News from KM3NeT

On April 8 – 15, the KM3NeT collaboration has successfully deployed five strings at the ARCA site and is operating now a configuration of six strings. Congratulations! Read the report from Simone Biagi (LNS Catania) who was one of the five KM3NeT members on board:

In the week between the 8th and the 15th of April, an intense sea campaign has been carried on at the KM3NeT/ARCA site, in the Mediterranean Sea 80 km off the Sicilian coast of Capo Passero. Main purpose of the operation was the installation and connection at 3,500 meters sea depth of a second-generation Junction Box (JB), together with 5 new Detection Units (DUs), or “strings”, for neutrino detection. The Junction Box is the result of a huge effort from KM3NeT members themselves for the electronics design and production. The JB mechanical structure, instead, is the result of a close collaboration with MacArtney/Teledyne, a company specialized to build deep underwater instrumentation. The Junction Box is part of the ARCA submarine infrastructure, which comprises two main electro-optical cables (MEOC) for power distribution and data transfer onshore, together with a series of inter-link cables used to connect JBs and DUs. Each Junction Box provides connection for 12 Detection Units.

Five KM3NeT members were onboard the “Miss Marilene Tide” boat, operated by the FUGRO company. The Junction Box and the 5 Detection Units were loaded onboard in the Malta harbor at La Valletta. A brand new ROV (Remotely Operated Vehicle), designed to work until a depth of 4,500 m, was used to support the submarine activities, like installations, positioning and connections. A large onshore team, composed of KM3NeT experts, was either present in the ARCA onshore laboratory located in Portopalo or connected by remote (due to COVID related travel limitations). These experts provided an immediate feedback to the offshore activities, with extensive functional tests immediately performed after installation and connection of each part. As soon as the ROV was recovered from the sea water, the detector was put in “Run”, in order to start to collect data useful for calibration and commissioning.

The 5 Detection Units installed are now taking physics data together with the first KM3NeT DU which was deployed as early as in December 2015. After all these years invested to consolidate the technology and set up a powerful construction organization, the 6 DUs of ARCA, together with the 6 already in operation in ORCA, represent the first core toward the full construction of KM3NeT. New deployment campaigns are foreseen in the next months at the two installation sites. The mass construction of the detector (both in ARCA and ORCA) is really on its way.

Useful video links:

Deployment of ARCA Junction Box:

Deployment of ARCA Detection Unit: and

Detection Unit unfurling:


image001 copy copyThe 5 Detection Units on the deck waiting for deployment


image003 copyThe night “splash” of the Junction Box


image005 copyThe first Detection Unit entering the water. It takes about 4 hours to reach the sea bottom, where it is installed with a precision of a few meters, with the support of the ROV


image007The ROV arm (right) releasing the unfurling mechanism. The LOM (Launcher of Optical Modules) with the optical modules is going up to release, rotating, the Detection Unit. The LOM will be recovered on board later. This installation procedure is described in the paper: “Deep-sea deployment of the KM3NeT neutrino telescope detection units by self-unrolling”, S. Aiello et al. (The KM3NeT Collaboration), 2020 JINST 15 P11027 and arXiv:2007.16090


image009 copyAfter the “unfurling” of the Detection Unit, a visual inspection is performed with the help of the ROV, from the anchor on the sea bottom up to the top buoy. This frame is taken from a HD video, showing a Digital Optical Module in its final position at high depth


 image011 copy copyDuring the entire week, the team of KM3NeT experts gave a prompt and fast support from the INFN Portopalo lab. Due to the impossibility to travel from abroad for COVID restrictions, some colleagues worked remotely through an almost 24/7 active video call. (Look for them on the screen!)


image013 copyThe offshore team in front of the empty LOMs used for Detection Unit deployment, tired but happy, at the end of the campaign. From left: Nunzio Randazzo, Klaus “Captain” Leismüller, Giorgio “Sea dog” Riccobene, Simone Biagi, Edward Berbee


image015 copyA nice six-string event recorded by ARCA, as seen in the online monitoring; x-axis represents time, y-axis is the vertical position referred to the seabed. The horizontal distance between DUs is of the order of 80 m.


News from Baikal

The 2021 campaign was successfully completed, with eight clusters operating. Apart from the first cluster, about 99% of the OMs are alive and deliver data. The first (and oldest) cluster has two full sections and several single OMs which went out of operation over the years. Due to lack of time, the cluster could not be repaired during this expedition but will be the first on the repair-list for the next season.


New GNN Board

After the recent changes in the management of KM3NeT and IceCube, and following the suggestions of the new spokespersons, the GNN Board now includes the following persons:

  • Igor Belolaptikov (JINR Dubna, Russia)

  • Rosa Coniglione (INFN/LNS, Italy)

  • Paschal Coyle (CPPM, France)

  • Zhan Djilkibaev (INR Moscow, Russia)

  • Grigorij Domogatsky (INR Moscow, Russia)

  • Albrecht Karle (Univ. Wisconsin, USA)

  • Uli Katz - chair (Univ. Erlangen-Nürnberg, Germany)

  • Antoine Kouchner (APC/University Paris, France)

  • Christian Spiering (DESY-Zeuthen, Germany)

  • Maurizio Spurio (Univ. Bologna/INFN, Italy)

  • Ignacio Taboada (Georgia Tech Atlanta, USA)

  • Julia Tjus (University Bochum, Germany)

Thanks to Darren Grant, Mauro Taiuti and Shigeru Yoshida for contributing to the GNN Board during the last 4 years!



News from IceCube

IceCube is celebrating its "First Decade of Discovery" this year. December 18, 2020 marked the 10th anniversary of the last DOM being deployed, and on May 13, 2021, it will be 10 years since the start of IceCube’s first fully configured physics run---and the beginning of IceCube's 10th anniversary celebrations.


While IceCubers wish they could meet in person to celebrate this milestone together, this year's celebration will be mostly virtual and will consist of content on IceCube social media and website as well as a collaboration-wide memory collection project and some virtual events. To join IceCube's celebrations, keep an eye on IceCube's website and social media over the next five months leading up to the fall IceCube collaboration meeting in September.



In this edition of GNN Monthly, two technical IceCube papers will be highlighted, the one posted on March 1, the other already on January 27. I had overseen the January 27 paper and add the review here retrospectively.

The first paper A Convolutional Neural Network based Cascade Reconstruction for the IceCube Neutrino Observatory has been submitted to JINST and posted at 2101.11589.pdf (

Real-time analyses and alerts require powerful and fast reconstruction methods. Deep neural networks can be extremely powerful, and their usage is computationally inexpensive once the networks are trained. These characteristics make a deep learning-based approach an excellent candidate for the application in IceCube.  The paper presents a reconstruction method based on convolutional architectures and hexagonally shaped kernels (according to IceCube’s hexagonal top view). The CNN (Convolutional Neural Network) method is robust towards systematic uncertainties in the simulation and has been tested on experimental data. In comparison with standard reconstruction methods in IceCube, it can improve the reconstruction accuracy, both in angle and in energy (see the next two figures), while reducing the time necessary to run the reconstruction by two to three orders of magnitude (third figure).



Cascade angular resolution in IceCube as a function of neutrino energy for the standard reconstruction method and the newly developed CNN-based method. The shaded area and lines denote the 20%, 50%, and 80% quantiles.  At higher energies, the resolution can be improved by up to 50%. Systematic uncertainties are not included.



Correlation plots between true and reconstructed deposited energy for the standard reconstruction (left) and CNN (right). See the paper for more detailed information.



Per-event runtimes as a survival function of the fraction of events exceeding a specified runtime. In contrast to the standard reconstruction, the CNN based method is able to run on multiple cores as well as on a GPU. Runtimes shown for the CNN are for application on a GPU.

The paper also discusses possible improvements which can better address IceCube’s irregular geometry.

The second paper A muon-track reconstruction exploiting stochastic losses for large-scale Cherenkov detectors is posted at 2103.16931.pdf (

IceCube’s best-performing muon track directional reconstruction is based on a maximum likelihood method using the arrival time distribution of light at the PMTs. A known systematic shortcoming of this method is to assume a continuous energy loss along the muon track. However, at energies >1 TeV the light yield from muons is dominated by stochastic radiative processes producing electromagnetic showers.

The paper discusses a generalized ansatz where the expected arrival time distribution is parametrized by a stochastic muon energy loss pattern.

The next figure illustrates the track reconstruction chain depicting the underlying assumed photon arrival time distribution (or PDF, for probability density function) in the respective likelihood reconstruction. The PDF approximation quality and CPU time requirements broadly increase with more sophisticated reconstructions later in the chain.




The previous steps start with the plane wave seed (“Gaussian PDF”) and end with the SplineReco which takes into account the layered ice. The principle behind the new algorithm (SegmentedSplineReco) is illustrated in the next figure.


image027Schematic view of SegmentedSplineReco. The incoming muon track in red is first reconstructed by the black line, representing the initial track hypothesis for SegmentedSplineReco. It follows an energy reconstruction which results in a series of cascades along the muon track (yellow stars), placed at the centre of each segment of length ℓ. The energy information of each segment is used to define the final PDF at each DOM.


This more realistic parameterization of the loss profile leads to an improvement of the muon angular resolution of up to 20% for through-going tracks and up to a factor 2 for starting tracks over existing algorithms (see the next figure).  The figure demonstrates the improvement in angular resolution for three event classes. SplineReco-optimized events (left side) belong to a class of muon tracks that has passed certain quality cuts based on SplineReco, which to some extent mimics events that are usually found on the final analysis selections used in IceCube. These cuts have not been applied for the other two event classes (middle, right) which are subdivided in through-going tracks and starting track events.


Median angular resolution as a function of MC muon energy calculated at the interaction vertex for three IceCube all-sky simulations: SplineReco-optimized events, through-going and starting tracks. The SegmentedSplineReco reconstruction is compared to SplineReco (black line). The three different likelihood models for SegmentedSplineReco are compared: the standard unbinned likelihood ? (blue line), the extended unbinned likelihood ?ext (green line) and the unbinned likelihood for the first hit per DOM ?1st (red line).


Conferences and Workshops


Conferences and Workshops relevant for the GNN community can be found at



VLVNT (Valencia, May 18-21)

Grossman Meeting (July 5-9):

ICRC (Berlin, July 12-23):

EPS-HEP (Hamburg, July 26-30)

TAUP (Valencia, Aug 30 – Sept 3):


Not yet in the lists above is the XXXII INT. SEMINAR of NUCLEAR and SUBNUCLEAR PHYSICS "Francesco Romano”, a school for graduate students and young researchers:  (June 7-11), including Mauro Taiuti (Genova) and Ralf Ulrich (Karlsruhe) as lecturers.