5.2.2. Number of history and Bank

Table 5.2.2 maxcas

Value

Explanation
(D=10)
Number of histories per batch. The upper limit is 2147483647.
Table 5.2.3 maxbch

Value

Explanation
(D=10)
Number of batches. The upper limit is 2147483647.
Table 5.2.4 maxbnk

Value

Explanation
(D=10000)
Maximum number of secondary particles that can be temporarily stored in computer memory. (D=1000000 when [weight window] is defined.)
When a large number of secondary particles are generated by high-energy spallation reactions or when a large number of particle splitting occurs using [importance] or [weight window], a warning like jbnk = 0, ibnk = 1 may appear. In such cases, increasing this value may speed up the calculation. Storing the data of one particle requires about 300 bytes of memory. So, please note that if the value is increased too much, the computer may run out of memory and an error may occur.
Table 5.2.5 ihistout

Value

Explanation
(D=0)
Interval of histories for outputting calculation progress to the terminal. If set to 0, progress is output only at the end of each batch. Unlike the standard process at the end of a batch, this function does not perform the summarization or output of tally information, making it useful for monitoring the simulation progress in real-time while minimizing the computational load. Note that when performing shared-memory parallel computing (e.g., OpenMP), the history numbers may not necessarily be output in ascending order due to the timing of processing in each thread.
Table 5.2.6 iwwbnk

Value

Explanation
(D=1)
Option to suppress the particle splits as the number of secondary particles stored in memory approaches maxbnk when [weight window] is set.
1: enabled, otherwise disabled.
Table 5.2.7 igeomem

Value

Explanation
(D=10000)
Maximum reserved memory size for a single cell definition. When you define a complex cell using too much #, you may get an error message that suggests to increase this parameter. When you get the message, you must reduce the number of # used in a cell definition, or increase this parameter to the suggested value or more.
Table 5.2.8 xsmemory

Value

Explanation
(D=1.0)
Relative size of memory space allocated for nuclear data. In general, this value is not necessary to be changed. If you get an error to request for increasing this value, please follow the request and send your input file to PHITS office.
Table 5.2.9 timeout

Value

Explanation
(D=-1.0)
Cut-off CPU time for PHITS calculation (in units of second).
Table 5.2.10 istdev

Value

Explanation
(D=0)
Control parameter for statistical uncertainty calculation type and restart mode.
=-2
Restart calculation mode is activated; if there is no past tally result, a new calculation is started with istdev=2.
=-1
Restart calculation mode is activated; if there is no past tally result, a new calculation is started with istdev=1.
=0
istdev is automatically set to 1 for shared-memory parallel computing with italsh=1, and 2 for the other cases.
=1
A new calculation is begun. Statistical uncertainties are estimated from the standard error of each batch result.
=2
A new calculation is begun. Statistical uncertainties are estimated from the standard error of each history result.
Table 5.2.11 italsh

Value

Explanation
(D=0)
Control parameter to share tally variable among multi threads for the OpenMP shared-memory parallel computing. This is useful for parallel computing requiring a large tally memory space because this option enables such a simulation with a smaller physical memory by sharing the tally variables. On the other hand, this option induces a process required to prevent memory corruption which may slow down the computational speed. Test with a small number of histories is recommended before full simulation.
=0
Tallies are not shared.
=1
Tallies are shared.
Table 5.2.12 ireschk

Value

Explanation
(D=1)
Control parameter for tally consistency check.
=0
Check consistency between new and old tally settings.
=1
No check. This option is useful when setting a very complex tally whose settings are not fully written in the tally output file.

In the distributed-memory parallel calculation, the number of batches should be an integer multiple of \(N_{PE}-1\) , where \(N_{PE}\) is the total number of Processing Element (PE). Otherwise, PHITS will automatically convert maxbch to an integer multiple of \(N_{PE}-1\) , and adjust maxcas such that the total number of histories becomes close to or the same as the given total number of histories. In this case, some comments will be output at the end of an input echo.

Time limit of a PHITS calculation can be activated by assigning timeout a positive value. If the CPU time reaches timeout in the middle of a batch, PHITS will finish the calculation of all histories in that batch, and then terminates the code. In the distributed-memory parallel computing, timeout is ruled by the sum of all CPU time. Note that the function of timeout can be used only when the computer’s CPU time is correctly retrieved.

To calculate statistical uncertainties, two modes can be selected –“batch variance mode” and “history variance mode”– which calculate standard deviations of tally results by batch and by history, respectively. In both modes, standard deviation \(\sigma\) is calculated as

\[\sigma = \sqrt{ \frac{\displaystyle\sum_{i=1}^N\left(x_iw_i/\bar{w}\right)^2 - N\bar{X}^2} {(N-1)} }\]

where \(N\) is the total batch number (istdev=1) or the total history number (istdev=2), \(x_i\) is the tally results, \(w_i\) is the source weight of each sample, \(\bar{X}\) is the mean value of the tally results, and \(\bar{w}\) is the source weight of \(N\) samples. PHITS calculates the absolute value of the error, and then outputs the relative error obtained by dividing it by the mean of the tally values, \(\bar{X}\) .

The standard error of the mean is given as \(\sigma/\sqrt{N}\) . Then, PHITS outputs the ratio calculated by the following formula as relative error in the tally output file.

\[(r.err) = \frac{\sigma}{\sqrt{N}\bar{X}}\]

The standard errors calculated by the batch variance mode vary by combination of maxcas and maxbch, even for the same total history number. In principle, a larger maxbch provides more reliable statistical uncertainties in batch variance mode but may require a longer computational time. Setting the value of maxbch greater than 10 is recommended to obtain reliable results. By contrast, in history variance mode, the standard error depends only on the total history number and is independent of the combination of maxcas and maxbch. Thus, it is recommended to select the history variance mode (istdev=2) in most cases. Note, however, that computational time occasionally becomes extremely long in history variance mode, especially in cases in which there are tallies using many memories, e.g., xyz mesh tally with its very fine structure.

In tallies for calculating the standard error of deposition energies by individual history, such as [t-deposit] with output=deposit and [t-deposit2], the standard errors instead of the standard errors are output as relative errors. The relative standard errors can be estimated as \(1/\sqrt{K}\) where \(K\) is the number of histories contributing to the tally result. This calculation procedure is independent of the istdev parameter.

Statistical errors are generally output in the r.err column, which is the rightmost column of tally results. In the case of 2D-plots, such as tallies with axis=xy or rz, errors are output into another file, '*_err' where '*' indicates the file name specified in the tally. For example, when file=tally.out, the name of the error file is “tally_err.out.” This error file has the same format of the conventional tally output file; hence, a graph for the error in the 2D-plot can be obtained by ANGEL through a conversion process. See Section 4.9 in more detail.

Note that as what PHITS calculates is the standard error, it means the true value has a 68% of chance falling within the confidence interval. Furthermore, PHITS does not estimate systematic uncertainties contributed by the nuclear reaction models.

When istdev < 0, the restart calculation mode is activated; however, if there is no past tally result, a new calculation is started with istdev=|istdev|; in other words, the batch and history variance modes are selected for istdev = -1 and -2, respectively. When restarting a calculation, the variance mode is automatically determined from the past PHITS calculation. The tally parameters in the restart calculation must be consistent with the original tally parameters (i.e., parameters that already exist in the previous tally results must remain unchanged); however, new tally sections can be added in the restart calculation.

Table 5.2.13 nrandgen

Value

Explanation
(D=1)
pseudo-random number generator option.
=0
Linear Congruential Generator is used. Period length is \(2^{46}\) . See the reference [1] for more details.
=1
xorshift64 is used. Period length is \(2^{64}-1\) . See the reference [2] for more details.
Table 5.2.14 irskip

Value

Explanation
(D=0)
Random number control.
irskip > 0
Skip the first irskip number of histories and the random number seeds for debugging. (The total number of history is reduced by irskip.)
irskip < 0
Begin calculation with the No. irskip+1 random number seeds for executing parallel computing manually. (The total number of history is unchanged.)
Table 5.2.15 rseed

Value

Explanation
(D=0.0)
Initial random number option.
rseed=0
As the initial random seed, 6.647299061401E+12 is used. When nrandgen=0, this value is used. When nrandgen=1, 64-bit data corresponding to this value is used.
rseed \(\ne\) 0
rseed is used as the initial random seed. When nrandgen=0, absolute integer value of rssed is used. When nrandgen=1, 64-bit data corresponding to rssed is used, which can be set to a real and negative value.
Table 5.2.16 itimrand

Value

Explanation
(D=0)
Option for time dependent initial random number.
=0
Use a value determined by rseed.
=1
Get an initial random seed from starting time.
Table 5.2.17 bitrseed

Value

Explanation

64-bit data of the initial random seed. Default is bit data corresponding to 6.647299061401E+12.