ipinv2d ipinv2d.inp

Format of the control file ipinv2d.inp:

IREST  NITER
CHIFACT
OBS_IP.DAT
DCINV2D.CON
FDMESH.DAT
TOPO.DAT
INI_MOD.CHG
REF_MOD.CHG
ALPHAS, ALPHAX, ALPHAZ
W.DAT

Control parameters:

irest

restarting control parameter:
0 - begin inversion from scratch
1 - restart inversion from the previous iteration. This requires that the files ipinv2d.chg, ipinv2d.out, and ipinv2d.log be present.

niter

number of iterations to be performed.

chifact

misfit parameter such that the final = N × chifact where N is the number of data. Usually chifact = 1.0 .
When a numerical value is input, the program will find the chargeability model which has a misfit = N × chifact. If the misfit plateaus at a higher value, the program will terminate and print advisory messages to both the screen and log file.

When NULL or null is input, the program will start with a very small value for chifact and adjust it to a higher value when the plateauing of data misfit is detected. The final misfit will be 20% higher than the estimated asymptotic value of the data misfit function.

obs_ip.dat

observed apparent chargeability data and electrode configuration.
If the error standard deviation is provided, the program will use the user-provided error estimates in the inversion. If the standard deviation is missing, the program assigns error estimates to the data and then proceeds to the inversion.

dcinv2d.con

conductivity file from the DC inversion. If NULL or null is entered, the inversion will be carried out using a uniform halfspace conductivity.

fdmesh.dat

finite difference mesh. This must be the same mesh file used by DCINV2D to produce dcinv2d.con. When NULL or null is entered, or ncell aspr are entered, a mesh will be generated automatically. However, the last two options can only be used if NULL or null is entered for the conductivity model (see above).

topo.dat

topography file. This must be the same file used in the DC resistivity inversion if dcinv2d.con is used as the background conductivity model.

ini_mod.chg

initial chargeability model. This is either a file name or a real value.

ref_mod.chg

reference chargeability model. This is either a file name or a real value.

alphas

coefficient for the smallest model component.

alphax

coefficient for the x-derivative term.

alphaz

coefficient for the z-derivative term.

There are two options for this input line:

1. The user supplies the values for the three coefficients and the program sets (_s , _x , _z) = (ALPHAS, ALPHAX, ALPHAZ).
2. When NULL or null is entered on this line, alphas is calculated based on array geometry and alphax and alphaz are set to 1.0 .

The program constructs a model using a standard model objective function and the three coefficients (_s , _x , _z) determine the relative importance of the three components in the model objective function. (See equations (8) and (9) )

w.dat

user supplied weighting functions. If NULL or null is entered, default values of unity are used.

If the IP inversion is performed using the conductivity model recovered from the DC inversion, the files fdmesh.dat and topo.dat must be identical to the ones used in the DC inversion.

Both ini_mod.chg and ref_mod.chg are stored in model.chg format. If either of the initial or reference conductivity model is constant, the file name can be replaced by the constant value in the parameter file.

Output files:

ipinv2d.chg

chargeability model of latest iteration. This file is overwritten at the end of each iteration.

ipinv2d.out

value of the objective function and the value of the ridge regression parameter as functions of iteration.

ipinv2d.log

log file containing detailed information about each iteration.

ipinv2d.pre

predicted apparent chargeability data from the inverted model in the latest iteration. The predicted data are in the format of obs_ip.dat with the field corresponding to the data error removed. This file is overwritten at the end of each iteration.

Examples of IPINV2D Control File:

The following is an example control file for IPINV2D. The inversion starts from scratch and the number of iterations is limited to 15. The chifact is set equal to unity, the initial and reference models are zero, and the default weighting matrix is used. The finite difference mesh is located in fdmesh.dat, and the observed chargeability data is in obs_ip.dat.

0  15            !!! restart, number of iterations
1.0              !!! chifact
obs_ip.dat       !!! observed potential data file
dcinv2d.con      !!! conductivity file
fdmesh.dat       !!! finite difference mesh
topo.dat         !!! topography file
0.0              !!! initial model (file name or numerical value)
0.0              !!! reference model (file name or num. value)
0.001 1. 1.      !!! alphas, alphax, alphaz (NULL for defaults)
null             !!! weighting matrix (NULL for default)

The following input file allows a preliminary inversion of IP data independent of the DC inversion:

0  15            !!! restart, number of iterations
null             !!! chifact
obs_ip.dat       !!! observed potential data file
null             !!! conductivity file
null             !!! finite difference mesh
topo.dat         !!! topography file
0.0              !!! initial model (file name or numerical value)
0.0              !!! reference model (file name or num. value)
null             !!! alphas, alphax, alphaz (NULL for defaults)
null             !!! weighting matrix (NULL for default)

NOTE-1: To have the program display a sample input file, type: ipinv2d -inp

NOTE-2: If a special weighting (w.dat) is supplied by the user, care should be taken so that the final weighting matrix, W_m^TW_m , constructed by the program is positive definite. This property can be destroyed when the combination of the cell weighting coefficients and component coefficients (ALPHAS, ALPHAX, ALPHAZ, see the description of special weighting w.dat) makes the smallest model component too small or even negative. Under such circumstances, the program will usually print in the log file the indices of the rows which are not diagonally dominant. In the most severe situation, the program will output an error message indicating that the matrix may not be positive definite. When these messages appear, the user should stop the program execution and redesign the special weighting.

NOTE-3: IPINV2D will terminate before the specified maximum number of iterations is reached if the expected data misfit is achieved and if the model norm has plateaued. However, if the program exits when the maximum number of iterations is reached, the file ipinv2d.out should be checked to see if the desired (usually equal to the number of data) has been reached and that the model norm has plateaued. If not, the inversion should be restarted. If the desired misfit level is not achieved, but the model norm has plateaued and the model is not changing between successive iterations, then the user may want to adjust the target misfit to a higher value. Also an investigation as to which data are most poorly fit can be informative. It may be that the assigned standard deviations to specific data are unrealistically small. The program restarts using the information in ipinv2d.out and ip2dinv.chg. Users are referred to Technical Note TN001 for more insight on these matters.

NOTE-4: We stress that the default errors are only an initial guess and they facilitate the preliminary inversion of the data. The user will want to alter these error estimates for the final inversion used for interpretation. The data with the estimated errors are written in the file ipinv2d.log using the format of obs.dat. They can be copied to a file for finer adjustment of the error estimates.

This is especially important with IPINV2D. The apparent chargeability pseudo-section is usually characterized by a fluctuating background of very low values that are dominated by noise. The estimated errors need to account for the variation in this background. Although our implementation of error estimation has been tested on available data sets and has yielded reasonable results, it can still underestimate the noise for an atypical set of data. If this occurs, the inversion could produce a model that has excessive structure when the observed data are overfit, or it could fail completely if the misfit is dominated by the contribution from data points in the background region. The former case can be diagnosed by observing excessively high amplitude of the recovered model, while the latter is indicated by an unreasonably large misfit. In these situations, the inversion must be rerun with increased error estimates.

NOTE-5: The default conductivity of a uniform halfspace for IP inversions should only be used for preliminary examination of the data. When there is little structure in the background conductivity, the inversion using this default mode can yield a reasonable chargeability model and it is justifiable to fit the data close to the expected misfit value. However, when the background conductivity deviates greatly from a uniform halfspace, reproducing the data to within the assumed errors will certainly result in overfitting the data. If the halfspace conductivity is assumed, then it is prudent to assign a value greater than 1.0 for chifact when the background conductivity is structurally complex. The judgement can be made based upon the complexity of the apparent resistivity pseudo-section.