Saguaro: Multidisciplinary Research Innovations

• Saguaro is a collection of innovative, interdisciplinary research projects spanning optical follow-up of gravitational waves, high-resolution IGM simulations, galaxy evolution, edge blockchain, and speculative ML decoding.
• Key contributions include a rapid multimessenger system for gravitational-wave alerts, parameter-rich IGM models for reionization studies, and a benchmark high-z AGN/galaxy system for astrophysical insights.
• The suite also advances distributed ledger protocols for edge computing and introduces asynchronous speculative decoding, significantly improving inference throughput and operational efficiency.

Saguaro is a term denoting a collection of distinct, high-impact research projects and artifacts in astronomy, astrophysics, edge computing, and machine learning, each unified by substantial technical innovation and broad community utility. These include: (1) SAGUARO, an advanced infrastructure for electromagnetic follow-up of gravitational wave events; (2) SAGUARO, a simulation suite for small-scale intergalactic medium (IGM) physics; (3) WISEA J123635.56+621424.2 (“the Saguaro”), an archetypal AGN/galaxy system informing high-redshift galaxy interpretation; (4) Saguaro, an edge-enabled hierarchical permissioned blockchain; and (5) Saguaro, an optimized algorithm for speculative speculative decoding in LLM inference. Each project has advanced the state of scientific or technical methodology, as annotated below.

1. SAGUARO: Optical Follow-Up of Gravitational-Wave Triggers

SAGUARO (“Searches after Gravitational Waves Using ARizona Observatories”) is an end-to-end multimessenger program targeting rapid identification and vetting of optical counterparts to LIGO/Virgo/KAGRA-detected gravitational wave (GW) events. Leveraging the Steward Observatory 1.5 m Mt. Lemmon telescope (with a 5 deg² field of view and a median G-band 5$\sigma$ depth of ~21.1 AB mag), SAGUARO achieves rapid triggered and serendipitous tiling, supported by a multi-telescope Arizona network for prompt photometric and spectroscopic follow-up (Paterson et al., 2020, Hosseinzadeh et al., 2023, Lundquist et al., 2019).

Key systems include:

• A web-based Target and Observation Manager (TOM), integrating alert ingestion (via Kafka/Hopskotch), survey field probability rasterization (via HEALPix MOC), greedy adjacency-aware scheduling, and catalog cross-matching (Minor Planet Center, Milliquas, Gaia, PanSTARRS, GLADE+, HECATE, SDSS DR12, PS1-STRM), enabling real-time decision support and reporting (Hosseinzadeh et al., 2023).
• A scalable, deeply augmented image-subtraction pipeline; initially based on ZOGY difference imaging and a random forest “real/bogus” triage using 10×10 pixel cutouts, and for O4 further advanced with CNN-based classifiers, utilizing 16×16 cutouts of the significance image S_corr. The O4 CNN achieves TPR ≈ 0.94 and FPR ≈ 0.02 at threshold θ = 0.5; the nightly vet list reduction is ~4× compared to prior classifiers (Hosseinzadeh et al., 2023).
• Kilonova decision support, operating with multi-epoch photometry, rapid host galaxy association (via P_cc as in Bloom et al. 2002), and model comparison against a bank spanning AT2017gfo, NSBH, and GW190425-like kilonovae.

Processing latency from GW alert to candidate vetting is routinely <1 hr; detection sensitivity extends to m_AB ≈ 21.0–21.3, and area coverage is routinely 1.7–86% of published localizations (Paterson et al., 2020, Hosseinzadeh et al., 2023, Lundquist et al., 2019).

2. SAGUARO: High-Resolution Simulations of the IGM During Reionization

SAGUARO (Simulating IGM Evolution and Environments At High Resolution) denotes a grid of coupled 3D radiative-hydrodynamics simulations resolving the IGM at the Jeans scale (sub-kpc) across a multi-dimensional parameter space of reionization histories (Cain et al., 26 Mar 2026). The suite comprises 236 runs with boxes of 2 or 0.25 $h^{-1}$Mpc at 1024³ resolution, systematically varying reionization redshift ($z_{re}=5$, $7$, $9$, $15$), photoionization rate ($\Gamma_{HI}/10^{-12}\ \mathrm{s}^{-1}=0.03$–$30$), and large-scale overdensity ($\delta/\sigma = -\sqrt{3}, 0, +\sqrt{3}$), as well as modules exploring X-ray pre-heating, baryon–dark matter streaming, recombination radiation, and warm dark matter cosmologies.

Scientific outputs include:

• Quantitative mapping of self-shielding ($\Delta_{ss}\propto (1+z)^{-2.7}\Gamma_{HI}^{0.6}$), mean free path scaling ($h^{-1}$0, $h^{-1}$1, $h^{-1}$2), and Ly$h^{-1}$3 damping wing absorption signatures (persistent for $h^{-1}$4 Myr post-reionization).
• Recalibration of clumping factors for accurate reionization photon budgets; the relative differences between absorption-based and recombination-based clumping exceed $h^{-1}$5, with significant implications for global reionization timeline modeling.
• Dark matter model dependence: the halo mass function at $h^{-1}$6–$h^{-1}$7 matches large volume benchmarks to $h^{-1}$8, with warm dark matter models sharply suppressing small-$h^{-1}$9 halo abundance.

This parameter survey directly informs subgrid models, Ly$z_{re}=5$0 forest power spectrum interpretation, and 21 cm reionization forecasts (Cain et al., 26 Mar 2026).

3. The Saguaro: A Benchmark System for High-Redshift Galaxy Nuclei

WISEA J123635.56+621424.2, designated “the Saguaro”, is a $z_{re}=5$1 compact, IR-bright AGN/star-forming galaxy in GOODS-North, empirically linked as an analog to high-redshift “Little Red Dot” (LRD) galaxies observed by JWST (Rinaldi et al., 23 Jul 2025). This system’s nuclear SED (spectral slopes $z_{re}=5$2, $z_{re}=5$3), broad H$z_{re}=5$4 (FWHM $z_{re}=5$5 km$z_{re}=5$6s$z_{re}=5$7), high bolometric luminosity ($z_{re}=5$8), large $z_{re}=5$9, and Eddington ratio $7$0, exemplify the “V-shaped” SEDs characteristic of LRDs.

JWST/NIRCam imaging and bulge–disk decompositions (via GALFITM) reveal the Saguaro as a face-on spiral host with an exponential disk ($7$1, $7$2 kpc, decreasing with wavelength) and a nuclear point source. Surface-brightness dimming analysis, applying $7$3, demonstrates that at $7$4 only $7$5 of host light would survive above typical detection thresholds, rendering the host invisible and mimicking an LRD. Stacked UV profiles of 99 photometric LRDs uncover consistent faint, diffuse envelopes, indicating that high-$7$6 LRDs often mark only the nucleus of a more extended galaxy suppressed by cosmological dimming (Rinaldi et al., 23 Jul 2025).

4. Saguaro: Hierarchical Permissioned Blockchain for Edge Computing

Saguaro is a permissioned blockchain system designed for hierarchical edge architectures, combining geography-aware, low-latency cross-domain transaction protocols with scalable ledger aggregation and mobile consensus (Amiri et al., 2021). The architecture features:

• Height-1 domains (edge clusters) maintaining linear blockchains with consensus among $7$7 (CFT) or $7$8 (BFT) replicas; upper-level (fog/cloud) domains summarize and aggregate ledgers as DAGs for efficient abstraction.
• Two core cross-domain transaction protocols:
  • A coordinator-based protocol, where the lowest common ancestor domain orchestrates transaction ordering and commit; the round complexity is detailed via pseudocode and yields $7$9–$9$0 higher throughput than flat approaches (e.g., AHL, SharPer) at similar latencies for cross-domain workloads.
  • An optimistic protocol, allowing edge domains to speculatively execute and later reconcile transactions, favoring low-conflict, high-locality regimes.
• Lazy propagation and data aggregation compress communication and storage demands; mobile device transactions are executed with a single additional roundtrip without requiring cloud intervention.

Experimental results yield end-to-end latencies of $9$1–$9$2 ms and throughput up to $9$3 ktx/s (crash-only), robust to increasing mobile/cross-domain fractions, and scalable across geographically distributed cloud-edge deployments (Amiri et al., 2021).

5. Saguaro: Speculative Speculative Decoding for LLMs

In neural LLM inference, Saguaro is the first fully asynchronous speculative decoding algorithm, formalizing the speculative speculative decoding (SSD) paradigm to overlap speculation and verification steps for autoregressive decoders (Kumar et al., 3 Mar 2026).

Key innovations include:

• A speculation cache holding $9$4 pre-fetched blocks corresponding to the most probable verification outcomes (chosen by geometric fan-out, with $9$5), selected by a constrained optimization given the statistics of draft token acceptance.
• Cache-aware draft sampling (hyperparameter $9$6) trading off acceptance rate and cache-hit probability; monotonic $9$7 improvement as $9$8 at the expense of acceptance.
• Batched, adaptive fallback speculators, using fast neural or n-gram models for large batch sizes per the derivation $9$9.
• Theoretical and empirical throughput shows up to $15$0 speedup over standard speculative decoding (SD) and up to $15$1 over autoregressive baselines, with throughput scaling nearly linearly with cache-hit probability and batch size.

End-to-end implementation uses custom CUDA kernels, PagedAttention, and sparse attention masks; mask construction dominates overhead (Kumar et al., 3 Mar 2026).

6. Impact and Community Implications

The suite of Saguaro projects represents high-leverage advances for multiple research communities:

• Astronomical time-domain surveys and multimessenger astrophysics benefit from the combination of real-time alert brokering, robust transient classification, and follow-up orchestration. The SAGUARO platform’s design and open-source tools provide a model for managing the projected data volume of Rubin Observatory (~$15$2 transients/night) (Hosseinzadeh et al., 2023).
• Theoretical cosmology and large-volume simulations gain essential subgrid calibrations and parameter dependencies from SAGUARO’s suite of IGM simulations, critical for interpreting next-generation Ly$15$3 and 21cm surveys (Cain et al., 26 Mar 2026).
• Galaxy evolution studies are informed by the Saguaro galaxy as a low-$15$4 archetype revealing selection biases in JWST high-$15$5 point-source samples (Rinaldi et al., 23 Jul 2025).
• Decentralized cloud/edge systems have an efficient, fault-tolerant, and scalable blockchain infrastructure supporting cross-domain and mobile use cases (Amiri et al., 2021).
• The ML systems community obtains rigorous methodology and implementation blueprints for lossless, asynchronous, speculative decoding—directly impacting large-model inference throughput and latency (Kumar et al., 3 Mar 2026).

The Saguaro artifacts are already integrated into community workflows and serve as reference points for next-generation infrastructure and methodology development across astrophysics, cosmology, distributed systems, and machine learning.