Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion .github/workflows/periodic-h0-pilot.yml
Original file line number Diff line number Diff line change
Expand Up @@ -43,9 +43,11 @@ jobs:
run: python -B -S experiments/periodic_h0/hessian_grid_certificate.py --verify experiments/periodic_h0/results/hessian1
- name: Replay exact integer H0 and independent connectivity witnesses
run: python -B -S experiments/periodic_h0/exact_barcode_certificate.py --verify experiments/periodic_h0/results/exact_h0_1
- name: Replay the ideal Gaussian spectral tail and conditional coupling boundary
run: python -B -S experiments/periodic_h0/gaussian_tail_certificate.py --verify experiments/periodic_h0/results/gaussian_tail1
- name: Execute the complete fixed pilot
run: |
python -B experiments/periodic_h0/run_pilot.py --output "$RUNNER_TEMP/periodic-h0-replay"
python -B experiments/periodic_h0/run_pilot.py --verify "$RUNNER_TEMP/periodic-h0-replay/observations.json"
- name: State scientific scope
run: echo 'Tests, exact finite-field barcode bounds and a fresh discrete periodic H0 pilot passed. Ideal Gaussian/infinite-field error control, coefficient confirmation and scientific acceptance remain separate.' >> "$GITHUB_STEP_SUMMARY"
run: echo 'Tests, exact finite-field barcode bounds, ideal Gaussian tail bounds and a fresh discrete periodic H0 pilot passed. Coupling the historical coefficients to the ideal ensemble, lifetime-law confirmation and scientific acceptance remain separate.' >> "$GITHUB_STEP_SUMMARY"
9 changes: 6 additions & 3 deletions experiments/periodic_h0/CONFIRMATION_READINESS.md
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,8 @@ paired differences, finite-sample error, and the last refinement remain relevant
| Certified numeric nodal input | [New seed34000 / 128² snapshot](NODAL_CERTIFICATE.md), every node independently enclosed | Nodal error and abstract finite-field diagram bound supplied for this snapshot only; historical samples and computed barcode endpoints remain uncertified |
| Sharper finite-field spatial bound | [Hessian/grid certificate](HESSIAN_GRID.md), with a 256² derivative grid and a separate 1024² dyadic sample grid | Exact abstract-filtration error evaluated for deterministic new samples; no transfer of historical histogram counts |
| Exact fine-grid H0 endpoints | [Integer barcode and independent connectivity verification](EXACT_H0.md) | One C38 grid and its frozen finite polynomial only; exact endpoints and deterministic bin-count sandwich |
| Infinite-field truncation | Variance diagnostics through mode64 in pilot32 | No uniform realized or probabilistic tail enclosure |
| Ideal infinite-field truncation | [Exact Gaussian shell/theta bound](GAUSSIAN_TAIL.md), including all modes beyond64 and the normalization correction | Uniform probabilistic tail supplied for the explicitly defined ideal field |
| Coupling the stored coefficients to the ideal field | Explicit low-mode sup-norm/coefficient contract in the same proof | rho remains unknown; no full-field probability or count bound for the frozen polynomial |
| Lifetime range | Larger bins exhibit numerical stability; shortest bins still drift | No certified finite-lifetime range for the leading law |
| Remainder | Source-bound qualitative O(1) theorem with its hypotheses | No applicable numerical remainder constant or cutoff supplied here |
| Sampling precision | Eight coupled fields, with per-field counts and paired differences | Descriptive precision only; no simultaneous confidence claim |
Expand Down Expand Up @@ -60,7 +61,9 @@ available and **before** its fields are observed. Its specification must record:
No held-out seeds have been consumed by this delivery. Assigning a sample size
or a pass threshold now would conceal missing analytic inputs. The barcode
computation on the certified fine dyadic grid now has exact integer
endpoints and separate connectivity checks. The next scientific inputs are a
justified model, tail and remainder budget.
endpoints and separate connectivity checks. A quantified ideal-field tail is
now available. The next scientific inputs are a certified low-mode coupling
and a numerical remainder budget; neither is supplied by a historical seed
label or an ideal-tail probability alone.
The finite-polynomial derivative input is now available. The source-bound manuscript
and literature work can advance in parallel without waiting for that budget.
242 changes: 242 additions & 0 deletions experiments/periodic_h0/GAUSSIAN_TAIL.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,242 @@
# From a finite polynomial to an ideal Gaussian field

[Evaluated bounds](results/gaussian_tail1/RESULTS.md) ·
[Exact finite barcode](EXACT_H0.md) · [Remaining inputs](CONFIRMATION_READINESS.md)

This note supplies a uniform probabilistic bound for the omitted modes of a
specified ideal Gaussian field. It also states the additional coupling needed
to apply that bound to a certified finite polynomial. The stored PCG64/NumPy
coefficients have no such coupling certificate. Their total infinite-field
barcode error therefore remains unknown. No new random samples are used.

## 1. Exact ensemble and normalization

Set L=24, beta=pi²/L², alpha=2 beta, and

```
S_c,M = sum_{j=-M}^M exp(-c j²), S_c = sum_{j in Z} exp(-c j²).
R = {(x,y) in Z² : x>0 or (x=0 and y>0)}.
R_K = {k in R : ||k||_infinity <= K}.
```

On one probability space let A_0 and every A_k,B_k, k in R, be independent
standard real Gaussian variables. Define

```
F(x) = A_0/S_alpha
+ sqrt(2)/S_alpha sum_{k in R} exp(-beta |k|²)
[A_k cos(2 pi k·x/L) + B_k sin(2 pi k·x/L)].
```

For each derivative order the expected sum of absolute coefficient majorants
is finite, since polynomially weighted Gaussian-decay lattice sums converge.
Tonelli's theorem gives almost-sure absolute uniform convergence of each
derivative series. The intersection over integer derivative orders still has
probability one; F has a smooth version. Finite sums and their L² limits are
centered Gaussian and have covariance

```
E F(x)F(y) = sum_{k in Z²} exp(-alpha |k|²) exp(2 pi i k·(x-y)/L)
/ S_alpha².
```

This has variance one. The Fourier coefficient of
`sum_{n in Z²} exp(-|h+Ln|²/2)` is
`(2 pi/L²) exp(-2 pi²|k|²/L²)`: integrate over a period, unfold to R²,
and use the Gaussian Fourier integral. Dividing the periodized covariance by
its value at zero cancels the prefactor and gives the expression above.
This is the normalized periodized Bargmann–Fock covariance. The underlying
Fourier/Poisson identity is also given in
[DLMF §1.8(iv)](https://dlmf.nist.gov/1.8.iv); no external tail theorem is used.

Let F_K retain the DC term and modes R_K with this infinite normalization.
Let F64,K use the **same ideal Gaussian variables**, but replace S_alpha by
S_alpha,64. Thus

```
F_K = gamma F64,K, gamma = S_alpha,64/S_alpha in (0,1).
```

The mathematical comparison ensemble follows the convention in
`experiment.py`: its denominator D64 is S_alpha,64², and its complex
coefficient is `exp(-beta |k|²)(A_k-iB_k)/(sqrt(2) S_alpha,64)`.
This identification of formulas does not certify the law or numerical
accuracy of the implementation. Changing the generator's maximum cutoff
also changes how it assigns random draws to modes; equal seed labels at
different maximum cutoffs do not establish a common coefficient bank.

## 2. A simultaneous tail event

There are exactly 4m representatives in the shell `||k||_infinity=m`.
For each pair define R_k=sqrt(A_k²+B_k²). Integrating the two-dimensional
Gaussian density in polar coordinates gives

```
Pr(R_k > v) = exp(-v²/2), v >= 0.
```

Fix integer K>=1 and t>0. On the event E_K,t, require simultaneously

```
R_k <= t+j, ||k||_infinity=K+1+j, j=0,1,2,... .
```

The countable union bound, and `(t+j)²/2 >= t²/2+t j`, imply

```
Pr(E_K,t^c)
<= sum_{j>=0} 4(K+1+j) exp(-(t+j)²/2)
<= eta_K,t
:= 4 exp(-t²/2) [(K+1)/(1-exp(-t)) + exp(-t)/(1-exp(-t))²].
```

The permissible probability upper bound is min(1,eta_K,t). This bound does
not use independence between different tail modes; independence of tail
variables from retained variables will matter for conditional statements.

## 3. Uniform norm on that event

The inequality `|A cos(theta)+B sin(theta)|<=sqrt(A²+B²)` holds at every x.
The exact shell identity is

```
sum_{k in R, ||k||_infinity=m} exp(-beta |k|²)
= exp(-beta m²) [S_beta,m + S_beta,m-1].
```

For example, subtracting full square sums gives
`S_beta,m²-S_beta,m-1² = 2 exp(-beta m²)(S_beta,m+S_beta,m-1)`;
the half-plane contains one from each equal ± pair. Therefore on E_K,t,

```
||F-F_K||_infinity <= tau_K,t
:= sqrt(2)/S_alpha sum_{m=K+1}^infinity
(t+m-K-1) exp(-beta m²) [S_beta,m+S_beta,m-1].
```

This is a uniform bound on the whole continuous torus, not a grid-point
variance calculation. It holds with probability at least 1-min(1,eta_K,t).

## 4. Infinite remainders, including modes beyond64

For c>0 and M>=0, successive ratios of exp(-c j²), j>=M+1, are at most
`q_c=exp(-c(2M+3))`. Hence the **two-sided** theta remainder satisfies

```
0 < S_c-S_c,M <= 2 exp(-c(M+1)²)/(1-q_c) = T_c,M.
```

For M>=K the part of the amplitude sum above M is bounded by

```
U_K,t,M = 2 S_beta exp(-beta(M+1)²)
[(t+M-K)/(1-q_beta) + q_beta/(1-q_beta)²].
```

Indeed m=M+1+j gives weight t+M-K+j, and the exponential ratio is at most
q_beta^j. Use S_beta,m+S_beta,m-1<=2S_beta and sum the geometric series and
its first moment. The implementation sums shells through M=64 and includes
this positive infinite remainder. It uses upper bounds for numerator terms
and a lower bound for S_alpha. Nothing sets an omitted sum to zero.

The normalization mismatch has the separate deterministic bound

```
1-gamma = (S_alpha-S_alpha,64)/S_alpha
<= T_alpha,64/S_alpha,64 = delta_norm.
```

Computing the numerator by a remainder formula avoids subtracting nearly
equal interval enclosures. For this model the evaluated delta_norm is less
than 3×10^-64. It is a relative multiplier, not itself a sample error.

## 5. The additional coupling premise

Let P be a specified real finite polynomial with cutoff K<=64 and certified
norm upper bound B_P. Suppose a comparison on the same probability space
supplies the pathwise premise

```
||F64,K-P||_infinity <= rho.
```

On E_K,t, the triangle inequality and F_K=gamma F64,K give

```
||F-P||_infinity
<= tau_K,t + gamma rho + (1-gamma)||P||_infinity
<= tau_K,t + rho + delta_norm B_P.
```

A sufficient coefficient-level premise for rho is

```
|A_0/S_alpha,64 - c_0|
+ 2 sum_{k in R_K} |exp(-beta |k|²)(A_k-i B_k)/(sqrt(2) S_alpha,64)-z_k|
<= rho,
```

where `P=c_0+2 Re sum z_k exp(2 pi i k·x/L)`. This is an explicit
obligation, not an assertion that rounded pseudorandom coefficients meet it.

If an event C establishing this premise is measurable with respect to the
retained Gaussian variables and Pr(C)>0, independence gives
`Pr(E_K,t | C)=Pr(E_K,t)>=1-min(1,eta_K,t)`. If C can depend on tail variables,
that conditional conclusion is unavailable. A separate bound Pr(C^c)<=p
would instead give an unconditional success bound at least 1-p-eta_K,t by
the union bound. We supply neither C nor p for the stored polynomial.
Conditioning a fixed polynomial on rho=0 is generally conditioning on a
probability-zero event and is not justified here.

## 6. Consequence for a genuinely coupled barcode

Assume an exact finite barcode Q already satisfies
`d_B(D(P),Q)<=epsilon_finite`. Uniform function error gives shifted sublevel
inclusions. The q-tame matching theorem and continuous finite-polyhedron
scope used in [APPROXIMATION.md](APPROXIMATION.md) apply to both smooth
functions on the torus. On the stated coupling and tail events,

```
d_B(D(F),Q) <= epsilon_finite + tau_K,t + rho + delta_norm B_P.
```

The endpoint-safe half-open bin sandwich from that note applies with this
total error; a clean upper count still requires a>2 epsilon_total. One
essential H0 class is separate and no longest finite bar is discarded.
This is a samplewise implication. A high-probability diagram bound alone
does not control an expected bar count on the exceptional event; an
appropriate count-moment/tail argument would be an additional input.

The C39 object is one K24 rounded polynomial and a certified 1024² dyadic
barcode. Its finite error is available and B_P<16, but rho is not. The new
certificate records both rho and the full-field diagram bound as null.
The K32 examples illustrate a possible different ideal truncation; they
are not a certificate for an uncomputed K32 sample or the historical pilot.

## 7. Arithmetic and replay boundary

`finite_certificate.py` supplies the existing rational Machin bounds for pi
and integer square-root bounds. In `gaussian_tail.py`, exp(-x) is enclosed
for rational 0<=x<=4096 by reducing x=2^s y with y<=1/8. The alternating
Taylor polynomial P49(y) is a lower bound and P48(y) an upper bound.
Terms decrease in magnitude. Rounding the lower endpoint down and the
upper up to denominator 2^256 preserves enclosure. Repeated squaring,
with outward rounding after each square, encloses exp(-x). Zero is exact.
All endpoints remain nonnegative. Decay bounds use the lower pi endpoint
for upper exponentials and the upper pi endpoint for lower exponentials.
Every subsequent operation uses rational endpoint monotonicity.

The certificate includes the exact rational shell sums, positive infinite
remainders, theta enclosures and probability bounds. Decimal display values
are rounded upward. Tests include direct independent alternating sums,
half-lattice shell identities, multiplicity checks, positive omitted-tail
checks, invalid-input/cached-type controls, and rejection of invented
coupling or scientific-promotion fields. The wrapper binds the unchanged
C39 certificate, original receipt and consumed sources before adding the
new result. Exact replay checks the displayed report as well as its hashes.

The result closes this explicit ideal-field uniform-tail calculation.
It does not certify a random-number generator, the historical FFT arrays,
a persistence intensity coefficient, an asymptotic remainder window,
weighted-Palm event identification, higher homology or dimensions, or
independent human review. No novelty or priority claim is made.
21 changes: 15 additions & 6 deletions experiments/periodic_h0/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -9,15 +9,22 @@ derivative bounds for the rounded Fourier coefficients of the eight refinement
fields. Its [results](results/certificate8/RESULTS.md) provide a certified spatial
interpolation budget. A [new nodal certificate](NODAL_CERTIFICATE.md) now bounds
every numerical sample of one retained field on a 128² grid and combines that
error with its spatial budget. Historical samples, computed barcode endpoints
and the infinite-field tail retain their separate open obligations.
error with its spatial budget. Historical samples retain their separate
uncertified status.

A [sharper Hessian/grid certificate](HESSIAN_GRID.md) recovers cancellation
between Fourier modes and evaluates a smaller spatial budget on a 1024² grid.
Its [compact result](results/hessian1/RESULTS.md) defines new exact dyadic
samples by deterministic replay, avoiding a million-value data file. These
samples are distinct from the historical NumPy arrays; computed barcode
endpoints and lifetime-law confirmation remain open.
samples are distinct from the historical NumPy arrays. Their
[exact integer barcode](results/exact_h0_1/RESULTS.md) now supplies deterministic
bin bounds for the frozen finite polynomial.

A new [Gaussian tail calculation](results/gaussian_tail1/RESULTS.md) bounds
the omitted modes of the ideal full field. At cutoff24, its uniform error is
at most 0.00130835 with failure probability below 1.27×10⁻¹². Applying that
result to our stored polynomial still requires a certified coefficient
coupling; the [proof](GAUSSIAN_TAIL.md) states the missing premise explicitly.

This small experiment samples a truncated version of the periodized Gaussian field on a side-24 square torus and measures finite ordinary superlevel H₀ bars using GUDHI. It accompanies the [manuscript draft](../../docs/research-translation/20260930/MANUSCRIPT.md), [statement crosswalk](../../docs/research-translation/20260930/CROSSWALK.md) and [original protocol](../../docs/research-translation/20260930/EXPERIMENT.md). It is numerical evidence about the declared discretized ensemble, not a continuum certificate or a new mathematical acceptance.

Expand Down Expand Up @@ -100,8 +107,10 @@ majorants for the specified rounded finite polynomials are supplied. One new
grid have certified nodal bounds. The latter has a sharper spatial bound;
its [exact barcode and finite-field bin bounds](results/exact_h0_1/RESULTS.md)
are now supplied by an integer sweep and independent connectivity verification.
The relation to ideal Gaussian coefficients, an enclosed infinite-field spectral tail, a justified finite-lifetime
asymptotic window and numerical remainder constants remain open. Marginal
The [ideal Gaussian spectral tail](GAUSSIAN_TAIL.md) is now enclosed with an
explicit probability budget. The relation of the stored coefficients to that
ideal ensemble, a justified finite-lifetime asymptotic window and numerical
remainder constants remain open. Marginal
one-standard-error bars do not include these errors. The coefficient is a float64
comparison value from a separate exact enclosure. This experiment does not
estimate typed contacts, weighted-Palm events or candidate-minus-elder defects,
Expand Down
Loading
Loading