Stop 7 of 9 · Why does the Dirac operator on T³ see eight corners?

The eight doublers — hopping, Dirac and their sum

The hopping H = −2t T° moves an excitation from site to site; its energy −2t Σ cos vanishes on a whole plane of momenta. With the Kogut–Susskind signs the three axes anticommute and the energy becomes a length, ±√(Σ cos² + m²): it vanishes at one point only, where the eight doublers sit, the corners (±π/2)³. The square is the sum of the three axes, D² = T_x² + T_y² + T_z² + m².

Now close a fourth, small axis: a ring of three positions, the fibre w. Its turning modes act as a mass of ½. Two things follow: no energy is ever smaller than ½ (a mass gap), and, because the other axes are closed too, no state of definite energy moves faster than 1/φ = 0.618… of the light cone. Further down, slide the mass and watch both appear.

The torus: the hopping state

size |Ψ|² · colour phase

The spectral star

⟨cos⟩ on each axis, spread of the cosine

The eight doublers, spectral side: hopping, Dirac and their sum

follows the torus, the mode and Ψ(s) above · drag any panel to turn all

Everything here is the torus above, at the same time: its modes, the occupied mode (○), Ψ(s) (●, at ⟨cos⟩ on each axis, at ⟨cos²⟩ in the squares) and the turn. Every mode (a, b, c) of the torus sits at (cos θa, cos θb, cos θc) in the cube of cosines. The hopping E = −2t Σ cos vanishes on a whole plane, Σ cos = 0. With the Kogut–Susskind signs the three axes anticommute and the energy becomes a length, E = ±√(cos²θa + cos²θb + cos²θc + m²): it vanishes at one point only, the origin, where the eight doublers (±π/2)³ sit. The square root turns the zero plane of the hopping into the zero point of Dirac. Below, their sum: in the squares (cos², cos², cos²) the Dirac energy squared is the sum of the three axes, D² = Tx² + Ty² + Tz² + m². The fibre of 3 positions gives the mass ½: the origin opens into the gap |E| ≥ ½.

Hopping: E = −2t Σ cos, zero on a plane

Dirac: E = ±√(Σ cos² + m²), zero at one point

Sum: the squares, E² − m² = cos² + cos² + cos²

Lean (nrs3-tritoro): TorusDirac3.dirac3_sq (Dm² = Tx² + Ty² + Tz² + m² on T³, even sizes), the anticommutations Tx_Ay, Tx_Az, Ay_Az, eps_Tx, eps_Ay, eps_Az, and dirac3_wave (the eight waves (±i)x(±i)y(±i)z are zero modes of D₀, and the mass sends each to the opposite corner). The spectrum of each ring is the Lean TorusSpectrum.eq_cos_of_eigenvector. The fibre as fourth axis is TorusFiberDirac.dirac_sq; its rotating modes give the mass ½ and the gap (TorusFiberDirac.dirac_doublers, TorusFiberDirac.abs_energy_ge).

The mass gap and the speed limit

Take one closed axis and a mass m. Each state has a momentum k and an energy on one of two bands, E = ±√(cos²k + m²). Its velocity is the slope of its band. With m = 0 the bands touch (the doublers) and the slope reaches the light cone. With a mass a gap opens, the bands round off, and no slope can exceed r = √(1 + m²) − m. The fibre gives m = ½, and then r = 1/φ: the same number as the largest tension of a minimum-uncertainty state of four positions.

The hopping band of every ring

Each ring of d positions has hopping eigenvalues −2t cos(2πm/d), m = 0, …, d − 1. They fill [−2|t|, 2|t|] and touch both ends: m = 0 gives −2t, and for even d the alternating mode gives +2t. ○ the occupied mode · ● ⟨H⟩ of Ψ(s), with the spread σH. At s = 0 the state is one site and ⟨H⟩ = 0 with the spread open. At s = 1 the dot sits on the mode and the spread closes. Press Tour to stay on one torus, Hop to pause.

The octahedron of the band

drag to turn · dashed: the band edge

Put a vertex at ±1 on every axis: that octahedron is the whole cosine band, |λ| ≤ 1 on each ring, the same bound for every size (TorusSpeed.abs_energy_le_one). The inner octahedron has vertices at the occupied |cos(2πm/d)|. On the canonical 3 × 3 × 3 the modes are 1 and −1/2, so a single axis occupies either the outer vertex or 1/2. The star page draws δ, the unresolved share. This octahedron draws the resolved spectrum.

cos on x, y, z–
E = −2t Σ cos–
|E| / (6|t|)–
σH of the torus–

Energy spread against position spread

σH against σP, one panel per axis

The complement of the tension plot. As s runs from the site to the mode, σH falls and σP rises. At s = 1 the mode is an eigenvector of H, so σH = 0 and the position fills the ring. At s = 0 the site is an eigenvector of P, so σP = 0 and the energy fills the neighbours. The product stays positive in between: concentrating one observable opens the other.

family–
mode (mx, my, mz)–
E at the mode–
the complement–
axispositionsmcos(2πm/d) ε = −2t cos⟨H⟩(s)σH(s)σP(s)

What is exact here

Theorem trail
StatementName
anticommutationTorusDirac3.Tx_Ay, TorusDirac3.Tx_Az, TorusDirac3.Ay_Az
D_m² on T³ and on the open cubeTorusDirac3.dirac3_sq, OpenDirac.dirac_sq
the eight doublersTorusDirac3.dirac3_wave
the spectrum of the ringTorusSpectrum.eq_cos_of_eigenvector
the fibre as fourth axisTorusFiberDirac.dirac_sq, TorusFiberDirac.dirac_tensor
mass gap |E| ≥ ½FiberDirac.abs_energy_ge, TorusFiberDirac.abs_energy_ge, TorusFiberDirac.dirac_doublers
speed limit √(1 + m²) − |m|, any spaceDiracSpeed.abs_velocity_le, DiracSpeed.speedLimit_half, DiracSpeed.speedLimit_sinh
1/φ on the three axes of T³ × wTorusFiberDirac.abs_velocity_le_inv_goldenRatio_axes
the same 1/φ as minimum uncertaintyMinUncertaintyFiber.isGreatest_minUncertaintyTension

← Stop 6The pathStop 8 →