Project eAST proposal

Motivation

Large-scale detector systems for the EIC are designed by larger communities. The simulation efforts typically start as standalone exercises for each detector component with various levels of maturity, analytical calculations, simplified Monte Carlo exercises (fast simulations), Geant4-based Monte Carlo approaches (full simulations), and are later being extended in several different frameworks.

It is critical in the longer term to perform studies where the information from various detector elements and also the interaction region, support structures and other dead material is taken into account, which is only possible in a comprehensive simulation. This is essential to understand, e.g., the performance at the edges of the detector system or the effect of combining different technologies for the electro-magnetic calorimeter. For studies of the physics reach and detection capabilities, it is important to be able to switch detector options with varying levels of detail, combining full simulations for some detector components with fast simulations for the rest of the detector system. It is also critical to build a sustainable effort for the entire time scale of the experiments with common tunes and commonly validated results for both fast and full simulations.

However, such detailed studies are time consuming, both from people and computing view. To ease leveraging new and rapidly evolving computing technologies, we plan to implement a common and integrated approach for fast and full simulations in Geant4 with a plug and play modular approach.

Requirements

A comprehensive and centrally maintained simulation tool based on Geant4 for both fast and full simulations with a library of potential detector options has to be developed. The initial focus on the development will be:

After validation of the Geant4 simulations, they will be simplified (replacing some physics aspects) for fast simulations. This approach will allow us to use everywhere the same geometry (very important to reduce debugging and development time) and to combine full simulation for a subset of detectors with fast simulations for the rest of the integrated detector. Geant4 itself and various NP and HEP experiments already provide sub-systems for fast simulations. This will reduce the development time of the fast simulations substantially.

The detector simulation tool will be modular, and its development will be targeted. It will provide an interface to the output of Monte Carlo event generators but no further work on generators. The simulation of detector effects and detector responses (digitization) will be clearly separated. There will be a common geometry interface between the detector simulation and reconstruction.

The timely development of a comprehensive, unified and centrally maintained detector simulation tool for both fast and full simulations to serve the needs of the detector collaborations or groups is no small task. It must take place in the context of strong development teams in the labs together with important contributors in the universities, and go beyond legacy software to a new common effort to be successful. This effort builds on a sustainable common project based on Geant4 and tailored to needs of NP experiments as they exist today and will evolve to the 2030s.

Geant4 has the capability to support both fast and full simulation, which we want to use for large-scale detector systems. Geant4 through its multi-threaded reengineering has already been able to support high concurrency heterogeneous architectures, with excellent results in the memory economizations achieved. Leveraging and evolving this capability as heterogeneous architectures become ever more prevalent is of great importance for data processing and analyses for large-scale detector systems at the EIC.

Work program and deliverables

Create CAD interface to the detector simulation tool

Fast and detailed simulations need to be able to implement updates on detector layout fast, where these updates need to be based on detailed design drawings with parametric surfaces in 3D. This requires as crucial first step compatibility with the CAD applications being used at BNL and JLab, for easy exchange. The most flexible format for Geant4 to import CAD geometry is GDML. STEP is the file format commonly available to many CAD systems. Several tools exist that convert STEP files to GDML files. We will work with experts at BNL and JLab to conduct a survey if STEP is available for the variety of CAD applications used. These break down into two broad categories: mesh vs. nurbs. Mesh CAD applications represent a structure using a collection of flat polygons that emulate curved surfaces that can be used to represent volumes. The granularity of the polygon surfaces can be scaled up or down to create smoother or coarser surfaces, respectively. Examples are AutoCad and Sketchup which are excellent for expansive structural/architectural representations of components and buildings. Nurbs models use parameters that describe the shape and extent of a surface, the curvature of each element, and the thickness of the material. In effect they can produce very close approximations of real-life entities but are difficult to exchange or import by non-nurbs systems. Examples are CREO, SolidWorks, and Siemens NX which are used for mechanical design. GDML also supports various shapes and extruded solids, some of them may correspond to nurbs. Lastly, CAD files often only contain sparse information about the composed material, often just the name of the materials. On the other hand, detailed detector simulations require the full composition of each material, often down to the isotope composition level. We will develop a mechanism, e.g., through a macro file, to associate such additional information needed to the detector simulation.

Deliverables: 1) Development of a CAD Interface for Detector Simulations. 2) Macro file that allows material composition information to be easily ported to detector simulations.

Create an initial version of the fast and full detector simulation tool

The detector simulation tool has to be capable of doing both fast and full simulations in one application, easily configurable for each detector component. Also, the detector simulation tool has to easily integrate (“plug-in”) already existing standalone simulation applications. These requirements will be fulfilled by utilizing the “region” mechanism of Geant4. Each detector component is represented as a region, where geometry description including different levels of detail, physics options including fast simulation and unique physics model configurations, and detector responses based on geometry and physics options are taken care of. We will develop an initial version of such a detector simulation tool.

Deliverable: First running version of detector simulation tool.

Communicate with detector study groups

The existing standalone simulation applications that are to be adapted to the new detector simulation tool have to be examined and converted to be “plug-able” using the “region” mechanism. We will communicate with detector study groups and work with them to drive the efforts of integration. Through these interactions the requirements to the detector simulation tool will be refined as needed. The detector simulation tool may also integrate beam-test configurations if applicable. The deliverable will be a prototype integration of an existing simulation, the specific target to be chosen in the first month on the basis of importance and available developer effort during the initial funded period. Other detector components will be added by the detector community based on the experience from the prototype.

Deliverable: Documented prototype integration of an existing simulation.

Develop and deliver a common physics list

Geant4 offers several alternative physics models. To make comparisons over different detector configurations, one has to use a common set of physics models that is appropriate to NP detectors. We will develop and deliver such a physics list. We will consult with detector study groups and advise on validations of the physics models with beam-test results.

Deliverable: Documented common physics list.

Integrate with overall software efforts

We will compile the requirements of detector simulation to the software infrastructure (e.g., common geometry and data formats) to be shared across the whole software chain including physics event generation, simulation, digitization, reconstruction and analysis programs.

*Deliverable: Requirements document. *

Deliver a detector simulation tool extensible to heterogeneous architectures

In the future, simulation jobs may run on computers with heterogeneous hardware configurations, e.g., with GPGPU and/or FPGA, and with cutting edge IT technologies such as AI/ML. Enabling the effective use of such technologies requires design and implementation choices in the detector simulation tool. We will ensure the extensibility of the detector simulation tool through the use of a tasking mechanism (either PTL or TBB). We will also support the developers of detector components of the thread-safety of their code.

Deliverable: Proven use of a tasking mechanism.

Project leader

This project aims to take advantage of the proven value for detailed nuclear and particle physics experiment simulations by the HEP-developed Geant4 software and the growing prospects of use of advanced computing techniques. Any large-scale detector system will benefit from the project on “Fast and full simulations in Geant4 for large-scale detector systems with a plug and play modular approach”, with the foreseen EIC detectors prime examples. For example, EIC detector simulation development must be able to progress rapidly to provide the EIC user community proper flexible and forward-looking tools to meet the simulation needs of the detailed detector design and EIC science performance studies. This requires a common simulation framework and effort if the EIC community is to be properly served by a unified effort drawing on pooled expertise. This also requires simulation tools ready to take advantage of the growing use of heterogeneous computing environments and the prospects they offer. This work will seed a new common effort through the leadership of an expert of unique stature and technical expertise, who has the trust and confidence of the full community to establish a common effort, while also being a technical expert. The “Fast and full simulations in Geant4 for large-scale detector systems with a plug and play modular approach” will ensure that any work on AI/ML and running on computers with heterogeneous hardware configurations can be directly applied.

In Dr. Makoto Asai we identify the individual who can uniquely fill this role. Dr. Asai has been both a Geant4 project leader and deep technical expert for over 20 years. He is well known to and respected by the EIC community with which he has collaborated for many years on EIC Geant4 simulation needs. He is the designer and principal developer of Geant4’s capability to support both fast and full simulation, which we want to use for the EIC. He led Geant4 through its multi-threaded reengineering to support high concurrency heterogeneous architectures, with excellent results in the memory economizations achieved. Leveraging and evolving this capability as heterogeneous architectures become ever more prevalent is of great importance for the EIC, and together with the integrated fast simulation support opens the door to leveraging AI/ML in the unified simulation. Dr. Asai is also an expert in the geometry and modular detector description tools that will be the basis of unified geometry in detector simulations. He is also an expert in the Geant4 physics models that have to be tuned for NP experiments, having presided over their development and integration for much of the past 20 years. No other individual has an array of attributes better tailored to the leader we are looking for. Finally and crucially, Dr. Asai is available for this work if we act promptly.

Appendix

Notes from the March 25 EIC software meeting discussion (Torre)

Notes from the March 2 EIC software meeting discussion (Markus)