Testing component analysis of remanent magnetization curves with a series of synthetic mixtures: insights into the reliability of unmixing natural samples

Abstract:

Isothermal remanent magnetization (IRM) unmixing is an important quantitative method to reconstruct underlying magnetic components. There are two commonly applied unmixing approaches, i.e., model-distribution-based and endmember-based approaches. While both approaches are known to be subject to uncertainty, it remains difficult to evaluate the reliability of these methods due to the lack of an independent method to validate estimated results. For this purpose, we performed unmixing on a series of synthetic mixture samples of andesite powder and cultivated magnetotactic bacteria (Magnetospirillum gryphiswaldense strain MSR-1). Hysteresis loops, IRM acquisition curves, backfield curves, and first-order reversal curves (FORCs) were obtained. In our experiments, endmember-based unmixing can quantify the underlying endmembers better than the method using lognormal distributions. Moreover, the results by the endmember approach are consistent with that obtained from FORC analysis based on principal component analysis (FORC-PCA). However, when the datasets of samples or the variability of their magnetic properties are limited, the endmember model may fail to identify pure endmembers while the model-distribution-based unmixing may work. We demonstrated that by combining the two approaches it could provide improved unmixing results.