The code in this repository is designed to conduct 1D particle transport calculations using Markov chains and optimise geometry layout using genetic algorithms.
- Linux (should work on MAC, but never tested on Windows yet)
- C++ compiler with a minimum standard of C++17
- ROOT
- libtbb-dev
- libboost-program-options
The last two packages are standard and available in any modern Linux flavours. However, if ROOT is not available for your specific distribution, you can easily install it by following the instructions provided on this page.
git checkout https://github.com/kbat/mag.git mag
cd mag
mkdir build
cd build
cmake ..
make -j$(nproc)
The compiled code consists of two separate executables: solver
(mag-solve) which is intended for particle transport calculations,
and optimiser (mag-optimise) which is specifically designed for
geometry optimisation tasks.
List supported materials:
mag-solve -mat
The output shows material number, name and mass density:
Supported materials:
3 Stainless304 7.96703
11 Water 1
38 W 19.413
23 Lead 11.1837
47 B4C 2.50608
48 Poly 0.91
49 Concrete 2.33578
The material numbers can be chosen arbitrarily. Currently, they correspond to the material numbers used in CombLayer.
Run 20 layers of Tungsten followed by 4 layers of polyethylene with incident 3 GeV electrons:
mag-solve -layers 20 W 4 Poly -sdef e 3e3 1.0 -o spectra.root
Each layer is 1 cm thick.
The output shows the layer configuration and dose rates for each transported particle type:
24 layers: W W W W W W W W W W W W W W W W W W W W Poly Poly Poly Poly
Dose rates: e: 7.50763e-06 n: 0.00752361 p: 0.000113037 |: 2.62634e-25 total: 0.00764416
The particle designators adhere to the MCNP notation:
| ID | Description |
|---|---|
| e | electron (e+ and e-) |
| n | neutron |
| p | photon |
| | | muon (μ+ and μ-) |
| h | proton |
Optimise materials of 10 layers to minimise the fitness function:
mag -nlayers 10 -ngen 2
With 10 layers and 5 known materials, the code runs approximately 180 configurations for each generation, with the exact number depending on the available number of cores on your machine.
The output for each generation consists of the runtime line followed by a list of five best solutions, e.g.
Generation: 1
2.09588 sec
Fitness Dose Mass Complexity Materials
7.88793 3.23793 9.1 1 48 48 48 48 48 48 48 48 48 48
8.91059 3.36255 10.6961 2 48 48 48 48 48 48 48 48 48 47
8.9364 3.38836 10.6961 2 47 48 48 48 48 48 48 48 48 48
9.84813 3.50205 12.2922 2 48 48 48 48 48 48 48 48 47 47
9.88769 3.54161 12.2922 2 47 47 48 48 48 48 48 48 48 48
Generation: 2
2.15738 sec
Fitness Dose Mass Complexity Materials
7.88793 3.23793 9.1 1 48 48 48 48 48 48 48 48 48 48
8.91059 3.36255 10.6961 2 48 48 48 48 48 48 48 48 48 47
11.961 3.81885 15.4843 4 48 47 47 48 48 48 48 48 47 47
12.2642 3.82206 15.4843 7 48 48 48 47 47 48 47 48 47 48
13.2258 3.98564 17.0804 7 48 47 48 48 47 47 48 47 47 48
- The fitness function is the figure of merit for the optimisation algorithm defined in Optimiser::getFitness. Currently, it's hard-coded as a linear combination of dose, mass and complexity.
- Dose is equivalent dose beyond the layers. The flux-to-dose conversion factors are defined for neutrons in Solver::getNeutronFTD and in the other similar methods for the other particles.
- Mass is the sum of densities of all layers. This is probably misleading, but I still call it mass because this value is proportional to the total mass (since all layers have the same thickness of 1 cm). Actually, since our problem is 1D, the layer is infinite in the directions perpendicular to the beam, therefore its mass is infinite.
- Complexity is the number of material changes in the material list.
- Materials is the list of material numbers for each layer. Available materials can be printed with
mag-optimise -mat.