Harald MartensSend a message to

Harald Martens

Harald Martens

Friends, chemometricians: Svante was such a good man, - knowledgeable, smart, warm, kind and fun. At the sad news of his death, it feels as if time stops for a moment. Many of us chemometricians have fond personal memories of him. In the hectic years 1980-83 he and I had almost daily telephone conversations, about algorithms and models for multivariate data modelling, and about science in general. It was an inspiration to talk with him, as well as with his father Herman. In the 70ies Svante’s method focus had been mostly on qualitative classification / SIMCA modelling, if I have understood him correctly. My focus had instead been on unmixing and quantitative multivariate calibration of instruments. But once we met – it must have been around 1978 - we quickly discovered that as chemists, we had ended up with many of the same mental and algebraic models. Svante and I had only two serious fights in the beginning of the 80ies: 1) In 1982, when we at the Norwegian Food Research Institute almost had to terminate our Food Research and Data Analysis conference because anarchist Svante did not want to understand the strict Norwegian rules about consuming private drinkables in the hotel. 2) In 1983, when he and I tried to agree on a joint chemometric notation and terminology (at Bruce Kowalski’s meeting in Cosenza, outdoors, with wine). In the end we did - almost. After that, our contact remained pure joy, at least for me. Like Barry Wise, I was helped by the fact that Svante was generous wrt supporting others, even potentially competing software developers. For instance, he let me work with a copy of his original BASIC version for the SIMCA software, to see if I could include also multivariate calibration there. I really did my best. The bilinear modelling was easy to implement, even though the BASIC programming language was a bit limiting. But the cross validation gave me problems: His interactive component-by-component cross validation was formally correct and easy to use, but difficult to apply for the linearity optimization and RMSEP-assessments that I needed. So I had to go back to my old Fortran code for Unmixing with PCA, with the cross-validation that I learnt from my other statistics friends, particularly Tormod Næs. This code later became The Unscrambler, which over the years has been close to my heart, although I was never involved commercially. Svante – along with Bruce Kowalski, Luc Massart, Edmund Malinowski and other pioneers – set the tone for the Chemometrics culture: Real-world oriented, multifaceted, both academic and business-oriented, relaxed, friendly and fun. Thanks to these founders of Chemometrics, our field developed good methods, and never became academically snobbish or myoptic. As an older and wiser philosopher of science, Svante has been an inspiration for me personally. What he once presented as his First Law of Science (as I recollect it), still largely applies: “There is an inverse relationship between how deeply a scientific field is rooted in mathematics, and how much fun they have at their conference dinners”. I think this reflected his opposition against the overly theoretical, self-conscious ethos in classical sciences. Many chemometricians have met that scholastic ghost. On the other hand, I also remember that he jokingly formulated another good sentence – something like: “Anything that cannot be solved on an 8-bit computer is not worth academic investigation”. Of course, he did not mean that literally. I think this was his reaction against the tendency of REPLACING thinking and trying to understand, BY fancy, massive computing – already in the 80ies. This is valid today, more than ever. The news of his death is emotionally painful, for me and surely for many others. We chemometricians shall miss him. My warm thoughts now go to his family, near friends and former colleagues. Trondheim, Norway Jan 10, 2022 Harald Martens

More...