The origins of Bayesian Optimization don’t lie in chemistry - they lie deep underground.
The origins of Bayesian Optimization don’t lie in chemistry - they lie deep underground.
In the 1950s, South African mining engineer Danie Krige was trying to solve a problem:
How can you estimate the distribution of ore in a mine when you can only afford to take a limited number of samples?
He pioneered a statistical method, later formalized by Georges Matheron in France, known as Kriging.
It allowed geologists to predict the value of unexplored locations based on sparse sampling data - a breakthrough for mining and oil drilling.
But Kriging didn’t stop there.
Statisticians realized that if you combine this kind of probabilistic model of the landscape with a strategy for choosing the next best sample, you get what we now call Bayesian Optimization.
The parallels to chemical R&D are striking:
- In mining, every new borehole costs millions.
- In chemistry, every experiment consumes time, materials, and labour costs.
- In both, the search space is vast - and testing everything is impossible.
That’s why Bayesian Optimization, born in the oil fields and mines of the 20th century, has become a powerful tool for chemical reaction and process optimization today.
At ReactWise, we’re bringing this methodology back to the lab - helping chemists explore their experimental “landscapes” with the same efficiency that once guided miners to ore and drillers to oil.
From Kriging in the mines to Bayesian Optimization in the lab, the story is the same - Smarter sampling, better decisions, faster discovery.
