Gravity / The doublers
The lattice has four more massless waves.
At κ = 1 the lattice operator vanishes on the light cone at long wavelengths, as gravity requires, and also at four corners of the Brillouin zone, the waves with one coordinate equal to π. These four doublers are massless, each on a cone of its own. At one loop, in the stated diagram model, they make the native graviton’s limit diverge; a weight built from the ledger’s cost of curvature gives all four a gap.
Four doublers at κ = 1: provedGapped by the curvature weight: proved under two identificationsA massless graviton: open
- Zero modes, graviton sector
- Zero modes, doubler sector
- Leading-order cones
Live · solved from Appendix B
Solving the lattice equations on this plane
Each curve is where the gauge-fixed κ = 1 operator, assembled in your browser from the Hessian printed in Appendix B, has a zero mode on the plane k = a e₀ + b(1,1,1,1), found sector by sector under the permutations of axes 1, 2 and 3. Near the origin the graviton’s branch follows the light cone, which opens around record time (1,1,1,1); near πe₀ the doubler’s branch follows its own cone, which opens around a direction close to e₀. Shading marks each cone’s timelike directions. Away from the two points the branches are lattice modes of the same symmetry.
- Record time
- −4/3q: timelike for the graviton
- 190hᵀKh: spacelike for the doubler
- Shared
- none30q + hᵀKh is positive definite (Theorem 6.15)
The corners
Sixteen corners, four doublers.
The corners of the zone are πδ with δ a string of four bits, so they carry the same labels as the records of a cell. Your browser solves the operator at all sixteen and counts its zero modes.
- 1111·
- 1110·1101·1011·0111·
- 1100·1010·1001·0110·0101·0011·
- 1000·0100·0010·0001·
- 0000·
The kernel has dimension 10 at the origin, where it is the constant strains; 1 at each axis corner, where it is a doubler; and 0 at the other eleven corners. At the corner πec the kernel is spanned by vc(d) = 2|d|−3 on the seven classes with dc = 0 and 0 on the other eight. The Bloch matrix and the vertex-move columns annihilate it separately, so it is a zero mode for every gauge weight, and it is not a vertex move.
Check · solving
Earlier paper, Proposition 6.1 and Section 6.2
The doubler at πe₀, in position space
Exact · Proposition 6.1
The field is (−1)x₀v₀(d): at every vertex it changes the squared lengths of the seven edges in the cube orthogonal to axis 0 by ¼, ½ and 1 for one, two and three steps (line weight), and the sign flips with each step along axis 0 (solid ink +, dashed red −). Edges that step along axis 0 do not change. As a check, 3·(¼)² + 3·(½)² + 1² = 31/16 = |v₀|². Drawn at one vertex of each layer.
On a finite torus
An odd torus carries no doubler; an even one does.
On a torus of side N the lattice’s waves are k = 2πm/N. The corners πec are among them exactly when N is even. At κ = 1, on every odd side not divisible by 105, the paper proves that the flat operator’s only zero modes are the constant strains and the vertex moves; on every even side a doubler at an axis corner is a zero mode that is neither. Your browser solves every wave of the torus you pick.
- A wave with no zero mode beyond the gauge
- A wave with one
Live · Theorems 4.2 and 4.4
Each wave’s smallest eigenvalue over its largest, log scale
The side of the torus, N
- Side
- 5cells along each of the four axes
- Waves
- 624 of 624solved, every m but 0
- Null waves
- nonewith a zero mode beyond the gauge
- At axis corners
- 0of them, at πec, the doublers
- Elsewhere
- 0of them, found here by computation
- Smallest gap
- 1.7 × 10−6the least ratio among the other waves
Solving every wave of this torus.
No doubler. No wave of this torus carries a zero mode beyond the constant strains and the vertex moves.
Doublers. The four axis corners are waves of every even torus, and each carries a doubler, as the paper proves. The other null waves are found here by computation; the paper’s statement is that even sides fail.
What is computed. At every wave of the torus other than the origin, the gauge-fixed operator A(k) = B(k) + ΓΓ† of Appendix B, which makes the vertex moves cost something, is solved for its fifteen eigenvalues; a wave counts as null when the smallest is below 10−11 of the largest. On the sides 3 to 9 every null wave sits near 10−16 and every other wave above 10−7, so the line is not delicate. The paper proves the odd sides by exact arithmetic: one integer polynomial with 741 terms, the gauge-fixed Bloch determinant, has no zero at a torsion point other than 1 whose order is odd and not a multiple of 105; for odd multiples of 105 the question is open. This figure checks the same statement in floating point on small tori; it is not a proof.
Where the other null waves lie, computed here and not taken from the paper. On sides 4 and 8 the eight are k = π(½, 1, 1, 1), its images under permuting the four axes, and their negatives. On side 6 the 86 have coordinates that are multiples of π/3, in five families under permuting the axes and k → −k; two of them are the waves ±(π/3)(1, 1, 1, 1) along the time diagonal, each with two zero modes. On side 10 there are none: only the four axis corners are null. Each family is confirmed singular in exact arithmetic, and every one of these null waves is a genuine zero mode, not a strain or a vertex move.
Earlier paper, Theorems 4.2 and 4.4
With JavaScript on, this figure solves every wave of a torus of side 3 to 10. On the odd sides 3, 5, 7 and 9 no wave carries a zero mode beyond the gauge. On side 4, 12 waves do: the four axis corners and eight more; on side 6, 90 waves: the four corners and 86 more; on side 8, again the four corners and eight more; on side 10, only the four corners.
How it works
Why a real-time lattice has doublers.
1 · The cone runs out
A Lorentzian symbol vanishes on a cone, and a cone is unbounded.
On a lattice that cone runs out to the edge of the Brillouin zone, so the operator can vanish far from the origin. A Euclidean lattice operator that is positive away from the origin has no such zeros. Lattice fermions have doublers for a different reason; these belong to real-time discretizations, and at κ = 1 they sit exactly at the axis corners because the postings are null.
Earlier paper, Section 6.2
2 · A cone of its own
Each doubler is massless and propagates on its own cone.
Near the corner, the doubler’s effective operator is (√3/640) hᵀKh at leading order, where h is the offset from the corner and K has one negative and three positive eigenvalues, with det K = −1013375 = −55²·335. The two kinds of cone point different ways: record time is timelike for the graviton and spacelike for the doubler, and no direction is timelike or null for both.
Earlier paper, Proposition 6.2 and Theorem 6.15
3 · No mass from the construction
Nothing in the construction gives the doublers a mass.
A flat background that keeps every posting null shifts the four zeros, at first order, by amounts that sum to zero, so it cannot lift all four, and any nonzero shift of that kind makes at least one doubler tachyonic. A strain that moves all four the same way changes the posting lengths: raising κ above 1 gives all four a mass, and it leaves κ = 1. In the whole family κ > ½ the doublers occur exactly at κ = 1, one of four equivalent readings of κ = 1.
Earlier paper, Proposition 6.3 and Theorem 6.10; quantum gravity paper, Theorems 2.1 and 2.2
One loop
At one loop the doublers make the native limit diverge.
In the stated diagram model, the zero-momentum anisotropy of the graviton’s one-loop self-energy grows like log(1/ε) as the regulator ε is removed after the volume, so the ordered limit does not exist. The logarithm comes from the four doublers.
The doubler term D(ε)
Live · quadrature of Proposition 6.5
10⁻⁶
- Area
- 12.931of the integrand against ln τ
- D(ε)
- 8.13× 10⁻³
- Per decade
- 1.45× 10⁻³, added for every factor of ten in ε
D(ε) = (5/(44√335 π²)) ∫₀^∞ e−ετ τ²(1 + τ²)−3/2 dτ, computed here by quadrature; its asymptote, dotted, is (5/(44√335 π²))(log(1/ε) + log 2 − 1 − γ), so the anisotropy is Δ(ε) = (5/(44√335 π²)) log(1/ε) + r₀ + o(1), about 6.3 × 10⁻⁴ log(1/ε). Each axis corner is a pinch: the doubler’s eigenvalue vanishes quadratically there, so no deformation of the integration contour avoids the iε prescription. Every other degenerate zero contributes finitely, so nothing cancels the corners’ logarithm. The logarithm is slow: at every regulator a computation can reach, the anisotropy is dominated by its finite part; the divergence is a property of the limit.
Earlier paper, Propositions 6.5, 6.6 and 6.9, Theorems 6.7 and 6.8
The repair
A weight from the cost of curvature gaps all four.
On flat deformations
0
The weight vanishes on constant strains and vertex moves, the lattice’s flat directions.
Quantum gravity paper, Theorem 5.5
On long waves
|k|⁴
It is of fourth order in the wave number on long waves, so it leaves the classical limit alone.
Quantum gravity paper, Theorem 5.5
On the doubler v₀
0.12255
(10 + 3√3)/124: what the weight gives the doubler at πe₀, per |v₀|². Positive on all four doublers: a gap. The doubler is a zero mode of the action but not a flat deformation: its linearized deficits do not all vanish, which is what the weight reads.
Quantum gravity paper, Theorem 5.3 and Section 5.3, by exact computation
Free coefficients
0
None, once the channel count is fixed, given the scaling that makes the phase native.
Quantum gravity paper, Theorem 5.4
What it rests on
Two named identifications: the cost of a hinge record reads its area, as the action does; and the weight of a geometry is the Gibbs weight e−C of its ledger cost. With them, the ledger’s cost of the curvature at each hinge enters the measure at the action’s scale, and at Gaussian order it acts through 𝒟 = ½ Σh (Ah/nh) δ′hδ′h⊤.
Quantum gravity paper, Theorem 5.4
What it gives
With the weighted operator, in the same diagram model with the Feynman regulator, the real part of the anisotropy that diverges for the native path sum has a finite ordered limit, and so do the real parts of each probe’s contribution and of each diagram. A finite limit is not a massless graviton, which is the open target.
Quantum gravity paper, Theorem 7.4 and Target 6.8
Papers cited: the earlier paper and the quantum gravity paper (full titles).
Words on this page, in plain terms: doubler · zero mode · anisotropy · Brillouin zone · light cone · κ (kappa) · ordered limit · curvature weight · one-loop diagram model.
The gravity portal · Newton’s law from the lattice · Next: the quantum question · The doublers in the full account · One loop in the full account · The earlier paper (PDF) · The quantum gravity paper (PDF) · Glossary · Search every page
With JavaScript on, this page solves the lattice operator live: where it vanishes on one plane of waves, its zero modes at the sixteen corners of the zone, and the one-loop doubler term.