# Silver-Ratio Pell Lattices: Algebra, Hyperbolic Geometry, Nonreciprocal Transport, Exceptional Points, and Discrete Symmetry
*Results marked [hand] were derived by hand. Numerical confirmation is pending (see §11).*
## 1. The Pell transfer matrix
T = [[2,1],[1,0]] has characteristic equation λ² − 2λ − 1 = 0 and eigenvalues λ± = 1 ± √2. The dominant eigenvalue δ = 1 + √2 is the silver ratio.
- Tr T = 2 and det T = −1, so T ∈ GL(2,ℤ). T² = [[5,2],[2,1]] has trace 6 and determinant 1, so it is a hyperbolic element of SL(2,ℤ).
- Tᴺ = [[P(N+1), P(N)], [P(N), P(N−1)]], with P₀ = 0, P₁ = 1, P(n) = 2P(n−1) + P(n−2).
- Tr Tᴺ = Q(N), the Pell–Lucas numbers (2, 2, 6, 14, …).
- Invariant:
```latex
Q_N^2 - 8P_N^2 = 4(-1)^N
```
- In the ring ℤ[√2], δᴺ = (Q_N + 2√2·P_N)/2, so the invariant is the norm N(δᴺ) = (−1)ᴺ. Ratios P(N+1)/P(N) are the continued-fraction convergents of δ = [2; 2, 2, …].
- The Lyapunov exponent is ln δ per step.
- The partition function of the transfer-matrix chain is Z(N) = Tr Tᴺ = Q(N) ~ δᴺ, giving a free energy per site set by ln δ.
## 2. Hyperbolic geometry and geodesic flow
Orientation-preserving elements of GL(2,ℝ) with positive determinant act on the upper half-plane ℍ = {z : Im z > 0} by Möbius maps. Since det T = −1, T acts as the orientation-reversing glide z ↦ 2 + 1/z̄, and T² acts by the Möbius map z ↦ (5z + 2)/(2z + 1).
- **Fixed points.** T² has fixed points 1 ± √2, the quadratic irrationals in ℚ(√2). They are the endpoints of a geodesic in ℍ: a semicircle centered at 1 with radius √2.
- **Geodesic flow.** Iterates Tⁿ move a base point along (and asymptotically onto) this geodesic. Per step of T the hyperbolic distance advances by 2 ln δ ≈ 1.763, twice the Lyapunov exponent. For T² the translation length is 4 ln δ ≈ 3.526.
- **Fuchsian group.** The cyclic group ⟨T²⟩ is a discrete subgroup of SL(2,ℝ) acting properly discontinuously on ℍ. The quotient ℍ/⟨T²⟩ is a hyperbolic cylinder with a single closed geodesic of length 4 ln δ = 2 arccosh 3.
- **Modular surface.** Because T² ∈ SL(2,ℤ) has trace 6, it corresponds to a closed geodesic on the modular surface ℍ/PSL(2,ℤ). The periodic continued fraction [2; 2̄] is the standard encoding of this geodesic (Lagrange's theorem).
- **Pell identities.** The integer matrix entries P(N), Q(N) give exact coordinates for the orbit, and Q² − 8P² = 4(−1)ᴺ is the norm relation in ℤ[√2].
## 3. Nonreciprocal chain
Take on-site energy ε = 0, hoppings t_R = 1 and t_L = −1, and energy E = 2. The stationary equation becomes ψ(n+1) = 2ψ(n) + ψ(n−1), whose single-step transfer matrix is T.
- The periodic-boundary spectrum is E(k) = 2i sin k, purely imaginary. Driving at the real energy E = 2 lies in a gap, so the response is evanescent decay with localization length 1/ln δ ≈ 1.135 sites, not a conventional skin effect.
- For N sites the total transfer matrix is Tᴺ.
- Attach leads with hopping J = 1 and on-site energy 1, so E = 2 sits inside the lead band at k = π/3. With the nonreciprocal bond at each interface and A = e^{ik} [hand]:
```latex
\Delta = P_{N-1}A^2 + 2J P_N A + J^2 P_{N+1}
t = \frac{2iJ(-1)^N \sin k\, A}{\Delta},\qquad
r = -\frac{A^2\,[\,2J P_N\cos k + P_{N-1} + J^2 P_{N+1}\,]}{\Delta}
```
- Hand checks at J = 1, k = π/3: |t|² = 1/4, 3/61, 3/364 for N = 1, 2, 3. Successive ratios are about 0.168 against δ⁻² = 0.172, so |t|² ∝ δ^(−2N).
- Flux is not conserved (|r|² + |t|² > 1 at N = 2), as expected for nonreciprocal transport.
- Platforms: topolectrical LC ladders with op-amp nonreciprocal links (voltage attenuation in powers of δ), and coupled ring resonators or directional-coupler acoustic lattices (fractional transmission at the locked energy).
## 4. Exceptional points
**Uniform chain.** With staggered imaginary gain/loss iγ(−1)ⁿ the Hamiltonian is anti-Hermitian and diagonalizable, so there are no exceptional points. With real ±γ the exceptional points sit at γ = 2|cos(jπ/(N+1))|, which scales like 1/N.
**Pell-barrier PT dimer [hand].** Two defects with on-site energies 2√2 ± iγ are separated by N barrier sites (t_R = 1, t_L = −1). Each defect alone binds a state at E = 2 that decays as δ⁻ⁿ. Eliminating the barrier, the eigenvalue condition evaluated at E = 2 is exactly:
```latex
F(2,\gamma) = (-1)^N\,2\sqrt{2}\,\delta^{-(N+1)} + \gamma^2 P_{N+1}
```
- **N odd:** F = 0 at γ_c = √(1 − δ^−2(N+1)) / P(N+1) ≈ 1/P(N+1). For N = 1, γ_c = 0.4926 against 1/P₂ = 0.5. For N = 3, γ_c = 0.08331 against 1/P₄ = 0.08333. The bound pair E = 2 ± w is real below γ_c, and the sign change of F(2, γ) is consistent with the pair merging there.
- **N even:** F(2, γ) > 0 for all real γ, so there is no real-γ threshold.
- Operating below γ_c (N odd) keeps the bound pair real and avoids lasing instability.
- Context: for PT impurities in a hopping chain, the critical strength equals the hopping amplitude between the impurities (Joglekar et al., PRA 84, 024103). Increasing the gain-loss separation lowers the threshold exponentially. The Pell formula is the exact closed form for an anti-reciprocal barrier.
## 5. Hilbert-space formulation
Define |ψₙ⟩ = (Pₙ, Pₙ₋₁)ᵀ, so |ψₙ₊₁⟩ = T|ψₙ⟩. Since T is real symmetric,
```latex
\frac{\langle\psi_n|T|\psi_n\rangle}{\langle\psi_n|\psi_n\rangle} \to \delta \quad (n\to\infty)
```
This is the Rayleigh quotient, equivalent to power iteration. T is non-unitary, so the norm grows as δⁿ and the dynamics describe an amplifying system. The spectral decomposition |ψₙ⟩ = c₁λ₁ⁿ|e₁⟩ + c₂λ₂ⁿ|e₂⟩ is the matrix form of Binet's formula.
The method applies to any linear recurrence. A small library could take the coefficients, build the companion matrix, and return eigenvalues and closed forms: (2, 1) gives the silver ratio ≈ 2.414, and (1, 1) gives the golden ratio ≈ 1.618. Tribonacci and higher-order recurrences extend the same way, and symbolic computation and visualization are natural additions.
## 6. Tensor-network benchmark
The Pell matrix is a convenient exactly solvable test case for matrix-product-state and transfer-operator methods.
- Used as a bond-dimension-2 transfer operator, T has eigenvalues δ and −1/δ, so |λ₂/λ₁| = δ⁻². The correlation length is ξ = 1/(2 ln δ) ≈ 0.567 sites, with sign alternation from the negative subleading eigenvalue.
- Because T is non-unitary and non-Hermitian in the chain setting, it tests algorithms that must handle left and right eigenvectors, biorthogonal normalization, and non-positive transfer operators.
- The closed forms for Tᴺ (§1) give exact reference values for truncation and extrapolation studies.
## 7. Hyperbolic continuation of SU(2) rotations
A rotation continues under θ → −iη to a boost:
```latex
e^{i\theta\,\mathbf{n}\cdot\sigma}=\cos\theta+i\sin\theta\,(\mathbf{n}\cdot\sigma)\;\to\;e^{\eta\,\mathbf{n}\cdot\sigma}=\cosh\eta+\sinh\eta\,(\mathbf{n}\cdot\sigma)
```
- At η = ln δ, cosh η = √2 and sinh η = 1. For n = x̂ this gives B = [[√2, 1],[1, √2]], with det B = 1 and eigenvalues δ^(±1).
- B² has trace 6, so T² is conjugate to B². T itself differs by a parity sign, reflecting det T = −1 and the (−1)ᴺ in the invariant.
- Pairing a clock step θ = π/12 with the rapidity η = ln δ is a choice of convention, with rescaling factor 12 ln δ/π.
- A similarity transformation alone cannot connect a unitary rotation to a hyperbolic step, because it preserves eigenvalues. The analytic continuation is what does it.
## 8. Discrete symmetry: E₈, D₄, the 24-cell, and the number 24
**E₈ → D₄ × D₄.** E₈ contains SO(16), which contains SO(8) × SO(8). The 240 roots of E₈ partition exactly as:
```latex
240 = 24 + 24 + 64 + 64 + 64
```
- The first two 24s are the roots of the two D₄ factors. The three 64s are the products 8_v⊗8_v, 8_s⊗8_s and 8_c⊗8_c, which are permuted by the triality of D₄.
- At the level of the Lie algebra, 248 = (28,1) + (1,28) + (8_v,8_v) + (8_s,8_s) + (8_c,8_c).
- The 24 roots of D₄ are the vertices of the 24-cell, the self-dual regular polytope in ℝ⁴ with 24 octahedral cells.
**Hurwitz quaternions.** The 24 unit Hurwitz quaternions form the binary tetrahedral group in SU(2) and coincide with the vertices of the 24-cell. Their rotation angles are multiples of π/3 and π/2.
**The number 24.**
- The Pisano period of 9 is 24, and the digital roots of Fibonacci numbers repeat with period 24. Pell numbers mod 9 also repeat with period 24 (P₁₂ ≡ 0 and P₁₃ ≡ −1 mod 9).
- The rotation exp(iπ/12 n·σ) has order 24 and generates the cyclic group ℤ₂₄. Since π/12 is not a Hurwitz angle, this is a different group from the binary tetrahedral group. The match between 24 and 24 is a coincidence of order, and any deeper link is an open question.
## 9. Modular checksum and comparison with other invariant-based methods
The Lucas–Fibonacci identity reduced mod 9 gives a 24-periodic residue sequence:
```latex
L_n^2 - 5F_n^2 = 4(-1)^n
```
Tagging computational states with their position in this cycle gives a lightweight arithmetic invariant check.
- A random error is caught with probability about 8/9.
- It is a consistency check, not an error-correcting code like Reed–Solomon, Hamming, Turbo or LDPC codes, which add redundancy to recover lost or corrupted data. It is complementary to them.
- It belongs to the family of residue codes (arithmetic checks) and cyclic invariant checks.
**Comparison with elliptic-curve cryptography.**
- Both bind valid states to an algebraic curve: ECC uses y² = x³ + ax + b over a finite field, and the checksum uses the hyperbola L² − 5F² = ±4.
- Solutions of a Pell-type conic x² − dy² = 1 form a group under multiplication in ℤ[√d], analogous to the elliptic-curve group law. Lucas-sequence cryptosystems (for example LUC) exploit this structure.
- The difference is purpose and security. ECC security rests on the hardness of the discrete logarithm, while the mod-9 checksum provides no secrecy or authentication, and valid residues are easy to forge. It detects accidental drift only.
**Pell-matrix coding.** Encode (x, y) as M = [[x, dy],[y, x]], with determinant x² − dy². For solutions of the specific Pell equation the determinant is exactly 1, and a corrupted pair shifts it. The mod-9 sequence uses the evaluation of the identity as a real-time integrity check in the same spirit.
## 10. Geometric notes
**Klein bottle in ℝ⁴.** With a > b > 0:
```latex
x=(a+b\cos v)\cos u,\quad y=(a+b\cos v)\sin u,\quad z=b\sin v\cos\tfrac u2,\quad t=b\sin v\sin\tfrac u2
```
Under u → u + 2π, (z, t) → −(z, t), which is equivalent to v → −v, the Klein identification. Dropping t gives the familiar self-intersecting 3D immersion, while the 4D surface has no self-intersection. The crossing is an artifact of projecting a non-orientable surface into three dimensions.
**Cut-and-project.** Let Λ ⊂ ℝ^(d∥+d⊥) be a lattice, decompose the space as E∥ ⊕ E⊥ with orthogonal projections π∥ and π⊥, and let W ⊂ E⊥ be a compact window.
```
for each γ in Λ:
if π⊥(γ) in W: # cut
keep π∥(γ) # project
```
The result is a quasicrystal point set (for example Penrose tilings from ℤ⁵). Window-selected sets are uniformly discrete and non-overlapping, so unlike the Klein bottle crossing, the lower-dimensional output has no overlaps. What it shares with the Klein case is that the higher-dimensional lattice is simple while the lower-dimensional slice looks complicated.
## 11. Verification and open items
Verification script for the dimer threshold (not yet run):
```python
import numpy as np
d = 1 + np.sqrt(2)
def pell(n):
a, b = 0, 1
for _ in range(n):
a, b = b, 2*b + a
return a
def H(N, g, B=14):
L = 2*B + N + 2
h = np.zeros((L, L), complex)
for n in range(L - 1):
h[n, n+1] = 1
h[n+1, n] = -1
h[B, B] = 2*np.sqrt(2) + 1j*g
h[B+N+1, B+N+1] = 2*np.sqrt(2) - 1j*g
return h
def near(N, g):
e = np.linalg.eigvals(H(N, g))
return e[abs(e - 2) < 0.5]
def threshold(N):
lo, hi = 0.0, 0.6
for _ in range(50):
mid = (lo + hi) / 2
if np.max(abs(near(N, mid).imag), initial=0) > 1e-5:
hi = mid
else:
lo = mid
return lo
for N in (1, 3, 5, 7):
P = pell(N + 1)
print(N, threshold(N), np.sqrt(1 - d**(-2*(N+1))) / P, 1 / P)
```
Open items:
1. Confirm the dimer threshold γ_c numerically for N = 1, 3, 5, 7.
2. Verify the closed-form t and r against a direct linear solve for several N and k.
3. Check the double-root condition ∂F/∂E = 0 at γ_c.
4. State and test precisely which perturbations of the chain preserve the Pell structure of the transfer matrix.
5. Determine whether a unitary-clock-to-lossy-bus interface built on the §7 continuation preserves fidelity. This needs an explicit model with a decoherence mechanism.
6. Establish whether any structural link exists between the 24-step modular cycle and the Hurwitz/24-cell group beyond equality of order.
7. Test the §6 correlation-length benchmark in an MPS code.
## 12. Related literature
- **Metallic-mean chains.** The silver mean arises from the a = 2 inflation rule (octonacci) and is analyzed with transfer matrices. The n = 2 chains relate to Pell numbers and can be built by cut-and-project. Those chains are aperiodic sequences, whereas the chain here is uniform.
- **Nonreciprocal transfer matrices.** An exact mapping of the Hatano-Nelson model to continuous nonreciprocal systems has been derived by a transfer-matrix approach. A generalized transfer-matrix method treats reflectionless transport, lasing and coherent perfect absorption in nonreciprocal systems.
- **PT defect pairs.** A PT-symmetric defect in a Hermitian lattice creates localized states that meet at an exceptional point (Mostafavi et al., PRR 2, 032057).
- **Modular periods.** The Pisano and digital-root periods are tabulated in OEIS A001175 and A030132.
**What changed in the restored sections:**
- **Geodesic section:** restated for the cyclic group generated by T², since T has determinant −1. I did not restore the "algebraic skin effect" wording or the claim about wave-chaos poles locked to ℚ(√2), because neither is supported.
- **Tensor-network section:** reduced to the correlation-length benchmark, which I could derive. The claims about new silver-ratio bond-dimension scaling and entanglement spectra were not restored.
- **ECC comparison:** I added the point about security and forgeability.
- **E₈ detail:** I added the 24 + 24 + 64 + 64 + 64 decomposition and the 248-dimensional version.