← Back to the observatoryZeta Observatory / Methods
Read the geometry through the mathematics

A field guide to
the observatory.

Every surface is a coordinate view of a specified complex function. Colors, heights, arrows, and moving signals have different mathematical meanings. This reference keeps those meanings visible while you explore.

Actual function evaluationExact coordinate mapsFinite numerical resolution

01 / Functions & coordinates

The spectral coordinate is s = σ + it. The real part σ runs across the critical strip; t is the imaginary height. The critical line is σ = ½. Animation time and the de Bruijn–Newman heat parameter τ are separate controls.

ζ(s) = Σₙ≥₁ n⁻ˢ (Re s > 1) ξ(s) = ½ s(s−1) π⁻ˢᐟ² Γ(s/2) ζ(s) X(w) = ξ(½+w), Ξ(t) = ξ(½+it) ξ(s) = ξ(1−s), ξ(s̄) = ξ(s)̄ The displayed ζ outside Re s > 1 uses analytic continuation. Its values are not obtained by summing a divergent Dirichlet series.

ζ has a simple pole at s = 1, trivial zeros at the negative even integers, and the nontrivial zeros studied by RH. ξ is entire: its prefactor removes the ζ pole and the trivial zeros, leaving precisely the nontrivial zeros. In particular, ξ(0) = ξ(1) = ½.

Hardy's real function is useful when taking a critical-line slice:

Z(t) = eⁱϑ⁽ᵗ⁾ ζ(½+it) ϑ(t) = Im log Γ(¼+it/2) − (t/2)log π Z(t) is real for real t. Its sign changes and zeros can be inspected without the rotating ζ phase.

02 / Height & rainbow phase

QuantityWhat it representsHow to read it
arg F(s)The complex argument, wrapped modulo 2π.A full color cycle around a contour records a phase winding. The rainbow seam is a color convention.
|F(s)|The magnitude of the selected function.Zeros have zero magnitude. A clipped peak is a display limit, not a finite value at a pole.
log₂(1 + |F|/q₉₅)A compressed height encoding normalized by a sampled magnitude reference.Useful for seeing shallow valleys and tall peaks together. It is not the untransformed magnitude.
Re F / Im FThe two Cartesian components.Their simultaneous vanishing identifies a candidate zero; a sign contour of only one component does not.
Real-axis / line sliceSamples along a specified σ or t.Compare the slice with the actual complex values, especially near a zero or a pole.

The renderer chooses q₉₅ as the area-weighted 95th percentile of sampled, positive magnitudes on the base surface. The weights use the plane, sphere, or embedded-torus area element, so concentrating samples near a puncture does not give that tiny region disproportionate influence. Lock height scale retains the current reference in logarithmic form. One common reference applies across each rendered surface; there is no per-height rescaling of ξ.

Log relief: h = log₂(1 + |F|/q₉₅) Magnitude relief: h = |F|/q₉₅ Real relief: h = asinh(Re F/q₉₅) / asinh(1) Imaginary relief: h = asinh(Im F/q₉₅) / asinh(1) Displacement = g · c · clamp(h, −2.5, 2.5) c = 2.1 on the plane; c = 0.46 on the sphere or torus. g is the height control. Flat mode has zero displacement. q₉₅ is a visual normalization, not an error bound. Changing the sampled window can change this reference and therefore the displayed relief.

The completed function decays rapidly as |t| grows, so distant ξ relief may remain shallow even when phase structure is visible. With heat playback paused, the point inspector freshly evaluates the unscaled function. Log-polar arithmetic retains phase and logarithmic magnitude when a raw number would overflow or underflow; the inspector then displays a scientific-notation magnitude and identifies this representation. Derivatives and bridge comparisons use analytic function derivatives and logarithmic ratios, never derivatives of the compressed height map.

03 / Phase vortices, amplitude & shear

Away from zeros and poles, set q(s) = F′(s)/F(s) and U = log|F|. In the ordinary (σ,t) chart, the Cauchy–Riemann equations give two perpendicular fields:

∇U = (Re q, −Im q) ∇arg F = (Im q, Re q) j = Im(dF/F) = *dU J = Im(F̄ dF) = |F|² j = ½*d|F|²

The phase field circulates around a zero and follows amplitude contours. It is not an attractive flow into the zero. Amplitude descent is the different field −∇U. Normalizing arrow lengths changes the speed shown, while preserving their directions away from singularities.

A zero of multiplicity m has phase circulation 2πm. A pole contributes the opposite sign. Numerical winding estimates require a contour that avoids zeros and poles and enough samples to resolve the phase change. A colorful singularity or a small sampled value alone does not certify a zero.

Estimated winding = (1/2π) Σⱼ Arg[F(sⱼ₊₁)/F(sⱼ)] The winding tool compares 128-point and 256-point contour sampling. Agreement is a consistency check, not a proof that no phase turn was missed. A contour through a zero or pole has no well-defined winding of this form.

The bridge shear

In centered coordinates w = x+iy = s−½, a useful transverse quantity is:

W(x,y) = Re[X′(w) X(w)̄] = ½ ∂ₓ|X(w)|² W / |X|² = Re(X′/X) (where X ≠ 0) On x=0, W vanishes by ξ symmetry. A zero there is not a positive shear certificate. One must examine transverse shifts or the continuous normalized limit.

In the horizontal f(z)=Ξ(z) convention, the same orientation is written Wₕ(t)=−Im[f′(t+ih)f(t+ih)̄]. The minus sign comes from rotating the spectral coordinate.

Wₕ(t)/h → f′(t)² − f(t)f″(t) as h → 0 This first modal expression is a local diagnostic. Its sign on finitely many heights is not an all-height theorem.

The switchable bridge-color field uses a finite shift, rather than a derivative:

Bₐ(s) = [|G(s+a)|² − |G(s−a)|²] / [|G(s+a)|² + |G(s−a)|²] G = ξ for the ordinary views; G = Fτ for the heat view. This is a bounded, signed amplitude comparison. Where the denominator vanishes its value is undefined. The display scale is not differentiated or substituted into this calculation.

04 / Mounting on a Riemann sphere

Inverse stereographic projection places a complex chart coordinate c = a+ib on a unit sphere with north direction +Y. The observatory uses the displayed spectral scale ℓ/3:

(X,Y,Z) = (2a, |c|²−1, 2b) / (1+|c|²) c = (X+iZ)/(1−Y) s = ½ + (ℓ/3)c c = ∞ maps to the north pole Y=1. The selected chart-scale control is ℓ; the sphere explicitly uses ℓ/3. Rotation of the sphere changes the camera view, not F(s).

A sphere can compactify the drawing of the complex plane. It cannot turn ξ into a meromorphic function at infinity: ξ retains an essential singularity there. Large spectral ranges are compressed near the chart pole, so pixel-sized regions may conceal many oscillations. The finite evaluation mask indicates where the renderer has stopped sampling.

The critical guide is now traced from the exact base geometry, independently of the sampling mask: one complete great circle on the sphere and two complete critical ovals on the torus. A solid raised trace follows evaluated values; dashed portions mark the base-coordinate continuation. An infinity ring remains a puncture, not an assigned function value. The half-angle sphere chart avoids cancellation when a finite high point approaches the north pole.

Curved meshes include additional logarithmically spaced knots near the punctures and critical curves. Standard, Fine, and Ultra settings increase the actual sampling density; point picking and exported coordinates retain the true nonuniform UV positions. Expanding the evaluated domain fills more of the surface, but infinitely many windings cannot fit into a finite image. Use Unwrap this point in the plane for a narrow, finely sampled spectral window around a selected tail point.

05 / The studied elliptic double cover

The torus uses the two-sheet construction from Math Explorer. Two copies of the w-sphere are cross-glued at cuts between four branch points. Its exact algebraic curve is:

Eᵣ: η² = (w²+r²)(w²+r⁻²), r > 1 w = (s₀−1)/s₀, s₀ = 1/(1−w) The four branch points w=±ir, ±i/r give genus one by Riemann–Hurwitz. This is a double cover of the sphere, not an identification of ζ values at arbitrary periodic boundaries.

The implementation evaluates the complex Jacobi functions through their convergent theta quotients. With parameter m=r⁻⁴, K=K(m), and complementary K′=K(1−m), the chart is:

z = 4K u + 2iK′(v−½), 0 ≤ u,v < 1 w = (i/r) sn(z|m) η = cn(z|m) dn(z|m) qₑ = exp(−πK′/K) sn(z|m) = [θ₃(0,qₑ)/θ₂(0,qₑ)] · θ₁(πz/2K,qₑ)/θ₄(πz/2K,qₑ) K in the code uses the parameter m. Some references use the modulus k=√m=r⁻². Confusing these conventions changes the surface.

The true common periods of w and η are 4K and 2iK′. A geometric torus in three-dimensional space displays this complex period rectangle. The embedding's circular shape is not an additional equation for ξ.

The visible spectral lens

s = ½ + L(s₀−½) L = 1: literal studied Cayley pullback L > 0: explicit affine spectral lens The lens makes a larger spectral range visible. It preserves the reflection s ↦ 1−s and the critical line but changes the finite branch values in the displayed s-plane. It is not a functional identity of ζ.
Feature(u,v) locationMeaning
Critical ovalsv = ¼ and v = ¾Two lifts of Re s = ½, each containing one infinity puncture.
p₊, p₋(0,¼), (½,¾)w=1, s=∞. Essential punctures for the completed function.
q₋, q₊(½,¼), (0,¾)w=−1, s=½. Finite anti-bridge symmetry anchors.
Branch points(¼,½), (¾,½), (¼,0), (¾,0)Projection ramification. Their coordinate stagnations do not imply zeros of ξ.
Deck exchange(u,v) ↦ (½−u,1−v) modulo 1Preserves s and reverses η. It exchanges the two lifts.

Two zero markers on the torus can represent one spectral zero. Every ordinary s-value has two preimages. The sheets carry the same pulled-back scalar function; their duplicate zeros do not count as additional zeros of ζ or ξ.

The anti-bridge anchors are not collision sites. The actual heat family has Hτ(0)>0 for every real τ. At the other end, apparent freezing near infinity is coordinate compression; it does not supply a zero-trapping force.

Transforming a field correctly

Phase textures pull back by evaluating F at the chart's s. For the flat conformal phase-flow interpretation, first set qz=(F′/F)·ds/dz. The corresponding UV velocities are du=Im(qz)/(4K) and dv=Re(qz)/(2K′). These can then be mapped along the torus's tangent vectors. Treating unchanged plane arrows as torus tangent arrows would omit the chart derivative.

05b / Projective atlas & point states

Click Projective above the viewport, then choose an elliptic torus or projective domain sphere. The linked panels below the 4:3 scene follow the inspector's selected point: one shows its projective state, and the other shows two sphere charts with crosswise branch cuts. The presets use r = 2 and L = 24; Original projective chart · L = 1 restores the literal construction.

s₀ = ½ + (s−½)/L, w = [s₀−1:s₀] s = ½ + L(1/(1−w)−½) Eᵣ ⊂ ℂℙ³: [X₀:X₁:X₂:X₃] = [1:w:w²:η] X₁² = X₀X₂ X₃² = X₀² + (r²+r⁻²)X₁² + X₂² The sphere displays the input projective line; the torus is a three-dimensional display embedding of Eᵣ. Its homogeneous ℂℙ³ coordinates are reported separately.
Marked pointCoordinateMeaning
p or p±w = 1Spectral infinity s = ∞; the pulled-back ξ function remains essential and unevaluated there.
q or q±w = −1Anti-bridge s = ½; a nonzero phase saddle for ξ, with zero circulation.
ω±w = ∞Regular quartic infinities [0:0:1:±1], mapping to finite s = ½−L/2. The reciprocal chart handles them.
Branch pointsw = ±ir, ±i/rProjection ramification; the two-sphere panel shows local sheets, joined crosswise at the indicated cuts.

Domain displays [s₀−1:s₀], switching between w and 1/w. Shifted pair displays [U:V] with U = G(s+a), V = G(s−a), using G = ξ or, in Heat mode, Fτ. Its sphere coordinates retain the relative phase and normalized amplitude difference:

n = (2 Re(ŪV), 2 Im(ŪV), |U|²−|V|²) / (|U|²+|V|²) R = V/U, n₃ = (1−|R|²)/(1+|R|²) Polar normalization avoids losing the state when raw values underflow. Either channel may vanish; if both vanish, the projective state requires local analysis and is marked unresolved.

Derivative disk uses [G:ℓG′], with ℓ > 0. Its Cayley coordinate is:

C = (G−ℓG′)/(G+ℓG′) 1−|C|² = 4ℓ Re(ḠG′) / |G+ℓG′|² Where defined, inside |C| < 1 corresponds to Re(G′/G) > 0. A simple zero maps to C = −1 regardless of its input location. Singular or ill-conditioned signs are shown explicitly.

The two ordinary lifts of a ξ zero have the same local winding in the holomorphic orientation. The shifted ratio R can instead have a zero–pole pair with opposite winding; their contour contributions can cancel without canceling any ξ zeros. The marked anti-bridge and geometric chart boundaries supply no RH proof. During buffered playback the point-state panels are held; pause and inspect to refresh their numerical values.

05c / Spinor & Hopf fiber

Open the Spinor view to lift the selected Domain, Shifted pair, or Derivative state to a normalized vector ψ = (U,V)ᵀ. Its common complex phase is retained in the amplitude dials and reference overlap, although it disappears from the projective Bloch point. The complex-line and Pauli-matrix descriptions identify this base with ℂℙ¹ ≅ S².

ψ = (U,V)ᵀ / √(|U|²+|V|²) n = ψ†σψ = (2 Re(ŪV), 2 Im(ŪV), |U|²−|V|²) [unit ψ] ψ ↦ exp(iχ)ψ leaves n unchanged Rₐ(θ) = exp(−iθ â·σ/2) = cos(θ/2)I − i sin(θ/2) â·σ Rₐ(2π)ψ = −ψ, Rₐ(4π)ψ = ψ The Bloch direction returns after 2π. The signed spinor amplitudes return after 4π. The chosen channel pair is a mathematical spinor state; this does not identify ξ with a physical particle.

Common phase explores the Hopf circle. Rotation and Axis use the standard spin-one-half SU(2) rotation. The displayed frame has columns determined by (ψ,Jψ), with J(U,V) = (−V̄,Ū); its normal is n. A common phase changes the frame without moving n. SU(2) mixing generally changes the channel balance |U|²−|V|², so a rotated display does not prove the original bridge balance positive.

05d / A chosen Euclidean twistor structure

Open the Twistor view to use the selected point Z = [X₀:X₁:X₂:X₃] of the elliptic curve Eᵣ above. We explicitly choose the standard Hermitian norm and the ordered quaternionic pairing (Z₀+jZ₁, Z₂+jZ₃), with complex coefficients on the right of j. These choices give the genuine Euclidean fibration ℂℙ³ → ℍℙ¹ ≅ S⁴ with ℂℙ¹ fibers, using Salamon–Viaclovsky's conventions, sections 2.2–2.4. The elliptic curve's projective embedding alone does not canonically select this twistor structure.

q₁ = Z₀+jZ₁, q₂ = Z₂+jZ₃, Σ|Zₖ|² = 1 A = Z₀Z̄₂ + Z̄₁Z₃, B = Z₁Z̄₂ − Z̄₀Z₃ x = (2 Re A, 2 Im A, 2 Re B, 2 Im B, |Z₀|²+|Z₁|²−|Z₂|²−|Z₃|²) ∈ S⁴ JZ = (−Z̄₁,Z̄₀,−Z̄₃,Z̄₂), J² = −I Fiber: [αZ+βJZ], α = cos(θ/2), β = exp(iφ) sin(θ/2) The five real coordinates have unit norm. The fiber controls change the ℂℙ³ representative while preserving its S⁴ base; the displayed invariance residual checks this numerically.

The identity behind the fiber is (q₁h) overline(q₂h) = q₁q̄₂ for a unit quaternion h. The ordinary common complex phase is included in this right action. The two fiber angles cover a full sphere of projective representatives. Its local frame uses the largest quartet component as a positive-real phase reference; where that component changes, the gauge may change.

Elliptic sheets and twistor fibers are different. The sheet selector first sends X₃ to ±X₃, choosing the source lift or its elliptic deck partner. The subsequent fiber motion generally leaves Eᵣ. The source and moved quartets therefore have separate residuals for X₁²−X₀X₂ and X₃²−X₀²−(r²+r⁻²)X₁²−X₂². A moved point outside Eᵣ inherits no spectral coordinate s or value ξ(s); the point inspector remains attached to the source.

North chart: y = (x₀,x₁,x₂,x₃)/(1−x₄) South chart: y = (x₀,x₁,x₂,x₃)/(1+x₄) These are ℝ⁴ chart coordinates. The panel draws three and retains the fourth separately. A projection pole is explicitly marked as infinity; the other chart is regular there.

The map at a spectral puncture still exists geometrically because Eᵣ is smooth there, while the loaded ξ value remains essential and masked. Neither twistor projection nor fiber motion supplies a new global zero-location or positivity theorem. Spinor and twistor panels use held projective data; drawing and buffered playback do not reevaluate the analytic field.

The links ?view=projective, ?view=spinor, and ?view=twistor select the corresponding starting point. The address bar subsequently tracks controls, the inspected point, and the camera automatically. Copy its URL or use Copy link to reopen that view, with playback paused.

05e / Exact Klein quotient & reflection annulus

The two added views use quotients of the existing elliptic curve Eᵣ, with r > 1. The Klein view identifies a free antiholomorphic involution on the whole curve. The annulus view quotients by the reflection with two fixed critical ovals. Identifying only the two infinity points would instead create a singular node.

R(w, η) = (1/w̄, η̄/w̄²), D(w, η) = (w, −η) A = D ∘ R: (w, η) ↦ (1/w̄, −η̄/w̄²) R(u,v) = (u, ½−v), A(u,v) = (½−u, ½+v) modulo 1 The common fundamental band is ¼ ≤ v ≤ ¾. In the displayed chart V = 2(v−¼), the Klein seam glues (u,1) to (½−u,0). The annulus retains both boundary circles.

A has no fixed points for r ≠ 1. A hypothetical fixed point would have |w| = 1 and η/w purely imaginary, while the curve equation gives (η/w)² = (r−r⁻¹)² + 4 cos²(arg w) > 0. Thus the quotient is a smooth nonorientable Klein bottle as a compact domain. The ξ-loaded surface remains punctured: p₊ and p₋ become one essential infinity puncture. The two finite q lifts become one distinct anti-bridge point. The critical ovals become one two-sided, nonseparating loop; cutting it gives an annulus, so it is not a no-escape wall.

The Klein surface is shown by a figure-eight immersion in three dimensions. Its visible self-intersections do not identify extra points; picking and sampling use the intrinsic chart. A global scalar displacement along a normal is unavailable on a nonorientable surface, so Klein relief is fixed to zero. The cylinder used for the annulus is a display of its intrinsic band; displayed lengths and areas are not the conformal metric used for energy.

Conjugate fields and boundary data

F(Ap) = F(Rp) = conjugate F(p), F = ξ or Fτ U = Re F: even across the reflection boundary V = Im F: odd across the reflection boundary V = 0, ∂ₙU = 0 on each regular annulus boundary Raw ζ has a functional-equation multiplier and is not an equivariant field under this simple conjugation law. The quotient modes therefore use completed ξ or the symmetric heat family.

On the Klein quotient, |F| and U are ordinary scalars; V and the local phase change sign under the orientation-reversing transition. The phase current is a twisted one-form. Its trajectories are shown within the selected local chart and stop at the seam. Local winding still records zero multiplicity; conjugate chart changes do not annihilate zeros. The panel compares actual reflected evaluations in logarithmic polar form and reports floating-point residuals, not certified error bounds.

What the energy comparison establishes

|∇U|² = |∇V|² pointwise for a holomorphic field E(torus cutoff) = 2 E(Klein cutoff) = 2 E(annulus cutoff) The factors use the intrinsic conformal metric and symmetry-matched cutoffs. They do not assert that a finite total energy exists at the essential puncture.

The equality of real and imaginary Dirichlet energies follows directly from the Cauchy–Riemann equations. Their difference is identically zero. The bridge even/odd energies use different boundary traces and kernels; they are not these two bulk energies. Folding the domain does not create a new positivity inequality or remove the Gram/Schur coupling.

The annulus does supply a positive reference operator. In uniformizing coordinates its circumference is ℓ = 4K and width b = K′. For κ = 2π|n|/ℓ > 0, the harmonic Dirichlet-to-Neumann block is κ times the matrix with diagonal coth(κb) and off-diagonal −csch(κb). Its symmetric and antisymmetric eigenvalues are κ tanh(κb/2) and κ coth(κb/2); the zero-mode eigenvalues are 0 and 2/b.

To turn that reference into a useful bound for ξ, one still needs an exact comparison with the theta bridge form and control of all signed coupling and infinity terms. Green's identity makes the punctures essential: removing their boundary flux while imposing the regular homogeneous conditions forces a constant harmonic field. The complete derivation and a symmetric entire counterexample with off-line zeros are recorded in KLEIN-ANNULUS.md.

05f / Infinity ends & positive display damping

ViewCoordinates or displayWhat remains explicit
Infinity endA cylinder and unrolled chart of q = 1 − w near either p±.The finite near and far cutoffs, selected sheet, and exact Klein partner.
Fixed strip endsx = σ − ½, with log-height ρ = log|t| and reciprocal height y = 1/t.The transverse offset x and the choice of upper or lower end.
Damped Klein displayPositive Gaussian attenuation of the completed field.The original samples, zeros, derivatives, inspector and bridge diagnostics.

The geometric curve is smooth at p±, where w = 1. The loaded ξ function has an essential singularity there. The end chart expands a neighborhood of that puncture:

q = exp(−τₑ + iθ), s = L/q + (1−L)/2 Klein partner: q ↦ −q̄/(1−q̄), s ↦ 1−s̄ τₑ = −log|q| is an end coordinate, separate from heat time. The chart uses τₑ ≥ 0, so |q| ≤ 1 stays away from the elliptic branch points. Its far edge is a finite cutoff, not infinity.

The strip view keeps x independent as |t| grows; it does not collapse different real offsets into the same limiting direction. Both charts show actual finite evaluations. Gray regions are outside numerical coverage, and hatching warns that phase may turn between samples. A partner can be outside the plotted window while still evaluable. The cutoff fluxes use the original field and retain their amplitude multipliers. In strip mode they cover the two height caps only, with the vertical sides omitted; they are not a global energy or tail bound.

What the Gaussian changes

hS(s) = exp(−|s−½|²/S²), F̃ = hS F, S > 0 arg F̃ = arg F and Im(dF̃/F̃) = Im(dF/F) wherever F ≠ 0 ∂s̄F̃ = −(s−½)F̃/S² The default scale is S = 120. The positive factor retains finite zeros and phase, but the weighted field is nonholomorphic.

The main surface and end panel provide separate damping controls. They change the display; raw numerical values, caches, exports, derivatives and signed bridge forms remain attached to the original function. For ξ, its theta integral gives growth bounded by exp(O(|s| log|s|)). The Gaussian therefore allows the weighted complex amplitude to extend smoothly and flatly by zero at the puncture. This added display zero is not a ξ zero, and no phase is assigned there. The original ξ singularity survives. Finite damped heat samples do not establish a global smooth-extension theorem for that family.

05g / The dual conormal family

Open Dual conormal to inspect tangent hyperplanes of the elliptic curve at the selected point or at exact infinity, anti-bridge, branch and reciprocal-chart anchors. A curve in ℂℙ³ has a family of tangent hyperplanes, rather than one preferred normal.

Q₁ = X₁² − X₀X₂ Q₂ = X₃² − X₀² − (r²+r⁻²)X₁² − X₂² W = α dQ₁ + β dQ₂, [α:β] ∈ ℂℙ¹ W·Z = 0, W·T = 0 Z is the source point and T its projective tangent representative. Incidence uses the complex bilinear pairing, without conjugation. A separate Hermitian norm chooses convenient display representatives.

The controls vary [α:β] in the stated quadric basis. The source sphere, family sphere and coordinate dials are coordinate displays, not a literal three-dimensional drawing of a complex hyperplane. Point and tangent residuals check the incidence numerically. The family is regular at spectral infinity and requires no ξ evaluation there; it neither assigns a field to the dual space nor resolves the essential singularity. This projective conormal construction does not by itself choose a Penrose spacetime interpretation.

The complete derivations include the exact end seam, strip metric, Gaussian smooth-extension proof, full-energy distinction and conormal conventions.

06 / Twin ladder, bridge & packets

The direct and reflected prime-power tones come from the two spectral arguments s and 1−s. A rung with prime p and exponent m has log-frequency λ = m log p:

Direct tone: p⁻ᵐˢ = exp(−σλ) exp(−itλ) Dual tone: p⁻ᵐ⁽¹⁻ˢ⁾ = exp[−(1−σ)λ] exp(+itλ) λ = m log p

At σ=½ the direct and dual tone magnitudes agree. Finite collections of these tones are useful for studying interference, delays, and wave packets. A finite prime sum is a specified signal, not the analytically continued zeta function. The Euler product itself converges absolutely only in Re s>1.

The exact analytic bridge is the functional equation:

ζ(s) = χ(s) ζ(1−s) χ(s) = πˢ⁻½ Γ((1−s)/2) / Γ(s/2) |χ(½+it)| = 1 ξ(s) = ξ(1−s) Removable values are understood by continuation. “Lossless” on the critical line means unit modulus of this scalar factor. Off-line amplitude scaling is not a physical dissipation law.

The ladder rails at σ=½±a are aligned at the same spectral height. The completed-function identity at that height is ξ(½+a+it)=conj ξ(½−a+it); a same-height bridge therefore links conjugate values. Both rails are traced through the current surface map.

The finite prime-power packet

The packet inset displays the magnitude of a specified finite sum against log-delay ℓ. With packet width h and independent transport clock T, its direct signal is:

ψσ(ℓ,T) = Σₚ≤pₘₐₓ Σₘ₌₁² (log p) p⁻ᵐσ · exp[−(ℓ−m log p)²/(2h²)] · exp(−imT log p) The dual uses p⁻ᵐ⁽¹⁻σ⁾ and the opposite carrier sign exp(+imT log p). The Gaussian window localizes the selected log-frequency atoms.

The log p coefficients correspond to the prime-power atoms in −ζ′/ζ on its convergent Euler-product half-plane. This Gaussian-windowed, finite packet is not itself ζ or ξ. The moving surface envelope is an illustration of packet transport. Its clock T is separate from τ in the actual heat equation, and it does not supply an infinite passive realization of ξ.

07 / Actual theta heat evolution

The heat layer uses the classical de Bruijn–Newman integral, with the convention used in the repository's torus heat-flow note and in Polymath's paper:

Φ(u) = Σₙ≥₁ (2π²n⁴e⁹ᵘ − 3πn²e⁵ᵘ) exp(−πn²e⁴ᵘ) Hτ(z) = ∫₀∞ exp(τu²) Φ(u) cos(zu) du H₀(z) = ⅛ ξ(½+iz/2) ∂τHτ = −∂z²Hτ

To retain the same σ,t axes as the ζ and ξ views, the renderer evaluates:

Fτ(s) = 8Hτ(−2i(s−½)) = 8Hτ(2t−2i(σ−½)) F₀(s) = ξ(s) ∂τFτ = ¼∂s²Fτ = −¼∂t²Fτ A τ increase is backward heat evolution in the real spectral t-coordinate under this convention. A generic smoothing filter or ordinary diffusion on the embedded torus would be a different operation.

For comparison with the theta-convolution bridge notes, the real kernel coordinate has a different scaling:

κ(u) = Φ(u/4) κτ(u) = exp(τu²/16) κ(u) Fτ(½+a) = ∫ℝ κτ(u) exp(au/2) du

The kernel variable u, spectral height t, torus parameters (u,v), and animation clock are different variables even when historical notation reuses a letter. The on-screen axes identify which space is being viewed.

Why the infinity end seems to freeze

For the literal lens ℓ=1, on a critical oval write w=eⁱθ. Then s=½+(i/2)cot(θ/2), and the pullback of H obeys:

fτ(θ) = Hτ(cot(θ/2)) ∂τfτ = −(1−cos θ)² ∂θ²fτ − (1−cos θ)sin θ ∂θfτ At θ=0 the second-derivative coefficient vanishes to fourth order. The degeneration records coordinate compression near infinity, not a new dynamical boundary condition.

Λ is the threshold for all zeros of Hτ to be real. Rodgers–Tao established Λ≥0. A lower positive bound would contradict RH; reducing an upper bound toward zero is the direction compatible with proving RH. This finite visualization does not establish either a bound or its global hypotheses.

07b / Dynamics, perturbations & zero trapping

The Dynamics laboratory turns the phase-field and finite-zero experiments into a playable, reproducible view. It uses three distinct kinds of evolution: probes moving through a fixed analytic field, a complex field changing under a specified operation, and a finite system whose zeros evolve under an explicit model. Each mode identifies its kind in the live inspector.

ModeWhat movesInterpretation
Vortex phase flowNumerical probes in the actual completed ξ field.Phase circulation follows amplitude contours away from singularities.
Dissipative relaxationProbes following amplitude descent.Observed probe capture leaves the underlying ξ function unchanged.
Seeded random walkProbes with drift and reproducible Gaussian increments in (σ,t).A bounded spectral-plane stochastic experiment.
Shifted-field cancellationThe field U − A exp(iφ)V, where U=ξ(s+a), V=ξ(s−a).Additive interference can move its zeros; these are zeros of the combined field.
Bounded heatA perturbed complex grid under spatial diffusion.A finite boundary-value experiment; distinct from the theta heat family above.
Zero trapping modelA finite interacting set with imposed transverse confinement and optional forcing.Tests the mechanics of a chosen model and its perturbations.
Heat collisionsRoots of a finite polynomial evolved by exp(−τD²).The coefficient evolution is exact; displayed root locations use numerical solving except the explicit cubic.

Phase current, dissipation and diffusion

U = log|ξ|, Φ = arg ξ, q = ξ′/ξ ∇U = (Re q, −Im q), ∇Φ = (Im q, Re q) b = −α∇U + β∇Φ Formal deterministic identity: dU/du = −α|∇U|² Random-walk increment: Δr = b Δu + √(2D Δu) N(0,I) The numerical integrator regularizes singular speeds and uses finite steps. The formal identity applies to the unregularized, continuous deterministic field away from zeros; a bounded boundary rule and stochastic noise add different effects.

The component signs follow from the Cauchy–Riemann equations. Phase circulation alone is tangent to log-amplitude contours. A small raw |ξ| is not a sufficient capture test, because ξ decays rapidly with height; the lab instead uses scale-independent local root diagnostics. Candidate counts and winding measurements remain finite numerical observations.

Additive cancellation and bounded spatial heat

The cancellation mode evaluates a shifted-pair interference field. At a fixed parameter setting it is a specified entire function. Adjusting its relative phase or amplitude changes that function. Play sweeps the relative phase at the selected phase rate; setting that rate to zero holds the field fixed. A common phase multiplier applied to one unchanged function would preserve its zeros, since the exponential never vanishes.

F(s) = ξ(s+a) − A exp(iφ) ξ(s−a) Spatial heat: ∂u f = D(∂σσ f + ∂tt f) Spatial heat evolves the real and imaginary grid components, with the selected boundary rule. Analytic ξ is harmonic in the continuum; the added perturbation supplies nonharmonic structure to relax. A finite-difference discretization also introduces truncation error.

Finite polynomial collisions and forced zero models

The six-real, Lehmer-type, eight-real and quartet starting polynomials follow the coefficient-heat experiments in Math Explorer. An explicit cubic provides a collision whose roots are known directly:

Pτ(z) = exp(−τD²)P₀(z) P₀(z) = z(z²+b²) ⇒ Pτ(z) = z(z²+b²−6τ) Roots: 0 and ±√(6τ−b²) At τ=b²/6 the cubic has a multiple root. Before and after that time the pair changes between imaginary and real positions. This is a finite-polynomial mechanism, not a computed violation of RH.

For the actual theta family, zero motion is obtained by differentiating the changing function at a simple zero. In this app’s s coordinate it is dρ/dτ = −F″/(4F′). The full de Bruijn–Newman zero interaction is an infinite, appropriately summed system; see Rodgers and Tao’s explanation. The lab’s finite trapping system exposes its imposed confinement and forcing as controls. Future capture in such a system does not establish that the original ξ zeros are globally trapped.

Reproduce an experiment

The lab uses a seeded random generator and fixed simulation steps. Mode, parameters, overlay choices, seed and step index are included in the view URL. Play changes the address at a bounded cadence; Pause and Copy view link record the exact current step. Opening that URL replays deterministically to the selected step and remains paused. Parameters reset the trajectory; display overlays retain it. The actual theta heat view continues to use the buffered evaluator described above.

07c / Bridge research views

The Research workspace makes the bridge calculations interactive. It uses the same completed-function evaluator as the renderer. Exact polynomial and finite-oscillator models identify themselves in their scope notes. Parameter changes update the view URL; frame-based experiments reopen paused at the shared frame.

ViewQuantityScope
Bridge shearW = Re(F′ conjugate(F)), S = Re(F′/F), phase and finite bridge.The smooth W vanishes at every zero; S becomes singular there. The displayed sign alone is no zero barrier.
Modal signsFε(w)=w²+2ε; W=2x(x²+y²+2ε); L₁(y)=4y²+8ε.An exact local model with an explicit moving negative region.
Even & odd energyBoundary forms E₊ and E₋; their difference compared with the direct bridge derivative.Finite quadrature and a reported residual, with no certified infinite tail.
Gram & SchurFinite complex Hermitian bridge/ratio-Pick matrices, the positive-real-node logarithmic-derivative Pick kernel, eigenvalues and a frozen-block complement.Positive scaling preserves inertia. Near-singular blocks are unresolved; no arbitrary-order theorem is inferred.
Theta cancellationDN(ω)=Σd|Ncos((ω/2)log(d/√N)), with prime-factor product and theta truncations.Individual arithmetic blocks can change sign without deciding the assembled form.
Prime modesSigned amplitudes (log p)p−m/2cos(γm log p), finite phase evolution, solo/mute.An explicit finite oscillator basis, not established eigenfunctions of a zeta operator.
Ratio pathsRa(s)=F(s−a)/F(s+a), traced along the chosen path.The optional symmetric quartet multiplies a separate model function. It does not alter actual xi.
Pulse responseS=(F−αF′)/(F+αF′) sampled on the critical line and applied to a finite periodic Fourier transform.Norm and tail energies are finite-window observations. They do not prove an infinite-line causal or tail inequality.
Heat-zero motionComplex roots of the actual theta family, derivative velocities and contour diagnostics.Local numerical continuation, with explicit unresolved states and a separate exact quartic calibration.

In the main renderer, Logarithmic shear uses the diverging color value tanh Re(F′/F). This compresses large signed values; it is not the uncompressed W. The selected field supplies both F and its derivative. The atlas probe can be opened on that surface and then mounted on the sphere or elliptic torus.

07d / What a no-escape condition would require

Circulation around a fixed zero constrains phase-flow probes. The zero itself moves only when the analytic family changes. Near a zero of multiplicity m, the leading singular circulation has tangential speed proportional to m/r and zero radial velocity. A regular analytic factor adds bounded corrections, including a possible radial component: the exact invariant is log|F|, not Euclidean distance from the zero. Adding a descent term attracts probes, but does not establish a law for the zero.

Fτ(s) = 8Hτ(−2i(s−½)), ∂τF = ¼Fss Fτ(ρ(τ)) = 0, F_s(ρ) ≠ 0 ⇒ ρ′ = −Fss/(4F_s) x = Re ρ − ½, M = x Re(Fss/F_s) (d/dτ)(x²) = −M/2 A proved M≥0 along an entire simple-zero branch interval makes x² nonincreasing. At a moving tube boundary |x|=d(τ), M≥−4dd′ is an instantaneous inward/tangent test. Equality only at contacts can still allow higher-order tangential crossing; a barrier requires strict inward bounds or a suitable uniform neighborhood comparison estimate. At d=0, M=0 is automatic. Multiple-root collisions require separate analysis, and finite samples do not prove these bounds.

The heat-zero view evaluates these diagnostics and also displays sampled winding and perturbation margins on a contour. A genuine Rouché barrier requires sup∂Ω|Fτ−Fτ₀| < inf∂Ω|Fτ₀| over the entire boundary. Certified use needs bounds between samples and throughout the parameter interval. If a reflection-symmetric disk contains exactly one zero counted with multiplicity, symmetry forces that zero onto the critical line.

The exact quartic calibration is Fτ(s)=P−τ(−2i(s−½)), where Pu(z)=z⁴+(12u−5)z²+12u²−10u+4. It obeys the same heat normalization and xi reflection symmetries. Initially its roots lie on Re s=½. At τ=−¼ they lie at Re s=¼ and ¾, after a double-root collision. Its central value remains positive. This demonstrates why circulation and symmetry alone do not force confinement; it is a polynomial model, not an off-line xi zero.

The de Bruijn strip theorem gives a global statement when all zeros initially satisfy |Re ρ−½|≤d₀: for increasing τ, the bound contracts to d²≤max(d₀²−(τ−τ₀)/2,0). It is a forward result with a global initial hypothesis. It cannot be run backward to infer the zero locations at τ=0.

08 / Numerical scope & reproducibility

The browser uses ordinary IEEE-754 double precision. ζ uses Euler–Maclaurin summation with 12 Bernoulli corrections and functional-equation continuation into the left half-plane. ξ uses the completed function with reflected evaluation and removable endpoint values. No nontrivial zeros are inserted into those evaluations.

Inside the original numerical window, the heat integrator shifts its complex contour to reduce oscillatory cancellation. It uses a finite theta sum and trapezoidal quadrature, with a finer grid available as a convergence diagnostic. At τ=0 it evaluates ξ directly. Outside the original window, the heat engine uses the equivalent Gaussian convolution of the entire completed function, evaluated in log-polar arithmetic:

Fτ(s) = (1/√π) ∫ℝ e^(−u²) ξ(s + √τ u) du τ > 0: √τ is real τ < 0: √τ = i√(−τ) Expanding the Gaussian moments gives exp[(τ/4)∂s²]ξ, matching the stated heat normalization. The extended integrator uses 32-node Gauss–Hermite quadrature, with 64 nodes for convergence diagnostics. A small refinement difference is numerical evidence, not an interval error enclosure. High-ordinate heat grids require many ξ evaluations; Responsive density reduces the sample count, while Fine and Ultra can take substantially longer.
SettingNumerical kernel
Evaluation windowExtended: −128 ≤ σ ≤ 129, |t| ≤ 2,000; heat |τ| ≤ 1. The Original window option retains −12 ≤ σ ≤ 13, |t| ≤ 120 for comparison. Independent reference checks sample these domains; they do not certify every point.
Heat cutoffReal contour half-length 1.5; 384 steps per half-line, refined to 768 for a diagnostic.
Theta series12 integer terms in the complex heat kernel.
Elliptic chartJacobi theta quotients; AGM complete elliptic integrals; both common periods retained.
Independent checksReference special-function values, normalization, periodicity, symmetry, and coordinate inverse checks.
CertificationNumerical exploration. There is no rigorous interval or all-height proof attached to an image.

Independent accuracy checks

The extended-domain checks compare logarithmic magnitude, phase, and the analytic logarithmic derivative against independent high-precision references, including points whose ordinary floating-point values overflow or underflow. Their measured discrepancies and sample counts are recorded in the numerical report below. The original checks remain as regressions.

The machine-readable numerical report records nine original tests and five extended kernel tests. The original fixtures contain 33 function/derivative reference points and 13 heat reference points computed independently with 80-decimal mpmath arithmetic. The largest observed scaled discrepancies are approximately 2.04×10⁻¹³ for ζ/ξ values and 3.93×10⁻¹³ for heat values. These are measurements on the checked finite sample, not uniform error bounds.

ζ error metric: |computed − reference| / max(1, |reference|) ξ and heat error metric: |computed − reference| / max(|reference|, exp(−π|t|/4)(1+|t|)⁷ᐟ⁴) The envelope in this test metric avoids an undefined relative error at a zero. It is not a display rescaling and does not establish relative accuracy where the true value is extremely small.

The heat references use independent, unshifted real-axis quadrature; the app uses a shifted contour. Other checks cover functional/conjugate symmetry, the τ=0 identity, the PDE sign and factor ¼, the ζ pole, ξ's removable endpoints, and nonzero points beside reference zeros. The separate elliptic tests compare chart values to 60-decimal Jacobi-function references and check both periods, both lifts, and the fixed landmarks.

Zero markers and point probes

For ζ and ξ at heat time zero, the nontrivial markers use the first 25 positive reference zero heights, together with their negative-height partners. This finite marker list is not every zero in the full numerical window. The ζ view also distinguishes negative-even trivial zeros and the pole at s=1.

At nonzero heat time, markers come from sampled sign changes on the critical line at positive heights up to t=60, followed by bisection and reflection. This detects selected real-axis roots of Hτ. It does not detect off-line zeros, roots without a resolved sign change, or certify that the displayed list is complete.

Outside buffered playback, a pointer probe first obtains the chart coordinates (u,v), applies the sphere or elliptic map again, and freshly evaluates the selected function at the resulting (σ,t). It therefore reports numerical function values rather than interpolation of the mesh's complex vertex values. During playback the inspector holds these numerical readouts. Pressing Inspect pauses the sequence and resumes fresh point and derivative evaluation. The drawn surface itself remains a finite mesh.

Near a zero, relative value errors and logarithmic derivatives can become large. Near a pole or a chart puncture, clipping and sampling limits matter. At a branch point, the chart derivative can vanish even though the function does not. These cases have different mathematical meanings and should be inspected with the point probe.

CSV exports include true nonuniform (u,v), spectral (σ,t), log |F|, argument in radians, and explicit validity and raw-representability flags. Raw value columns are blank if their magnitudes cannot be represented as ordinary floating-point numbers. To make an observation reproducible, retain the selected function, computed extent, spectral bounds, geometry and lens scale, heat parameter, layer settings, numerical resolution, and probe coordinates. Compare suspected features at greater resolution and with a second numerical method before treating them as research evidence.

09 / Computed cache & buffered playback

The observatory separates numerical sampling from displaying a field. A prepared frame contains the actual double-precision complex values and their log-polar representation, together with its critical traces, landmark values, root markers, requested overlay curves, and 361-point critical-line inset. Reusing that frame preserves the same numerical samples. It changes latency and repeated work, not the mathematical function, numerical precision, or accuracy claims.

Where a field comes from

For an exact numerical setting, the loader checks the memory cache, the browser's saved cache, the shipped starter pack, and finally the numerical worker. The cache key includes the engine version, function, heat parameter when applicable, chart and spectral bounds, density, computed extent, bridge offset when relevant, and numerical overlay choices. Camera movement and display controls such as phase rotation, opacity, and height style can reuse a field.

ResourceBudget or contentsBehavior
Memory cache128 MiBLeast recently used computed views are evicted when the budget is reached.
Saved field cacheUp to 256 MiB in IndexedDBBest effort. Available browser storage can reduce this budget; storage failure falls back to memory and computation.
Active heat clip192 MiBEvery frame must fit before Play begins. An oversized clip asks for fewer frames or Responsive density; it does not silently lower the requested density.
Original starter pack78 independently evaluated fields; 155 exact keys; 59.32 MB compressedFiles load on demand, with shared geometry between heat frames. Optional ladder records allow both overlay settings to reuse the same numerical field.

The deployment build adds 38 default projective fields while preserving the original pack: ζ and ξ on each projective geometry, plus 17 heat frames per geometry at r = 2, L = 24 and Standard density. Their keys are distinct from ordinary charts. Custom settings can require fresh calculation.

The complete starter-pack size is not the initial download size. The pack includes ζ and ξ on all three default mounts, the bridge, pole, phase-atlas, and vortex presets, and the following Standard-density heat sequences:

τⱼ = −0.30 + 0.05j, j = 0,…,16 Default plane: −2 ≤ σ ≤ 3, 8 ≤ t ≤ 38 Sphere / torus: lens L = 24; torus r = 1.6 Heat-preset plane: −0.75 ≤ σ ≤ 1.75, 10 ≤ t ≤ 32 The sphere and torus use the extended finite computation window. Each heat field was evaluated independently at its displayed τ. The machine-readable manifest lists the exact numerical keys and assets.

View the starter-pack manifest. The binary format preserves typed-array bytes, NaN, infinity, and signed zero. Gzip compression does not quantize the values. A browser that cannot load a compressed asset can compute the same field in its worker instead.

Prepare, play, inspect

  1. In Cache & playback, choose the heat range, 9, 17, or 33 exact frames, and Responsive, Standard, or Current mesh density. The default 17-frame Standard sequence matches the shipped pack.
  2. Choose Prepare heat frames, or press Play to prepare and then start. All frames load or compute before playback begins. Custom settings reuse exact matches and compute any missing frames.
  3. Play traverses the stored frames forward and then reverse. The direction, frame index, and actual τ remain visible. The frame slider selects a prepared frame; it does not interpolate between τ values.
  4. Pause or press Inspect to resume numerical point inspection. During playback and after frame scrubbing, the held inspector avoids presenting a value from another heat time as a fresh measurement.

Preparation locks one common height reference for the sequence, preserving an existing lock when appropriate. Keep this lock enabled when comparing amplitudes: independent per-frame normalization would change the apparent relief. “Forward” means increasing τ and “reverse” means decreasing τ; these are playback directions through the same prepared family. The PDE sign remains the one in the heat normalization above.

The heat playback step performs no ζ, ξ, or heat-kernel evaluation and loads no additional fields. Surface samples, critical traces, landmarks, roots, overlay curves, and the critical-line inset come from the current buffered frame. The renderer still transforms, colors, and draws them. A requested frame rate can therefore be limited by rendering speed. The Canvas fallback draws a smaller motion mesh and restores stationary detail when playback and other motion pause; the underlying numerical samples remain available.

The prime-power packet panel is separate. Its independent clock T continues to drive the explicit finite sum while packets are enabled, and its own Pause control stops that clock. Buffered heat playback does not turn this signal into ζ or into a heat solution.

Precache modes prepares the nine chart/function combinations for the current numerical settings. Clear cache clears computed fields and the active clip while retaining the saved-view notebook and notes. Browser persistence is convenient reuse; exported session JSON remains the portable record of a chosen view.

Reproducible preparation

node scripts/build-precache.mjs node scripts/build-projective-precache.mjs Run from the app's source directory. The shared sampler computes every missing frame, writes lossless gzip assets, and updates dist/precache/manifest.json. Existing entries are reused only for the current numerical and codec version.

The implementation separates sampler.js and critical-samples.js from the thin sample-worker.js wrapper. field-cache.js defines exact versioned keys and bounded caches; field-service.js selects a source; playback.js prepares a complete bounded clip; cache-codec.js preserves the stored data. The build script invokes the same sampler used for a custom browser calculation. Codec, cache, sequence, and baked-frame tests complement the mathematical checks above; they do not strengthen those finite accuracy bounds.

10 / Interpreting a possible RH clue

Useful observation

A recurring sign pattern, a stable contour, or a candidate cancellation can suggest a precise inequality. Record its exact function and coordinate dependence.

Remaining mathematical task

Convert the observation into an identity or bound with the required domain, precision, and tail control. A finite picture cannot provide those quantifiers.

The recent Math Explorer notes distinguish exact geometric identities, finite positivity results, and the unresolved arbitrary-depth positivity problem. The fourth-stage Pick and later theta-ratio checkpoints do not claim a general preservation theorem or an RH proof.

In particular, the phase circulation of a fixed contour counts enclosed zeros without determining whether they lie on the critical line. Positive boundary kernels can have signed off-diagonal correlations. Two-sheet gluing and compact-looking pictures do not remove the essential punctures or their limiting boundary terms. These are specific mathematical distinctions that help keep an attractive pattern useful.

11 / Sources & research checkpoints

Additional bridge views follow the sign-switching and boundary-energy checkpoint. Heat-root collision conventions can be compared with Tao’s polynomial heat-flow derivation and DLMF’s argument principle.

The mathematical definitions below were consulted directly. Repository links point to immutable research commits so the conventions remain reproducible.