Gravity / A finite sum
With a stated cutoff, the sum over geometries is finite.
Quantum gravity as a sum over histories gives every spacetime geometry a phase and adds them up. That needs four things: which geometries, which action weighs them, which measure adds them, and a reason the sum exists. The finite-path-sum paper gives all four on the recognition lattice and states the status of each. The geometries form a cone of Lorentzian lattices, the action is Regge’s, the measure is a model built from ledger states, and with a stated regulator the sum is a finite integral whose expectations are defined at every phase scale except a discrete set. The regulator is stated, not derived.
Finite at a stated regulator: provedThe measure and the regulator: a modelThe sum without a regulator, and κ: open
- Admissible, κ > 1
- The null stratum, κ = 1
- Admissible, κ < 1: not causal
- Boundary rays
Live · exact, Definition 2.1
The uniform fields α|d| − β|d|², one ray each
51.3°
- The ray
- (1, 1.25)(α, β): the metric αI − βJ
- κ = β/α
- 1.25the ray is the flat lattice at this κ, and its multiples
- A posting
- 0.866 cits speed, v² = 3/(4κ − 1)
- Triangles
- 10 of 10timelike, H < 0
- Tetrahedra
- 5 of 5with P < 0; the simplex has G < 0: yes
- Edges
- 3 of 3classes of two to four postings timelike; the posting edges timelike
- The ball
- 8.9 × 10−8the explicit radius r(κ) around the flat field
Admissible for every κ > 1. Every simplex is Lorentzian and every edge timelike: the causal sector, with the flat lattice at κ = 5/4 inside it.
The null stratum, κ = 1. Every simplex is Lorentzian and the posting edges are null: the light-speed reading.
Admissible for every κ < 1, but not causal. Every simplex is Lorentzian, but the posting edges are spacelike: a posting would outrun light.
On the boundary of the cone. Not admissible: .
Not admissible. Here .
What is computed. On the field ℓ(x, d) = α|d| − β|d|² every Kuhn simplex has the same numbers, so one simplex is checked: H on its ten triangles, P on its five tetrahedra and G on the simplex, each the determinant of the doubled Gram matrix built from squared lengths, and the sign of each of the four edge classes. A field is admissible when every simplex is Lorentzian nondegenerate, every edge of two or more postings is timelike, and the posting edges have the sign of 1 − κ. By the cone property each ray is wholly inside or wholly outside. The shading is the same test on every half degree; the paper checked its Figure 1 on 3600 directions, and this page’s tests repeat that check. That no ray outside the two sectors is admissible is that computation, not a theorem. The explicit ball, every field within 8.9 × 10−8 of the flat lattice edge by edge at κ = 5/4, is crude by design: it gives a neighbourhood of the flat field, not a sharp one.
Finite-path-sum paper, Section 2, Definition 2.1, Theorems 2.2 to 2.5 and 2.11, and Figure 1
With JavaScript on, this figure tests any ray of the slice live. On the flat lattice at κ = 5/4 every triangle is timelike, every tetrahedron has P < 0 and the simplex has G = 16(1 − 4κ) < 0; the posting edges are timelike, so it is admissible for every κ > 1. The admissible rays are β > α > 0 (κ > 1), β = α (the null stratum) and α/2 < β < α (κ < 1, not causal).
The four requirements
Each requirement, at its own status.
Proved means a theorem with all its hypotheses stated; conditional, proved from a premise that is not derived; a model, a stated construction; open, a precise problem stated in the paper.
1 · The geometries
Proved An open cone of admissible geometries for κ ≠ 1, with an explicit ball around the flat lattice, and a null stratum at κ = 1; every admissible simplex has signature (3, 1); κ ≥ 1 is causal.
Model or conditional The fixed Kuhn graph, real squared lengths, the family n(1 − κn), time on the body diagonal; one diagonal in every cell, given a clock without folds.
Open The value of κ (Problem 2.14).
2 · The action
Proved Regge’s action, smooth and homogeneous on the cone, its first variation by Schläfli’s identity, the hinge sum for N ≥ 3, and continuity up to the boundary.
Conditional The Lorentzian sign and κ ≥ 1, given name-blindness and a causal reading of the cell; the weight past a null hinge, given a normal form.
Open The native phase scale; the continued action (Problems 3.6 and 3.10).
3 · The measure
Proved Finite, nonzero, with a density for every admissible family; the exact kernel at the balance; no Faddeev–Popov factor; the family, the radius and the weight change it; a measure on the stratum at κ = 1.
Model The ledger chart, the account family, the box radius and a flat bare weight.
Open Whether the bare weight is flat; whether admissible families agree up to sets of measure zero; whether the action is constant on displaced flat lattices; refinement (Problems 4.5, 4.8 and 4.11).
4 · Finiteness
Proved A finite integral at a stated regulator, also at κ = 1 (Finiteness).
Model The regulator: a bound on squared lengths, or on the ledger box.
Open The Abel limit of the uncut sum; reflection positivity at the cutoff; a derived cutoff (Problems 5.5, 5.7 and 5.12). Unitarity is not addressed.
Finite-path-sum paper, Section 1 and Table 1
One diagonal in every cell
A clock without folds puts the same diagonal in every cell.
Read one cell of the lattice as the records of four postings: a history is a chain of records, each holding one more posting than the one before. The histories of a cell are exactly its 24 Kuhn simplices, and a tick clock on the cell is the number of postings. Every cell carries its cut along the same diagonal when one global clock, without folds, is a tick clock on every cell; a clock that also advances along the axes makes it the Kuhn cut, with time on the body diagonal. This is conditional: the record reading of a cell, the global clock and the absence of folds are premises. The paper draws the two-dimensional analogue, live here.
- The Kuhn cut
- The reflected cut
- A fold
Live · exact, Theorems 2.7 and 2.8
Squares labelled with the clock, each cut from its lowest corner
The clock
- The Kuhn cut
- 9 of 18squares
- Reflected
- 9 of 18squares
- Ahead
- 6ticks, the most any corner shown runs ahead of the origin
No fold. Every square is read from its lowest corner and carries the Kuhn cut, along (1, 1). In four dimensions such a clock is T(0) + x0 + x1 + x2 + x3.
A fold at x0 = 0. Right of it every square carries the Kuhn cut, left of it the reflected cut. The two squares beside the fold read the same clock correctly and disagree on the diagonal.
Alternating cuts. The clock is a tick clock on every square, but it never runs more than two ticks ahead of the origin here, four in four dimensions, so it cannot advance along the axes.
What is drawn. Each square is read from the corner where the clock is lowest and cut along the diagonal from that corner, and the clock is checked to rise by one at every step away from it, so that it is a tick clock on the square. The paper’s Figure 2 is the fold clock |x0| + x1; in four dimensions the same construction gives its counterexample. On the torus an integer clock must fold, so coherent diagonals there come from the cyclic clock x0 + x1 + x2 + x3 mod N or from the lift to the infinite lattice. Of the 384 symmetries of the unit four-cube, the 48 that reflect no axis or all four keep the cut, and central inversion among them reverses the clock: it is the lattice’s time reversal.
Finite-path-sum paper, Theorems 2.6 to 2.10 and Figure 2
With JavaScript on, this figure draws the cuts for three clocks. With the fold clock |x0| + x1 the nine squares right of x0 = 0 carry the Kuhn cut and the nine left of it the reflected cut; with x0 + x1 all eighteen carry the Kuhn cut; with (x0 mod 2) + (x1 mod 2) the cuts alternate and the clock never runs more than two ticks ahead.
The action
Regge’s action, and where its sign comes from.
1 · Area times deficit
On the cone, the field equations are Regge’s equations everywhere.
The action adds, over every simplex and each of its ten triangles, the triangle’s area times its share of the deficit angle; on a torus of side N ≥ 3 this is exactly Regge’s sum of area times deficit over the triangles of the lattice. On the admissible cone it is smooth, every area is real and positive, and S(cℓ) = cS(ℓ). By Schläfli’s identity the variation of the angles drops out, so the first variation is the variation of the areas times the deficits, at every admissible field and not only near flat space.
Finite-path-sum paper, Theorem 3.1
2 · Why not sinh
Only a deficit function with a constant derivative keeps that property.
Replace the deficit by a function of it, f(ε). The first variation gains a term in f′(ε) times the change of the deficits; Schläfli’s identity removes it only when f′ is constant on the deficits, and for two hinges it vanishes for every datum exactly when f′ is constant. The variant f = sinh fails on the lattice itself: on the torus of side 4 there is a configuration on the null stratum, with a positive deficit at one hinge, at which its first variation is not the area form. The test selects Regge’s action; it does not supply independent physics.
Finite-path-sum paper, Theorems 3.2 and 3.3
3 · The sign
The tick order fixes the time axis; the minus sign is a premise.
A tick clock on one cell makes time run along the body diagonal, with no interval assumed. Given two premises, name-blindness and a causal reading of the cell, it fixes κ ≥ 1 and the overall sign. The minus sign itself comes from the premise that history steps are causal: name-blindness and the tick order alone also admit the Euclidean interval. Inside the admissible set every simplex has signature (3, 1) and the Euclidean Regge action vanishes identically, so the Lorentzian form is the only real Regge form that sees the geometry.
Finite-path-sum paper, Theorems 2.4 and 3.4
4 · The phase
The action does not fix the phase scale.
Each geometry is weighted by eiaS, with the phase scale a in the role of 1/ħ in lattice units. Stretching every squared length by c turns a into ac, so the action alone cannot fix a. The native candidate, from the posting action scale of the companion paper, is a = ln 2/(2π) ≈ 0.110 per unit of Regge action, with squared lengths in units of λrec²; that the ledger produces it is open.
Finite-path-sum paper, Proposition 3.5 and Problem 3.6
5 · Null hinges
The real action does not see a null hinge.
Where a triangle turns null its angle has a square-root branch point, but the angle leaves 0 or π at the same rate as the area that multiplies it vanishes. So the action is continuous up to the boundary of the cone, and the boundary has measure zero. Past a null hinge the action has to be continued. Under a stated normal form the weight there is suppressed, enhanced or of modulus one according to the sign of εaδ0, where δ0 is the limiting deficit and the prescription ε is a stated parameter, not a derived one. One complex action that continues S everywhere is open.
Finite-path-sum paper, Theorems 3.7 to 3.9 and Problem 3.10
Into a null hinge, and past it.
- End angle, α/π
- Middle angle, 1 − β/π
- Area of a span-2 triangle
Live · exact, Theorems 3.8 and 3.9
(a) The flat family as s falls to 1/2
0.750
- Area
- 0.354of a span-2 triangle, ½√(2s − 1)
- End angle
- 0.148πα at v0v1v2, falling to 0; over the area: 1.31
- Middle angle
- 0.295πbelow π, β at v1v2v3
- Identity
- 16.000 = 16.000sin²α · P3P4 = 4A²(−G)
- Closed forms
- all three agreewith the angles and the area computed here
(b) Past the hinge: the modulus of the weight, a = 1
The prescription, ε
Slot angles tending to π, k
- Deficit
- πthe limiting deficit δ0 = (2 − k)π
- At H = 4
- 4.81the modulus of the weight
Enhanced. With εδ0 > 0 the weight grows past the hinge: an irregular light cone, enhanced for this prescription and suppressed for the other.
Suppressed. With εδ0 < 0 the weight decays past the hinge: suppressed for this prescription and enhanced for the other.
Modulus one. With δ0 = 0, two angles tending to π, the weight keeps modulus one for both prescriptions: a regular light cone, like those of the flat family at s = ½.
What is computed. (a) comes from the Gram matrix of one Kuhn simplex of the flat lattice at s: each slot angle is the angle between the parts of the edges to the two other vertices that are orthogonal to the triangle’s plane, in the simplex’s metric. The paper’s closed forms are sin²α = (2s − 1)/(2(3s − 1)) and β = arccos(−s/(3s − 1)), with area ½√(2s − 1). At s = ½ the angle formula is 0/0: the end angle has no continuous extension there, while the action is continuous, and for N ≥ 3 every limiting deficit at that point is zero. (b) holds under a normal form, which is an assumption: a hinge term that continues as (δ0/4)u plus an even part real on the real axis, with u = √(−H). The prescription ε is a stated parameter, not a derived one.
Finite-path-sum paper, Theorems 3.8 and 3.9 and Figure 5
With JavaScript on, this figure computes the approach live. At s = 0.75 the span-2 area is 0.354, the end angle 0.148π and the middle angle 0.295π below π; all three fall to zero as s falls to ½. Past a null hinge, at a = 1 and H = 4, the weight has modulus eπ/2 ≈ 4.81 for ε = +1 and one angle tending to π, e−π/2 for three, and 1 for two.
The measure
A model measure, pushed from ledger states.
A measure on squared lengths could simply be chosen. The paper builds one from ledger states instead, says plainly that it is a model, and proves how each choice in it changes the sum, so that what is forced and what is chosen can be told apart.
1 · The chart
A ledger state reads a squared length on every edge.
A ledger state says how the accounts move: one move per posting direction, and a shift of the accounts at every vertex. Along an edge they add up to an account move, and the chart reads its squared length as the sum of the squares of the move, the ledger cost’s second derivative along it, minus κ times the square of the total moved. At the balance the chart reads exactly the flat lattice. The chart is a modelling reading; it is not derived.
Finite-path-sum paper, Definition 4.1 and Theorem 4.2
2 · The measure
Lebesgue measure on a box of states, cut to the cone, pushed through the chart.
The ledger measure is Lebesgue measure on the box of states within R of the balance, restricted to the states that read an admissible geometry, and carried to squared lengths by the chart. For κ ≠ 1 it is finite, puts no mass outside the cone, and is not zero for R > 0. For every family of accounts at which the chart is onto somewhere, it has a density.
Finite-path-sum paper, Theorems 4.2 and 4.3
3 · The choices
The accounts, the box and the weight each change the sum.
With too few accounts the measure lives on a set of measure zero. The box radius is not a normalization convention: a small box reads only fields near the flat lattice and a large one reads fields far from it. And weighting the states by the ledger’s own cost, in place of the flat weight, changes the expectations. Which bare weight the ledger carries is open.
Finite-path-sum paper, Theorems 4.4 and 4.10, Problems 4.5 and 4.11
4 · No gauge factor
No Faddeev–Popov factor.
Moving the vertices of a flat lattice inside flat space changes edge lengths but not the geometry. At the balance the ledger’s translations map exactly onto those vertex moves. The chart’s own redundancy is its kernel, whose dimension the paper gives exactly, and the vertex moves are not part of it. The integrand is unchanged along them only to second order at the balance, and for boxes of radius at least one the cut breaks the invariance outright, so no orbit volume factors out of the sum.
Finite-path-sum paper, Theorems 4.6 and 4.7
Finiteness
Along a ray the plain sum never settles; a cutoff makes it finite.
The action is homogeneous, S(cℓ) = cS(ℓ), and the cone of geometries is unbounded, so the sum over all of it is not absolutely convergent. Along one ray the weight is eiλc with λ = aS(ℓ): the plain integral runs round a circle forever, or grows like R on a ray where S = 0, such as the flat one. Damping each ray, Abel’s method, makes every ray with aS ≠ 0 converge; whether the damped sum over the whole cone converges is open. The paper takes the other branch: a stated cutoff.
- The integral up to R
- The damped limit
- The circle about i/λ
Live · exact, Theorem 5.6
I(R) = ∫0R eiλc − εc dc, in the complex plane
1.00
0.00
20.0
- I(R)
- 0.91 + 0.59ithe integral up to the cutoff
- Its size
- 1.09|I(R)|
- R → ∞
- none: it circlesthe damped integral to infinity, 1/(ε − iλ)
No limit as R → ∞. The integral runs round the circle of radius 1/λ about i/λ forever.
No limit. On a ray where the action is zero the integral is R itself, and grows without bound.
With damping it converges to 1/(ε − iλ), and as the damping is removed that tends to i/λ.
With damping it is 1/ε, which diverges as the damping is removed: on the flat ray, where S = 0, the Abel integral diverges.
What is drawn. I(R) = ∫0R e(iλ − ε)c dc from its closed form, (e(iλ − ε)R − 1)/(iλ − ε), and R itself when λ = ε = 0. The paper’s Figure 7(b) draws the undamped integral at aS = 1 and aS = 0. The cone’s true radial density is cD − 1 with D = 15N⁴, one real squared length per edge; with it the truncations diverge too, and Abel damping gives m!/(ε − iλ)m + 1 at power m. Any divergence of the damped sum must come from directions near the zero set of S, which contains the flat ray; the paper expects one there, as an inference, not a theorem.
Finite-path-sum paper, Theorem 5.6, Problem 5.7 and Figure 7
With JavaScript on, this figure draws the sum along one ray live. Undamped, at aS = 1 the integral runs round the unit circle about i forever, and at aS = 0 it grows like R; with Abel damping ε it converges to 1/(ε − iaS), which tends to i/aS as ε → 0 when aS ≠ 0 and diverges when aS = 0.
Bounded
CNL
On the box where every squared length is at most L in size, the action is at most CNL, with CN = 120√13 πN⁴, degenerate simplices included: the weight cannot become singular there.
Finite-path-sum paper, Theorem 5.1
Finite
π/2
With the cutoff L the sum is a finite integral, and when |a|CNL < π/2 its normalization Z is nonzero and ⟨1⟩ = 1. The ledger box of radius R has such a window too.
Finite-path-sum paper, Theorems 5.2 and 5.3
Entire
Z(a)
The normalization is an entire function of the phase scale, nonzero at a = 0, so its real zeros are isolated and expectations are defined at every phase scale except a discrete set. Only the product aL matters.
Finite-path-sum paper, Theorem 5.4
The native scale
4 × 10−5
At N = 4 the window |a|CNL < π/2 at the native scale a ≈ 0.110 ends near L = 4.1 × 10−5 (computed here from CN), while the flat field at κ = 5/4 has size 16: every box that holds it lies outside. The entire-function theorem covers the native scale up to a discrete set instead.
Finite-path-sum paper, Theorems 5.3 and 5.4
The conformal factor
Four negative directions in Lorentzian signature, one in the Euclidean cell.
In Euclidean gravity the conformal mode makes the action unbounded below, and its algebraic root is the DeWitt supermetric on strains. On the lattice backgrounds g = I − κJ its sign structure is exact: negative on a four-dimensional space of strains for every κ > 1/4, which includes every Lorentzian Kuhn background, and on a line, the conformal mode, for κ < 1/4, which includes the Euclidean cell κ = 0. Its bearing on the Lorentzian sum is indirect, because the Lorentzian weight has modulus one; it matters for the Euclidean continuation and for reflection positivity.
- A negative norm
- A positive norm
Live · exact, Theorem 5.8
The norms of the ten orthogonal strain modes, signed log scale
1.25
- n
- −1Tg−1T = 4/(1 − 4κ), which sets the time–space shears
- Index
- 4negative directions, from the eigenvalues of the form on all ten strains
Index 4. The conformal strain and the three time–space shears are negative: the classical signature (6, 4) of the DeWitt metric on Lorentzian metrics.
Index 1. Only the conformal strain is negative, as on Riemannian backgrounds; the time–space shears are positive.
Degenerate. At κ = 1/4 the background has det g = 1 − 4κ = 0.
What is computed. The DeWitt form Gg(h, k) = tr(g−1hg−1k) − ½ tr(g−1h) tr(g−1k), on the ten modes of the paper’s proof. In the frame T = (1, 1, 1, 1), w1 = (1, −1, 0, 0), w2 = (1, 1, −2, 0), w3 = (1, 1, 1, −3) the inverse background is diagonal, diag(n, 2, 6, 12), and the modes are the conformal strain h = g, the time–space shears T ⊗ wi + wi ⊗ T, the space–space shears wi ⊗ wj + wj ⊗ wi and three trace-free diagonal modes. Their norms are −4; 4n, 12n, 24n; 24, 48, 144; 2, 6, 12. The index is counted separately, from the eigenvalues of the form on all ten dimensions of strains.
Finite-path-sum paper, Theorem 5.8 and Figure 8
With JavaScript on, this figure computes the DeWitt norms live. At κ = 5/4, n = −1 and the norms are −4; −4, −12, −24; 24, 48, 144; 2, 6, 12: index 4. At κ = 0, n = 4 and the time–space shears are 16, 48, 96: index 1.
The lattice has a static resonance, a zero mode at one static wave for every κ > 1/2 that is not a vertex move. A numerical survey reads the branch that crosses zero there as the conformal mode, negative near k = 0 and turning positive at the resonance. That reading is evidence, not a theorem; the paper proves two exact facts that bracket it, and the reading itself is its Problem 5.10.
Finite-path-sum paper, Proposition 5.9 and Problem 5.10
At κ = 1
At the light-speed reading, the same finite sum on a thinner set.
κ is the speed of one posting: v² = 3/(4κ − 1), so κ = 1 is a posting at the speed of light and κ = 5/4 is v = (√3/2)c. The cell’s premises leave every κ ≥ 1 open, and deriving κ is the paper’s Problem 2.14. At κ = 1 the posting edges are null, and the configuration space loses a dimension per posting edge.
1 · The stratum
The null stratum has measure zero, so it needs its own measure.
At κ = 1 the admissible fields are those with every Kuhn simplex Lorentzian, every edge of two or more postings timelike and every posting edge null: a set inside the subspace where all posting edges are null, open there, with an explicit relative ball of radius 1/(144 · 37³) about the flat field, and of Lebesgue measure zero. A finite account family built by averaging over the torus’s translations carries a measure on it that equals the subspace’s own measure near the flat field.
Finite-path-sum paper, Theorems 2.5 and 4.12
2 · The sum
Finite, with a nonzero normalization, and a regulator still needed.
With the subspace’s measure cut at L above the size of the flat field, the sum on the stratum is finite and nonzero, and its normalization is nonzero throughout |a|CNL < π/2. The stratum is a cone too, so without a cutoff its measure is infinite and the weight is not integrable. Every clause of the action carries over to κ = 1. Proved only off the stratum: the explicit ball, the classification of the frontier, the reasons for the fixed graph, the gauge and kernel statements, the statements on dilations and on translation-invariant measures, and consistency under restriction.
Finite-path-sum paper, Theorem 5.13 and Section 6
What is open
Three problems decide the next step.
Deriving the posting speed fixes κ, and with it the reading on which every theorem is applied. Deriving the bare weight and the regulator from ledger dynamics turns the model measure into a derived one. And reflection positivity at the regulator decides whether the lattice sum has a positive transfer matrix. Underneath all of them is the derivation of the lattice itself.
Each open problem is listed with its requirement in the table above.
Finite-path-sum paper, Section 6 and Figure 9
Paper cited: the finite-path-sum paper (full title).
Words on this page, in plain terms: path sum · admissible geometry · null stratum · phase scale · regulator · Kuhn lattice · squared edge length · κ · posting · Regge action · deficit angle · vertex move · tick.
The gravity portal · The lattice · The quantum question · Glossary · Search every page