in s. Replacing s by k-s and exchanging the signs leaves the expression unchanged. The pole statements and entire regularization follow directly.

Substitute s=kw into Theorem 11.4. A linear change of variable divides each residue by k.

The reciprocity equation is linear in the two kernels. Formula (11.4), including its constant terms, is also linear. A finite average can be interchanged with its convergent integrals and rational terms.

Evaluating \widehat f(\xi',\eta) at \eta=0 gives the Fourier transform in x' of \int f(x',u)\,d u, by Fubini. For the second identity, apply one-dimensional Fourier inversion in u at u=0, then Fourier transform in x'. Schwartz decay justifies all operations.

The Gaussian satisfies g(0)=\int_\mathbb R g=1 and \widehat g=g. Tensor factorization of the Fourier transform proves all three identities.

Absolute convergence in the product half-plane permits Fubini, so the double integral is the product of the two integrals in Theorem 4.3. Their meromorphic continuations give (12.3) globally. The residues of \Lambda_{\zeta} at 1 and 0 are 1 and -1, giving (12.3).

For the last identity, expand F(1-s_1,z) near z=1 as a_{-1}(1-s_1)/(z-1) plus the remaining Laurent terms. Substituting z=1-s_2 changes this term to -a_{-1}(1-s_1)/s_2. Applying -\operatorname{Res}_{s_2=0} yields a_{-1}(1-s_1), which is R_1U_+F. The argument applies coefficientwise wherever the relevant meromorphic Laurent germs are defined, and hence by continuation in the remaining variable.

Apply the rank-one reflection to each factor and repeat Theorem 12.3 in the final coordinate.

Define \overline f([y])=[f(y)]. The implication makes it well-defined, and surjectivity of q_1 gives uniqueness. The quotient topology makes it continuous because its composite with q_1 is continuous. Equivariance follows by evaluating the identity on representatives.

An axis-preserving signed permutation has a signed n\times n permutation block and an independent sign in the last coordinate. Projection onto the first block is a homomorphism. In contrast, a permutation exchanging the first and last axes has, after deleting the last row and column, a zero first row and column; the remaining matrix is not a signed permutation matrix.

A compatible sequence x=(x_n) maps to (d_nx_n), which is compatible by the bonding identity. Applying the inverse maps gives the inverse sequence. Both maps are continuous because their coordinates are continuous in the product topologies. Equation (?) makes \widehat d identify the positive seed embedding with the negative one, so it descends through the pushout. Define D by \widehat d on the positive side and \widehat d^{-1} on the negative side. Then D^2=\mathrm{id} and every identified seed point is fixed.

For the topological statement, an inverse limit is the closed subset of the product cut out by p_nx_{n+1}=x_n. Compactness follows from Tychonoff. Surjectivity gives the finite intersection property for these constraints, hence nonemptiness. The seed lifts are embeddings because their level-three coordinate is an embedding. Gluing two compact Hausdorff spaces along homeomorphic closed subspaces gives a compact Hausdorff quotient: the equivalence relation is the diagonal together with the two closed graphs of the identification, so it is closed. This proves the assertions.

For the first assertion,

For the second, the first identity of (?) gives T_n^-J_{n+1}\theta_{n+1}=J_n\theta_n. The middle identity gives its Mellin state as R_n\mathcal M_n^+\theta_n, and the last identity makes these reflected states compatible with the negative transitions. The inverse maps prove involutivity.

For the third, the supplied identifications make the maps equal on representatives that are to be identified over the seed. They descend by the quotient universal property, as in Lemma 13.2. At a fixed geometric point and self-dual fiber state, (?) becomes its scalar functional equation. Rank one gives the normalization s\mapsto1-s.

Identify (x,z)_j with (x,g_{ij}(x)z)_i. The cocycle makes these identifications consistent. The overlap identity makes the local graphs of F_i agree, giving the section and its uniqueness. Applying duality in the local trivializations gives the global identity by the same overlap condition.

A nonvanishing holomorphic factor does not change local order. Pullback by a biholomorphism transports each local order to the corresponding point. Apply these facts to the assumed identity.

Hilbert–Schmidt summability gives \sum_j\kappa_j^2<\infty. On compact z-sets, all sufficiently late factors have |z\kappa_j|\le1/2, and \log((1-z\kappa_j)e^{z\kappa_j})=O(|z|^2\kappa_j^2). The tail product converges normally to a nonvanishing analytic function; finite initial factors account for its zeros, exactly the real numbers 1/\kappa_j. The zero-free E does not alter zeros in (15.3). A zero s_0 of \xi gives a real zero z_0=(s_0-1/2)/i, so \Re s_0=1/2.

A spanning tree has five edges, leaving seven edges outside it. Contracting the tree gives a wedge of seven circles, so the fundamental group is free on seven generators and its abelianization is \mathbb Z^7. Rank-one unitary holonomy assigns one unit complex number to each generator. Equivariance includes actual fiber maps covering the group action and obeying composition; invariance of an isomorphism class alone does not specify those maps.

Two affine points determine a unique nonzero difference and the corresponding line. An affine point and a direction determine one affine line, and two direction points determine the line at infinity. This gives the projective plane of order two.

Affine transformations a\mapsto Ma+b preserve the incidence and act on directions by d\mapsto Md. There are 4\cdot6=24 such transformations, acting faithfully on the four affine points, hence as all of S_4. Conversely, any incidence automorphism preserving the line at infinity acts faithfully on the affine points and belongs to this same group.

The rotational cube group has order 24 and acts faithfully on its four body diagonals. Indeed, a linear orthogonal map preserving all four diagonal lines acts by signs on representatives; the linear relation among any four such representatives has all coefficients nonzero, forcing the signs to agree. The map is therefore I or -I, and only I preserves orientation. This gives S_4, with kernel \{\pm I\} for the full cube group. For the remaining labels, represent a diagonal by a sign triple modulo overall sign. For axis i, the sign of the product of the other two coordinates partitions the four diagonals into two pairs. Signed permutations transport these three partitions exactly as they transport axis classes. These are the three parallel classes of the affine plane.

Multiplicativity implies the identities on basis products. Conversely, they, the unit relation, and e_i^2=-1 preserve every basis product. Bilinearity gives preservation of every product, and signed permutation maps are invertible. The group law gives the claimed action.