Intro
This library implements advanced derivative techniques to allow for implementation of EOS without any hand-written derivatives. The name TEQP comes from Templated Equation of State Package. A paper about teqp is published in Ind. Eng. Chem. Res.:
Why?
- Implementing an EOS is an error-prone and boring exercise. Automatic differentiation packages are a mature solution for calculating derivatives
- Algorithms that use the EOS can be implemented in a very generic way that is model-agnostic
-
teqp
is very fast because it is written in C++, see the profiling in the paper in Ind. Eng. Chem. Res.
So far the following EOS are implemented:
- cubic:
- van der Waals
- Peng-Robinson
- Soave-Redlich-Kwong
- SAFTish:
- PC-SAFT (+dipoles, +quadrupoles, but no association)
- SAFT-VR-Mie (mixtures, but no association)
- cubic plus association (CPA) for pure fluids
- model fluids
- Exp-6 (modified Buckingham) from Kataoka
- square-well from Espíndola-Heredia et al.
- Two-center Lennard-Jones models (+dipoles, +quadrupoles)
- multiparameter
- multi-fluid model in the form of GERG
- ammonia+water model of Tillner-Roth and Friend
What is teqp not?:
- A feature-rich property library like NIST REFPROP. It is not intended to be, and other packages will be written that allow for the development of new algorithms. You are also welcome to write your own based upon
teqp
. - Written in a new and exciting programming language like Clapeyron.jl or FeOS. C++ is a mature and stable language that can be readily embedded into other environments or code bases via shared libraries.
Written by Ian Bell, NIST, with the help of colleagues and collaborators
Changelog
-
0.15.4 :
- Fixed a serious bug with get_Ar0n (https://github.com/usnistgov/teqp/issues/41)
-
0.15.3 :
- Fixed a serious bug in the calculation of virial coefficients and their temperature derivatives for SAFT models with a hard sphere term
-
0.15.2 :
- Fixed properly the kmat loading, and added additional testing
-
0.15.1 :
- Fixed evaluation of PC-SAFT at zero density (for calculation of virial coefficients)
- Fixed kmat reading when kmat is empty (in python shim functions)
-
0.15.0 :
- Added SAFT-VR-Mie model of Lafitte et al.; verified against calculations from Clapeyron.jl (see notebook in notebooks folder)
- Added L-J EOS from Kolafa-Nezbeda
- Added L-J EOS from Johnson
- Added Mie {11-13}-6 EOS of Pohl et al.
- Added the quadrupolar and dipolar contributions to the PC-SAFT model (for the hard chain variant only)
-
0.14.3 :
- Fix the ability to provide BIP and departure information as JSON-formatted string to build_multifluid_model
- Added dpsatdT_pure function to get dp/dT along vapor pressure curve
-
0.14.2 :
- Fix missing
get_meta
andset_meta
functions for cubics (should have been in 0.14.0) - Attach methods for multifluid mutants as well
- Fix missing
-
0.14.1 :
- Fix typo in loading of two-center LJF models, should be "2CLJF-Dipole" for "kind"
-
0.14.0 :
- Massive re-working of the c++ interface. There is now a "public interface" of teqp defined in the
teqp/cpp/teqpcpp.hpp
header, with the model contained in a private type-safe variant. This results in much faster incremental compiles in C++ (in general). - [Breaking]: Functions
get_Ar10iso
andget_Ar00iso
removed, superceded byget_Ar00
andget_Ar10
. Functionget_Psiriso
removed. - [Breaking]:
rho
keyword argument to isochoric derivatives replaced withrhovec
- [Breaking]:
convert_CoolProp_format
static method of IdealHelmholtz hoisted to free functionconvert_CoolProp_idealgas
- [Breaking]: Format of ideal-gas Helmholtz term is changed to include gas constant R everywhere
- Two-center Lennard-Jones models exposed to Python
-
get_kmat
method added to PC-SAFT and cubic models -
p_termination
andcrit_termination
are flags that can be set onTVLEOptions
for tracing of isothermal VLE. They default to very large and small values, respectively.
- Massive re-working of the c++ interface. There is now a "public interface" of teqp defined in the
-
0.13.0 :
-
Added 2-center Lennard-Jones models to the C++ code (thanks to Sven Pohl)
-
Added routine for converting the ideal-gas format of CoolProp. Exposed to Python as the
convert_CoolProp_format
method of theIdealHelmholtz
class
-
-
0.12.0 :
- Added Exp-6 (modified Buckingham) from Kataoka and square-well from Espíndola-Heredia et al.
-
0.11.0 :
-
Move all module-level free functions to bound methods of instances in Python
-
Update the C++ packaged interface of teqp to enable almost-zero-cost rebuilds of the exposed methods of C++ interface
-
Merge sphinx docs branch into main
-
Add the hardcoded Tillner-Roth and Friend model for ammonia + water (see AmmoniaWaterTillnerRoth class)
-
Add method for dp/dT along critical locus
-
Add method for pure-fluid endpoint solving for critical locus (opt-in)
-
-
0.10.0 :
-
Add isobar tracing for VLE of binary mixtures (exposed to Python)
-
Add
IdealHelmholtz
class for ideal-gas Helmholtz energy contribution (exposed to Python) -
Bugfix: Fix order of coefficients in one departure term. See f1da57. Had been wrong in CoolProp for many years.
-
-
0.9.5 :
- Bugfix: Fix the eigenvector orientiation as well when taking the temperature derivative in critical curve tracing. See b70178f7
-
0.9.4 :
- Expose the a and b parameters of cubic EOS. See 84ebc0fb
-
0.9.3 :
- Bugfix: Fixed stopping condition in
mix_VLE_Tx
(ifdx
was negative, automatic stop, missingabs
). See d87e91e
- Bugfix: Fixed stopping condition in
-
0.9.2 :
- Bugfix:
kmat
can be set also when specifyingsigma
ande/kB
with PC-SAFT
- Bugfix:
-
0.9.1 :
- Transcription error in a coefficient of PC-SAFT
-
0.9.0 :
-
Add ability to obtain ancillaries for multifluid model (
see teqp/models/multifluid_ancillaries.hpp
) or thebuild_ancillaries
method in python -
Enable ability to use multiprecision with PC-SAFT
-
-
0.8.1 :
- Replace the
get_Ar20
function that was erroneously removed
- Replace the
-
0.8.0 Significant changes include:
-
kij can be set for PC-SAFT and cubics (PR & SRK)
-
Added Lennard-Jones EOS from Thol et al.
-
Partial molar volume is now an available output
-
Added solver for pure fluid critical point
-
Added 2D Chebyshev departure function
-
Starting work on a C++ wrapper in the hopes of improving compile times for C++ integration
-
-
0.7.0 Significant changes include:
-
get_Arxy
generalized to allow for any derivative -
Local stability tests for critical points can be enabled
-
Critical curve polishers much more reliable
-
Add a method for dp/dT along isopleth of phase envelope of mixture.
-
Estimation is not enabled by default by the
estimation
flag. If that is desired, useforce-estimation
-
-
0.6.0 Add VLLE from VLE routine based upon https://pubs.acs.org/doi/abs/10.1021/acs.iecr.1c04703
-
0.5.0 Add VLE polishing routine 50b61af0, fix bug in critical curve tracing misalignment of eigenvectors f01ac7be, assorted issue fixes, especially support for javascript
-
0.4.0 Add VLE tracing, code coverage and valgrind testing, fugacity_coefficients, generalize loading of multifluid models
-
0.3.0 Add integration options to the tracing of the critical curve; fix
__version__
-
0.2.0 Add fluid files to the python package
Install
For all users, you should be able to install the most recent version from PYPI with
pip install teqp
Tests
Build (cmake based)
Be aware: compiling takes a while in release mode (multiple minutes per file in some cases) thanks to the use of generic typing in the models. Working on making this faster...
For example to build the critical line tracing example in visual studio, do:
mkdir build
cd build
cmake ..
cmake --build . --target multifluid_crit --config Release
Release\multifluid_crit
On linux/OSX, similar:
mkdir build
cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
cmake --build . --target multifluid_crit
./multifluid_crit
Random notes for future readers:
- When building in WSL via VS Code, you might need to enable metadata to avoid pages of configure errors in cmake: https://github.com/microsoft/WSL/issues/4257
- Debugging in WSL via VS Code (it really works!): https://code.visualstudio.com/docs/cpp/config-wsl