pith. machine review for the scientific record. sign in

arxiv: 2605.00479 · v1 · submitted 2026-05-01 · 🧬 q-bio.QM

Recognition: unknown

Reduced-Precision Stochastic Simulation for Mathematical Biology

Tom Kimpson , Mark B. Flegg , Jennifer A. Flegg
Authors on Pith no claims yet

Pith reviewed 2026-05-09 15:23 UTC · model grok-4.3

classification 🧬 q-bio.QM
keywords stochastic simulation algorithmreduced precisionmixed precisionstochastic roundingmathematical biologySSApropensity functionsensemble statistics
0
0 comments X

The pith

Mixed-precision arithmetic lets the stochastic simulation algorithm run faster while matching full-precision statistics on standard biological models.

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

The paper tests whether lowering the number of bits used in floating-point calculations can reduce the cost of exact stochastic simulations of chemical reaction networks without changing the resulting probability distributions. Two strategies are compared on five classic models: birth-death processes, the Schlögl reaction, the telegraph model, dimerisation, and the repressilator. Mixed precision keeps propensity calculations in 16-bit format but stores accumulators in 32-bit format and reproduces the reference 64-bit distributions according to Kolmogorov-Smirnov and Wasserstein metrics. Uniform 16-bit arithmetic introduces systematic bias with ordinary rounding, yet stochastic rounding together with propensity normalisation removes the bias and restores agreement above p = 0.05. The approach therefore offers a hardware-level route to smaller memory footprints and modest speed gains while preserving the statistical fidelity required for biological modelling.

Core claim

Mixed-precision SSA computes propensities in 16-bit floating point while retaining 32-bit accumulators and yields ensemble statistics that are statistically indistinguishable from 64-bit references across all five tested models; uniform 16-bit arithmetic with deterministic rounding produces measurable biases, which are eliminated when stochastic rounding and propensity normalisation are applied.

What carries the argument

Mixed-precision SSA with stochastic rounding: propensities are evaluated in 16-bit format while accumulators remain in 32-bit format, and rounding errors are randomised rather than truncated to prevent systematic drift in the sampled trajectories.

If this is right

  • Memory required per variable drops by a factor of two to four relative to 32-bit or 64-bit storage.
  • Wall-clock time on conventional CPUs improves by up to roughly 1.5 times even without native 16-bit hardware support.
  • The method complements existing algorithmic accelerations such as tau-leaping.
  • The reduced data movement maps directly onto GPUs and TPUs that already provide fast 16-bit arithmetic.
  • Ensemble sizes or simulation durations can be increased on fixed hardware budgets while keeping the same statistical quality.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same low-precision pattern could be applied to other Monte Carlo methods used in systems biology, such as spatial SSA or hybrid deterministic-stochastic integrators.
  • Hardware vendors could expose 16-bit stochastic rounding as a standard primitive, making the speedup available without code changes on future accelerators.
  • Real-time or on-device simulation of small genetic circuits becomes feasible on embedded or mobile processors that support half-precision arithmetic.

Load-bearing premise

The statistical agreement found on the five small canonical models will continue to hold when the same arithmetic is applied to larger, more complex biological networks.

What would settle it

A Kolmogorov-Smirnov p-value below 0.05 or a Wasserstein distance clearly larger than the reference values when the mixed-precision code is run on any of the five models with substantially increased molecule counts or reaction steps.

Figures

Figures reproduced from arXiv: 2605.00479 by Jennifer A. Flegg, Mark B. Flegg, Tom Kimpson.

Figure 1
Figure 1. Figure 1: Steady-state distribution of the birth–death process ( view at source ↗
Figure 2
Figure 2. Figure 2: Bimodal stationary distribution of the Schlögl system ( view at source ↗
Figure 3
Figure 3. Figure 3: Telegraph mRNA marginal distributions. Top row: bimodal parameterisation ( view at source ↗
Figure 4
Figure 4. Figure 4: Telegraph model validation. (a) OFF-state dwell-time distribution for the bimodal parameterisation ( view at source ↗
Figure 5
Figure 5. Figure 5: Steady-state distributions for monomer A (left) and dimer D (right) across all six mixed-precision modes. All configurations overlap closely with the FP64 baseline, consistent with the numerical stability conferred by the conservation law. Uniform-precision modes (not shown) are visually similar for most configurations, though uniform BF16 RTN shows moderate deviation; see view at source ↗
Figure 6
Figure 6. Figure 6: Stationary protein A distributions for the repressilator. (a) Mixed-precision modes overlap closely with the view at source ↗
Figure 7
Figure 7. Figure 7: Repressilator horizon sweep under uniform precision ( view at source ↗
read the original abstract

The stochastic simulation algorithm (SSA) is widely used to perform exact forward simulation of discrete stochastic processes in biology. However, the computational cost, driven by sequential event-by-event sampling across large ensembles, remains a computational barrier. We investigate whether reduced-precision floating-point arithmetic can accelerate SSA without degrading statistical fidelity, drawing on the success of reduced-precision methods in weather and climate modelling. We evaluate two strategies across five canonical models (birth--death, Schl\"{o}gl, Telegraph, dimerisation, repressilator): (i) mixed precision, computing propensities in 16-bit while maintaining accumulators in 32-bit; and (ii) uniform precision, performing all arithmetic in 16-bit. Mixed-precision SSA produces ensemble statistics that closely match the 64-bit reference for all models, as measured by Kolmogorov--Smirnov tests and Wasserstein distances. Under uniform precision, deterministic rounding introduces systematic biases across several models, with catastrophic failures in some cases. Stochastic rounding (SR) and propensity normalisation eliminate these biases, restoring distributional fidelity across all models tested (KS $p > 0.05$). Our results establish mixed-precision SSA with SR as a viable acceleration strategy for mathematical biology: 16-bit formats shrink per-variable data size by $2$--$4\times$ relative to \texttt{fp32}/\texttt{fp64}, yielding comparable reductions in memory footprint and up to $\sim 1.5\times$ wall-clock speedup on CPU hardware that lacks native 16-bit arithmetic. As a hardware-level acceleration, mixed-precision SSA complements algorithmic methods such as tau-leaping and maps naturally onto modern GPU and TPU architectures with native 16-bit arithmetic.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper investigates the use of reduced-precision (16-bit) floating-point arithmetic in the stochastic simulation algorithm (SSA) for discrete stochastic processes in mathematical biology. It evaluates mixed-precision (16-bit propensities with 32-bit accumulators) and uniform 16-bit precision strategies on five canonical models (birth-death, Schlögl, Telegraph, dimerisation, repressilator). Mixed-precision SSA matches 64-bit reference ensemble statistics closely per Kolmogorov-Smirnov tests and Wasserstein distances. Uniform precision with deterministic rounding introduces biases, but stochastic rounding combined with propensity normalisation restores fidelity (KS p > 0.05). The work claims this yields 2-4x memory reduction and up to 1.5x wall-clock speedup on CPU, positioning it as a hardware-level acceleration complementary to tau-leaping and suited to GPU/TPU architectures.

Significance. If the fidelity results generalise, the approach could meaningfully accelerate ensemble SSA runs for biological models on low-precision hardware, reducing memory footprint and runtime while preserving statistical properties. The empirical use of KS tests and Wasserstein distances to quantify distributional agreement, plus explicit comparison to 64-bit baselines, provides a concrete starting point. However, the restriction to small models (1-3 species) limits immediate impact; broader adoption would require evidence on scaling to networks with tens of species and hundreds of reactions where propensity summation errors could accumulate.

major comments (2)
  1. [Abstract / Results] Abstract and Results (five-model evaluation): The central claim that mixed-precision SSA 'establishes a viable acceleration strategy for mathematical biology' rests on statistical agreement for only five small canonical models (birth-death: 1 species; Schlögl: 1; Telegraph: 2; dimerisation: 2; repressilator: 3 species/6 reactions). No tests are reported on larger networks where summed low-precision propensities could produce larger relative errors or where rare-event/long-time statistics might diverge despite KS p > 0.05 on the small set. This is load-bearing for the viability claim.
  2. [Methods] Methods / Implementation details: The abstract reports KS p-values and Wasserstein distances but the manuscript provides no full pseudocode, random-number generator details, or raw data for the 16-bit exponential sampling and propensity calculations. Without these, independent reproduction and assessment of whether the observed fidelity is robust to implementation choices is not possible.
minor comments (2)
  1. [Results] Clarify the exact 16-bit format (bfloat16 vs. fp16) and any hardware-specific intrinsics used for the reported CPU speedups, as this affects portability claims.
  2. [Abstract] The abstract states 'up to ~1.5x wall-clock speedup on CPU hardware that lacks native 16-bit arithmetic'; provide the baseline compiler flags and measurement methodology for this figure.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the scope and reproducibility of our work. We address each major point below and propose targeted revisions.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results (five-model evaluation): The central claim that mixed-precision SSA 'establishes a viable acceleration strategy for mathematical biology' rests on statistical agreement for only five small canonical models (birth-death: 1 species; Schlögl: 1; Telegraph: 2; dimerisation: 2; repressilator: 3 species/6 reactions). No tests are reported on larger networks where summed low-precision propensities could produce larger relative errors or where rare-event/long-time statistics might diverge despite KS p > 0.05 on the small set. This is load-bearing for the viability claim.

    Authors: We agree that the evaluation is confined to five small canonical models and that this constrains the strength of the broad viability claim for mathematical biology as a whole. These models were selected because they are standard benchmarks in the SSA literature and permit rigorous statistical comparison via KS tests and Wasserstein distances. We acknowledge that propensity summation errors could accumulate in networks with tens of species and hundreds of reactions, and that rare-event statistics might behave differently. In the revised manuscript we will (i) temper the abstract and conclusion language to emphasize that the results demonstrate feasibility on representative small systems, and (ii) add a new paragraph in the Discussion that explicitly discusses potential scaling limitations, references relevant numerical-stability literature, and identifies validation on larger networks as a priority for future work. No new large-scale experiments will be added at this stage. revision: partial

  2. Referee: [Methods] Methods / Implementation details: The abstract reports KS p-values and Wasserstein distances but the manuscript provides no full pseudocode, random-number generator details, or raw data for the 16-bit exponential sampling and propensity calculations. Without these, independent reproduction and assessment of whether the observed fidelity is robust to implementation choices is not possible.

    Authors: We accept that the current manuscript lacks sufficient implementation detail for independent reproduction. The revised version will include: (i) complete pseudocode for the mixed-precision SSA (16-bit propensities, 32-bit accumulators) and the uniform-precision variant with stochastic rounding and normalisation; (ii) explicit specification of the random-number generator (Mersenne Twister with fixed seeds) and the emulation of 16-bit arithmetic; and (iii) a public GitHub repository containing the raw simulation trajectories, analysis scripts, and exact parameter sets used for the five models. These additions will allow readers to verify the reported KS p-values and Wasserstein distances. revision: yes

Circularity Check

0 steps flagged

No circularity; purely empirical validation on fixed models

full rationale

The paper reports direct numerical experiments: mixed- and uniform-precision SSA runs on five named canonical models are compared to 64-bit reference trajectories via Kolmogorov-Smirnov and Wasserstein metrics. No equations are derived, no parameters are fitted and then re-predicted, and no self-citations are invoked to justify uniqueness or ansatzes. All claims reduce to observable statistical agreement on the tested instances; the derivation chain is empty.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The work rests on the standard mathematical assumptions of the Gillespie SSA and IEEE floating-point rounding behavior; no new free parameters, axioms, or invented entities are introduced.

pith-pipeline@v0.9.0 · 5608 in / 1083 out tokens · 36983 ms · 2026-05-09T15:23:54.811874+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

48 extracted references · 41 canonical work pages

  1. [1]

    Gillespie

    Daniel T. Gillespie. Exact stochastic simulation of coupled chemical reactions.The Journal of Physical Chemistry, 81(25):2340–2361, 1977. doi:10.1021/j100540a008

    work page doi:10.1021/j100540a008 1977

  2. [2]

    Anderson

    David F. Anderson. A modified next reaction method for simulating chemical systems with time dependent propensities and delays.The Journal of Chemical Physics, 127(21):214107, 2007. doi:10.1063/1.2799998

    work page doi:10.1063/1.2799998 2007

  3. [3]

    Gillespie, and Linda R

    Yang Cao, Daniel T. Gillespie, and Linda R. Petzold. Efficient step size selection for the τ-leaping simulation method.The Journal of Chemical Physics, 124(4):044109, 2006. doi:10.1063/1.2159468

    work page doi:10.1063/1.2159468 2006

  4. [4]

    Haseltine and James B

    Eric L. Haseltine and James B. Rawlings. Approximate simulation of coupled fast and slow reactions for stochastic chemical kinetics.The Journal of Chemical Physics, 117(15):6959–6969, 2002. doi:10.1063/1.1505860

    work page doi:10.1063/1.1505860 2002

  5. [5]

    Gillespie, and Linda R

    Yang Cao, Daniel T. Gillespie, and Linda R. Petzold. The slow-scale stochastic simulation algorithm.The Journal of Chemical Physics, 122(1):014116, 2005. doi:10.1063/1.1824902

    work page doi:10.1063/1.1824902 2005

  6. [6]

    In-datacenter performance analysis of a tensor processing unit,

    Norman P. Jouppi, Cliff Young, Nishant Patil, David Patterson, et al. In-datacenter performance analysis of a tensor processing unit. InProceedings of the 44th Annual International Symposium on Computer Architecture (ISCA), pages 1–12, 2017. doi:10.1145/3079856.3080246

    work page doi:10.1145/3079856.3080246 2017

  7. [7]

    A Study of BFLOAT16 for Deep Learning Training

    Dhiraj Kalamkar, Dheevatsa Mudigere, Naveen Mellempudi, Dipankar Das, Kunal Banerjee, Sasikanth Avancha, Dharma Teja V ooturi, et al. A study of bfloat16 for deep learning training.arXiv preprint arXiv:1905.12322, 2019. URLhttps://arxiv.org/abs/1905.12322

    work page Pith review arXiv 1905

  8. [8]

    Tim N. Palmer. More reliable forecasts with less precise computations: A fast-track route to cloud-resolved weather and climate simulators?Philosophical Transactions of the Royal Society A, 372(2018):20130391, 2014. doi:10.1098/rsta.2013.0391

    work page doi:10.1098/rsta.2013.0391 2018

  9. [9]

    Scale-selective precision for weather and climate forecasting.Monthly Weather Review, 147(2):645–655, 2019

    Matthew Chantry, Tobias Thornes, Tim Palmer, and Peter Düben. Scale-selective precision for weather and climate forecasting.Monthly Weather Review, 147(2):645–655, 2019. doi:10.1175/MWR-D-18-0308.1

    work page doi:10.1175/mwr-d-18-0308.1 2019

  10. [10]

    Düben, David A

    Andrew Dawson, Peter D. Düben, David A. MacLeod, and Tim N. Palmer. Reliable low precision simulations in land surface models.Climate Dynamics, 51(7–8):2657–2666, 2018. doi:10.1007/s00382-017-4034-x

    work page doi:10.1007/s00382-017-4034-x 2018

  11. [11]

    Monte carlo arithmetic: Exploiting randomness in floating-point arithmetic

    David Stott Parker. Monte carlo arithmetic: Exploiting randomness in floating-point arithmetic. Technical Report CSD-970002, Computer Science Department, University of California, Los Angeles, Los Angeles, CA, 1997. URLhttps://searchworks.stanford.edu/view/4234742. 18 Reduced-Precision Stochastic Simulation

    work page arXiv 1997

  12. [12]

    Connolly, Nicholas J

    Michael P. Connolly, Nicholas J. Higham, and Théo Mary. Stochastic rounding and its probabilistic backward error analysis.SIAM Journal on Scientific Computing, 43(1):A566–A585, 2021. doi:10.1137/20M1334796

    work page doi:10.1137/20m1334796 2021

  13. [13]

    Matteo Croci and Mike B. Giles. Effects of round-to-nearest and stochastic rounding in the numerical so- lution of the heat equation in low precision.IMA Journal of Numerical Analysis, 43(3):1358–1390, 2022. doi:10.1093/imanum/drac012. URLhttps://arxiv.org/abs/2010.16225

    work page doi:10.1093/imanum/drac012 2022

  14. [14]

    Higham, Théo Mary, and Mantas Mikaitis

    Matteo Croci, Massimiliano Fasi, Nicholas J. Higham, Théo Mary, and Mantas Mikaitis. Stochastic round- ing: Implementation, error analysis and applications.Royal Society Open Science, 9(3):211631, 2022. doi:10.1098/rsos.211631

    work page doi:10.1098/rsos.211631 2022

  15. [15]

    Paxton, Matthew Chantry, and Tim Palmer

    Tom Kimpson, Ellie A. Paxton, Matthew Chantry, and Tim Palmer. Climate-change modelling at reduced floating-point precision with stochastic rounding.Quarterly Journal of the Royal Meteorological Society, 150(S1): e4435, 2024. doi:10.1002/qj.4435

    work page doi:10.1002/qj.4435 2024

  16. [16]

    On the use of scale-dependent precision in Earth system modelling

    Tobias Thornes, Peter Düben, and Tim Palmer. On the use of scale-dependent precision in Earth system modelling. Quarterly Journal of the Royal Meteorological Society, 143(703):897–908, 2017. doi:10.1002/qj.2974

    work page doi:10.1002/qj.2974 2017

  17. [17]

    Adam Paxton, Matthew Chantry, Milan Klöwer, Leo Saffin, and Tim Palmer

    E. Adam Paxton, Matthew Chantry, Milan Klöwer, Leo Saffin, and Tim Palmer. Climate modeling in low precision: Effects of both deterministic and stochastic rounding.Journal of Climate, 35(4):1215–1229, 2022. doi:10.1175/JCLI-D-21-0343.1

    work page doi:10.1175/jcli-d-21-0343.1 2022

  18. [18]

    Elowitz, Arnold J

    Michael B. Elowitz, Arnold J. Levine, Eric D. Siggia, and Peter S. Swain. Stochastic gene expression in a single cell.Science, 297(5584):1183–1186, 2002. doi:10.1126/science.1070919

    work page doi:10.1126/science.1070919 2002

  19. [19]

    Gutenkunst, Joshua J

    Ryan N. Gutenkunst, Joshua J. Waterfall, Fergal P. Casey, Kevin S. Brown, Christopher R. Myers, and James P. Sethna. Universally sloppy parameter sensitivities in systems biology models.PLoS Computational Biology, 3 (10):e189, 2007. doi:10.1371/journal.pcbi.0030189

    work page doi:10.1371/journal.pcbi.0030189 2007

  20. [20]

    Babtie, Paul Kirk, and Michael P

    Ann C. Babtie, Paul Kirk, and Michael P. H. Stumpf. Topological sensitivity analysis for systems biology. Proceedings of the National Academy of Sciences, 111(52):18507–18512, 2014. doi:10.1073/pnas.1414026112

    work page doi:10.1073/pnas.1414026112 2014

  21. [21]

    Efficient formulation of the stochastic simulation algorithm for chemically reacting systems.The Journal of Chemical Physics, 121(9):4059–4067, 2004

    Yang Cao, Hong Li, and Linda Petzold. Efficient formulation of the stochastic simulation algorithm for chemically reacting systems.The Journal of Chemical Physics, 121(9):4059–4067, 2004. doi:10.1063/1.1778376

    work page doi:10.1063/1.1778376 2004

  22. [22]

    Gibson and Jehoshua Bruck

    Michael A. Gibson and Jehoshua Bruck. Efficient exact stochastic simulation of chemical systems with many species and many channels.The Journal of Physical Chemistry A, 104(9):1876–1889, 2000. doi:10.1021/jp993732q

    work page doi:10.1021/jp993732q 2000

  23. [23]

    Thompson, and Steven J

    Alexander Slepoy, Aidan P. Thompson, and Steven J. Plimpton. A constant-time kinetic Monte Carlo algorithm for simulation of large biochemical reaction networks.The Journal of Chemical Physics, 128(20):205101, 2008. doi:10.1063/1.2919546

    work page doi:10.1063/1.2919546 2008

  24. [24]

    Karr, Jayodita C

    Jonathan R. Karr, Jayodita C. Sanghvi, Derek N. Macklin, Miriam V . Gutschow, Jared M. Jacobs, Benjamin Bolival, Nacyra Assad-Garcia, John I. Glass, and Markus W. Covert. A whole-cell computational model predicts phenotype from genotype.Cell, 150(2):389–401, 2012. doi:10.1016/j.cell.2012.05.044

    work page doi:10.1016/j.cell.2012.05.044 2012

  25. [25]

    Simulation of stochastic reaction- diffusion processes on unstructured meshes.SIAM Journal on Scientific Computing, 31(3):1774–1797, 2009

    Stefan Engblom, Lars Ferm, Andreas Hellander, and Per Lötstedt. Simulation of stochastic reaction- diffusion processes on unstructured meshes.SIAM Journal on Scientific Computing, 31(3):1774–1797, 2009. doi:10.1137/080721388

    work page doi:10.1137/080721388 2009

  26. [26]

    Agent-based models in translational systems biology.WIREs Systems Biology and Medicine, 1(2):159–171, 2009

    Gary An, Qi Mi, Joyeeta Dutta-Moscato, and Yoram V odovotz. Agent-based models in translational systems biology.WIREs Systems Biology and Medicine, 1(2):159–171, 2009. doi:10.1002/wsbm.45

    work page doi:10.1002/wsbm.45 2009

  27. [27]

    Shiryaev.Limit Theorems for Stochastic Processes

    Willem Hundsdorfer and Jan Verwer.Numerical Solution of Time-Dependent Advection-Diffusion-Reaction Equations, volume 33 ofSpringer Series in Computational Mathematics. Springer, 2003. doi:10.1007/978-3-662- 09017-6

    work page doi:10.1007/978-3-662- 2003

  28. [28]

    Gillespie

    Daniel T. Gillespie. Approximate accelerated stochastic simulation of chemically reacting systems.The Journal of Chemical Physics, 115(4):1716–1733, 2001. doi:10.1063/1.1378322

    work page doi:10.1063/1.1378322 2001

  29. [29]

    Anderson

    David F. Anderson. Incorporating postleap checks in τ-leaping.The Journal of Chemical Physics, 128(5):054103,

  30. [30]

    doi:10.1063/1.2819665

    work page doi:10.1063/1.2819665

  31. [31]

    R-leaping: Accelerating the stochastic simulation algorithm by reaction leaps.The Journal of Chemical Physics, 125(8):084103, 2006

    Anne Auger, Philippe Chatelain, and Petros Koumoutsakos. R-leaping: Accelerating the stochastic simulation algorithm by reaction leaps.The Journal of Chemical Physics, 125(8):084103, 2006. doi:10.1063/1.2218339

    work page doi:10.1063/1.2218339 2006

  32. [32]

    An exact accelerated stochastic simulation algorithm.The Journal of Chemical Physics, 130(14):144110, 2009

    Eric Mjolsness, David Orendorff, Philippe Chatelain, and Petros Koumoutsakos. An exact accelerated stochastic simulation algorithm.The Journal of Chemical Physics, 130(14):144110, 2009. doi:10.1063/1.3078490

    work page doi:10.1063/1.3078490 2009

  33. [33]

    IEEE Std 754-2019 (Revision of IEEE 754-2008)

    IEEE standard for floating-point arithmetic, 2019. IEEE Std 754-2019 (Revision of IEEE 754-2008). 19 Reduced-Precision Stochastic Simulation

    2019

  34. [34]

    Kolmogorov

    Andrey N. Kolmogorov. Sulla determinazione empirica di una legge di distribuzione.Giornale dell’Istituto Italiano degli Attuari, 4:83–91, 1933

    1933

  35. [35]

    Nikolai V . Smirnov. Table for estimating the goodness of fit of empirical distributions.The Annals of Mathematical Statistics, 19(2):279–281, 1948. doi:10.1214/aoms/1177730256

    work page doi:10.1214/aoms/1177730256 1948

  36. [36]

    Vaserstein

    Leonid N. Vaserstein. Markov processes over denumerable products of spaces, describing large systems of automata.Problemy Peredachi Informatsii, 5(3):64–72, 1969

    1969

  37. [37]

    Optimal transport , volume 338 of Grundlehren der math- ematischen Wissenschaften [Fundamental Principles of Mathematical Sci- ences]

    Cédric Villani.Optimal Transport: Old and New, volume 338 ofGrundlehren der mathematischen Wissenschaften. Springer, 2009. doi:10.1007/978-3-540-71050-9

    work page doi:10.1007/978-3-540-71050-9 2009

  38. [38]

    Chemical reaction models for non-equilibrium phase transitions.Zeitschrift für Physik, 253(2): 147–161, 1972

    Friedrich Schlögl. Chemical reaction models for non-equilibrium phase transitions.Zeitschrift für Physik, 253(2): 147–161, 1972. doi:10.1007/BF01379769

    work page doi:10.1007/bf01379769 1972

  39. [39]

    Düben, and Tim N

    Milan Klöwer, Sam Hatfield, Matteo Croci, Peter D. Düben, and Tim N. Palmer. Fluid simulations accelerated with 16 bits: Approaching 4× speedup on A64FX by squeezing ShallowWaters.jl into Float16.Journal of Advances in Modeling Earth Systems, 14(2):e2021MS002684, 2022. doi:10.1029/2021MS002684

    work page doi:10.1029/2021ms002684 2022

  40. [40]

    Markovian modeling of gene-product synthesis.Theoretical Population Biology, 48(2):222–234, 1995

    Jean Peccoud and Bernard Ycart. Markovian modeling of gene-product synthesis.Theoretical Population Biology, 48(2):222–234, 1995. doi:10.1006/tpbi.1995.1027

    work page doi:10.1006/tpbi.1995.1027 1995

  41. [41]

    Gillespie

    Daniel T. Gillespie. Stochastic simulation of chemical kinetics.Annual Review of Physical Chemistry, 58(1): 35–55, 2007. doi:10.1146/annurev.physchem.58.032806.104637

    work page doi:10.1146/annurev.physchem.58.032806.104637 2007

  42. [42]

    A synthetic oscilla- tory network of transcriptional regulators,

    Michael B. Elowitz and Stanislas Leibler. A synthetic oscillatory network of transcriptional regulators.Nature, 403(6767):335–338, 2000. doi:10.1038/35002125

    work page doi:10.1038/35002125 2000

  43. [43]

    Dissecting the Graphcore IPU architecture via microbenchmarking, 2019

    Zhe Jia, Blake Tillman, Marco Maggioni, and Daniele Paolo Scarpazza. Dissecting the Graphcore IPU architecture via microbenchmarking, 2019

    2019

  44. [44]

    arXiv preprint arXiv:2502.01070 , year=

    Jiwoo Kim, Joonhyung Lee, Gunho Park, Byeongwook Kim, Se Jung Kwon, Dongsoo Lee, and Youngjoo Lee. An inquiry into datacenter TCO for LLM inference with FP8.arXiv preprint arXiv:2502.01070, 2025. doi:10.48550/arXiv.2502.01070. Documents hardware-accelerated stochastic rounding as a distinctive feature of Intel Gaudi HPUs

    work page doi:10.48550/arxiv.2502.01070 2025

  45. [45]

    Algorithms for stochastically rounded elementary arithmetic operations in IEEE 754 floating-point arithmetic.IEEE Transactions on Emerging Topics in Computing, 9(3):1451–1466,

    Massimiliano Fasi and Mantas Mikaitis. Algorithms for stochastically rounded elementary arithmetic operations in IEEE 754 floating-point arithmetic.IEEE Transactions on Emerging Topics in Computing, 9(3):1451–1466,

  46. [46]

    doi:10.1109/TETC.2021.3069165

    work page doi:10.1109/tetc.2021.3069165 2021

  47. [47]

    Federico Brogi, S. Bna, G. Boga, Giorgio Amati, T. Esposti Ongaro, and M. Cerminara. On floating point precision in computational fluid dynamics using OpenFOAM.Future Generation Computer Systems, 152:1–16,

  48. [48]

    doi:10.1016/j.future.2023.10.006. 20

    work page doi:10.1016/j.future.2023.10.006 2023