AMADEUS: A Multifaceted Research Label

• AMADEUS is a context-dependent research label applied in fields like astroparticle physics, nuclear experiments, AI, and scientific writing.
• In neutrino astronomy, it denotes an acoustic detection system with precise environmental calibration for ultra-high-energy neutrino feasibility studies.
• In nuclear physics and AI, AMADEUS signifies specialized antikaon experiments, advanced language models, and structured writing strategies, reflecting its flexible interdisciplinary use.

AMADEUS is a recurrent research name rather than a single scientific object. In arXiv usage it designates, among other things, an acoustic test system embedded in the ANTARES deep-sea neutrino telescope, an experimental program on low-energy antikaon–nucleus interactions at DAΦNE/KLOE, a scientific-writing strategy for non-native users of English, and several recent computational systems, including a Brazilian-Portuguese large-language-model family and a retrieval-augmented framework for role-playing agents (Graf, 2010, Piscicchia et al., 2013) 0611013. The shared name is therefore nominal rather than disciplinary, and its interpretation depends entirely on context.

1. Scope and principal uses of the name

Across the literature, the term is attached to projects with distinct expansions, infrastructures, and problem domains. In astroparticle physics it expands to “ANTARES Modules for the Acoustic Detection Under the Sea.” In hadron and nuclear physics it appears as “Antikaon Matter At DAΦNE: Experiments with Unraveling Spectroscopy,” and also as “Anti-kaonic Matter At DAΦNE: An Experiment with Unraveling Spectroscopy.” In scientific-writing pedagogy it denotes a strategy rather than a detector or collaboration. In recent AI work, it appears as “Amadeus-Verbo” and as the name of a modular retrieval-augmented framework (Collaboration et al., 2010, Marton et al., 2017) 0611013.

Domain	AMADEUS usage	Representative papers
Astroparticle physics	ANTARES acoustic-neutrino test system	(Collaboration et al., 2010, Lahmann, 2011)
Hadron and nuclear physics	DAΦNE/KLOE antikaon program	(Piscicchia et al., 2013, Marton et al., 2017)
Detector R&D for antikaon physics	Trigger, TPC, and channel studies	(Scordo et al., 2013, Lener et al., 2013, Piscicchia et al., 2013)
AI and NLP	Brazilian-Portuguese LLMs; RAG for role-playing agents	(Cruz-Castañeda et al., 20 May 2025, Park et al., 4 Aug 2025)
Scientific writing pedagogy	Learn-by-doing writing strategy	[0611013]
Travel-industry analytics	Operational context for customer-search clustering	(Chatterjee et al., 2020)

This multiplicity is not accidental. The record suggests that AMADEUS has functioned as a reusable project label for collaborations that wanted a memorable identity, while the scientific content remained domain-specific.

2. AMADEUS in acoustic neutrino detection

In neutrino astronomy, AMADEUS is the dedicated acoustic sub-system of the ANTARES neutrino telescope in the Mediterranean Sea. It was built as a feasibility study for acoustic detection of ultra-high-energy neutrinos in the deep sea and as a hybrid opto-acoustic technology demonstrator for future very large volume neutrino telescopes (Graf, 2010). The installed system comprises six acoustic clusters, each with six acoustic sensors, distributed over two ANTARES lines at water depths between 2050 and 2300 m; cluster spacings range from 14.5 m to 340 m, the sensors record broadband signals up to 125 kHz, their typical sensitivity is around $-145\ \mathrm{dB\ re\ 1\,V/\mu Pa}$ including preamplifier, and on-shore filtering reduces the stored data volume to about 10 GB per day (Collaboration et al., 2010).

Its physical basis is the thermo-acoustic model. A UHE neutrino interaction deposits energy locally in a small volume, inducing rapid heating, expansion, and a bipolar pressure pulse with cylindrical symmetry around the cascade axis and a thin “pancake” emission pattern transverse to that axis. For a 1 EeV cascade, the quoted peak-to-peak pressure amplitude is of order $10\ \mathrm{mPa}$ at about 200 m vertical distance, the spectral energy density peaks around 10 kHz, and the attenuation length in sea water is about 5 km at 10 kHz and about 1 km at 20 kHz. This long attenuation length motivates sparse instrumentation on scales not realistic for optical Cherenkov arrays; simulations cited for AMADEUS-like conditions indicate that $O(100)$ sensors per km$^3$ can yield effective volumes of order $100\ \mathrm{km}^3$ at several hundred EeV, while the “break-even” sensitivity of acoustics relative to optical methods occurs around 50 EeV (Graf, 2010).

The system’s central empirical contribution is environmental characterization. Ambient-noise studies over roughly two years found a mean noise level of $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$ in the 1–50 kHz band, with 95% of samples below $2\sigma_{\text{noise}}$; restricting to 10–50 kHz reduces the RMS to about 7.5 mPa. Under these conditions, and for a source at 200 m, the reported threshold is approximately 1 EeV or below for 50% of the time. AMADEUS also demonstrated acoustic positioning at the level relevant to detector calibration; comparison of compass heading with heading derived from acoustic triangulation gave an RMS deviation of $1.7^\circ$, corresponding to positioning uncertainty of a few centimeters at storey scales (Lahmann, 2011). The papers are explicit that the array is too small to be a competitive neutrino telescope; its role is R&D and environmental qualification rather than flux measurement (Graf, 2010).

3. AMADEUS in low-energy antikaon–nucleus physics

In hadron and nuclear physics, AMADEUS is a program at the DAΦNE $e^+e^-$ collider at LNF-INFN, embedded in the KLOE detector, and devoted to low-energy interactions of negatively charged kaons with nucleons and nuclei. DAΦNE operates at the $\phi$-meson resonance, so the decay $10\ \mathrm{mPa}$0 provides nearly monochromatic charged kaons with momentum $10\ \mathrm{mPa}$1, a back-to-back topology, and a natural trigger geometry. The dedicated setup was designed for installation in the central region of KLOE, between the 6 cm beam pipe and the 50 cm inner wall of the drift chamber, and to use cryogenic gaseous targets, initially $10\ \mathrm{mPa}$2 and later $10\ \mathrm{mPa}$3, together with KLOE’s tracking, calorimetry, and 0.52 T magnetic field (Piscicchia et al., 2013).

The physics program addresses four tightly coupled questions: the existence and properties of deeply bound kaonic nuclear states, the nature of the $10\ \mathrm{mPa}$4 in a nuclear environment, low-energy $10\ \mathrm{mPa}$5 cross sections on light nuclei for kaon momenta below $10\ \mathrm{mPa}$6, and the general pattern of threshold and subthreshold kaon–nucleus interactions. The motivation is that the $10\ \mathrm{mPa}$7 interaction in the isospin-$10\ \mathrm{mPa}$8 channel is strongly attractive, which raises the possibility of dibaryon and tribaryon kaonic clusters such as $10\ \mathrm{mPa}$9, $O(100)$0, $O(100)$1, and $O(100)$2; at the same time, the existence and interpretation of such states are explicitly described as controversial, and conventional multi-nucleon absorption plus final-state interactions can mimic bound-state signals (Piscicchia et al., 2013).

A major step in the program was “Step 0,” the analysis of KLOE data from 2004–2005, corresponding to $O(100)$3, using the detector materials themselves as active targets. This provided samples of $O(100)$4 absorption on H, $O(100)$5He, $O(100)$6Be, and $O(100)$7C, both at rest and in flight. One result was a detailed decomposition of $O(100)$8 captures into 2NA-QF, 2NA-FSI, 3NA, and 4NA + background components. In the same channel a fit including a $O(100)$9 component found a best fit at binding energy $^3$0, width $^3$1, and yield $^3$2 per stopped $^3$3, but the improvement was only at the $^3$4 level, so the spectrum was described as compatible with such a contribution without constituting evidence (Marton et al., 2017).

4. Instrumentation and channel studies in the antikaon program

AMADEUS at DAΦNE generated a substantial detector-R&D program. For the kaon trigger, a compact system of two layers of BCF-10 double-clad scintillating fibers read out at both ends by Hamamatsu S10362-11-050-U SiPMs was developed for operation in the KLOE magnetic field. A prototype tested at the PSI $^3$5M-1 beam line reached a kaon-relevant time resolution of $^3$6; for 440 MeV/$^3$7 protons the measured double-layer efficiency was $^3$8, while the corresponding MIP efficiency was $^3$9, giving more than 90% rejection of MIP-like background (Scordo et al., 2013).

For inner tracking, the collaboration developed a GEM-based TPC. The foreseen AMADEUS TPG was specified as 20 cm long with inner diameter 8 cm and outer diameter 40 cm, targeting spatial resolution better than $100\ \mathrm{km}^3$0 in $100\ \mathrm{km}^3$1-$100\ \mathrm{km}^3$2, better than $100\ \mathrm{km}^3$3 in $100\ \mathrm{km}^3$4, and material budget below about $100\ \mathrm{km}^3$5. A $100\ \mathrm{km}^3$6 prototype with drift gap up to 15 cm, tested with low-momentum pions and protons, achieved detector efficiency up to 99%, spatial resolution in the drift direction as good as $100\ \mathrm{km}^3$7, and an extrapolated $100\ \mathrm{km}^3$8 resolution of about 9% for 100 hits, sufficient for strong $100\ \mathrm{km}^3$9 and $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$0 separation in the relevant momentum range (Lener et al., 2013).

The early KLOE analyses also established the spectroscopy channels on which the physics case rested. In the $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$1 channel, KLOE data showed that the $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$2 could be studied through the pure $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$3 final state $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$4, with the detector’s photon reconstruction making it possible to isolate at-rest and in-flight capture components; the observed $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$5 spectrum contained a significant excess above the at-rest kinematic limits for $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$6He and $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$7C, interpreted as evidence for in-flight $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$8 absorption (Piscicchia et al., 2013). In the charged $\sigma_{\text{noise}} = 25^{+7}_{-5}\ \mathrm{mPa}$9 channel, the reconstructed $2\sigma_{\text{noise}}$0 invariant-mass spectrum exhibited a second low-mass component not explained by the standard simulated backgrounds, and the missing-mass analysis of the residual nuclear system was presented as possible evidence for $2\sigma_{\text{noise}}$1 internal conversion (Scordo et al., 2013).

5. AMADEUS in AI, language technology, and industrial analytics

In 2025 the name appeared in “Amadeus-Verbo,” a family of Qwen2.5-based LLMs specialized for Brazilian Portuguese. The report describes base-tuned, fine-tuned-instruct, and merged variants in sizes 0.5B, 1.5B, 3B, 7B, 14B, 32B, and 72B parameters. The training regime used full-parameter supervised fine-tuning, maximum sequence length 8192, two epochs, and 78,840 examples from a larger instruction corpus of around 600k Brazilian-Portuguese examples; evaluation used a Brazilian-Portuguese adaptation of the LM Evaluation Harness on tasks such as ASSIN2-RTE, ASSIN2-STS, BlueX, ENEM, FaQuAD-NLI, hate-speech detection, TweetSentBR, and OAB exams. The reported pattern is that PT-BR-specialized models often match or exceed the original Qwen2.5-Instruct baselines, with merged models frequently offering the best trade-off between general and localized capabilities (Cruz-Castañeda et al., 20 May 2025).

A second AI usage is a retrieval-augmented framework for role-playing agents, also named AMADEUS. It is composed of Adaptive Context-aware Text Splitter (ACTS), Guided Selection (GS), and Attribute Extractor (AE), and is evaluated on CharacterRAG, a benchmark containing persona documents for 15 fictional characters totaling 976K written characters and 450 question–answer pairs. ACTS assigns character-specific chunk length and hierarchical context; GS filters retrieved chunks by whether they support inference of persona-relevant attributes; AE extracts “Belief and Value” and “Psychological Traits” to stabilize generation, especially for out-of-knowledge questions. With GPT-4.1 as the generator, the reported CharacterRAG scores were ACC 92.67, ACC$2\sigma_{\text{noise}}$2 9.26, and HS 2.89, outperforming Naive RAG, CRAG, and LightRAG on the reported setup (Park et al., 4 Aug 2025).

The name also appears in travel-industry machine learning as the operational environment for a multi-objective consensus clustering framework for flight-search recommendation. In that work, Amadeus customer search data are represented by nine features, the framework optimizes ensemble diversity while automatically determining the number of clusters, and evaluation combines Adjusted Rand Index with an external “Amadeus business metric,” defined as the difference between the estimated booking probability of the flight recommendations output by the search engine before and after optimization (Chatterjee et al., 2020). This suggests a distinct, corporate usage in which “Amadeus” denotes the deployment context rather than the algorithmic object.

6. Pedagogical and terminological significance

Outside physics and AI, AMADEUS was also introduced as a scientific-writing strategy. The 2006 paper “Developing strategies to produce better scientific papers: a Recipe for non-native users of English” states that the AMADEUS strategy had been used to produce scientific writing tools for non-native users of English for 15 years, emphasized a learn-by-doing approach, and provided a 9-step recipe for students to compile writing material according to a procedure described as efficient in scientific writing courses [0611013]. In this usage, AMADEUS denotes neither a detector nor a software platform in the modern ML sense, but a pedagogical method centered on structured practice.

Taken together, these literatures show that AMADEUS is best understood as a context-dependent scientific name. In ANTARES it denotes an acoustic-neutrino technology demonstrator; at DAΦNE/KLOE it denotes a low-energy antikaon program with dedicated trigger, TPC, and hyperon-spectroscopy developments; in AI it names both a Portuguese LLM suite and an attribute-aware RAG architecture; in pedagogy it names a writing strategy. This suggests that the encyclopedia treatment of AMADEUS must be intrinsically disambiguating: its scientific identity is fixed not by the name itself, but by the host collaboration, acronym expansion, and experimental or computational stack to which the name is attached.