Start-2-end simulations model the entire beam line from the generation of the photo electrons to the transport of the FEL radiation light to the user station. Because the beam dynamics changes significantly along the beam line a series of 'expert' codes are used. The interface is the electron beam or radiation field distribution.
Fig.1: Typical structure of a beam line for start-end simulation.
The components of a start-end simulation, shown in Fig. 1, are specialized on a particular problem of the beam dynamics, which are:
In a rf photo gun electrons are generated and then accelerated to relativistic energies. This wide range in energy imposes a significant limitation in modelling this problem. In addition the beam parameters are sensitive to the space charge field and the effects by the image charges on the cathode. While the tracking is done in the lab frame, the space charge field is calculated in the rest frame of the electron bunch, where it is mainly an electrostatic problem. After obtaining the solution, the field is transformed back into the lab frame, where it is applied to the particles.
The linac has typically the majority of beam line components, such as rf accelerating structures, quadrupole and bending magnets. Is of importance to use a fast code to explore the vast parameter space of all components, in particular with respect to jitter and tolerance studies. If the beam energy is large enough, space charge effects are negligible and particles can be tracked in large steps by using the method of transport matrices. However the results should be compared with a code that include space charge to check the validity of the tracking.
Although a bunch compressor can be modeled with codes, used for the linac, high brightness beam are sensitive to any emittance degrading effects, such as coherent synchrotron radiation in a bunch compressor. For a self-consistent simulation, the radiation field and the space charge field has to be included, which automatically makes the calculation CPU intensive. There can be multiple bunch compressors along the beam lines, so that an exchange back and forth between the compressor code and the linac code is required. Numerically it is not a problem to reduce the number of macro particle to keep the calculation time for the bunch compressor in reasonable limits. However the inverse process (going from the bunch compressor to the linac code) involves the increase the number of macro particles, which out introducing new correlation in phase space or washing out existing ones.
The length scales in a FEL is extreme, which range from a hundred meter long device, which a spacial resolution down to a sub-Angstroem level, including particle distribution and radiation field. The FEL codes are highly specialized to serve this purpose. External distribution from the linac code can be used, but the longitudinal distribution must be discarded and newly filled to provide the correct statistic for the spontaneous radiation level. In addition the total number of used macro particle can be several order of magnitude larger than those from the linac code in the case of VUV and X-ray FELs.
Several designs are based on multi-stage FELs, where beam manipulation devices are placed between undulator modules. The manipulations can be done on both, radiation and electron beam. In contrast to the bunch compressor/linac exchange, here it is essential not to change the total number of macro particles. Otherwise the modulation degree on the radiation wavelength become invalid. The range of manipulating are multifold, including monochromator for the radiation field or buncher/debuncher for the electron beam, to name the most importants.
The output of the FEL has to be transported to the user endstation, which might also include some beam manipulations (collimation, damping, filtering). The problem arises with frequency-dependent devices (e.g. monochromator), which requires the entire radiation field to be Fourier transformed, applying the greens filter and then transformed back. The field data on the 3D/2D grid, provided by the FEL codes, are typically huge, up to a few GBytes for an X-ray FEL. Filtering and reduction of the FEL output is necessary. Currently these codes are only in their initial stage.
At PBPL, we use only a subset of codes for the start-2-end simulations. The gun is simulated with Parmela, the beam transport through the linac with Elegant and the FEL interaction with Genesis 1.3. The bunch compressor dynamics is included with Elegant, based on an analytical model. Multi-Stage FELs as well as beam transport haven't been yet included in our studies.
We primarily simulate two FEL experiments: LCLS and VISA. In both cases the FEL simulation requires the most time, from 1 day for VISA up to more than a week for LCLS. The simulations would benefit from a parallel version of Genesis, using the full advantage of the 16 node Beowulf cluster in our group. The conversion to a MPI version of Genesis 1.3 is in progress.
The used version of Parmela and Genesis 1.3 are not SDDS compliant to simplify the interface with Elegant, but scripts have been written to do the conversion. The support of Genesis for SDDS is in progress, while it is uncertain for Parmela.