Can $\pi$ generate itself? A Monte Carlo analysis of 314 trillion digits

Alessandro Razeto; Nicola Rossi

REVIEW 3 major objections 4 minor 13 references

Can $\pi$ generate itself? A Monte Carlo analysis of 314 trillion digits

T0 review · 3 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Using pi's own digits as the random-number source, a Monte Carlo simulation recovers pi to seven decimal places.

desk verdict A cute scale-up of a known Monte Carlo curiosity, but the manuscript's internal 14-vs-16-digit inconsistency makes the headline estimate unreproducible as written. read the letter →

arxiv 2608.06438 v1 pith:I6XE224P submitted 2026-08-06 math.GM

Alessandro Razeto , Nicola Rossi This is my paper · ORCID

classification math.GM MSC 11K1665C0511K45

keywords pidigitsMonteCarloestimationnormalitypseudorandomnumbersstatisticalrandomnessdecimalexpansion314trillion

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Can pi's own decimal expansion supply the randomness needed to compute pi? Using the record 314-trillion-digit dataset, the authors map consecutive digit blocks into coordinates inside the unit square and run the simplest Monte Carlo estimator, counting how many points fall inside a quarter circle. The result, $\hat\pi = 3.14159313 \pm 0.00000035$, reproduces the first seven decimal digits of $\pi$, with a measured deviation of about $1.4\sigma$ and a $p$-value of $0.17$. The authors read this as empirical evidence that the digit stream behaves as a statistically uniform random source at the available scale, while stating clearly that this proves neither normality nor independence.

What carries the argument

The load-bearing object is the hit-or-miss Monte Carlo estimator for $\pi$: sample points uniformly in the unit square, keep those with $x^2+y^2\le1$, and estimate $\pi=4N_{\rm in}/N$. The paper's twist is that the uniform samples are generated by decoding consecutive digits of $\pi$ themselves, with the block length chosen so that the expected occupancy is about one point per cell of a square grid. The quoted error bar is the binomial standard deviation $\sigma=4\sqrt{p(1-p)/N}$ with $p=\pi/4$, giving $\sigma\approx 3.5\times10^{-7}$, and the goodness-of-fit tests check that the per-file estimates scatter by the expected amount.

What would settle it

Shift the segmentation of the digit stream by one digit and recompute the same Monte Carlo estimator: under the paper's uniformity assumption, estimates from the original and shifted segmentations should agree within roughly $2\sigma$. A systematic gap many times $\sigma$ would show that the apparent agreement is specific to the chosen block boundaries rather than a property of the digit stream itself.

Watch

Extended reading notes

Core claim

The central claim is that the decimal expansion of pi can successfully act as its own source of pseudorandom numbers in a large-scale Monte Carlo simulation. The digit stream is split into blocks; each block is read as the fractional coordinates $(x,y)$ of a point in the unit square, and the point counts as a hit when $x^2+y^2\le 1$. From roughly $2.24\times10^{13}$ points the hit fraction gives $4N_{\rm in}/N = 3.14159313 \pm 0.00000035$, matching the exact value of $\pi$ to seven decimal places. The paper's statistical checks, a Student $t$-test on the 628 per-file estimates and a Pearson $\chi^2$ test on digit-bin histograms, show no evidence of bias, and the authors emphasize that this is a practical, finite-scale demonstration, not a proof of normality or digit independence.

Load-bearing premise

The quoted uncertainty and p-values treat every digit-block-derived point as an independent, uniform random draw. Since the digit stream is deterministic, independence can only hold approximately; if strong correlations persist across blocks, the effective sample size is smaller than $N$ and the error bars are underestimated.

Editorial extensions

If this is right

No detectable statistical bias appears in pi's digits up to 314 trillion places, so frequency-based normality tests are corroborated by a practical stochastic computation at this scale.
The estimator already sits at the $\sigma\sim10^{-7}$ precision limit set by the dataset size; larger future digit records should extend the recovered decimals of pi as $N^{-1/2}$.
The same digit-to-coordinate mapping provides a generic testbed for the practical randomness of any long decimal expansion, not only pi.
Deterministic digit streams such as pi's can serve as reproducible seeds for physics Monte Carlo simulations, as the paper suggests for spigot-generated constants.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A stricter version of the paper's test would split the digit stream: estimate the hit fraction from the first 90% of the digits and use it to predict the last 10%, removing the circularity of measuring randomness and target on the same data.
Running the identical estimator on the first 314 trillion digits of a proven-normal constant or a high-quality pseudorandom generator would calibrate how unusual a $p$-value of 0.17 really is.
Because only non-overlapping blocks are tested, correlations at lags shorter than one block remain invisible; an autocorrelation scan of the inside-circle indicator over lags 1 to 1000 block lengths would probe exactly that gap.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

The paper is a clean empirical exercise: use the digits of pi as the random source for the standard quarter-circle Monte Carlo estimator of pi, applied to the new 314-trillion-digit dataset. What is genuinely new is the scale and the 628-chunk consistency analysis – no one has run this on the full record stream. Credit where due: the estimator is standard, the authors are clear that the digits are deterministic and that independence cannot hold, and the philosophical section does not oversell. The reported result, pi-hat = 3.14159313 ± 0.00000035, matches what N^{-1/2} precision would predict and is a legitimate empirical test of the digit stream's random-like behavior.

The soft spot is not the acknowledged non-independence; it is the internal contradiction in the sample construction. Section 2 derives 14 digits per point (7 per coordinate) and N = 22.4e12. But Eq. (7) is arithmetically wrong: sqrt(314e12) is about 1.772e7, whose log10 is 7.25, so the ceiling is 8, not 7. Section 3 states that each point uses 16 digits (8 per coordinate). The reported total N = 22,428,557,114,832, however, is exactly 314e12/14, which cannot arise from 16-digit blocks (that would give 19.6e12). The two descriptions are mutually incompatible. If the code used 7-digit coordinates, the points lie on a 10^7 by 10^7 lattice and the expected hit probability is the lattice fraction, not pi/4, with a deterministic O(10^-7) boundary correction comparable to the quoted sigma; if it used 8-digit coordinates, the quoted N is impossible. Either way, the central estimate and its uncertainty are not pinned down by the manuscript. This is a reproducibility issue, not a cosmetic typo, because a reader cannot tell which procedure generated the numbers.

The underlying idea probably works, and the discrepancy may be fixable, but as submitted the paper is not self-consistent. I would send it to peer review because the scale is new and the flaw is concrete and correctable, but I would not accept until the block-length discrepancy is resolved and the code or data are made available. The citation pattern is unremarkable; the only references are standard.

This paper is for readers interested in pi digit statistics and Monte Carlo methods, not for those seeking a theoretical advance. Worth a look in reading group if someone wants to debug the inconsistency, but I wouldn't cite it in its current form.

Referee Report

3 major / 4 minor

Summary. The paper proposes to use the publicly available 314-trillion-digit decimal expansion of pi as the random-number source for a classic Monte Carlo estimate of pi. Digits are blocked into coordinate pairs in the unit square, points are classified inside or outside the quarter unit circle, and the fraction of inside points is multiplied by four to estimate pi. The authors report \hat\pi = 3.14159313 \pm 0.00000035 from approximately 22.4e12 points, with t-test and chi-square p-values of 0.17, and interpret the result as evidence that the digit stream behaves as a statistically valid pseudorandom source at this scale. They also discuss the philosophical question of whether pi can 'generate itself' from its own decimal expansion.

Significance. If the construction were exactly as stated, the paper would provide a clean, large-scale empirical test of the random-like behavior of pi's digits in a nontrivial stochastic computation, using a uniquely large public dataset. The reported aggregate counts and statistical p-values are internally consistent with a 14-digit-per-point binomial calculation, and the estimator formulas in Eqs. (14)-(16) are standard. The main strength is the direct use of the complete record dataset, which makes the test more stringent than earlier digit-frequency analyses. However, the manuscript currently contains a load-bearing internal inconsistency about the number of digits per point, and an unresolved lattice-discrepancy issue, so the experiment as written is not reproducible and the quoted uncertainty is not fully justified.

major comments (3)

[Section 2, Eqs. (5)-(8)] Equation (7) is arithmetically wrong. From Eq. (6), N_x = sqrt(314e12) = 1.772e7, whose base-10 logarithm is about 7.25, so ceil(log10 N_x) = 8, not 7. If the code really used 7-digit coordinates, the points lie on a 10^7 by 10^7 lattice with 10^14 cells, not on the N_x by N_y grid required by Eq. (5). If the intended grid has 3.14e14 cells, each coordinate requires 8 digits and Eq. (8) would give N = 314e12 / 16 = 19.625e12, not 22.4e12. The derivation of the sample size and the optimal block length therefore needs to be corrected or explicitly restated to match the actual implementation.
[Section 3, Eqs. (10)-(11) and Eq. (15)] The paper contradicts itself on the block size. Section 3 states that each worker consumes blocks of sixteen decimal digits, with eight digits per coordinate, but the reported sum of counts in Eq. (15), 22,428,557,114,832, equals 314e12 / 14, not 314e12 / 16 = 19.625e12. The subsequent histogram test in the same section also uses 10^7 bins and mean 2 N_i / 10^7, which corresponds to 7-digit coordinates, not 8-digit coordinates. These descriptions cannot all be true. The actual block construction must be stated unambiguously, and the quoted N_i and aggregate counts must be recomputed or reconciled with that construction.
[Section 3, final paragraph, and Eq. (16)] If the actual implementation used 7-digit coordinates, the success probability is not exactly pi/4. On a 10^7 by 10^7 lattice, the expected fraction of points inside the quarter circle is the number of lattice points in the quarter circle divided by 10^14, which differs from pi/4 by a deterministic boundary term of order 10^-7. This is comparable to the quoted sigma = 0.00000035 in Eqs. (14) and (16). The manuscript does not quantify or correct this lattice bias, so the uncertainty and p-values are not centered on the true sampling distribution unless the code used 8-digit blocks, in which case the reported N is impossible. The authors need to either confirm the exact digit-to-coordinate mapping and account for the lattice correction, or provide evidence that the bias is negligible at the stated precision.

minor comments (4)

[Abstract and Eq. (14)] The abstract reports pi approx 3.141593, while Eq. (14) gives 3.14159313 +/- 0.00000035; these should be made consistent so that the claimed 'first sequence of decimal digits' is precisely defined.
[Section 3, paragraph 4] The text contains the typo 'additional data data'; it should read 'additional data'.
[Section 3, final paragraph] The claim that the 628 p-values are 'uniformly distributed between 0 and 1' is not accompanied by a figure, histogram, or summary statistic. A visual or quantitative support would make this verification check reproducible.
[Section 2, Eq. (9)] The scaling statement sigma ~ 1/sqrt(N) is correct in order of magnitude, but the precise prefactor for the pi/4 Bernoulli estimator is 4 sqrt(p(1-p)) with p = pi/4, which gives sigma = 3.47e-7; stating the exact relation would improve clarity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Monte Carlo estimate is a genuine empirical test and is not forced by construction.

full rationale

The paper's central estimate is constructed from the digit stream by a fixed, parameter-free mapping (Eqs. 3-4 and 10-11). The expected success probability p = pi/4 in Eq. (1) is a geometric constant, not a fitted parameter; the digits of pi enter only as the random input. Consequently, the estimates in Eqs. (14)-(15) are genuine empirical statistics whose deviations from pi carry information about the statistical quality of the digit stream. No parameter is fitted to the target value, and no load-bearing claim rests on self-citation: all cited works are external mathematical literature or the public dataset. The philosophical section explicitly frames the outcome as 'self-extraction' rather than a derivation, and the limitations regarding normality and independence are acknowledged in the abstract and Sec. 1. I also note an internal inconsistency between the 14-digit construction of Sec. 2 and the 16-digit description in Sec. 3, and an apparent rounding error in Eq. (7); however, these are reproducibility/correctness issues, not circularity, because the reported estimate is not forced by construction to equal pi. The result is therefore a self-contained empirical test rather than a circular derivation.

Assumptions & free parameters 1 free parameters · 4 assumptions · 0 invented entities

The analysis introduces no new entities. It relies on the integrity of the external dataset, the empirical assumption of practical independence of digit blocks, and standard probability facts. The hand-chosen block size is the only design parameter, and it influences the sample count and error scale but is not fitted to force agreement with pi.

free parameters (1)

Digit block size (nx = ny) = 7 digits per coordinate (14 per point)
Chosen so that the grid has roughly one point per cell (Nx * Ny ≈ N); not fitted to the target value, but a hand-picked design choice that sets the sample size and error scale.

assumptions (4)

domain assumption The 314-trillion-digit public dataset is a correct and complete decimal expansion of pi.
Invoked in Section 1.1; the analysis trusts the StorageReview/Backblaze data integrity and the YCD binary decoding.
domain assumption Consecutive digit blocks can be treated as statistically independent uniform samples at the scale of interest.
Used throughout Sections 2 and 3 to model the point count as Binomial(N, pi/4); the authors acknowledge strict independence cannot hold for a deterministic sequence.
standard math The central limit theorem applies to the 628 chunk means for the one-sample t-test.
Assumed in Section 3 when computing the p-value of 0.17; each chunk contains a very large number of points.
standard math The quarter-circle geometry gives success probability p = pi/4 for a uniformly drawn point.
Used in Section 1, equation (1), as the calibration of the Monte Carlo estimator.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Can $\pi$ generate itself? A Monte Carlo analysis of 314 trillion digits." pith.science (2026). https://pith.science/paper/I6XE224P

@misc{pith2026260806438,
  author       = {Pith},
  title        = {Pith review of: Can $\pi$ generate itself? A Monte Carlo analysis of 314 trillion digits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I6XE224P}},
  note         = {Machine review of arXiv:2608.06438}
}

abstract

At the end of 2025, a record computation of $\pi$ reached 314 trillion decimal digits, providing the largest numerical dataset ever generated for this constant. We exploit this unprecedented dataset to investigate whether the digits of $\pi$ themselves can serve as a source of pseudorandom numbers for estimating $\pi$ through the simplest Monte Carlo method. Our results go beyond the normality hypothesis by providing empirical evidence of a high degree of statistical randomness in the available digits, although not of digit independence, which cannot hold for a deterministic sequence. By optimizing the mapping of the digit sequence into Monte Carlo samples, we obtain the highest precision allowed by the dataset. As predicted, the method successfully reproduces the first sequence of decimal digits, demonstrating that the largest available dataset of $\pi$ digits can be used to recover $ \pi \approx 3.141593 $ through Monte Carlo simulation.

Figures

Figures reproduced from arXiv: 2608.06438 by the authors.

**Figure 2.** Schematic representation of the grid optimization procedure. Consecutive blocks [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

[1]

D. H. Bailey and J. M. Borwein,Pi: The Next Generation, Springer International Publishing, Cham (2016)

work page 2016
[2]

J. M. Borwein, D. H. Bailey, and R. Girgensohn,Experimentation in Mathematics: Computational Paths to Discovery, A. K. Peters, Wellesley, MA (2004)

work page 2004
[3]

Les probabilit´ es d´ enombrables et leurs applications arithm´ etiques,

E. Borel, “Les probabilit´ es d´ enombrables et leurs applications arithm´ etiques,”Ren- diconti del Circolo Matematico di Palermo,27, 247–271 (1909)

work page 1909
[4]

Kuipers and H

L. Kuipers and H. Niederreiter,Uniform Distribution of Sequences, Wiley, New York (1974)

work page 1974
[5]

Bugeaud,Distribution Modulo One and Diophantine Approximation, Cambridge University Press, Cambridge (2012)

Y. Bugeaud,Distribution Modulo One and Diophantine Approximation, Cambridge University Press, Cambridge (2012)

work page 2012
[6]

Random Generators and Normal Numbers,

D. H. Bailey and R. E. Crandall, “Random Generators and Normal Numbers,” Experimental Mathematics,11(4), 527–546 (2002)

work page 2002
[7]

On the Rapid Computation of Various Polylogarithmic Constants,

D. H. Bailey, P. Borwein, and S. Plouffe, “On the Rapid Computation of Various Polylogarithmic Constants,”Mathematics of Computation,66(218), 903–913 (1997)

work page 1997
[8]

314 Trillion Digits ofπ,

A. J. Yee, “314 Trillion Digits ofπ,”y-cruncher / NumberWorld, 2025. Available at: https://www.numberworld.org/y-cruncher/

work page 2025

Show all 13 references

[9]

314 Trillion Digits of Pi Stored in the Backblaze Cloud,

Backblaze, “314 Trillion Digits of Pi Stored in the Backblaze Cloud,” Back- blaze Technical Blog, 2025. Available at:https://www.backblaze.com/blog/ backblaze-now-serving-314-trillion-digits-of-pi/, accessed on Aug 5, 2026

2025
[10]

Digit Statistics of the First 22.4 Trillion Decimal Digits of Pi,

P. Tr¨ ub, “Digit Statistics of the First 22.4 Trillion Decimal Digits of Pi,” arXiv:1612.00489 (2016). 8

2016 arXiv
[11]

The Monte Carlo Method,

N. Metropolis and S. Ulam, “The Monte Carlo Method,”Journal of the American Statistical Association,44(247), 335–341 (1949)

1949
[12]

M. H. Kalos and P. A. Whitlock,Monte Carlo Methods, Second Edition, Wiley-VCH, Weinheim (2008)

2008
[13]

G. S. Fishman,Monte Carlo: Concepts, Algorithms, and Applications, Springer, New York (1996). 9

1996

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

D. H. Bailey and J. M. Borwein,Pi: The Next Generation, Springer International Publishing, Cham (2016)

work page 2016

[2] [2]

J. M. Borwein, D. H. Bailey, and R. Girgensohn,Experimentation in Mathematics: Computational Paths to Discovery, A. K. Peters, Wellesley, MA (2004)

work page 2004

[3] [3]

Les probabilit´ es d´ enombrables et leurs applications arithm´ etiques,

E. Borel, “Les probabilit´ es d´ enombrables et leurs applications arithm´ etiques,”Ren- diconti del Circolo Matematico di Palermo,27, 247–271 (1909)

work page 1909

[4] [4]

Kuipers and H

L. Kuipers and H. Niederreiter,Uniform Distribution of Sequences, Wiley, New York (1974)

work page 1974

[5] [5]

Bugeaud,Distribution Modulo One and Diophantine Approximation, Cambridge University Press, Cambridge (2012)

Y. Bugeaud,Distribution Modulo One and Diophantine Approximation, Cambridge University Press, Cambridge (2012)

work page 2012

[6] [6]

Random Generators and Normal Numbers,

D. H. Bailey and R. E. Crandall, “Random Generators and Normal Numbers,” Experimental Mathematics,11(4), 527–546 (2002)

work page 2002

[7] [7]

On the Rapid Computation of Various Polylogarithmic Constants,

D. H. Bailey, P. Borwein, and S. Plouffe, “On the Rapid Computation of Various Polylogarithmic Constants,”Mathematics of Computation,66(218), 903–913 (1997)

work page 1997

[8] [8]

314 Trillion Digits ofπ,

A. J. Yee, “314 Trillion Digits ofπ,”y-cruncher / NumberWorld, 2025. Available at: https://www.numberworld.org/y-cruncher/

work page 2025