Chapter 6
INSTALLATION AND EXECUTION
6.1 Installation
ERIS runs on Windows, OS X, and Linux. URLs of their files are as follows:
- Windows
- OS X, 64 bit application
- Linux, 64 bit x86_64
- Linux, 64 bit Arm64
Once the above archive files are downloaded and extracted, the executable file “ERIS" (or ERIS.exe) can be directly executed without any additional installation process. To install ERIS to interoperate with RIETAN-FP, place the extracted files in a specific directory. On Windows, extract the zip files and move the resulting files to C:\(\backslash \)Program Files\(\backslash \)RIETAN_VENUS\(\backslash \). On macOS, mount ERIS.dmg by double-clicking its icon and move all the files to folder /Applications/RIETAN_VENUS/. Both versions can be run in combination with a shell script, MPF_multi.command, written in bash for MPF. MPF_multi.command and its manual are included in distribution files of the RIETAN-FP–VENUS package.
6.2 Execution
6.2.1 Interactive mode
Run ERIS by double-clicking its icon or entering the name of its executable binary file in a command line. Then, the program asks questions that must be answered by the user. In what follows, these questions are represented by boxed texts with gray background.
# The name of the MEM data set file, *.fos or *.mem.
Type the absolute path of an input file, for example, C:\(\backslash \)users\(\backslash \)user_name\(\backslash \)sample\(\backslash \)sample.fos (Windows).
# The type of the MEM data set file. # 0. (a) X-ray diffraction data or (b) neutron diffraction data of a compound # containing no element with a negative bc value. # 1. Neutron diffraction data of a compound containing at least one element # with a negative bc value.
This option is set for the backward compatibility of input files because the file format of *.mem is different depending on dataset types.
# From which densities will you start? # 0. Uniform densities. # 1. Restart from densities recorded in the 3D densities file (*.pgrid).
If you choose the second option, initial densities are input from *.pgrid. This option is useful when you restart MEM analysis after precedent iterations. Note that this option is different from the choice of prior density \(\tau _k\) in Eq. (2.1) in strict sense, although on the use of ZSPA algorithm, use of non-uniform prior is approximated by restarting calculation from \(\tau _k\). For rigorous analyses using non-uniform prior, use L-BFGS or Cambridge algorithms and specify the prior data on the later question. MEM analysis can only be restarted from densities recorded in *.pgrid output by ERIS.
# Optimization algorithm. # 0. 0th order single-pixel approximation (ZSPA). # 1. The Limited-memory BFGS algorithm (L-BFGS). # 2. The Cambridge algorithm (Obsolete).
Read 2.3 for details in the above two algorithms.
# Which initial Lagrangian multiplier will you use (LM_type)? # 0. The value input by user. # 1. The value calculated by ERIS. # 2. The value written in the MEM data set file.
This question appears only on use of the ZSPA algorithm, and option 2 appears only when the format of the MEM dataset file is *.mem.
# The initial Lagrangian multiplier, lambda
This question is asked only when 0 is input in the previous question.
# A parameter x to impose weighting factors on the basis of # lattice-plane spacing.
Parameter \(x\) is included in Eq. (4.1). Input \(x=0\) in classical MEM analysis where no weight is imposed in the \(F\) constraint. If \(x > 0\), it is regarded as \(x\) in Eq. (4.1). As described in 4, typically \(x = 0\) in neutron diffraction, \(x = 1\)–2 in X-ray powder diffraction, and \(x = 4\) is optimum in single-crystal X-ray diffraction.
# The coefficient, t, to adjust the Lagrangian multiplier.
This question appears only on use of the ZSPA algorithm. Input a value between 0.05 and 0.1 for fast computation. If \(\lambda \) is too large, MEM iterations diverge and never reach the solution. If \(\lambda \) is too small, the convergence of the MEM equation is ensured whereas the computation time may increase considerably, with no convergence achieved on specification of a relatively small maximum number of cycles. The coefficient \(t\) allows us to find the best \(\lambda \) value automatically through multiplying \(\lambda \) by \(1 + t\) in every cycle of MEM iterations until \(\lambda \) becomes too large to converge the MEM equation, which is different from the manner of changing \(\lambda \) in PRIMA [9].
# The coefficient, E, to adjust estimated standard uncertainty of # structure factors.
For details in \(E\), see 5.1 and 5.2.
# Will you save a feedback data file? # 0. Yes (output only individual F data). # 1. Yes (output all the F data including those for grouped reflections # and estimated for unobserved reflections). # 2. No.
In MPF, option 1 is usually selected to include \(F (\bm {h}_j)\)’s of high-\(Q\) reflections whose profiles are partly lacking.
# Will you save a e(GAUSS) distribution data file? # 0. Yes (raw data file, *_eps.raw). # 1. No.
When option 0 is selected, a text file storing histogram of \begin {equation*} \frac {|F_\mathrm {o}(\bm {h}_j )| - |F(\bm {h}_j )| }{\sigma (\bm {h}_j )} \end {equation*} will be output. See also section 8.1.
# Fractions of lambda for generalized constraints. # (8 parameters; l_2, l_4, l_6, l_8, l_10, l_12, l_14, l_16)
Eight parameters, \(\lambda _n\) (\(n = 2-16\)), for the fractions of generalized constraints with orders of \(n\) (see 3.4).
# Which prior\index{prior} densities will you use? # 0. Uniform densities. # 1. Densities recorded in the 3D densities file (*_prior.pgrid).
This option appears only on the use of the L-BFGS algorithm. When option 1 is selected, \(\tau _k\) in Eq. (2.1) will be read in from 3D densities file. To use non-uniform prior densities in the ZSPA algorithm, restart the calculation using a 3D data file (*.pgrid) storing prior densities.
# Will you save a preferences file? # 0. No. # 1. Yes (*.prf is output).
If *.prf is output, all the above parameters are stored in it and it can be reused as a template file for subsequent MEM analyses.
After answering all the questions, the conditions of MEM analysis are displayed in the screen. Before starting MEM iterations, ERIS asks the maximum number of iterations.
# Maximum number of cycles.
The iterations will continue until the convergence is achieved (CONSTR \(\le \) 1). When the convergence is reached, ERIS terminates after outputting CPU times.
6.2.2 Running ERIS using condition files
If the entire conditions of MEM calculations in filename.prf have already been input by modifying a template file with a text editor, ERIS can be executed in several different ways.
Let the name of *.prf be sample.prf. Then, ERIS can be run from the command line by typing
ERIS sample.prf
to read in sample.prf. When no environment variable PATH is set to a directory under which the program is placed, type the full path of the program, e.g.,
"C:\(\backslash \)Program Files\(\backslash \)RIETAN_VENUS\(\backslash \)ERIS.exe" sample.prf
if ERIS.exe is stored in folder “C:\Program Files\RIETAN_VENUS\” on Windows. Of course, sample.prf has to be placed in the current folder in the above case. A simple batch file (Windows) or a shell script (OS X or Linux) where such a command is recorded would be more convenient.
The most practical and convenient way of carrying out MPF by ERIS is the use of MPF_multi.command because \(E\) in Eq. (5.1) can be automatically changed during a series of MPF analyses. For details in this bash script, read the manuals of MPF_multi.command, MPF_multi_Win.pdf (Windows) or MPF_multi_Mac.pdf (OS X), included in a distribution file, documents.zip.1
