---

No wizardry with GRINSP predictions, 
only a cute (?) algorithm. 

---

Introduction

The concept of inorganic crystal structure prediction by using geometrical restraints is not new :

- Zeolite researchers have documented more than 1000 hypothetical structures by using classical physical model building [1] during the past 60 years.
- Simulated annealing is a rapid generator of hypothetical 4-connected framework structures and others. More than 5000 hypothetical zeolite structures were reported in ref [2]. More than 1.000.000 are now in the Hypothetical Zeolites Database.
- Many recent works in inorganic structure prediction (as well as organic and organometallic) have produced huge quantities of hypothetical compounds (using commercial packages as CERIUS, etc), no room here for citing them all.
- Systematic enumeration is now based on advances in mathematical tiling theory [3-4].

Where are these predicted structures ? A few of them are inside of the ICSD, for instance some theoretical SiO₂ structures [5]. More than 1.000.000 zeolite models are inside of the Hypothetical Zeolites Database. But what about predicted compounds other than SiO₂ ? The main purpose of GRINSP(*) is to generate hypothetical inorganic structures MxM'yXz which will be documented in a searchable database : PCOD (Predicted Crystallography Open Database), a subset of the COD. If you are a good "predictor" and want to deposit your predicted structures in PCOD, this is already possible (CIF files only), visit the upload page.

GRINSP does not work by applying simulated annealing to a starting random configuration. Version 2.00 works schematically as follows, by using the Monte Carlo method :

• Manual selection of the constraints on cell parameters, of restrained interatomic distances, of the  type(s) of coordinations, and of the space groups. Then the Monte Carlo process starts.
• Random selection of the cell parameters inside of the predefined range.
• Random positioning of a first cation M (or M') of the future MxXy (or MxM'yXz) compound on a general or special position, itself selected randomly.
• Random positioning of the next cations (random choice of M or M') in respect of the distance restraints with the atoms already accepted, on a general or special position, itself selected randomly.
• If a model fulfills all distance and coordination criteria, place the X atoms at M-M midpoints, refine the atomic positions and cell parameters so as to improve an R factor.
• Continue to try to predict structures in that way till a certain number of cells are tested.
• Find if the predicted structures are new or were already described (using CS - Coordination Sequences), keep those with best R factors.

Currently, there are some limitations in GRINSP 2.00 which proved to be efficient for a maximum number of 192 M/M' atoms on up to 1 to 6 different general or special positions. It was shown to be able to predict many zeolites (ABW, ACO, AFI, BIK, EDI, FAU, GIS, LTA, SOD... but not all of them) or the compact SiO₂ phases (quartz, cristobalite, tridymite, etc), polymorphs for B₂O₃, MF₃ (M = Al, Fe, Cr...), hypothetical phases in binary systems B₂O₃/SiO₂, B₂O₃/ReO₃, SiO₂/ReO₃, titanosilicates, etc (see the PCOD). It is up to you to try GRINSP with other systems, and even the above ones have not been completely explored.

Further work is needed for improving the GRINSP efficiency :

• Introduction of different linkage modes than by corners (edges, faces...)
• Adding the possibility for insertion of big cations K/Sr/Ba/Cs/etc as spheres in the holes/tunnels
• Considering bond valence as an alternative to pure geometrical restraints for the model final refinements
• Increase speed by not recalculating always everything (distances)
• Increase the box size for the CS (coordination sequence) calculations (729 cells is not always enough...)
• Make a parallel version, working on dual-core processor or biggest multi-processor machines
• Produce a database of calculated powder patterns which would allow early identification of new phases and even structure solution before indexing or in spite ot non-indexation
• Etc !

[1] J.V. Smith, Chem. Rev. 88 (1988) 149-182.
[2] M.W. Deem and J.M. Newsam, J. Am. Chem. Soc. 114 (1992) 7189-7198.
[3] O. Delgado Friedrichs, A.W.M. Dress, D.H. Huson,, J. Klinowski, A.L. Mackay, Nature 400 (1999) 644.
[4] M.D. Foster, O. Delgado Friedrichs, R.G. Bell, F.A. Almeida Paz, J. Klinowski, Angew Chem. Int. Ed. 42 (2003) 3896-3899.
[5] M.B. jr Boisen, G.V. Gibbs, M.S.T. Bukowinski, Phys. & Chem. of Minerals (Germany) 21 (1994) 269-284.

(*) If you find "GRINSP" unpronounceable, suggest another name to alb@cristal.org , other possible names, not retained, were "INORGOD" (only God can predict...) or "INORGURU", "AUGUR", "PREDINORG"...

GRINSP is distributed in two packages :

• grinsp.zip version for monoprocessor
• pgrinsp.zip version for dual-core processors (see details below)

Both packages contain, in appropriate directories :

• grinsp.exe : GRINSP version 2.00, executable for Win95/98/NT/XP.
• grinsp.pdf : a copy of the text published in the J. Appl. Cryst. (2005).
• grins.zip: contains the GRINS satellite program for model optimization after cation/anion substitution.
• cutcifp.zip : contains the CUTCIFP satellite program for cutting multiple CIFs.
• cif2con.zip : contains the CIF2CON satellite program reading a multiple CIF and creating a .con file with coordination sequences
• connect.zip : contains the CONNECT satellite program for analysis of coordination sequences (.con file)
• framdens.zip : contains the FRAMDENS satellite program for listing the compounds with smallest densities.
• index.zip : Containing this help file, index.html, and additional image files.
• connectivity.txt : The coordination sequences (CS) of known zeolites and dense SiO₂.
• distgrinsp.txt : file containing the geometrical restraints for a few given atom pairs (you may add your own data there).
• wyckoff.txt : file containing the general and special position codes for all standard space groups. Note that the true multiplicity is regenerated by GRINSP after application of the Bravais translations.
• examples.zip : Some example files.
• grinsp.ico : an icon for GRINSP, a few tetrahedra coming out from a wizardry hat.
• the pgrinsp package contains an additional file : libguide40.dll, to be installed in the same directory as grinsp.exe

Unzip all these files in any directory named at your convenience, and run the program (no DLL needed, nothing to change in the autoexec.bat file...). Note that connectivity.txt, distgrinsp.txt and wyckoff.txt have to be absolutely in the same directory as grinsp.exe and the .dat file defining your expected predictions. 

The FORTRAN source codes for GRINSP and the satellite programs (GRINS, etc) ready for compilation ("console application") by the Intel Visual Fortran 9.1 compiler can be found into the .zip files, with .f extension.

sio2.dat : File ready for the prediction of zeolites by GRINSP.
titanosilicates.dat : File ready for the prediction of titanosilicates by GRINSP.
test129.dat : File for the prediction of t-AlF₃ in the P4/nmm space group by GRINSP.
GaF3.dat and TiP.dat : Files ready for GRINS.

Examples.zip contains also the .imp files, if you wish to compare with your results (note that this is Monte Carlo, you may obtain the same models in a different sequence).

October 2006 : PGRINSP package

- GRINSP inside of the PGRINSP package is the version for parallel computing able to exploit dual-core processors (tested only with Intel Pentium D...), using OpenMP directives as available inside of the Intel Visual Fortran compiler 9.1.

- You will note some changes in the order of the solutions displayed on the screenbox during the run. If say ncells =10000 cell tests are required per space group, one processor will work on ncells = 1 to 5000 and the other will work on ncells = 5001 to 10000. Otherwise, no change in datafile or output files. 

- Speed : between 1.7 and 1.8 times faster with 2 processors than with only one.

- Quad cores in 2007 ;-)

- 80 cores in 5 years ???

Trick : if you cannot work well with your PC when GRINSP is running, decrease the GRINSP priority to lower than normal. This is done with the task administrator (Ctrl Alt Supp). Go to "process", select the GRINSP process with the mouse right button, select "priority" and decrese it.

NOTE : The libguide40.dll has to be installed in the same directory as grinsp.exe
 

April 2006 : Improvements in GRINSP version 2, if compared to version 1.

- Larger models can be predicted (limit now at 192 M/M' atoms instead of 64). Structures as
             complex as faujasite can be produced.

- Bugs corrected (the coordination sequences are no longer giving strange results)

- More user-friendly, parameter file simplified :
       + a range of space groups can be examined inside of the same run, instead of only one,
       + space groups are specified by their number instead of the Hermann-Mauguin description
       + no range angles to provide

- More details inside of the output CIFs : 
      + better analysis of the formula and Z, 
      + output of the M/M' starting Wyckoff positions (before optimization) so that retrieving the 
             true space group from the P1 description is facilitated,
      + output of the FD (Framework Density).

- GRINS allowing for the computation of isostructural compounds is considerably improve, and 
              can read multiple CIFs of a previous series from GRINSP : titanosilicates, once 
              modelled can lead fast to isostructural titanophosphates, vanadophosphates,
              gallophosphates, etc.

- Satellite programs are provided for global analysis of the (sometimes) huge lists of predicted 
              structures:
      + CUTCIFP can read a multiple CIF and provide single CIFs having names changes at your 
         convenience,
      + CIF2CON can read a multiple CIF and provide a .con file containing the connectivity 
          sequences (CS) for further use and identification of unique models,
      + CONNECT can read a .con file, identify unique models, compare to a previous list of 
         models characterized by their CS (stored into the file connectivity.txt), and provide a sorted 
         classification by decreasing order of the R factor. 
      + FRAMDENS reads a multiple CIF and produces the list of compounds ordered according
         to their framework densities (number of cations for 1000 A³), allowing to point at the best
         models with largest tunnels/cavities.

2009

The PCOD is updated with a lot of results from GRINSP.

2018

Hundreds of drawings of these previous 2009 results are made and classified according to their 0D, 1D, 2D or 3D character. Se the What is new page.

Verify first that the working directory contains grinsp.exe, connectivity.txt, distgrinsp.txt, wyckoff.txt and your parameter file with .dat extension (for instance SiO2.dat as below). The Windows PC version will run by clicking on grinsp.exe, opening a window shown below :

select your data file, for instance SiO2.dat from the example files (do not type the .dat extension). The entry data parameters are displayed, as well as a summary for every successful prediction. You may stop the program execution by typing K (capital letter) anytime. The program will store and sort the results, and stop at the next multiple of 50 runs (so the stop is not immediate but may need a few seconds), See now what are exactly the parameters into the starting .dat file:
 

Zeolites SG: 16-74       : text for this run
16 74                    : SG : Space groups range (two values between 1 230)
1 0 1 192                : npol, ncon, nmim, nmax
4                        : ncpol (coordination for each polyhedron-type)
Si  O                    : elements for a search in the distgrinsp.txt file
3. 30. 3. 30. 3. 30.     : min and max cell parameters a, b, c 
5. 35.                   : min and max framework density (FD)
1000 300000 0.02 0.12    : ncells, genmax, Rmax, Rdt0
6000 1                   : idls (MC cycles for distance refinement) and iref (1 or 0)
1                        : code for selecting models to output

Titanosilicates - SG: 188-194 
188 194
2 0 2 192      npol = 2 here instead of 1.    
6 4            note the two coordinations defined here (octahedron and tetrahedron)
Ti  O                    one line for the TiO6 octahedron
Si  O                    and one line for the SiO4 tetrahedron
3. 30. 3. 30. 3. 30. 
5. 35.
2000 300000 0.02  0.12
6000 1    
1    

text    A title for the prediction session, 80 characters max (format 20A4).

SG    The space group(s) range examined (two integers : SG numbers)
            examples :
               1 230     :   all space groups examined starting at 230, decreasing symmetry.
               74 74     :   only one space group examined : N°74
               195 230 :   cubic space groups examined stating at 230, decreasing to 195

          GRINSP contains all standard space groups 
                      (if you want non-standard SG, add them in Wyckoff.txt
 

npol, ncon, nmim, nmax : four integers (free format)

npol = number of different types of polyhedra you wish in the predicted structure
Warning : only npol = 1 or npol = 2 are allowed in GRINSP version 2.00. 

ncon = defines the degree of connection between polyhedra
          0 : every X atom is connected to two M/M' atoms (corner-sharing)
          > 0 : some X atoms are connected to only one polyhedron
          < 0 : there could be edge-sharing, etc
Warning : only ncon = 0 is working in GRINSP version 2.00

nmin = minimal mumber of M/M' atoms for saving a model
             (if you are tired to see these small stuctures with nT = 1, 2, 3 or 4 all the time) 
            For exploring a binary system, only solutions mixing M and M' will be retained 
            (meaning that the minimum will be nmin=2, anyway)

nmax = maximal mumber of M/M' atoms for saving a model (max = 192)
            if you are exploring small cell volumes, reduce nmax to appropriate values (20, etc)
            this will save computer time (avoiding to test Wyckoff positions corresponding to
            too much atoms).

ncpol = coordination for every of the npol different polyhedron-type
Warning ! only four values ncpol = 3, 4, 5 or 6 are allowed in GRINSP version 1.00
and a maximum of 1 or 2 values can be given (because npol = 1 or 2 maximum)

Elements : two elements for a given polyhedron - there may be two lines if a binary system is explored. This part is formatted as 2A4 : four cases per atom.

GRINSP will have to find the minimal/maximal/ideal interatomic distances for this polyhedron inside of the file distgrinsp.txt
After the same atom codes in distgrinsp.txt  is added the coordination, and then 4 lines corresponding to the prescribed interatomic distances : 
Si  O   4               : atom codes and coordination (here a SiO4 tetrahedron) 
2.60 3.60 3.070   : minimal/maximal/ideal first distance between atom pair 1-1 
1.30 1.90 1.610   : minimal/maximal/ideal first distance between atom pair 1-2 
2.20 3.00 2.629   : minimal/maximal/ideal first distance between atom pair 2-2 
4.40 6.00             : minimal/maximal second distance between atom pairs 1-1 

You may edit distgrinsp.txt and add there your own data (anywhere) : 5 lines per kind of polyhedron as above. Care that the elements are given in 2A4 format.

cell = one line giving the cell parameter ranges to explore for structure prediction
            6 values : amin, amax, bmin, bmax, cmin, cmax
            free format

Framework density, min and max.
     The framework density is the number of M/M' atoms for a volume of 1000A³.
     Solutions outside of this given range will be excluded. This allows to avoid retaining some 
     two-dimensionnal crystal structures, if you do not want them. Be careful to allow a sufficiently 
     large range ( 5 to 35 should be correct for zeolites)

ncells, genmax, Rmax, Rdt0

     ncells   = number of different cells examined per space group (use 200-20000)
                There is no maximum in fact, but testing 1000 cells may need 10 to
                60 minutes.
                These cells are proposed randomly by the Monte Carlo process.

     genmax  = number of Monte Carlo trials for a given cell
                use 10000-500000, the latter being time consuming...

     Rmax    = the structure predictions corresponding to the R factor lower
                than Rmax will be stored and sorted (use Rmax = 0.005-0.01
                if you wish regular polyhedra, up to 0.02 if you tolerate distortion,
                up to 0.03-0.05 if you expect trigonal prisms instead of octahedra, 
                or even pentagonal pyramids, etc) 
         The R value is analogous to the RDLS (see the Database of Zeolites
                structures), but is obtained by Monte Carlo, not by a 
                least-squared refinement. Use no more than Rmax = 0.02 if you wish to
                upload your predicted structure into the PCOD.

     Rdt0    = the candidate structures corresponding to the R factor lower
                than Rdt0 before optimization will be optimized 
                (use Rdt0 = 0.10-0.20). The more Rdt0 is small, the more the
                optimizations have chances of being successful (leading to
                R < Rmax). In general, one third to half of the candidate structures 
                have Rdt0 < 0.13 before optimization. Some statistics are given at
                the end of the .imp file. See the study about speed below.

Note that from genmax are calculated 'insistence factors' (IF). These IF are
6 values corresponding to the number of Monte Carlo trials during which to insist
before to change some parameters (selecting a new cation for completing its
environment, or selecting which new coordination for a new cation to add, or etc) :

   - value 1 is for insisting on placing a second atom
   - value 2 is for insisting on the completion of the cationic neighbouring of 
             a cation having already one previous neighbour
   - value 3 is for insisting on the completion of the cationic neighbouring of 
             a cation having already two previous neighbours
   - value 4 is for insisting on the completion of the cationic neighbouring of 
             a cation having already three previous neighbours
   - value 5 is for insisting on the completion of the cationic neighbouring of 
             a cation having already four previous neighbours
   - value 6 is for insisting on the completion of the cationic neighbouring of 
             a cation having already five previous neighbours

These insistence factors are fixed to be equal to IF1=genmax/320, IF2=genmax/160, 
IF3=genmax/80, IF4=genmax/40, IF5=genmax/20 and IF6=genmax/10 if the maximal 
coordination is larger than 4, and to be equal to IF1=genmax/80, IF2=genmax/40, 
IF3=genmax/20 and IF4=genmax/10 if the maximal coordination is smaller or equal 
to 4. So that the choice of genmax may be critical. The author uses generally 
genmax = 300000.

     idls     = number of Monte Carlo steps for the interatomic distances
                and cell improvement (use 20000) at the optimization stage

     iref     = code for cell improvement (iref=1) or not (iref=0)
                if iref = ,1, half of the idls Monte Carlo steps wll be used for
                the cell improvement.

     1 or -1

     if 1  : output of new solutions, having Coordination Sequences 
            not already existing into connectivity.txt, or having a better R.
             
 if -1 : output of all solutions (not only those being new).

GENERAL LIMITS:

10000 optimized structures (2000 in PGRINSP)
192 M/M' cations
5000 coordination sequence in connectivity.txt (2000 in PGRINSP)
576 anions
2 kinds of polyhedra

This satellite program can optimize previous GRINSP models, changing the M/X and/or M'/X couples (building isostructural compounds).

The organization of the .dat file is much simpler than for GRINSP. Details for the model to be transformed are obtained from a CIF previously built by GRINSP (the model description should be exclusively in P1 space group). GRINS can also cope with multiple CIFs, transforming for instance a long series of titanosilicates into gallophosphates (etc). 

Two examples are below with the GaF3.dat (Ga replacing Fe) and TiP.dat (Ti and P replacing Ti and Si) files :

GaF3.dat file :

Test : building GaF3 from FeF3
FeF3                ! the filename of the CIF containing previous model(s)

1          ! number of different polyhedra
6          ! coordination(s) of the polyhedra
Fe  F      ! couple M/X in the previous model(s)
Ga  F      ! new couple M/X for the isostructral compound
5          ! nruns : number of different tests (for finding the best)
5000 1 3   ! optimization steps, cell refined, lines in the original CIF
3300000    ! first filename

Test : building titanophosphates from titanosilicates
total-cif-best       ! the filename of the CIF containing previous model(s)
2          ! number of different polyhedra
6 4        ! coordination(s) of the polyhedra
Ti  O      ! couple M/X in the previous model(s)
Si  O      !    second couple in the previous model(s)
Ti  O      ! new couple M/X for the isostructral compound
P   O      !    second couple for the isostructural compound
5          ! nruns : number of different tests (for finding the best)
5000 1 2   ! optimization steps, cell refined, lines in the original CIF
2201000    ! first filename

The parameter "line" above corresponds to the number of lines between the line
containing "Probable space group" and the line starting by _cell_length_a
in the CIF containing the model(s). In the example below, there are 3 lines.
Some old results from GRINSP version 1 had not the cell formula unit, so that
there would be only 2 lines.  

data_PCOD2201011
_publ_section_title
;
Structure prediction by GRINSP 2.00 - 2006 (A. Le Bail)
TiPO5          PCOD2201011 R =  0.0051
Probable space group: P -4 21 C                           
;
_chemical_formula_sum  "Ti P O5"
_cell_formula_units_Z    4
_cell_length_a          6.4086
_cell_length_b          6.4086
_cell_length_c          7.7859
_cell_angle_alpha     90.000
_cell_angle_beta      90.000
_cell_angle_gamma     90.000
_cell_volume           319.77

Satellite software allowing to cut a multiple CIF into individual CIFs, changing all PCOD numbers at your convenience. Note that the building of multiple CIFs can be made by concatenation of single CIFs by using the command line :

In the example delivered into cutcifp.zip in the package, all the 89 CIFs produced by the SiO2.dat example were concatenated, then they were cutted by CUTCIFP with a change of their PCOD numbers in a continuous series from PCOD4500000 to PCOD4500088, then all these individual CIFs were again concatenated into only one multiple CIF that was then processed by CIF2CON below (use numbers with 7 digits). 

Satellite software allowing to extract all the coordination sequences (CS) from a multiple CIF. The created file is names with the .con extension.

In the example delivered into cif2con.zip, the multiple CIF file total.cif is processed by CIF2CON, giving the starting file number 4500000, producing a file named total.con. A part of the content of that file is shown below :

PCOD4500000 SiO2       R =  0.0092   19 P 21 21 21 filenumber plus some details (formula, R, SG number
  2                                                         number of different nodes 
  8  4                                                      number of equivalent atoms for each node
   4  12  30  48  76 114 152 196   0   0                    connectivity sequence for each node
   4  12  28  50  80 110 152 198   0   0                                      
PCOD4500001 SiO2       R =  0.0083   19 P 21 21 21
  4
  4  4  4  4                                                                  
   4  11  26  43  68 101 132 174 221 267                                      
   4  11  25  44  69  97 135 172 218 275                                      
   4  11  24  43  68  98 134 173 222 265                                      
   4  11  23  44  69  94 131 181 213 277                                      
PCOD4500002 SiO2       R =  0.0090   19 P 21 21 21
  1
 12                                                                           
   4  12  26  46  70 100 136 178   0   0                                      

Satellite software allowing to analyze a .con file built up by using CIF2CON, and to detect the models having the same coordination sequences (CS). A list of the best models (with best R values) is proposed at the end of the result file with .txt extension.

In the example delivered into connect.zip, the file total.con is analyzed by running CONNECT. Note that the connectivity.txt file is required. In that case, it contains all the connectivity sequences of the known zeolites. The 89 CIFs produced by GRINSP with the test case SiO2.dat are shown by CONNECT to be reduced into 48 distinct models. CONNECT produces the result into the file named total.txt. At the beginning of that file, are listed the 89 first lines of the connectivity sequences found into total.con. Then the analysis starts by comparison of the connectivity sequences :

  PCOD4500000                 is probably new
 PCOD4500001                 is probably new
 PCOD4500002                 is probably new
 PCOD4500003                 is probably new
 PCOD4500004                 is probably new
 PCOD4500005                 is probably new
 PCOD4500006                 is probably new
 PCOD4500007 is probably PCOD4500005
    R factors     7.8999996E-03  9.7000003E-03
    but with better R, or =...  7.8999996E-03  9.7000003E-03
    update made
 PCOD4500008                 is probably new
 PCOD4500009 is probably PCOD4500008
    R factors     6.3000000E-03  6.6999998E-03
    but with better R, or =...  6.3000000E-03  6.6999998E-03
    update made
Etc

    models to save :          48
 
   Models ordered according to R
PCOD4500051 SiO2            R =  0.0027   44 I M M 2     one model
PCOD4500057 SiO2            R =  0.0034   45 I B A 2     two times the same model     
 PCOD4500056 SiO2            R =  0.0035   45 I B A 2    note that next models are beginning by a space
PCOD4500050 SiO2            R =  0.0040   43 F D D 2     4 times the same model
 PCOD4500047 SiO2            R =  0.0062   43 F D D 2   
 PCOD4500048 SiO2            R =  0.0051   43 F D D 2   
 PCOD4500049 SiO2            R =  0.0048   43 F D D 2   
PCOD4500031 SiO2            R =  0.0041   36 C M C 21    2 times the same model 
 PCOD4500029 SiO2            R =  0.0065   36 C M C 21  
PCOD4500011 SiO2            R =  0.0045   20 C 2 2 21      3 times the same model
 PCOD4500008 SiO2            R =  0.0067   20 C 2 2 21  
 PCOD4500009 SiO2            R =  0.0063   20 C 2 2 21  
Etc

 ===============================================================================
        FINAL LIST OF UNIQUE PROPOSALS, sorted by R :
 ===============================================================================
 
 
   R     NT       Vol     FD       a        b        c      alpha   beta    gamma     MC      Run    File   Ident?
0.0013   8      361.02 22.1597   6.9675   6.9710   7.4328  90.000  90.000  90.000       2      89  700008 CRISTOBALIT
0.0027   8      368.59 21.7045   8.9135   4.8747   8.4830  90.000  90.000  90.000    7762    1514  440026 PCOD 710001
0.0032  12      450.62 26.6301   4.9265   8.6867  10.5298  90.000  90.000  90.000     153    1782  200048 QUARTZ     
0.0033  32     1945.19 16.4508  13.8758  10.1564  13.8026  90.000  90.000  90.000    1501    1406  220103 GIS        
0.0034   8      428.02 18.6906   5.0921  10.1408   8.2890  90.000  90.000  90.000     534    1201  520012 ABW        
0.0034  12      494.19 24.2820   5.0553  10.8859   8.9802  90.000  90.000  90.000     426    1602  450029 PCOD 740021
0.0034   8      335.84 23.8210   8.0582   5.0521   8.2494  90.000  90.000  90.000      65     385  260004 TRIDYMITE  
0.0040  16      596.60 26.8187   8.9009  14.2210   4.7132  90.000  90.000  90.000      52     223  430017 PCOD 700002
0.0041  16      836.90 19.1182   9.5155   9.9582   8.8321  90.000  90.000  90.000    1491     614  360019 PCOD 630014
0.0042  12      648.68 18.4991   8.6590   8.6618   8.6488  90.000  90.000  90.000    3902    1755  230051 SOD        
0.0045  12      587.34 20.4310   7.1288  11.7931   6.9862  90.000  90.000  90.000    3624    1082  200023 PCOD 200010
0.0046  16      737.58 21.6924   9.0896   9.5722   8.4773  90.000  90.000  90.000      51    1919  360053 PCOD 640003
0.0047  12      564.22 21.2682   7.3123  15.6050   4.9446  90.000  90.000  90.000    4281     231  630004 BIK        
0.0048  16      714.58 22.3909   5.2489  14.9206   9.1242  90.000  90.000  90.000     281      42  690001 PCOD 720011
0.0048  16      695.25 23.0132  15.8401   8.8016   4.9868  90.000  90.000  90.000    2553    1155  400013 PCOD 400005
0.0049  14      707.83 19.7789   6.3038   7.2929  15.3965  90.000  90.000  90.000    3582     429  230010 PCOD 230010
0.0051  12      559.51 21.4472   8.9773  10.3248   6.0365  90.000  90.000  90.000      72    1654  670004 PCOD 670003
0.0055  10      608.49 16.4342   9.7613   9.7443   6.3973  90.000  90.000  90.000   85191     993  180002 EDI        
0.0056   8      319.50 25.0391   5.2417   4.8038  12.6887  90.000  90.000  90.000      28    1246  240044 PCOD 240044
0.0058  40     1870.96 21.3793  10.3170  13.1406  13.8006  90.000  90.000  90.000     726     712  730001 PCOD 730001
0.0058  10      376.79 26.5401   6.4514   4.6984  12.4307  90.000  90.000  90.000    7235     521  230014 PCOD 230014
0.0059  20     1227.67 16.2910  13.7750  13.8734   6.4240  90.000  90.000  90.000   50834     744  240029 NAT        
0.0059  12      580.57 20.6694  13.0594   9.1383   4.8648  90.000  90.000  90.000    2122      62  200004 PCOD 200004
0.0060   8      306.70 26.0838   8.9958   7.2965   4.6726  90.000  90.000  90.000      28    1553  200042 PCOD 200002
0.0061  20      915.21 21.8530  14.3208   5.1496  12.4102  90.000  90.000  90.000   20941    1438  670002 PCOD 670002
0.0061  16      758.36 21.0981   8.7498  16.8682   5.1382  90.000  90.000  90.000     100    1557  360046 PCOD 360031
0.0063  20     1090.25 18.3444   9.5394  16.5853   6.8910  90.000  90.000  90.000  127096    1732  550007 PCOD 550007
0.0064  10      429.96 23.2580   8.6555   4.5462  10.9267  90.000  90.000  90.000    2032    1259  230038 PCOD 230002
0.0067  24     1211.09 19.8169  16.0871   5.2582  14.3174  90.000  90.000  90.000    3848     305  690003 PCOD 690003
0.0067  12      597.91 20.0698   7.5333   5.0628  15.6770  90.000  90.000  90.000     179     127  570002 JBW        
0.0068  16      933.34 17.1428  10.1442   9.3739   9.8153  90.000  90.000  90.000   36415    1751  440032 ACO        
0.0069  24     1144.78 20.9646  11.4090  11.1425   9.0052  90.000  90.000  90.000    4846    1317  370006 PCOD 370003
0.0070   8      326.70 24.4872   8.5680   8.1240   4.6935  90.000  90.000  90.000     715    1090  330027 PCOD 330014
0.0070  28     1544.08 18.1338   8.2739  15.1687  12.3030  90.000  90.000  90.000    4318    1755  720013 PCOD 720013
0.0071   6      323.33 18.5567   7.8229   5.2452   7.8800  90.000  90.000  90.000   44943     301  250001 PCOD 250001
0.0072  32     1588.64 20.1430   8.1375  13.9744  13.9703  90.000  90.000  90.000    1929     197  360008 PCOD 450025
0.0073  12      622.96 19.2629   5.1841  11.1652  10.7627  90.000  90.000  90.000    1733     818  190026 PCOD 620004
0.0074  20      807.61 24.7643   6.7322  13.1351   9.1330  90.000  90.000  90.000    1616     689  450013 PCOD 450013
0.0075  24     1168.17 20.5449  14.9664   8.7792   8.8907  90.000  90.000  90.000   18663    1669  440029 PCOD 460031
0.0076  16      734.95 21.7702   8.6236   5.0654  16.8249  90.000  90.000  90.000   17121     453  310008 PCOD 310008
0.0076  10      342.44 29.2024  12.3922   6.1469   4.4955  90.000  90.000  90.000       2     199  230005 PCOD 230005
0.0076  32     1756.93 18.2136  14.3148  14.3022   8.5816  90.000  90.000  90.000    5657    1296  450026 PCOD 450026
0.0081  12      521.25 23.0216  10.9829   9.1328   5.1967  90.000  90.000  90.000      80     803  390002 PCOD 390002
0.0082   9      501.64 17.9413   7.9015   7.8104   8.1285  90.000  90.000  90.000  147639    1351  250003 PCOD 250003
0.0083  16      824.98 19.3943   9.2286  13.2011   6.7717  90.000  90.000  90.000  165260     413  190017 PCOD 190017
0.0083  12      468.51 25.6133   6.6836  10.0733   6.9588  90.000  90.000  90.000    5405    1482  600010 PCOD 600010
0.0084  20     1169.31 17.1041   5.1393  12.5137  18.1818  90.000  90.000  90.000  177422    1479  190057 PCOD 190057
0.0085  16      789.34 20.2700  12.4578  12.0723   5.2485  90.000  90.000  90.000      49     178  400003 PCOD 400003
0.0085  16      829.08 19.2985   9.5249   6.6388  13.1115  90.000  90.000  90.000    8163     499  290005 PCOD 290005
0.0086  16      818.66 19.5442   8.9983  17.2871   5.2628  90.000  90.000  90.000    2228     659  360022 PCOD 360022
0.0087  16      663.31 24.1214  12.7331   4.8813  10.6720  90.000  90.000  90.000     109     415  460018 PCOD 460018
0.0090  20      916.61 21.8196  13.6697   6.6169  10.1337  90.000  90.000  90.000    1796    1276  560003 PCOD 560003
0.0090  12      498.71 24.0621   7.5668  13.2404   4.9778  90.000  90.000  90.000    9810     526  190019 PCOD 190019
0.0090  16      641.74 24.9322  11.5305   5.0206  11.0856  90.000  90.000  90.000   19313     974  260007 PCOD 260007
0.0092  12      491.74 24.4030   5.1404   9.8008   9.7606  90.000  90.000  90.000    3225     201  190010 PCOD 190010
0.0092   8      570.95 14.0117  13.1227   5.3287   8.1650  90.000  90.000  90.000   77244    1411  390005 PCOD 390005
0.0093  20      976.55 20.4802  11.7140  12.3496   6.7505  90.000  90.000  90.000   61708    1914  560004 PCOD 560004
0.0094  16      664.25 24.0874   8.8854   5.9774  12.5067  90.000  90.000  90.000    8482    1448  290016 PCOD 290016

0.0096  12      501.60 23.9236   5.2949   6.8507  13.8281  90.000  90.000  90.000       6     167  540001 PCOD 730002
0.0097  32     1634.85 19.5737  10.4747   9.2333  16.9035  90.000  90.000  90.000    6312    1585  700162 MON        
 
 14-Apr-2006      3 hour  9 min 26 Sec 
 
  Total CPU time elapsed in seconds :    31422.70    
 
  Total number of runs     :       118059
  Number of structure candidates  :        94963
  Number of them with Rdt0 <   0.1400000     is :        45625
  Number of them with Rdt0 <   0.1300000     is :        33563
  Number of them with Rdt0 <   0.1200000     is :        23689
  Number of them with Rdt0 <   0.1100000     is :        14837
  Number of them with Rdt0 <   0.1000000     is :         7815
  Successful optimizations :         1484
  Unique proposals         :           60

CONNECT produces also a new .con file with exclusively the best models.

CONNECT will help you to make a global analysis of a total list of CIF files (the lists studied here was corrsponding to a small time calculation restricted to orthorhombic space groups).

Satellite software allowing to read a multiple CIF and to provide a list (in a file with .out extension) of models ordered according to the framework densities (from the smallest to the largest). 

The example inside of framdens.zip is the same total.cif file as above.
Note that this version is only working for oxides, not for flourides or anything else (because the way to count the number of cations is by detecting the first O atom appearing in the list of atomic coordinates...). Moreover, it will not work also if the cation is Os... A part of the output file is below. Includes are the formula, the PCOD number the R factor, the volume and finally the framework density (FD) : 
 

   Models ordered according to FD
SiO2           PCOD4500040 R =  0.0092         570.95    14.01
SiO2           PCOD4500004 R =  0.0084        1169.31    17.10
SiO2           PCOD4500066 R =  0.0072         893.65    17.90
SiO2           PCOD4500020 R =  0.0082         501.64    17.94
SiO2           PCOD4500085 R =  0.0070        1544.08    18.13
SiO2           PCOD4500029 R =  0.0065         879.87    18.18
SiO2           PCOD4500055 R =  0.0076        1756.93    18.21
SiO2           PCOD4500061 R =  0.0063        1090.25    18.34
SiO2           PCOD4500019 R =  0.0071         323.33    18.56

From and input file name.dat, GRINSP and GRINS create several output files :

name.imp : summarizes all the results (including FD, the framework density)
name.con : contains the coordination sequences  (CS) of all predicted unique final models
filename.dat : atomic coordinates in STRUPLO/STRUVIR format, allowing a direct look
at the structure in VRML
filename.cif : atomic coordinates in the IUCr CIF format 

name will be the same as the starting name.dat file 
         (i.e. if you start with the parameters file SiO2.dat, then you will obtain SiO2.imp and 
          SiO2.con)
filename is determined by the space group number
         working with SG N°230, you will obtain filenames starting at 2300001.cif and .dat, and next.
         working with SG N°1, you will obtain filenames starting at 10001.cif and .dat, and next.

General prediction

If the aim is to explore the possible crystal structures showing corner-sharing in M₂X₃ (triangles) , MX₂ (tetrahedra, square plane), M₂X₅ (square pyramids or triangular bipyramids) MX₃ (octahedra, trigonal prisms, pentagonal pyramids) or mixte compounds M_aM'_bX_c, then it is recommended to make a complete search (SG: 1 to 230), going from high symmetry to low symmetry. The problem is that this will need long calculation times if you wish to attain some 10.000 different cells testes per space group. So that you may decompose the problem into parts. For instance making 10 times calculations for 1000 different cells for the 6 symmetries apart :
Cubic : 195-230
Hexagonal/trigonal/rhombohedral : 143-194
Tetragonal : 75-142
Orthorhombic : 16-74
Monoclinic : 3-15
Triclinic : 1-2

Anyway, this will need close to 230 days, so that you may choose as well to make a space group per day...

Finally you will have to gather all results and perform a global analysis by using the satellite programs (CRYS2CON, CONNECT, FRAMDENS). 

Or you may keep all models (several possibilities of cells and space group per structure type), if you wish.

Using GRINSP as a structure determination tool for zeolites or MX₃ compounds or others

If the aim is to explore a small range of cell parameters and if the symmetry is known, this is even more simple. Restrain the search to the known cell parameters (allow a +- 1 Angstrom tolerance). 

Two-dimensionnal models

GRINSP may generate two-dimensionnal structures as well as the three-dimensionnal ones, in some cases (with triangles and tetrahedra). These structures were not included into the PCOD, with one exception for a tubular B₂O₃ predicted model. The problem with these two-dimensionnal models is that GRINSP has no way to estimate the correct distance between the layers or the rods.

Parameters having large influence on the speed of calculations:
These parameters are mainly nmax, the max cell parameters, nruns, genmax, Rmax, Rdt0 and idls :

Zeolites SG: 127-133     : text for this run
127 133                  : SG : Space groups range (two values between 1 230)
1 0 1 192                : npol, ncon, nmim, nmax
4                        : ncpol (coordination for each polyhedron-type)
Si  O                    : elements for a search in the distgrinsp.txt file
3. 30. 3. 30. 3. 30.     : min and max cell parameters a, b, c 
5. 35.                   : min and max framework density (FD)
200 30000 0.02 0.20      : nruns, genmax, Rmax, Rdt0
20000 1                  : idls (MC cycles for distance refinement) and iref (1 or 0)
-1                       : code for selecting models to output

a- Searching essentially for EDI
with small cell range : 6. 8. 6. 8. 5.5 7.5 and nmax=20

nruns genmax Rmax Rdt0 idls  total time  candidates candidates   successful    unique
                             in seconds   (total)   Rdt < Rdt0  optimization   models
200   300000 0.02 0.20 20000     252       188         174           45          6
200   300000 0.02 0.15 20000     139       194         108           45          5
200   300000 0.02 0.10 20000      51       188          29           20          5
200    30000 0.02 0.10 20000      48       187          38           21          4
200    30000 0.02 0.10  6000      17       192          35           20          5

b- Searching all solutions in P-4m2
with larger cell range : 4 16 4 16 4 16 and nmax=64

nruns genmax Rmax Rdt0 idls  total time  candidates candidates   successful    unique
                             in seconds   (total)   Rdt < Rdt0  optimization   models
20000  30000 0.02 0.10  6000    1101       17355      1672          636         13
20000 300000 0.02 0.10  6000    1246       19132      1723          640         15
20000 300000 0.02 0.12  6000    2711       19093      3567         1031         22

c- with even larger cell range : 4 22 4 22 4 22 and nmax=192

nruns genmax Rmax Rdt0 idls  total time  candidates candidates   successful    unique
                             in seconds   (total)   Rdt < Rdt0  optimization   models
20000 300000 0.02 0.12  6000    2570       19396      2294          556         19

Conclusions : 
- Clearly, changing Rdt0 from 0.20 to 0.10 can save time in the ratio 5 to 1, without 
changing a lot the number of final unique models (in all cases, the EDI model was 
retrieved). 
- Decreasing genmasx from 300000 to 30000 has not a great influence (gain of less 
than 10% in time).
- All this means that the optimization second step is the more consuming time.
Discarding those rough models with large Rdt0 save much time, apparently without
reducing seriously the number of unique final models, since their optimization is
unsuccessful, anyway. However, Rdt=0.12 or 0.13 is probably a better choice than 0.10
which would lead to discard many models, as shown by the second series of tests above,
in spite of more than twice a longer time at Rdt0=0.12 than at 0.10.

Problems with the .dat file for being read by STRUVIR for building VRML files (the limit for one atom-type in STRUVIR being 99, you have to rename those in excess and change their numbers..., quite boring, but the whole prediction stuff is boring...). 

Sometimes the formula is not correct in the CIF.

The Coordination Sequence (CS) calculation in GRINSP is not absolutely infallible (but works much better for version 2.00 than for version 1.00)... 

Using a Rmax up to 0.05 or 0.10 or larger may make appear some strange structures and even some not so strange (trigonal prisms instead of octahedra, etc). 

Polyhedra with 5 vertices can be either square pyramids or triangular bipyramids. The latter cannot correspond to all M-X distances equal, and therefore cannot lead to low R values (requiring all distances equal to the ideal one).

Contact the author if you find bugs or have other 'problems' than the above ones... 

The final structure is still described in the P 1 space group, but you will find in the .imp file the general or special positions occupied by the cations in the selected working space group before the final refinement made in the P 1 space group. Apply PLATON there for checking the true symmetry, and CRYSCON for transforming the symmetry from P 1 to the real symmetry.