notesum.ai

Published at November 15

Physics-informed neural networks need a physicist to be accurate: the case of mass and heat transport in Fischer-Tropsch catalyst particles

cs.LG

cs.AI

physics.chem-ph

physics.comp-ph

Released Date: November 15, 2024

Authors: Tymofii Nikolaienko¹, Harshil Patel², Aniruddha Panda², Subodh Madhav Joshi², Stanislav Jaso³, Kaushic Kalyanaraman²

Aff.: ¹SoftServe Inc., Lviv, Ukraine; ²Shell India Markets Pvt. Ltd., Bengaluru, India; ³Shell Global Solutions International B.V., Amsterdam, The Netherlands

Arxiv: http://arxiv.org/abs/2411.10048v1

Model	Absolute error ( $\|[S]^{\operatorname{true}}-[S]^{\operatorname{predicted}}\|$ )			Relative error ( $\left\|\frac{[S]^{\operatorname{true}}-[S]^{\operatorname{predicted}}}{[S]^{% \operatorname{true}}}\right\|$ )
Model	Mean	Median	Max	Mean	Median	Max
Proposed	2.64e-05	2.64e-05	3.93e-04	6.55e-04	6.55e-04	5.19e-03
Baseline ¹³	2.31e-05	2.31e-05	1.90e-03	6.83e-04	6.83e-04	1.94e-02