Bringing reliability to physical property prediction
We intelligently use all available data and the best models to provide the most accurate property prediction and its uncertainty, so you know how reliable each prediction is
Problem: Garbage in, garbage out
Computer simulations are playing an increasingly important role in process and product development. But they rely fully on the accuracy of the physical property data used.
Experimental data is often unavailable or unreliable, so accurately predicting physical properties of molecules and mixtures is typically the most critical part of a good simulation.
Due to their ease of use and low computational cost, group contribution (e.g. UNIFAC) and QSPR methods are in widespread use despite their well-documented lack of accuracy across application areas.
Quantum chemical methods have gained traction but continue to be limited by the need for computational power and expert users. Also, no practically applicable quantum method consistently yields accurate predictions and they do not inform whether a specific prediction is reliable, even though errors can be very large.
Solution: Q-props™
Q-props™ is the first development in years to drastically improve the accuracy and reliability of physical property predictions.
Our novel scientific approach selects and qualifies all available data and applies state-of-the-art models such as quantum chemistry and molecular simulation in a fully automated, intelligent and transparent way.
Q-props™ always provides the best possible property prediction and a prediction-specific uncertainty, so you know how reliable each prediction is. Also, you get transparency on what data has been most critical in making a prediction and guidance on experiments that can reduce overall uncertainty most.
And Q-props™ is easy to use; it interfaces with leading process simulation tools, performs heavy calculations in the cloud and handles all expert choices automatically.
Example: Heat of formation for ethylbenzene and tert-butanol
The table below shows the heat of formation of ethylbenzene and tert-butanol as predicted using a group contribution method (Joback), quantum chemistry (G3MP2), and Q-props™, as well as the three experimental data points available from NIST. The graphs below the table shows the same data as well as visualizing how Q-props™ obtains its reliability measure:
| Method | Heat of formation, 25°C [kJ/mol] | |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
Ethylbenzene |
Tert-butanol |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Group Contribution |
28.1 |
-286.9 |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Quantum chemistry |
23.1 |
-313.3 |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
Q-props™ |
29.5 ± 2.9 |
-313.8 ± 3.2 |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
NIST webbook (3 data points exist for each compound) |
29.8 ± 0.8 49.0 ± 4.0 69.3 |
-312.6 ± 0.9 -313.0 ± 1.5 -309.7 |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Visualizations of the Q-props™ predictions using histograms:
(colored lines show the same data as the table above)
Ethylbenzene
Tert-butanol
What the example shows
Group contribution yields chemical accuracy (<1 kcal/mol deviation) in the case of ethylbenzene, but is far off in the case of tert-butanol. Conversely, the quantum chemical method (G3MP2) gives good results in the case of tert-butanol but is significantly off for ethylbenzene. Hence, “all models are wrong”.
However, Q-props™ gives an accurate prediction in both cases. And, notably, only Q-props™ gives an uncertainty estimate, informing how reliable each prediction is. Intelligently using available models and experimental data, Q-props™ gives consistently accurate and reliable predictions.
Q-props™ also gives a clear validation of what experimental data to trust – and what not to trust – when contradictory data exists.
(*Note: The experimental data is only shown to benchmark the different prediction methods and highlight how Q-props™ can be used to validate, and in-validate, experimental data. For the predictions above, Q-props™ used all the best experimental data available at the time, but of course not that of the compound being predicted.)
The results were produced with a proof-of-concept tool developed by Hafnium Labs in 2017. Since then, we have developed sophisticated tools based on our technology and built the world’s largest combined database of experimental data and millions of CPU-hours of simulation data.
Today, we cover a wide range of properties, achieve even higher prediction accuracy and better reliability measures. Our tools continuously improve with every model and experimental data point added – and you can integrate your own data as well.
Want to learn more about our tools?