top of page

Destination Mars: A Machine Learning Investigation of the Magnetosphere Loss Mystery

  • 6 days ago
  • 4 min read

Eolisa Space LLC today publicly releases "Destination Mars: A Machine Learning Investigation of the Magnetosphere Loss Mystery," a comprehensive, data-driven study examining the mechanisms behind the cessation of the Martian geodynamo and the collapse of the planet's global magnetic field.

The full study including an 11-page scientific manuscript, five Python analysis modules, eight publication-quality figures, and complete reproducibility documentation is available as an open-access data package under MIT license.

DOI: 10.5281/zenodo.19477330 Repository: Zenodo(https://doi.org/10.5281/zenodo.19477330)


THE QUESTION


Mars was not always the barren, irradiated desert we observe today. Geological evidence carved valley networks, layered sedimentary deposits, hydrated mineral signatures detected by orbital and surface instruments points to a past in which the planet maintained a thick atmosphere, liquid surface water, and potentially habitable conditions.

All of this required a functioning magnetosphere: a global magnetic shield generated by convective motion in the planet's liquid iron core. At some point between 3.7 and 3.5 billion years ago, that shield failed. Without it, the solar wind stripped the atmosphere. Surface pressure dropped. Liquid water became thermodynamically unstable. Mars transitioned from a world that might have sustained life to the cold, thin-aired planet it is today.

The question of why the magnetosphere collapsed remains one of the most consequential open problems in planetary science.


THE APPROACH

Rather than relying on qualitative argument or single-dataset analysis, Eolisa Space's Research Division applied machine learning methods across five publicly available NASA mission datasets:

  1. Mars Global Surveyor (MGS) magnetometer data — crustal magnetic field intensity evaluated on a global 5° × 5° grid (2,592 cells) using the Langlais et al. (2019) spherical harmonic model

  2. Mars Odyssey Gamma-Ray Spectrometer (GRS) — surface concentrations of thorium (0.17–0.97 ppm), potassium (0.09–0.57 wt%), and iron, mapped by Boynton et al. (2007)

  3. InSight seismometer — Mars interior structure including the 2025 discovery of a solid inner core (radius 613 ± 67 km, FeO-rich composition) by Durán et al. (Nature, 2025) and outer core constraints (1,799 ± 66 km) from Khan et al. (Nature, 2023)

  4. MAVEN — present-day atmospheric ion escape rates (~2–3 kg/s), calibrated against Jakosky et al. (2018)

  5. MOLA topography and the Robbins & Hynek (2012) impact crater database — 13 major impact basins cataloged with coordinates, diameters, depths, estimated impactor sizes, and impact energies

Five machine learning methods were deployed:

• K-Means clustering and DBSCAN for magnetic province identification

• Random Forest classification for demagnetization prediction

• Gradient Boosting regression for basin-level field prediction

• Principal Component Analysis for multivariate structure extraction

• Analytical radiogenic heating model with progressive core depletion


THE RESULTS

Result 1 — Two Magnetic Provinces K-Means clustering identifies K = 2 as optimal (silhouette S = 0.40), cleanly separating the magnetized southern highlands (mean |Br| = 134.3 nT) from the demagnetized northern lowlands (mean |Br| = 10.2 nT) — a 13.2× contrast.

Result 2 — Demagnetization Is Predictable A Random Forest classifier trained on 32 features achieves F1 = 0.9754 ± 0.003 (5-fold cross-validation) in predicting whether a surface cell is demagnetized. Basin distance features account for ~35% of total predictive importance. Thorium concentration ranks as the second most important feature (9.8%).

Result 3 — Thorium Anti-Correlates with Magnetism Surface thorium concentration and crustal magnetic field strength show an inverse relationship at the basin scale (Pearson r = −0.47). Regions enriched in radiogenic elements at the surface tend to have weaker residual magnetic fields — consistent with upward redistribution of heat-producing elements from depth.

Result 4 — A Dynamo Death Timeline The radiogenic heating model traces the progressive expulsion of thorium, uranium, and potassium from the Martian core following giant impact events: • 4.1 Gya: Hellas + Utopia impacts → ~15% core radiogenic loss • 3.9 Gya: Argyre + Isidis impacts → ~20% additional depletion • 3.8 Gya: Chryse and secondary impacts → ~10% further loss • 3.5 Gya: Cumulative depletion reaches ~50% The predicted dynamo cessation window (~3.7–3.5 Gya) is consistent with independent paleomagnetic constraints.

Result 5 — The Dominant Signal PCA reveals that the first principal component captures 54.2% of total variance, loading on the magnetic–thorium–potassium–latitude axis. The impact-driven magnetosphere destruction is not a secondary geological signal it is the primary one.


THE HYPOTHESIS

The data supports the following chain of causation:

Giant asteroids (55–165 km diameter) struck Mars between 4.1 and 3.8 billion years ago, forming the Hellas, Utopia, Argyre, and Isidis basins. Each impact delivered energy on the order of 10⁹ to 10¹⁰ megatons TNT sufficient to mechanically disrupt the core–mantle boundary. The resulting convective mixing expelled radiogenic elements from the liquid iron core into the mantle and crust. Without sufficient internal heating, core convection weakened, the geodynamo failed, the magnetosphere collapsed, and the solar wind stripped the atmosphere over the following billions of years.

The geochemical fingerprint of this process is preserved in the Mars Odyssey GRS thorium maps. The magnetic fingerprint is preserved in the MGS crustal field data. Machine learning reveals the statistical patterns connecting them.


OPEN SCIENCE

The complete study is publicly available:

• Manuscript: 11-page PDF with full methodology, results, and 15 peer-reviewed references

• Code: 5 Python modules (92 KB), fully documented

• Figures: 8 publication-quality scientific graphics

• Reproducibility: pip install -r requirements.txt && python code/run_full_analysis.py

• License: MIT (code), CC BY 4.0 (manuscript)

• DOI: 10.5281/zenodo.19477330

Any researcher, student, or institution can download the package and reproduce every result within minutes on a standard laptop.


CONTEXT

This is the second major public release from Eolisa Space in 2026, following the Sagittarius A* wormhole analysis (DOI: 10.5281/zenodo.18528732).

The Mars study extends the conceptual framework established in the original "Destination Mars: To Fathom Magnetosphere Loss Mystery" paper authored by Ahmad J. (Before Science Team Pioneer, Eolisa Space, February 2025). The present work provides the computational infrastructure to test that hypothesis against real NASA data.

Both studies reflect the same institutional principle: if you make a scientific claim, you release the data and the code. No exceptions.



Onur H. Evgin President, Eolisa Space LLC Office of the President

Phone: +1 (505) 381-8228

Address: 1209 Mountain Road Pl NE, Albuquerque, NM 87110, USA


 
 
bottom of page