--- title: 'Hermes: A Multidisciplinary Designation' url: https://www.emergentmind.com/topics/hermes-64b181a3-0e32-4a14-afb3-7d97651ef429 type: topic --- # Hermes: A Multidisciplinary Designation Hermes is not a single research object but a recurrent scientific designation used for multiple unrelated systems, instruments, surveys, codes, and frameworks. In the literature represented here, it denotes a far-infrared extragalactic survey, two different optical spectrographs, a CubeSat transient mission, Monte Carlo and radiative-transfer codes in astroparticle physics, several networking and distributed-systems architectures, a privacy-preserving vehicle-access system, a space-computing qualification project, a pre-training data-labeling substrate, and a Bayesian exoplanet population model [1203.2562] [1509.00129] [1310.5235] [2409.00989] [1307.4356] [2105.13165] [1401.4629] [2411.13668] [2001.09804] [2101.01950] [2302.06427] [2607.02266] [2606.02696]. Unqualified use is therefore intrinsically ambiguous; the meaning is fixed by domain, acronym expansion, and technical context. ## 1. Ambiguity, acronym expansions, and research domains The name appears in both camel-case and upper-case forms. In far-infrared astronomy, **HerMES** expands to **Herschel Multi-tiered Extragalactic Survey**. In optical instrumentation, **HERMES** expands either to **High Efficiency and Resolution Multi Element Spectrograph** at the Anglo-Australian Telescope or to **High Efficiency and high Resolution Mercator Echelle Spectrograph** at the Mercator telescope. In high-energy astrophysics it denotes **High Energy Modular Ensemble of Satellites** and later **High Energy Rapid Modular Ensemble of Satellites**. In machine learning and exoplanet science it expands to **a Multi-Granularity Labeling Substrate for Pre-training Data Mixtures** and **HiERarchical Modelling for Exoplanet Science**, respectively. Other expansions include a space-technology project for **qualification of High performance programmable Microprocessor and development of Software ecosystem** and a computing-continuum framework for **Heterogeneous Computing Continuum with Resource Monetization, Orchestration, and Semantic interoperability** [1203.2562] [1509.00129] [1310.5235] [1812.02432] [2409.00989] [2607.02266] [2606.02696] [2302.06427] [2512.08288]. | Domain | Meaning of Hermes | Representative papers | |---|---|---| | Far-infrared astronomy | Herschel Multi-tiered Extragalactic Survey | [1203.2562], [1109.5186], [1005.2396] | | Optical spectroscopy | AAT and Mercator HERMES spectrographs | [1509.00129], [1011.0258], [1310.5235] | | High-energy astrophysics | Modular/rapid modular ensemble of satellites | [1812.02432], [2409.00989] | | Astroparticle and diffuse-emission simulation | UHECR propagation and Galactic multi-messenger codes | [1307.4356], [1305.4364], [2105.13165] | | Computer systems | Photonic interconnect, proxy overlay, replication, privacy-preserving access | [1401.4629], [2411.13668], [2001.09804], [2101.01950] | | Platform ecosystems and inference | Space computing, continuum orchestration, data mixtures, exoplanet hierarchy | [2302.06427], [2512.08288], [2607.02266], [2606.02696] | A common misconception is that “Hermes” denotes a unified research lineage. The corpus instead shows repeated independent coinage across disciplines. This suggests that the term functions primarily as a mnemonic acronym rather than as a stable cross-field brand. ## 2. HerMES in far-infrared extragalactic astronomy HerMES, the **Herschel Multi-tiered Extragalactic Survey**, is a **Guaranteed Time Key Program** on the *Herschel Space Observatory* designed as a legacy survey of dusty, star-forming galaxies in the far-infrared and submillimetre regime where dust-reprocessed emission peaks. The survey uses **SPIRE** at **250, 350, and 500 $\mu$m** and **PACS** at **100 and 160 $\mu$m**. Its design is explicitly multi-tiered: the survey maps nested fields totaling approximately **$\sim 380~{\rm deg}^2$**, with field sizes ranging from **0.01 to $\sim 20~{\rm deg}^2$**, plus an additional wider SPIRE-only component of **$270~{\rm deg}^2$**. It was intended to detect of order **100,000 galaxies at $5\sigma$** in some of the best-studied fields in the sky, while addressing the evolution of the infrared luminosity density, the far-infrared luminosity function, clustering of dusty galaxies, and populations below the confusion limit through lensing and statistical techniques [1203.2562]. Survey design and data products were shaped by confusion-limited imaging. The SPIRE beam FWHM values quoted for the survey are **18.2 arcsec**, **24.9 arcsec**, and **36.3 arcsec** at **250**, **350**, and **500 $\mu$m**, respectively, and empirically estimated **$5\sigma$** confusion levels are **24.0**, **27.5**, and **30.5 mJy** at those same wavelengths. HerMES combined direct source extraction with lensed cluster observations, multi-colour $P(D)$ analyses, and prior-based deblending. The program also defined public data products including **SCAT**, **SMAP**, **PCAT**, **PMAP**, **SPCAT**, **CLUS**, and **XID**, alongside staged releases such as the **2010 July 1** Early Data Release and later releases through ESA and **HeDaM** [1203.2562]. The paper on single-band point-source catalogues from the Science Demonstration Phase established the technical reference methodology for HerMES SPIRE catalogues. It produced the first **single-band point-source catalogues** from about **20 deg²** in five fields. Extraction used two complementary steps: **peak finding** to identify local maxima and **Gaussian point-response-function fitting** with `sussextractor` to estimate flux densities. A Gaussian PRF was assumed with FWHM of **18.15**, **25.15**, and **36.3 arcsec** at **250**, **350**, and **500 $\mu$m**. Validation relied on injected artificial sources. A “good” detection satisfied $$ \Delta r < \mathrm{FWHM} $$ and $$ \frac{1}{2} < \frac{S_\mathrm{measured}}{S_\mathrm{injected}} < 2. $$ The reported flux densities at **50% good-detection completeness** ranged from **11.6–25.7 mJy** at **250 $\mu$m**, **13.2–27.1 mJy** at **350 $\mu$m**, and **13.1–35.8 mJy** at **500 $\mu$m**. The deepest field, GOODS-North, reached **11.6**, **13.2**, and **13.1 mJy**; the shallowest, Lockman-SWIRE, reached **25.7**, **27.1**, and **35.8 mJy** [1109.5186]. HerMES was also used to infer galaxy populations from SPIRE colours. A **500 $\mu$m$** flux-limited sample with **$S/N > 3$** and isolation criteria was constructed in **GOODS-N**, **Lockman-North**, **FLS**, and **Lockman-SWIRE**, yielding **21**, **38**, **242**, and **244** matched sources. In colour-colour and colour-flux space, comparison with template tracks from **Pearson et al. (2007)**, **Xu et al. (2001)**, **Lagache et al. (2003)**, and **Dale et al. (2002)** implied that most SPIRE-detected galaxies lie at **$1 80,000**, **spectral range > 380–880 nm**, **throughput > 25% in V**, **cycle time < 60 s**, and **radial-velocity stability < 5 m/s**. The realized instrument covers **377 to 900 nm** in a single exposure, distributes the format over **55 orders**, and provides **$R = 85{,}000$** in the **HRF** mode and **$R = 63{,}000$** in the **LRF** mode. Spectrograph-only peak efficiency reaches **28%**, while total peak efficiency including the telescope is **17.5%**. By the 2013 performance paper, HERMES had become the **work-horse instrument** of Mercator, used for about **80% of the observing nights**, and had collected **more than 42,000 science spectra** over **1250 observing nights** [1011.0258] [1310.5235]. A third astronomical use concerns high-energy transient monitoring from CubeSats. The 2018 HERMES project paper defined **High Energy Modular Ensemble of Satellites** as a modular X/gamma-ray transient monitor based on nano-satellites, with a **Technological Pathfinder** of **3 nano-satellites**, a **Scientific Pathfinder** of **6–8 satellites**, and a longer-term constellation of **hundreds of nano-satellites**. The detector followed the **“siswich”** principle, combining **Silicon Drift Detectors** and **GAGG:Ce** scintillators, aimed at an energy range of approximately **3–5 keV to 2 MeV**, **sub-microsecond time resolution**, and localization accuracy below a degree, with an explicit triangulation scaling argument for much finer localization at larger $N$ and longer baselines [1812.02432]. The later **HERMES Pathfinder mission**, now expanded as **High Energy Rapid Modular Ensemble of Satellites**, specifies a **constellation of six 3U CubeSats in low-Earth orbit** designed to detect and localize bright transients such as **GRBs** and to complement future **gravitational-wave** observatories. Each unit uses a hybrid **SDD / GAGG:Ce** payload with a **“double detection”** mechanism: direct X-ray detection in the SDDs and scintillation-light readout for harder photons. The anticipated payload figures include **peak effective area 52 cm²**, **field of view 3.2 sr FWHM**, **low-energy threshold $\le 3$ keV**, **time resolution 320 ns (68% c.l.)**, **time accuracy 181 ns (68% c.l.)**, **payload mass 1.55 kg**, and **payload power 1.8 W**. The mission requirements include detecting GRBs with peak flux **$\ge 0.5$–$1$ ph/s/cm²** in the **50–300 keV** band and detecting **$\ge 40$ long GRBs** and **$\ge 8$ short GRBs** simultaneously in at least **3 units** with efficiency **$\ge 40$–$50\%$** in each unit. With a **50% duty cycle** and two observing triplets, the reported expected annual rates are **147** long GRBs in **S-mode**, **131** in **X-mode**, **19** short GRBs in **S-mode**, and **0.3** in **X-mode** [2409.00989]. ## 4. Simulation codes in astroparticle and Galactic-emission studies In astroparticle physics, HERMES is a Monte Carlo code for the propagation of ultra-high-energy cosmic rays. The code propagates nuclei in the range $$ 10^{16}\ \mathrm{eV} \le E \le 10^{22}\ \mathrm{eV}, $$ within a general FRW or $\Lambda$CDM cosmology, including source evolution, magnetic deflections, and interactions with the **CMB**, **CIOB/CIRB**, and **URB/CRB**. The injected source spectrum is written as $$ Q(z,E)=\mathcal{H}(z)\,Q(0,E), \qquad Q(0,E)=\mathcal{L}\,\mathcal{N}\,E^{-\gamma}, $$ and the continuous-loss propagation equation is $$ \frac{1}{E}\frac{dE}{dz}=-\beta(z,E)\frac{dt}{dz}. $$ Implemented processes include **adiabatic losses**, **electron/positron pair production**, **photo-pion production**, and **photodisintegration**, together with secondary-particle production. The code supports **BSS** and **ASS** Galactic regular magnetic-field models and uses the **Giacalone and Jokipii** approach for turbulent fields. The papers report agreement with **CRPropa** and with published calculations for interaction lengths and GZK horizons, while presenting applications to energy spectra, horizons, and arrival-direction distributions using source catalogues such as **2MRS** and the **SWIFT-BAT 58-month AGN catalog** [1307.4356] [1305.4364]. A separate HERMES code, expanded as **High-Energy Radiative MESsengers**, is a public, modular numerical framework for diffuse Galactic multi-wavelength and multi-messenger emission. It generates sky maps and spectra from radio through gamma rays and neutrinos. The modeled processes include **Faraday rotation**, **dispersion measure**, **free-free emission and absorption**, **synchrotron emission**, **inverse Compton scattering**, **pion production and decay**, **bremsstrahlung**, **gamma-ray attenuation by pair production**, **neutrino production**, and **dark matter annihilation**. For example, the rotation measure is written as $$ \mathrm{RM}(l,b) = \frac{q^3}{8 \pi^2 \epsilon_0 m_e^2 c^3} \int_0^\infty ds\, n_e(\mathbf r)\, B_{\parallel}(\mathbf r), $$ and the synchrotron brightness temperature follows $$ T_b(\nu,l,b)=\frac{c^2}{2\nu^2k_B}I_\nu(\nu,l,b). $$ The software uses a **modern C++** core, Python bindings via **pybind11**, documentation via **Doxygen** and **Sphinx**, parallelization with **C++11 threads**, HEALPix-compatible skymaps, and a public **GNU GPL v3** release. Demonstration outputs include **408 MHz** radio maps, **10 GeV** $\pi^0$-decay and inverse-Compton gamma-ray maps, very-high-energy absorption in the inner Galaxy above **$\sim 10$ TeV**, and neutrino maps consistent with hadronic gamma-ray production [2105.13165]. These two uses of HERMES are often conflated because both concern high-energy astrophysics. The distinction is categorical: one simulates **propagation of UHECRs through cosmological space**, the other simulates **diffuse Galactic radiative and multi-messenger emission** from an assumed local cosmic-ray environment. This suggests that identical naming can mask very different forward models and observables. ## 5. Computer systems, networking, and secure access In computer architecture, **Hermes** is a hierarchical silicon-photonic interconnect for **future thousand-core many-core chips**. The design combines a **broadcast sub-network** for low-latency coordination and multicast with a **circuit-switch linear sub-network** for high-throughput point-to-point transfers. Its physical primitive is an **SoI-based adiabatic coupler** with reported **96%** power efficiency, **48–52%** splitting ratio, and **45–55%** splitting under worst-case process variation across the **C+L** band. The broadcast topology is a **2-ary folded butterfly**, while the hierarchical organization uses **32-core local domains** and a **global domain**. The paper claims scalability to at least **1024 cores**, with hierarchy reducing both power and latency to **$\mathcal{O}(\sqrt{N})$** [1401.4629]. In networking, **Hermes** is a **general-purpose proxy-enabled networking architecture** built on an overlay of dynamically reconfigurable proxies. It delegates networking responsibilities from applications and services to **dependent proxies**, **standalone proxies**, and optional **assisting components**, while using HTTP as a semantic envelope for routing and processing. Supported techniques include **TCP over HTTP** with `CONNECT`, **UDP over HTTP** with `CONNECT-UDP`, **IP over HTTP** with `CONNECT-IP`, and direct TCP/UDP forwarding. The prototype uses **Envoy**, **Envoy Mobile**, a Go control plane, **Apache CouchDB**, **OPA**, and custom components for tunnel interfaces and split tunneling. Reported evaluations include **0%** success for direct TCP over an intermittent path versus **100%** success with the Hermes overlay, and latency reduction for policy-driven IP routing from about **382 ms** to **206 ms**, then to **79.68 ms** after moving ingress and server placement [2411.13668]. In distributed storage, **Hermes** is a **broadcast-based reliable replication protocol** for in-memory datastores that provides local reads and fully concurrent writes while preserving **linearizability**. It uses logical timestamps and cache-coherence-inspired invalidations. A write executes through the **INV / ACK / VAL** sequence, with timestamps of the form $$ TS = [v, c_{id}], $$ ordered lexicographically. Reads complete only from the **Valid** state, while writes never abort; interrupted writes are made fault-tolerant through **replayable writes**. In the authors’ **HermesKV** implementation over RDMA with five replicas, throughput reaches **770 MReqs/s** at **1%** writes, and at **5%** writes the tail latency is reported as **3.6X lower** than **CRAQ** and **ZAB** [2001.09804]. In applied cryptographic systems, **HERMES** is a **scalable, secure, and privacy-enhancing vehicle access system** for sharing vehicles. It securely outsources access-token generation to a set of untrusted servers using secure multiparty computation, while concealing vehicle secret keys, booking details, access-token contents, and user and vehicle identities. The booking details are defined as $$ BD^{u_o,u_c} = \left\{\mathsf{hash}(\mathit{Cert}^{u_c}),\, ID^{veh_{u_o}},\, L^{veh_{u_o}},\, CD^{u_c},\, AC^{u_c},\, ID^{BD}\right\}. $$ The protocol computes an encrypted access token and an authentication tag, publishes them on a public ledger, and permits offline vehicle access through OBU verification. Evaluations using **HtMAC-MiMC** and **CBC-MAC-AES** report approximately **1.83 ms** for access-token generation for a single-vehicle owner, approximately **11.9 ms** for a rental branch with over a thousand vehicles, throughput of **546** and **84** access-token generations per second in those cases, and a vehicle-side token-processing time of approximately **62.087 ms** [2101.01950]. ## 6. Space-grade computing and computing-continuum architectures In European space technology, **HERMES** is a coordinated project for the **qualification of High performance programmable Microprocessor and development of Software ecosystem**. Its main hardware target is the **NG-ULTRA FPGA**, described as a **28 nm FD-SOI**, radiation-hardened **SoC FPGA** with a **quad-core ARM Cortex-R52 processor running at 600 MHz** and **550k LUTs**. The project aims to bring the NG-ULTRA platform with its **ceramic hermetic package CGA 1752** to **TRL 6**, meaning **validated and demonstrated in a relevant environment**, under **ESCC** qualification logic and toward **ECSS level B** software compliance. The software ecosystem includes **Bambu** for **High-Level Synthesis**, **XtratuM NextGeneration** for multicore virtualization, and a **Generic Level 1 boot loader (BL1)**. Concrete functions attributed to BL1 include initialization of CPU#0 registers, caches, exceptions, **PLL clocks**, **DDR controller**, **Flash controller**, **SpaceWire controller**, and tightly coupled memories, together with **Memory Protection Unit** setup, eFPGA bitstream loading, integrity checks, and boot reporting [2302.06427]. A different HERMES paper proposes a framework for the **computing continuum**, expanded as **Heterogeneous Computing Continuum with Resource Monetization, Orchestration, and Semantic interoperability**. The framework treats cloud, fog, edge, and IoT as one hyper-distributed environment connected through three bridges: **resource monetization**, **orchestration**, and **semantic interconnect**. Key technical elements are **WebAssembly (WASM)** as a portable execution substrate, **lenses** as pluggable data-path processing modules, a distributed marketplace using **blockchain** and **smart contracts**, a hierarchy of lightweight orchestrators and agents, and a semantic layer based on ontologies, metadata, and knowledge bases. Unlike the space-qualification project, this HERMES paper is explicitly architectural rather than implementation-heavy: it does **not** report a formal experimental evaluation, benchmark suite, prototype implementation, or measured performance results [2512.08288]. The two projects operate at different abstraction layers. The space-computing HERMES focuses on a specific **rad-hard FPGA/SoC platform** and its software toolchain; the continuum HERMES focuses on distributed orchestration, data monetization, and semantic interoperability across heterogeneous infrastructures. A plausible implication is that the recurrence of the name in systems research reflects the broad appeal of framing a technical stack as an integrated ecosystem rather than as an isolated component. ## 7. Hierarchical data and inference frameworks In machine learning for data-mixture design, **HERMES** is a **labeling substrate for pre-training data mixtures** rather than a clustering algorithm per se. It addresses the claim that the bottleneck in mixture design is the **label system**, not the mixer. The core mechanism is a **Learned Semantic Transform (LST)** followed by **3-stage residual vector quantization (RVQ)**. For a frozen document embedding $e_i\in\mathbb{R}^{1024}$, the transformed representation is $$ h_i = \mathrm{normalize}(We_i + b), $$ and each document receives a code $(c_1,c_2,c_3)\in[256]^3$. Prefixes of this code are the labels: $b_1=c_1$ is coarse, $b_2=(c_1,c_2)$ is finer, and $b_3=(c_1,c_2,c_3)$ is finest. In the main experiments, $L_1$ yields **256 buckets**, $L_{12}$ about **65,408 observed buckets**, and $L_{123}$ about **129,955 observed buckets**. At coarse granularity HERMES sits on a **plateau** with KMeans-family methods on intrinsic clustering metrics, so the contribution is the reusable hierarchy rather than superior clustering at $K=256$. In **1B-parameter, 25B-token** pre-training, the strongest reported result occurs at **$L_{12}$** under **DoReMi** outer weights, where switching Stage 2 from max-entropy coverage to corrected-reader quality top-30% raises the **16-task** family-equal macro-average from **0.3969** to **0.4222**, a gain of **$+0.0253$** with **z-score $+2.09$**. At **$L_{123}$**, the same edge disappears, with **0.3988** versus **0.3986**, associated with median candidate-pool contraction from **2,271** to **429** documents, about **5.3×** [2607.02266]. In exoplanet science, **HERMES** expands to **HiERarchical Modelling for Exoplanet Science**, a multidimensional Bayesian framework motivated by ESA’s **Ariel** mission. Its use case is the three-dimensional relation among **stellar metallicity**, **planetary mass**, and **atmospheric metallicity**, with atmospheric water abundance as proxy. The core model is $$ y_i = \alpha_p + \beta_p\, m_{c,i} + \beta_s\, s_{c,i} + \epsilon_{\mathrm{int},i}, $$ where $y_i=\log X_{\mathrm{H_2O},i}$ and $$ \epsilon_{\mathrm{int},i} \sim \mathcal{N}(0,\varepsilon), \qquad \varepsilon \sim \mathrm{HalfNormal}(\sigma_y). $$ A key design feature is explicit treatment of stellar metallicity as an uncertain covariate: $$ s_{\mathrm{true},i} \sim \mathcal{N}\!\left(s_{\mathrm{obs},i}, \sigma_{s,\mathrm{meas},i}\right). $$ The simulations begin from the **11 May 2026** Ariel Mission Candidate Sample, reducing **977** confirmed planets to **858** after requiring host-star metallicity measurements and valid planetary-mass uncertainties. Using **NUTS** in **NumPyro/JAX** with **four chains**, **800 warmup steps**, and **800 posterior draws per chain**, the paper concludes that survey leverage remains a reliable predictor of slope precision, and that an **Ariel Tier 2 transit survey of at least 400 planets** can **robustly recover** the stellar-metallicity/planetary-metallicity correlation even when intrinsic astrophysical scatter is as large as **1.2 dex** [2606.02696]. Taken together, these frameworks show a recurrent “hierarchical” theme in modern uses of the name. In one case, hierarchy is a reusable label substrate over a large corpus; in the other, it is a Bayesian population model that propagates measurement error and intrinsic scatter. The shared label does not indicate shared methodology, but it does coincide with a common emphasis on multi-scale structure and reusable latent organization. Source: https://www.emergentmind.com/topics/hermes-64b181a3-0e32-4a14-afb3-7d97651ef429