Experimental data
--Nexp
Number of experiments (mandatory, even if there is only one)
Data file structure
For each experiment indexed by
k are requested 3 files
--FileInjk injection
profile
at the input of the column
--FileInik initial state
of
the column
--FileChromk observation at
the
output of the column
The file format is
t
1 c
1
...
...
t
N c
N
where t
i can be a space or time value, according to the
case,
and c
i is a *vector* of concentrations of size M if one has
M
components.
Outputs
--InterGraph
Graphical output, interactive graphics :
0 none
1 interactive graphics during computations. Makes use of
gnuplot,
and requires in the current directory a
ghost.out file, see
ghost.out.
--status
--status=
filename
The file
filename
contains a list of all parameters used by the currently used program,
including the default values.
Type of computation
--TypeTest
1 : Simulation test
2 : EO
optimization
3 : EO +
gradient optimization
4 : Gradient
optimization
5 : Gradient
comparison between direct and adjoint computations
Isotherms
--TypeIsotherm
Number
|
Name
|
Description
|
|
1
|
Langmuir
|
identification of Nstar and Kads
|
|
2
|
Langmuir
|
identification of Nstar, with constant
Kprim=Nstar*Kads
|
|
3
|
Langmuir
|
identification of Kads, with constant Nstar
|
|
7
|
Langmuir
|
identification of Nstar and Kprim=Nstar*Kads
|
|
4
|
Langmuir-Igor
|
identification of Nstar and Kads
|
|
5
|
Langmuir-Igor
|
identification of Nstar and Kprim
|
|
6
|
Langmuir-Igor
|
identification of Nstar_i and Kads
|
|
21
|
BiLangmuir
|
identification of Nstar_i, with constant Kads
|
|
22
|
BiLangmuir
|
identification of Kprim and Nstar_i
|
|
23
|
BiLangmuir |
identification of Nstar_i only
|
|
101
|
Lattice 1 comp |
identification on interaction
energies and Nstar
|
|
1011
|
Lattice 1comp |
identification on
exp(-beta*energies) and Nstar
|
|
102
|
Lattice 2 comp |
identification on interaction
energies and Nstar |
|
1022
|
Lattice 2 comp |
identification on
exp(-beta*energies) and Nstar |
|
103
|
Lattice 3 comp |
identification on interaction
energies and Nstar |
|
1033
|
Lattice 3 comp
|
identification on
exp(-beta*energies) and Nstar |
|
--Nstar
Must always be provided
--Kads, --Kprim
Kads vs Kprim : when identification with respect to kprim (resp. kads),
provide the values for Kprim (resp. Kads).
1
|
--Nstar=x
--Kads=(k1,...,kM)
|
x is a number
ki are numbers, i=1...M
M is the number of components
|
2
|
--Nstar=x
--Kprim=(k1,...,kM)
|
parentheses are mandatory, even if M=1, e.g.
--Kads=(0.034)
|
3
|
--Nstar=x
--Kads=(k1,...,kM)
|
|
7
|
--Nstar=x
--Kprim=(k1,...,kM)
|
|
4
|
--Nstar=(x1,...,xM)
--Kads=(k1,...,kM)
|
xi are numbers, i=1...M
|
5
|
--Nstar=(x1,...,xM)
--Kprim=(k1,...,kM)
|
|
6
|
--Nstar=(x1,...,xM)
--Kads=(k1,...,kM)
|
|
21
|
--Nstar1=x1
--Kads1=(k1,...,kM)
--Nstar2=x2
--Kads2=(j1,...,jM)
|
ji are numbers, i=1...M
|
22
|
--Nstar1=x1
--Kads1=(k1,...,kM)
--Nstar2=x2
--Kads2=(j1,...,jM)
|
|
Deterministic optimization
(gradient-like methods)
--TypeMethod
Method of deterministic (non-eo) optimization
1: ConjugateGradient with Polak-Ribiere rules from coool
2: ConjugateGradient with Polak-Ribiere rules and modified stopping
criteria (on gradients)
3: ConjugateGradient with Fletcher-Reeves rules
4: Steepest descent - the cheapest one from computational point of view
--EpsGrad
Tolerance on the gradient value. Not used when
TypeMethod=1.
--Epsof
Tolerance on the objective function value.
--TypeLS
Method of line search inside the deterministic optimization
1: CubicLineSearch from coool
2: LineSearch with Armijo-Goldstein rules
3: LineSearch (mysterious)
4: ConstantLineSearch (fixed search step - test version)
5: LineSearch with Wolfe rules - seems to be rather efficient
--TypeGrad
case 0: SimulEDP
case 1: SimulEDPandAdjoint
case 2: SimulEDPandDerived
Useful parameters for stochastic
optimization (EO)
--initBounds, --objectBounds
bounds for EO optimization (MUST be provided when required by
optimization)
e.g. for 1, provide Kads and Nstar
for 5, provide only Nstar_i
Langmuir : kads (resp. kprim), nstar
BiLangmuir : kads_1 (resp.
kprim_1), nstar_1, kads_2 (resp. kprim_2), nstar_2
Langmuir-Igor : kads (resp.
Kprim), nstar_i
The following three parameters are particularly important when
dealing with parameters with
various ranges of variation, e.g. such as nstars and kads... This is
related to the intervals in initBounds and objectBounds.
--Isotropic=1
--Stdev=1
Putting these two parametres to 1 allow
- Isotropic self-adaptive mutation
- One self-adaptive stDev (Sigma) per variable (see below sigmaInit): a
vector of sigma's will be used (as many as parameters to be
indentified).
--sigmaInit=0.3%
Initial value for Sigma(s). You can put any value of your choice (0.3
is the default). The % sign (no space between the numerical value and
the % !) means that each value in the vector of Sigma's will be scaled
according to initBounds.
--pMut=1
Should always be 1, according to modern techniques ! See EO tutorial
for more informations.
--seed
If a value is given to this parameter, it is used as a germ for the
initialization of the random genetic algorithm. This allows to repeat
the same computation, particularly useful for benchmarks (
is it machine-independant?). If no value is set the
initialization is "actually" random.
