ATAT SQS 与簇展开命令
mcsqs、关联函数、maps/mmaps、回归、模型预测及张量/CVM的15个命令。
阅读本篇目录
mcsqs、关联函数、maps/mmaps、回归、模型预测及张量/CVM的15个命令。 每条列出用途、输入输出、调用格式和使用条件；展开参数表可查看 ATAT 3.52 的原始英文选项与默认值。
[int] 表示整数，[real] 表示实数，[string] 表示文件名或文本；同名参数在不同命令中的含义可能不同。调用前应准备对应输入文件。
SQS、团簇和回归#
mcsqs#
以随机优化搜索匹配目标关联函数的特殊准随机结构。
输入与结果： rndstr.in、clusters.out → bestsqs.out、bestcorr.out、日志。
调用格式：
mcsqs -n=32 -ms=250000 -sd=2410
先用 -2/-3 等生成团簇；-n 是总原子数；-ms 限制步数；-ip 区分并发实例；-best 汇总；stopsqs 正常停。-T 是优化温度，不能当成材料温度。详见正文 SQS 两章。
展开本命令的完整参数表
Monte Carlo generator of Special Quasirandom Structures 3.52, by Axel van de Walle
-n=[int] nb of atom/unit cell (Default: 0)
If -n is not specified, generate clusters of the following sizes:
-2=[real] Maximum distance between two points within a pair (Default: 0)
-3=[real] Maximum distance between two points within a triplet (Default: 0)
-4=[real] Maximum distance between two points within a quadruplet (Default: 0)
-5=[real] Maximum distance between two points within a quintuplet (Default: 0)
-6=[real] Maximum distance between two points within a sextuplet (Default: 0)
-l=[string] Input file defining the random structure (default: rndstr.in)
-cf=[string] Input file defining the clusters (default: clusters.out)
-tcf=[string] Input file defining the target multibody correlations (default: internally calculated values for fully disordered state)
-tol=[real] Tolerance for matching correlations (default: 1e-3)
-wr=[real] Weight assigned to range of perfect correlation match in objective function (default 1)
-wn=[real] Multiplicative decrease in weight per additional point in cluster (default 1)
-wd=[real] Exponent of decay in weight as function of cluster diameter (default 0)
-T=[real] Temperature (default 1)
-pf=[string] Input file defining the optimization parameters (default: sqsparam.in)
-rc Read supercells from file sqscell.out (default: generate internally and write to sqscell.out) (Default: Off)
-ip=[int] Index of current process (for parallel operation) (Default: -1)
-best Collect best SQS among the outputs of prior parallel runs (Default: Off)
-crf=[string] Select correlation functions (default: trigo)
-sd=[int] Seed for random number generation (default: use clock)
-rt=[int] Read parameter file every (rt) step (default:10000)
-ms=[int] Maximum number of Monte Carlo steps (default: no limit)
-2d Generate only supercells in the plane of a,b axes (Default: Off)
-sig=[int] Number of significant digits to print in output files (default: 6)
-h Display more help (Default: Off)
gensqs#
用枚举寻找严格满足目标关联的 SQS。
输入与结果： lat.in、clusters.out、tcorr.out → 候选结构标准输出。
调用格式：
gensqs -n=8 > sqs8.out
-n 是严格指定的晶胞大小，不会列出周期更小的结构。目标过多或不相容可能没有输出。一般先用 mcsqs，gensqs 适合很小规模的枚举研究。
展开本命令的完整参数表
GENerate Special Quasirandom Structures 3.52, by Axel van de Walle
-n=[int] nb of atom/unit cell (Default: 0)
-sig=[int] Number of significant digits to print in output files (Default: 6)
-cf=[string] Input file defining the clusters (default: clusters.out)
-tc=[string] Input file defining the target correlations (default: tcorr.out)
-tol=[real] Tolerance for matching correlations (default: 1e-3)
-l=[string] Input file defining the lattice (default: lat.in)
-rc Read unit cells from file (Default: Off)
-pc Print number of supercells only, and quit (Default: Off)
-tp=[int] Total number of processes (for parallel operation) (Default: 1)
-ip=[int] Index of current process (0,...,tp-1) (for parallel operation) (Default: 0)
-crf=[string] Select correlation functions (default: trigo)
-h Display more help (Default: Off)
corrdump#
计算结构的团簇关联、生成团簇、输出对称信息或按 ECI 预测。
输入与结果： lat.in、str.out，或预先给定的 clusters.out → 关联表及团簇/对称文件。
调用格式：
corrdump -2=4.0 -3=2.8 > corr.out
-c 读取现有团簇；-clus 只建团簇；-ro 读取概率占据；-rnd 理想无序态；-eci 预测；-mi 表示系数已含重数。默认可能覆盖 clusters.out，扩展 SQS 关联检查要另用目录。
展开本命令的完整参数表
CORRelation DUMPer 3.52, by Axel van de Walle
-h Display more help (Default: Off)
-2=[real] Maximum distance between two points within a pair (Default: 0)
-3=[real] Maximum distance between two points within a triplet (Default: 0)
-4=[real] Maximum distance between two points within a quadruplet (Default: 0)
-5=[real] Maximum distance between two points within a quintuplet (Default: 0)
-6=[real] Maximum distance between two points within a sextuplet (Default: 0)
-l=[string] Input file defining the lattice (default: lat.in)
-s=[string] Input file defining the structure (default: str.out)
-pc Print composition only (Default: Off)
-pcm Print composition matrix only (Default: Off)
-fast Use fast algo to calculate correlations of 'simple' supercells (Default: Off)
-skipr Skip algorithm robust to relaxations (Default: Off)
-sym Just find space group (Default: Off)
-clus Just find clusters (Default: Off)
-c Read cluster file instead of writing it (Default: Off)
-cf=[string] File name of the cluster file (default: clusters.out)
-z=[real] Tolerance for finding symmetry operations (default: 1e-3)
-sig=[int] Number of significant digits printed (default: 5)
-noe Do not include empty cluster (Default: Off)
-nop Do not include point cluster(s) (Default: Off)
-noc Do not include any cluster(s) (Default: Off)
-eca Calculate Energy of Clusters of Atoms (Default: Off)
-wu=[string] Write Unrelaxed structure into specified file (Default: )
-rnd Print correlation of the random state of the same composition as the input structure (Default: Off)
-eci=[string] Predict quantity using ECI in specified file (Default: )
-mi Multiplicities are already included in ECI file (Default: Off)
-crf=[string] Select correlation functions (default: trigo)
-nb Print structure number (Default: Off)
-ro Read lattice file containing occupation variables (this code does not make use of them) (Default: Off)
getclus#
将团簇文件整理成便于阅读的摘要。
输入与结果： clusters.out，可选 ECI 文件 → 阶数、直径、重数及系数。
调用格式：
getclus -e > eci-table.out
无参数只看团簇；-e 读 eci.out，也可直接指定其他系数文件。摘要顺序与 clusters.out/eci.out 一致，不要分别排序后再配对。
展开本命令的完整参数表
Syntax: getclus [-e] [ecifile]
extracts, from clusters.out, the cluster sizes, lengths and multiplicities
if ecifile is given, it also prints the eci therein
if -e is specified, it prints eci from eci.out
maps#
为二元晶格体系自动选择结构并更新簇展开。
输入与结果： lat.in 与 编号/str.out、编号/energy → ECI、拟合、预测和基态候选。
调用格式：
maps -d
程序常驻，通过 ready 请求结构、stop 结束；-g/-mv 控制搜索范围；nbclusters.in 可人工选择簇族，改变后 touch refresh。energy 是结构晶胞总能；有多于两种可替换物种用 mmaps。
展开本命令的完整参数表
MIT Ab initio Phase Stability (MAPS) code 3.52, by Axel van de Walle
-h Display more help (Default: Off)
-l=[string] Input file defining the lattice (default: lat.in)
-z=[real] Tolerance for finding symmetry operations (default: 1e-3)
-c=[real] Exponent of the order of complexity (default: 3)
-t=[int] Time between disk reads in sec (default: 10 sec)
-m=[int] Maximum number of points in cluster (default 4)
-g=[int] Extend ground state search up to structures having at least that many atoms. (Default: 0)
-c0=[real] [c0,c1] is the concentration range where ground states must be correct. (Default: 0)
-c1=[real] (Default: 1)
-mv=[int] Maximum volume (number of atom per unit cell) allowed when generating structures (default: no limit)
-mw=[real] Maximum weight allowed when try to match ground states (Default: 16)
-2d Find supercells along a and b axes only (Default: Off)
-p=[string] Predictor plugins to use (examples: -p=cs or -p=cs_es) (Default: )
-ks=[string] same as -p (Default: )
-fa=[string] Select fitting algorithm (default: built-in)
-q Quiet mode (do not print status to stderr) (Default: Off)
-sig=[int] Number of significant digits to print in output files (Default: 6)
-d Use all default values (Default: Off)
mmaps#
面向多组分/多子晶格的自动簇展开。
输入与结果： 同 maps，增加多元组分区域和属性设置 → 多元拟合与化学势信息。
调用格式：
mmaps -d
-pn 指定要拟合的性质文件；-pa 表示已归一化；-cr 限定组分区域；-ig 不强制训练集基态一致。保持 atoms.out、团簇和 ECI 的顺序。Bayesian 拟合是另一个算法选项，不能与本文默认拟合结果混称。
展开本命令的完整参数表
MIT Ab initio Phase Stability (MAPS) code 3.52, by Axel van de Walle
-h Display more help (Default: Off)
-l=[string] Input file defining the lattice (default: lat.in)
-z=[real] Tolerance for finding symmetry operations (default: 1e-3)
-c=[real] Exponent of the order of complexity (default: 3)
-t=[int] Time between disk reads in sec (default: 10 sec)
-m=[int] Maximum number of points in cluster (default 4)
-g=[int] Extend ground state search up to structures having at least that many atoms. (Default: 0)
-cr=[string] Concentration region input file (Default: crange.in)
-he=[real] Highest predicted energy, above ground state hull, allowed when generating structures (default: no limit)
-mv=[int] Maximum volume (number of atom per unit cell) allowed when generating structures (default: no limit)
-mw=[real] Maximum weight allowed when try to match ground states (Default: 16)
-2d Find supercells along a and b axes only (Default: Off)
-p=[string] Predictor plugins to use (examples: -p=es or -p=es_cs) (Default: )
-ks=[string] same as -p (Default: )
-fa=[string] Select fitting algorithm (default: built-in, alternative: -fa=bayesian)
-xo=[string] input additional parameter for extension (Default: )
-crf=[string] Select correlation functions (default: trigo)
-pn=[string] Property to cluster expand (default: energy)
-ig Ignore whether cluster expansion predicts correct ground states (Default: Off)
-pa Quantity to expand is already per atom (Default: Off)
-q Quiet mode (do not print status to stderr) (Default: Off)
-sig=[int] Number of significant digits to print in output files (Default: 6)
-d Use all default values (Default: Off)
mapsrep#
用 gnuplot 查看二元簇展开的能量、误差和 ECI。
输入与结果： maps 输出文件 → 图形会话。
调用格式：
mapsrep
-e 也显示标有 error 的结构。需要 gnuplot；无桌面时设置文件终端或用随文绘图脚本。不会重新计算结构能量。
展开本命令的完整参数表
Display the output of maps in graphical form.
Syntax: mapsrep [-e]
the optional switch -e tells the code to plot structures flagged with error too.
Thanks to Ruoshi Sun and Mike Widom for updating the script for recent versions of gnuplot.
mmapsrep#
显示多元簇展开的组分投影与拟合信息。
输入与结果： mmaps 输出文件以及用于作图的组分列 → gnuplot 图形。
调用格式：
mmapsrep 1 2 3
三个参数是组分坐标列，不是元素原子序数。先核对文件列定义；-e 包含标记为 error 的结构。
展开本命令的完整参数表
Display the output of mmaps in graphical form.
Syntax: mmapsrep [-e] c1 c2 c3
where c1,c2,c3 are the column numbers of the compositions to plot
the optional switch -e tells the code to plot structures flagged with error too.
lsfit#
完成最小二乘、多项式回归及留一交叉验证。
输入与结果： -x 自变量矩阵、-y 目标、可选权重/正则项 → 系数或指定诊断量。
调用格式：
lsfit -x=x.in -y=y.in -1
-1 加常数列；-pw 定义幂次；-s 选列；-cv 交叉验证；-p 预测；-e 误差；-se 标准误；-colin 处理共线列。数据行必须严格对应，不把 CV 与训练 RMSE 混用。
展开本命令的完整参数表
Least-Square FIT 3.52, by Axel van de Walle
-x=[string] x file (Default: )
-y=[string] y file (Default: )
-w=[string] weight file (Default: )
-r=[string] regularization parameters file (Default: )
-pf Perfect Fit algorithm (Default: Off)
-pw=[string] number of powers of each column to regress on (Default: )
-s=[string] select columns to regress on (one number by column, 0: ignore, 1: use) (Default: )
-1 add a column of ones as regressor (Default: Off)
-colin Ignore colinear columns in x file (Default: Off)
-cv Print crossvalidation score (Default: Off)
-se Print standard errors (Default: Off)
-ty Print trues values of y (Default: Off)
-p Print predicted values of y (Default: Off)
-e Print prediction error (Default: Off)
-mat Print fitting matrix (Default: Off)
-sig=[int] Number of significant digits printed (default: 5)
cv#
从关联矩阵中选择簇并评估交叉验证。
输入与结果： corr.in、energy.in、clusters.out，可选 include.in/weight.in → 拟合、ECI 和诊断文件。
调用格式：
cv -d
未知能量在 energy.in 中写 x。-g/-w/-p 可强调基态。它会输出/替换模型文件，应在独立副本中操作，不覆盖正在运行的 maps。
展开本命令的完整参数表
Cross-Validation code 3.52, by Axel van de Walle
-g Favor cluster choices where the ground state line is correct (Default: Off)
-w=[real] Weight structures by w/(struct_energy-gs_energy+w) (Default: 1000)
-p=[real] Penality for structures that lie below the ground state line (Default: 0)
-d Use all default values (Default: Off)
-h Display Help (Default: Off)
clusterexpand#
用已有团簇集合拟合目录中某个标量、温度序列或张量性质。
输入与结果： lat.in、clusters.out、各目录 str.out 和指定性质文件 → 性质.eci 与 性质.ecimult。
调用格式：
clusterexpand -cv energy
默认按结构原子数归一化；-pa 表示输入已归一化；-e 跳过 error；-s 选择列；-g 用广义簇展开。各结构的温度点/张量排列必须一致。
展开本命令的完整参数表
Syntax: clusterexpand [-e] [-pa] [-s "1,0,1, ..."] filename
Cluster expands the quantity stored in */filename
Uses the clusters in clusters.out
Outputs to filename.eci filename.ecimult
ecimult contains the eci multiplied by multiplicity
the option -e causes the structures flagged with an error file to be ignored
the option -pa indicates that the quantity is per atom already
so that the code must NOT divide it by the number of atoms
the option -s lets you select which cluster to include (1) or exclude (0)
the option -g selects a generalized (tensor) cluster expansion
the -cv option prints the cv score.
clusterpredict#
用 clusterexpand 输出的模型预测单个结构的性质。
输入与结果： 结构目录中的 str.out、上层模型及性质.eci → 预测值。
调用格式：
clusterpredict energy
在独立预测目录运行；它会将结果写入与性质同名的文件，可能覆盖已有 energy。先复制 str.out，保留原 DFT 训练数据。与 clusterexpand 的单位约定配套使用 -pa，不能把总量与每原子量混用。
展开本命令的完整参数表
Syntax: clusterpredict [-pa] file
see clusterexpand for more info.
This command is to be run from within a specific structure directory containing a str.out file.
gce#
按晶体对称性构建张量等广义性质的簇基。
输入与结果： lat.in、str.out、gcetensor.in 等 → 广义 clusters.out 和关联/张量信息。
调用格式：
gce -2=2.8 > tensor-corr.out
gcetensor.in 定义张量阶数和对称置换；-st 对称化给定张量；-gce 选对象类型。该团簇文件与普通标量簇展开不通用，必须独立保存。
展开本命令的完整参数表
Generalized Cluster Expansion 3.52, by Axel van de Walle
-h Display more help (Default: Off)
-2=[real] Maximum distance between two points within a pair (Default: 0)
-3=[real] Maximum distance between two points within a triplet (Default: 0)
-4=[real] Maximum distance between two points within a quadruplet (Default: 0)
-5=[real] Maximum distance between two points within a quintuplet (Default: 0)
-6=[real] Maximum distance between two points within a sextuplet (Default: 0)
-l=[string] Input file defining the lattice (default: lat.in)
-s=[string] Input file defining the structure (default: str.out)
-pc Print composition only (Default: Off)
-sym Just find space group (Default: Off)
-st Symmetrize tensor given in tensor.in (Default: Off)
-clus Just find clusters (Default: Off)
-c Read clusters.out file instead of writing it (Default: Off)
-z=[real] Tolerance for finding symmetry operations (default: 1e-3)
-sig=[int] Number of significant digits printed (default: 5)
-noe Do not include empty cluster (Default: Off)
-nop Do not include point cluster(s) (Default: Off)
-rnd Print correlation of the random state of the same composition as the input structure (Default: Off)
-eci=[string] Predict quantity using ECI in specified file (Default: )
-mi Multiplicities are already included in ECI file (Default: Off)
-crf=[string] Select correlation functions (default: trigo)
-gce=[string] Select type of Generalized Cluster Expansion (default: tensor)
-nb Print structure number (Default: Off)
cvmclus#
生成簇变分法 CVM 熵表达式所需的簇、概率变换和系数。
输入与结果： lat.in、maxclus.in 的最大团簇 → clusters.out、clusmult.out 及 CVM 系数文件。
调用格式：
cvmclus -d
多个最大团簇以 end 分隔。它生成表达式系数，不自动最小化自由能，也不自行生成完整相图。
展开本命令的完整参数表
Cluster Variation Method CLUSter generator 3.52, by Axel van de Walle
-h Display more help (Default: Off)
-l=[string] Input file defining the lattice (Default: lat.in)
-m=[string] Maximal cluster (Default: maxclus.in)
-sig=[int] Number of significant digits printed (Default: 5)
-crf=[string] Select correlation functions (default: trigo)
-d Use all default values (Default: Off)
原始参数表来自 ATAT 3.52，作者 Axel van de Walle 及相应贡献者。中文说明由鸢算科技编写；算法出处与论文引用。
需要根据你的体系选择资源？联系讨论计算需求。实际可用软件与权限，以当前账户和运行环境为准。