Critical Assessment of Data using MTDATA
DATA ASSESSMENT COURSE
The aim of the two day course is be to introduce
to those attending the techniques involved in the derivation of
thermodynamic parameters for use in phase equilibrium calculations -
data assessment - and so give them the capability of building their own
databases or plugging gaps in existing databases. Particular emphasis
will be placed on the use of the ASSESSMENT module of MTDATA designed
to optimise the values of chosen thermodynamic parameters to give the
best possible agreement with experimental data.
The course covers:
The types of
experimental data required.
The typical steps taken during data assessment.
Data assessment tools available within MTDATA.
The use of the ASSESSMENT module of MTDATA.
Hands-on experience of data assessment.
ASSESSMENT module handbook and documented
No previous knowledge of data assessment is
necessary to attend the course but some familiarity with the basic
MTDATA commands used to set up phase equilibrium calculations is
If you are interested in reserving a place (or
places) on the course in September, or if you would like further
information about MTDATA courses, please contact John Gisby either by
or telephone (+44 (0)20 8943 7098).
Back to contents
DATA ASSESSMENT CLUB
At the MTDATA data assessment club meeting /
workshop, attended by existing and prospective users of the ASSESSMENT
module of MTDATA, NPL staff introduced new features of the ASSESSMENT
module and offered guidance in its use, illustrated by examples
appropriate to oxide and aqueous systems. Malcolm Rand of WintersHill
Consultancy provided a user's perspective, describing data assessment
work he had performed using MTDATA. A lively discussion followed,
featuring questions about the ASSESSMENT module and suggestions for
future enhancements. The possibility of setting up an MTDATA data
assessment club was also considered.
An informal MTDATA data assessment club, along the
lines of the MTDATA users' group, was generally held to be a good idea,
allowing users of the ASSESSMENT module to share ideas and experiences
with each other and with NPL. Annual meetings supplemented by a
newsletter or a bulletin board were recommended by those present.
Back to contents
QUESTIONS AND ANSWERS
The following are some of the questions asked and
suggestions made at the MTDATA data assessment club meeting / workshop
at NPL on 30th November 2000 with responses from NPL.
||Should the default
optimisation algorithm in MTDATA be changed to UPOLYT ?
- In general UPOLYT works best when
parameters are far from their final values and LSQNDN (Ellis) and
LCQNDF (Elsie) work best when parameters are close to their final
values. Automatic scaling of parameters within LSQNDN should eventually
perform a similar job to UPOLYT so giving good performance in both
regimes. LSQNDN will then become the default algorithm. It should be
noted that there is a restriction on the number of experimental data
items accepted by LSQNDN, currently 1000. No such limits apply with
UPOLYT except when examining results graphically.
||Can phase names be
supplied as subscripts to keywords in macros?
- Not at present. The keyword plus
subscript needs to be supplied, for example
- where the macro class.mac contains the
classify absent ********
||Can I plot experimental
property against calculated value in the ASSESSMENT module?
- Yes. In version 4.72 of MTDATA using the
plot residuals and plot correlations commands. See NEW FEATURE - CORRELATION AND
RESIDUAL PLOTS for more details and examples.
||I would like to be able
to identify experimental points by property or by reference
||What is the recommended
syntax for labelling experimental points?
- A facility for identifying experimental
points by property or reference was introduced after the initial
ASSESSMENT module handbook had been released. It is documented in the
on-line help of MTDATA version 4.72. The recommended syntax for labels
- Further details are given in the HINTS AND TIPS
||Do you plan to allow
component amounts to be set from user variables?
- Yes. In version 4.73 of MTDATA users will
be able to set component amounts and component masses from user
variables by means of commands like
set w(1) us_var(mass_of_Cu) !
||Do you plan to provide a
graphical user interface for entering experimental information?
- Other parts of MTDATA have a higher
priority for implementation within the graphical user interface at the
moment. Setting up experiments using standard MTDATA commands typed
into a text file allows a wide range of different types of experiment
to be considered. This flexibility could not be reproduced easily
within a simple GUI.
||I would like "site
fraction" to be available as an experimental property.
||I would like phase
boundary searches with composition as a variable to be available within
the ASSESSMENT module in the same way as phase boundary searches with
temperature as a variable.
||I would like "heat
capacity" and "entropy" to be available as experimental properties.
- All are being considered as future
enhancements to the ASSESSMENT module. Meanwhile, phase boundary
compositions can often be entered in experimental data files as
tie-line ends and heat capacities and entropies as combinations of
Gibbs energy and enthalpy values.
||Can I specify my own
value for the "penalty function"?
- The "penalty function" is the error
assigned to experiments where temporary changes in parameters during
the optimisation process make calculated results impossible to obtain.
Its default value of 1012 is designed to ensure
that such changes to parameters are quickly reversed. The value is
stored in a user variable with the name "penalty_function" so a new
value, for example zero, can be specified as follows
user_variable(penalty_function) 0 !
||Do you plan to write a
general function evaluator to supplement (or replace) the RATIO,
PRODUCT, DIFFERENCE, SUM, LOG10 and LN operators now available?
- This is a long term aim and preliminary
work has already started. Note that new EXP and SQUARE_ROOT operators
were added in version 4.72 of MTDATA.
||Can you arrange for the
list parameter command to list parameters in numeric order rather than
in the order they were selected for change?
- This will be done in version 4.73 of
MTDATA. It should make individual parameters easier to find in long
||Why do I get an error
message when trying to LINK the values of two parameters?
- The target parameter, that being linked
to, must have already been flagged as FIXED, OPTIMISED or STEPPED.
Specify the nature (or TYPE) of the target parameter first of all, then
link to it, and there should be no problem. For example
modify index 92 type
optimised ! modify index 98 link 92 !
parameters valid over different temperature ranges be LINKed together
- The LINKing of such parameters will be
automatic in version 4.73 of MTDATA.
||Is it possible to express
experimental uncertainties as percentages?
- This facility will also be available in
version 4.73 of MTDATA. The current plan is for a percentage
uncertainty to be entered as a negative number. An example is given
below in which the stated uncertainty on the measured activity value is
list property activity(Fe)
value 0.456 uncertainty -1 !
Back to contents
HINTS AND TIPS - LABELS OF
It is convenient to assign labels to individual
experiments being considered during an ASSESSMENT session, so enabling
them to be identified easily in graphical and tabular output. The
recommended label syntax is property-reference:condition,
where property indicates the
property determined experimentally, for example
indicates the source of experimental data, for example Smith or 92JON
and condition indicates conditions appropriate to the experiment
undertaken, for example 1473K, 10-5atm or w(Fe)=0.3.
When a diagram is produced using the PLOT command
of the ASSESSMENT module, comparing calculated and target data for
- the property
label is used to form the ordinate and in some cases abscissa
- (if the selected experiments share a common
- experiments with the same reference
and condition are indicated by
the same symbol
- points corresponding to different conditions
are joined up as separate lines in
plot value experiment
The diagram above compares calculated values of
the partial pressure of S2 above liquid
sulphides with experimental data from two references
indicated by BAL and MEY. The liquid sulphides in question contain a
series of different Fe to (Fe + Ni) ratios the values of which are used
as conditions in the experiment
labels. The actual experimental property
1/2 log p(S2) which appears in the
ordinate annotation. Note that each separate line refers to a different
Fe to (Fe + Ni) ratio or condition
and that each different reference
and condition combination has
been assigned a different symbol. Typical labels used were
log p(S2)-BAL:0.50 and
1/2 log p(S2)-MEY:0.20.
Back to contents
NEW FEATURE - CORRELATION AND RESIDUAL PLOTS
Users of the ASSESSMENT module of MTDATA (version
4.72) can now produce plots of calculated properties against
experimental values using the plot correlations
experiment ! command. An example of the type of
plot produced is shown below.
Different symbols are used for different
combinations of reference and property (see HINTS AND TIPS for the
recommended syntax for experiment labels). In the case of a perfect fit
all of the symbols plotted should lie along the line calculated value =
experimental value. This line is shown in red.
Using the new plot residuals
experiment ! command, diagrams can be produced in
which "errors" are plotted against a chosen abscissa variable. "Errors"
are defined as follows:
error = (calculated value -
experimental value) / experimental uncertainty
An example of this type of diagram can be seen
below. Different symbols are once more used to indicate different
combinations of reference and property. In the case of a perfect fit
all of the symbols should lie along the error = 0 line, shown in red.
Back to contents
For a current list of forthcoming events please
see the MTDATA News page.
Back to contents
If you would like further information about any of
the items featured in this newsletter, if you have comments about the
newsletter or if you would like to contribute an article to a future
newsletter, please contact:
NPL Materials Centre
National Physical Laboratory
TW11 0LW, UK
Tel: +44 (0)20 8943 7098
Fax: +44 (0)20 8614 0425
Back to contents
23 April 2010