Science-Based HT Formulation

Here is a numerical argument1 for why Science-Based HT is necessary.

You want to create a 5-component system. For a quick optimization you need to check Hi and Lo levels of each component. But for each component you have 3 possible raw materials. With our HT system we can easily do a systematic run to get close to the optimum. Well, no! Look at the numbers


N Components
N Levels
N Raw (Average)
N Removed (Average)

The key point is that if we have some objective criteria for removing a few of the potential raw materials, this has a dramatic effect on the numbers. With the default values, we would need 3888 experiments which, frankly, we are not going to do. If on science-based grounds we can remove an average of 1 possible raw material (i.e. we have eliminated just 5 of them) then the number of experiments drops to 512.

How can we make such decisions? One much-used technique for removing no-hoper materials is Hansen Solubility Parameters. If the HSP Distance between one potential component and the range of components with which it needs to interact favourably is calculated to be large then there is no reasonable chance that they will be compatible. So we don't need to bother to test that component. Just removing 1 such component gives an average removal of 1/5=0.2 and (check with the slider) we have saved over 1000 tests!

An example of a tool that allows an exploration of HSP Distances across multiple components can be found on the HSP-3DO page.

Similarly, using HLD, if we have an emulsion system and happen to know the rough part of surfactant space we need, then any surfactant with a Cc taking us outside that range will be a no-hoper.

An example of a tool that allows an exploration of HLD emulsion effects across a formulation range can be found on the HLD-Tubes page.

We don't have to worry about being too precise with our science-based elimination of no-hopers. There is a risk that our HSP or HLD estimate is wrong and we have removed a component that would have given stellar performance. But there is a certainty that we will not run 3888 experiments. So we balance science-based risk versus practical certainty.

Not to be confused with DOE

The well-known tricks for reducing the number of experiments within a Design of Experiments run are not the concern of this page, or this site. DoE is powerful, when done at the right stage of development. But it is fully covered by many other apps, programs and sites so there is nothing new that I can contribute.

As it happens, I've come across a lot of useless DoE experiments where the problem is that they are trying not to optimize a specific formulation but to understand the science of a process. The majority of these are useless because they are phenomenological rather than scientific. This means that although they can identify that in their examples, "scientific" parameters such as viscosity or temperature have such and such an effect, these observations are not linked to a refutable theory, so the results cannot be widened or translated to other systems. So Science-Based DoE is something that is sadly lacking. Maybe SB-HT and SB-DoE aren't so far apart after all, with each being rarer than should be the case.


The equations are N=NLevels^(NComponents-1)*NRaw^(NComponents-NRemoved)