Mars's Crustal and Volcanic Structure Explained by Southern Giant Impact and Resulting Mantle Depletion

Mars features a crustal dichotomy, with its southern hemisphere covered by a thicker basaltic crust than its northern hemisphere. Additionally, the planet displays geologically recent volcanism only in its low latitude regions. Previous giant impact models coupled with simulations of mantle convection have shown that the crustal dichotomy can be explained by post‐impact melt crystallization that emplaced a thick crust in the southern hemisphere. In this study, we show that the depleted residue left behind by the original post‐impact crustal formation can spread laterally, potentially persisting beneath the northern hemisphere to the present‐day. Such a large‐scale mantle province would concurrently explain both the prevalence of long‐term magmatism on Mars and its strong preference for localized equatorial regions.


Introduction
The origin of the martian dichotomy has been a long-standing problem since its discovery more than 50 years ago.Featuring a topographical contrast of 5.5 km between the southern highlands and the northern lowlands (Smith et al., 2001), the dichotomy manifests itself in both geological and geophysical observations, including crater distributions (Barlow, 1988;Robbins & Hynek, 2012) and crustal remnant magnetic fields (Acuña et al., 1999).Crustal thickness modeling suggests that the average northern and southern thicknesses are ≈37 km and ≈63 km, respectively (Wieczorek et al., 2022).There are a number of, often mutually exclusive models, claiming to reproduce Mars's crustal dichotomy.First, since impacts commonly strip away surficial material, it was long assumed that the northern basin had been generated by a giant impact (Marinova et al., 2008;Nimmo et al., 2008;Wilhelms & Squyres, 1984).Second, a giant impact in the south could have formed a hemispherical magma ocean, above which a very thick crust formed (Golabek et al., 2018;Leone et al., 2014;Reese et al., 2011).Third, a superplume could have produced enough melt to form the southern highland crust (Keller & Tackley, 2009;Roberts & Zhong, 2006;Zhong & Zuber, 2001), with Zhong (2008); Šrámek and Zhong (2012) further linking the formation and evolution of the Tharsis province to the migration of this dichotomy-forming superplume.More recently, studies have shown that a low-degree initial thermal perturbation would be amplified, the feedback of which could contribute to the formation of a crustal dichotomy without requiring an impact or a superplume (Bonnet Gibet et al., 2022;Watson et al., 2022).
All these models have their own advantages and limitations.Indeed, traditional crater-forming or crust-stripping impacts are commonly observed on all rocky planets.For craters of hundreds km size, the impact heating is often safely overlooked because it has negligible geological implications.In this case, impact melt production is often a result of extreme pressure changes relevant to impact shock waves, leaving behind a thin layer of impact melt within the impact crater.On the other hand, hydrodynamical simulation methods show that giant impacts can warm up the planetary mantles on the impacted hemisphere by thousands of degrees (Ballantyne et al., 2023;Nakajima et al., 2021;Yuan et al., 2023;Zhu et al., 2019), creating magma oceans of various extents.In quest of the dichotomy formation mechanism, collision simulations demonstrate that a northern giant impact, within a range of impact angles and velocities, can strip the right amount of primordial crust necessary to form the thin lowland crust (Marinova et al., 2008;Nimmo et al., 2008).However, these models either consider only the impact melt based on shock pressure, or neglect the subsequent evolution and crystallization product of the fully meltfilled impact site as well as surrounding locations.Considering the massive crust formation caused by magma ocean crystallization, it has been shown that a giant impact can lead to the emplacement of the crustal dichotomy (Golabek et al., 2011(Golabek et al., , 2018)).In the third hypothesis, degree-1 convection causes preferential crustal-production on one hemisphere, which results in the crustal dichotomy (Keller & Tackley, 2009;Šrámek & Zhong, 2012;Zhong & Zuber, 2001).Such a scenario, however, relies on a high Rayleigh number and a specific rheological stratification in the mid-Martian mantle to create the required plume within the expected dichotomy formation time of ∼400 million years (Frey, 2006).Often, the crustal thickness distribution produced by the mentioned upwelling contains an elongated ridge planform, instead of a hemispherical pattern, which is not observed on Mars (e.g., Figure 5c of Keller and Tackley (2009), and a number of cases in Šrámek and Zhong (2012)).
Geological investigations of Mars's surface show that both the exposed southern highlands and the buried northern basement rocks are Noachian in age (Edgar & Frey, 2008) (i.e., have been formed very early on in Mars's evolution).≈75 ancient volcanoes are mapped in the Tharsis Province and the circum-Hellas Volcanic Province and are thought to be associated with early Noachian activities (Xiao et al., 2012).In addition, ancient supervolcanoes, in the form of plains-style caldera complexes located in Arabia Terra (Michalski & Bleacher, 2013), are proposed to have produced 1,000-2,000 explosive eruptions between the mid-Noachian and the early Hesperian (Whelley et al., 2021), indicating that early Mars was more volcanically active than previously thought (Figure S6 in Supporting Information S1).The majority of the northern lowlands was later covered by sedimentary surficial deposits of Hesperian and Amazonian age (Tanaka et al., 2014).By the end of early Hesperian, the style of magmatism transitioned and has been restricted to a few localized volcanic edifices.Tharsis, being the largest and most active volcanic province on the planet, erupted most of its material in the Noachian to early Hesperian.Nonetheless, infrequent eruptions throughout the late Hesperian to Amazonian period persisted, and lava flows even as young as only 2 Ma have been discovered (Neukum et al., 2004).Similarly, while the surface age of the Elysium volcanic province spans more than 3 billion years, the youngest lavas were emplaced only a few million years ago in the Cerberus plains (Grott et al., 2013;Vaucher et al., 2009).The InSight mission revealed that Mars is still seismically active with some observed marsquakes, located mainly in low latitude regions near Cerberus Fossae, being associated with potential magmatic activity (Giardini et al., 2020;Stähler et al., 2022).
The spatial and temporal distribution of volcanic activity provides an opportunity to compare the crust production history with numerical models, and thus better constrain the dichotomy formation event and subsequent evolution of the planet.For this purpose, we use thermochemical mantle convection models to investigate the aftermath of a giant impact, and its relation to long-term volcanism and interior dynamics as a whole.

Mantle Convection Modeling
Thermochemical mantle convection is simulated using the code StagYY (Tackley, 2008) in two-dimensional spherical annulus geometry (Hernlund & Tackley, 2008).Using the parallel package PETSc (Balay et al., 1998), the code solves for conservation of energy, conservation of mass, and conservation of momentum (the Stokes equation) discretized using the finite difference approximation on a staggered grid.The computational domain is discretized using 1,024 (azimuthal) by 128 (radial) cells, with radial grid refinement near the top and the bottom boundaries.Tracers (markers) are used to track composition and other physical properties in the entire domain to avoid numerical diffusion.Tracers are advected using the fourth-order Runge-Kutta method with velocities that are interpolated from cell edges to tracer locations using second order, divergence-free interpolation.The computational domain is initially populated with an average of 30 tracers per cell.A free surface boundary condition is tracked inside the domain (Duretz et al., 2016), allowing for advection-based topography to form.Free slip boundary conditions are imposed at the core-mantle boundary.The core-mantle boundary temperature is initially set to 2400 K and evolves through time as heat is extracted from the core, assuming that it has a particular total heat capacity.The surface temperature is initially set to 210 K and evolves using a local radiative heat balance formulation (Lourenço, 2017), which allows it to be higher when the surface is molten.
Mantle composition is described as a mechanical mixture of "basalt" (mafic crust) and "harzburgite" (depleted mantle) end members.Based on the composition of the Martian crust (Taylor & McLennan, 2009) and primitive mantle, that is, bulk silicate Mars (Taylor, 2013), we choose the reference basalt fraction bs ref to be 0.35 (Text S1 in Supporting Information S1).Physical properties are assigned to materials using three phase systems, namely olivine, pyroxene-garnet, and melt.Each phase system except melt has phase changes with specified density jumps and reference temperatures and depths, as described and listed in Text S2 and Table S1 in Supporting Information S1.In our models, a pressure-, temperature-and melt-dependent viscosity, following a diffusion creep rheology, is used (Text S3 in Supporting Information S1).For simplicity, the effect of water on viscosity is omitted.
In each cell, when the melting temperature is exceeded, the melt fraction and cell temperature are updated toward thermodynamical equilibrium considering latent heat (taken to be 600 kJ/kg) consumption.Since basaltic melt is formed from the original rock composition, a more depleted residue (more harzburgitic in composition) is left behind, the melting temperature of which is increased accordingly.The melting temperature varies with composition in between three anchoring melting curves.While the solidus of our modeled martian primordial mantle (basalt fraction = 0.35) is taken from existing experimental work extended to 25 GPa (Duncan et al., 2018), the liquidus of that remains the same as the Earth's.The melting curve of the modeled martian crust (basalt fraction = 1) is set such that at 1 GPa, the melting temperature would be 50 K lower than that of "Earth crust.Such a 50 K temperature difference is consistent with that obtained from pMelts (Ghiorso et al., 2002) using the oxide compositions of martian crust and Earth MORB respectively.
Radioactive elements (Table S2 in Supporting Information S1) are partitioned during melting with a partition coefficient of 0.001.When the surface is solid, eruption or intrusion of melt is possible.For a melt fraction below the rheological threshold f rh , melt produced above the neutral-buoyancy depth of 600 km (Ohtani et al., 1995) is extracted from the cell, and is either erupted at the surface or intruded in the middle of any preexisting crust (Lourenço et al., 2020).The mass ratio between the erupted or intruded crust is governed by the "eruption efficiency," chosen to be 0.3 (similar to what is observed on Earth; Crisp, 1984).

Giant Impact as a Parametrized Thermal Anomaly
A parameterized giant impact (Monteux et al., 2007;Senshu et al., 2002) is imposed as a thermal anomaly at the start of our simulations.Since the aim of this paper focuses on the mantle dynamics and thermochemical aspect after an impact-induced regional magma ocean is formed, we consider a parameterized impact to be sufficient.The heat anomaly has a spherical geometry and is centered around a point located at a depth of half of the impactor radius.Its intensity depends on the radius of an isobaric core which depends on the impactor radius: where α ic = 3 1 3 ≈ 1.44.Within this radius, the temperature is increased due to the impact energy conversion: where β = v imp /v esc is the ratio between impactor velocity and escape velocity, γ imp = 0.3 is an efficiency factor of how much impactor energy is converted into heat, the fraction of energy in the isobaric core is a factor of 1/f imp of the total heat energy with f imp = 2.7, g is the surface gravity of the planet, R is the planetary radius, and c p is the specific heat capacity of the mantle.Outside this radius, the temperature anomaly is decreased with distance: where r a is the distance to the center of the anomaly and m = 4.4.This temperature anomaly is then superimposed onto the mantle adiabat.Away from the impact site (over 1.6 times the radius of the isobaric core) the initial temperature profile is constructed by the same mantle adiabat with initial boundary layers of 10 km at top and bottom of the domain.While we note that such a scaling law is best suited for smaller impacts, an approach like this has been used in several previous mantle convection studies to simulate a giant impact (Monteux et al., 2007;Roberts & Arkani-Hamed, 2012;Senshu et al., 2002;Watters et al., 2009).Moreover, the main goal of our study is to investigate the evolution and consequence of an impact-induced magma ocean, so this simplified method is sufficient.Here, we use an impactor radius of 1,250 km and impact velocity of 1.5 times the mutual escape velocity for our reference case.With the temperature exceeding the liquidus, we obtain a regional magma ocean filled with 100% melt as an extreme scenario.
In the case of high melt fraction (above the rheological threshold set to f = 0.35), the viscosity of silicate is expected to be on the order of 0.1-100 Pa⋅s, while that of the solid mantle is on the order of 10 21 Pa⋅s.A realistic model of magma dynamics would require solving the Navier-Stokes equation in a rotating system with a very high temporal and spatial resolution, especially right after the giant impact where a regional magma ocean is created.
Although this has been attempted, still using a viscosity much higher than expected for magma oceans (Maas & Hansen, 2019), simulations cannot yet be performed in realistic conditions for the complete cooling of a magma ocean using realistic parameters.Fortunately, parameterizations of rapid heat transport by turbulent flow have been proposed based on mixing length theory (Abe, 1997) and implemented in StagYY (Lourenço, 2017).In practice, this enhanced heat flux forces the radial temperature profile to follow a nearly adiabatic gradient in the regional magma ocean.As the magma ocean cools and crystallizes, that is, the melt fraction reduces to below the rheological threshold f rh , the turbulent heat flow treatment is progressively switched off (Text S5 in Supporting Information S1) and instead magmatism takes place, as described above.

Initial Crust and Depletion
An initial crust is present on the northern (non-impacted) hemisphere of our model.While the formation mechanism of an initial martian crust is unclear, we assume that it is not the final product of fractional crystallization of a whole mantle-depth magma ocean (because the heavy crust would have overturned under gravitational instability; Plesa et al., 2014), but instead formed from partial melting of the upper mantle.By choosing an initial temperature profile that crosses the solidus in the upper mantle, our code self-consistently computes the resultant melt fraction, which is then extracted to form crust in the first time step.This initial crust is enriched in heat producing elements (HPE), leaving behind a relatively depleted zone in the top 400 km (Figure S3a in Supporting Information S1), while the lower mantle retains the primitive HPE content.On the other hand, because a giant impact melts the mantle across all depths from the surface to the core-mantle boundary, the HPE originally in the lower mantle can be mixed into the magma pond.Therefore, depending on whether the primordial crust is excavated or remixed back into the magma pond at the impact site, the magma pond can have an HPE concentration of ≈0.7-1 times of that of the primitive mantle (Figure S3a in Supporting Information S1).
Detailed model parameters are listed in Table S3 in Supporting Information S1.

Results
Figure 1 illustrates the dynamics subsequent to the impact (see also Movie S1).As the impact-induced regional magma ocean solidifies and ceases within the first 4 million years, a thick crust is formed on top of a very voluminous layer of melt residue in the southern hemisphere (Figure 1b).The thermally and chemically buoyant depleted residue tends to rise and spread within the next 200 Myrs, driving a degree-1 flow pattern where the uppermost mantle migrates from the south to the north (Figure 1c).The northern lithosphere, being relatively undisturbed by the impact, remains undeformed and is underplated by the spreading depleted materials.
Dripping at the base of the southern crust recycles both crustal and depleted materials, stirs the mantle below and interrupts the degree-1 flow (Figure 1d).When the recycled materials reach the core-mantle boundary, mantle plumes are excited from the disturbed lower thermal boundary layer around the equatorial region, triggering another phase of crustal formation.Over a few hundred million years, the hot depleted material in the north cools down until its base destabilizes and partially mixes with the underlying mantle (Figure 1e), becoming neutrally buoyant (Figure 1f).By 500 Myr after the impact, the southern and northern hemispheres show a distinct difference in thermal and compositional structure, which is sustained throughout the remaining 4 Gyr of the model.
Due to a lack of heat producing elements, the melt residue allows the northern lithosphere to grow into a thick, reinforced and rigid thermo-compositional layer, which acts to insulate and thus heat up Mars's northern deep mantle.Equatorial magmatism occurs at the edge of this stable northern layer due to the combined effect of geometrical focusing of return flow generated by ambient sub-lithospheric downwellings and the slightly higher temperature of the northern mantle (Figure S2 in Supporting Information S1).
Figure 2a shows the latitudinal variation of crustal thickness and depleted layer thickness as a function of time.
For a better readability, crustal thickness above 120 km is shown in purple, while a depleted layer thickness (defined to be material having a harzburgite content >90%) above 150 km is shown in green.Initially, the thick crust in the south is only formed in the region ranging from the south pole up to ∼30°S latitude, which corresponds to the extent of the regional magma ocean.Within 10 s of Myrs the depleted material then spreads from the south to high-latitudes in the north (Figure 1c) and the low-latitude crust thickens on a timescale of 100 s of Myr (Figure 1d).Once emplaced, both the crustal dichotomy and the depleted layer are well preserved until presentday, with minimal lateral advection since 50 Myr after the impact event (see also Text S6 in Supporting Information S1).Figures 2b and 2c show the regional averages of crustal thickness and depleted layer thickness, respectively, as defined in Figure 2d.Not only does our model show that the thickness contrast between the southern and northern crust are established as early as the first few million years after the giant impact, but it also matches well with the current understanding based on geochemical evidence suggesting that the majority of crustal building occurs within the first billion years (Taylor & McLennan, 2009).The regionally averaged depleted layer thickness (Figure 2c) shows the accumulation in the north, and the loss of depleted material in the south, mainly due to spreading and recycling.Figure 2d shows the temporal evolution of melt production distribution, giving an insight into the magma source mechanisms for martian volcanism at different geological periods.After the formation of the dichotomous crust, we observe a migration of melt production from 30°S to 30°N within the first 500 Myrs of post-impact evolution.This trend can be compared to the initial emplacement of the Tharsis volcanic region, which first began in Thaumasia and then migrated northwards to the younger edifices like Olympus Mons.From 500 Myr to 1 Gyr after the giant impact, widespread melting occurs in the southern hemisphere as hot materials ascent following the stirring of the southern lower mantle.This may correspond to the magma sources of the numerous ancient volcanoes, proposed to be present in the southern highlands since the Noachian and part of the Hesperian periods (Broquet & Andrews-Hanna, 2023;Xiao et al., 2012).Starting from 1.5 Gyr, melt production decreases and mainly localizes at low latitude region until the present day (black boxes in Figure 2d).We interpret this to be the melt source for long-lived Martian volcanism in Tharsis and Elysium regions.Figure 2e shows the composition field and Figure 2f depicts the product of temperature and radially upward velocity, a convenient way to depict upand downwellings.Even though we do not observe active anchoring plumes that originate from the core-mantle boundary in our model, as hinted by recent studies (Broquet & Andrews-Hanna, 2022;Stähler et al., 2022), upwellings in a weaker and less stable form (e.g., Figure 1f, demonstrated by the mildly positive values in Figure 2f) are constantly seen rising from the mid-mantle on the 100s Myr time scale, including but not limited to the dichotomy boundary.With persistent scraping and entrainment of the edge of the reinforced lithosphere (Figures 2e and 2f and in Movie S1, near the dichotomy boundary), together with the magmatic intrusions and enriched radiogenic heating from the emplacement of major volcanic provinces, both the crust and lithosphere in the low latitude region become weakened and more susceptible to deformation.As hot and relatively enriched materials are delivered to the equatorial region from the northern mantle, the thinner lithosphere there allows for more decompression melting compared to other locations with a thicker lithosphere, causing more intrusion and weakening.This process provides a self-sustaining mechanism to produce melt at the low latitude region that continues from the Amazonian until geologically recent times, often in the presence of mild upwellings.It should be noted that, in Figure 2d, even though there is a stronger upwelling below the southern crust, no melt is formed because of a higher solidus resulting from a more depleted mantle, and a lesser extent of decompression melting in presence of a thicker lithosphere.In addition to Tharsis, several known volcanic regions are located close to the dichotomy boundary (within 30°of latitude), including Elysium rise and Elysium Planitia, Syrtis Major and Arabia Terra.Our model can provide a magma source region consistently present at the dichotomy boundary, which explains the location and, in the case of Tharsis and Elysium, even the longevity of martian volcanism.
Mantle convection sets the boundary condition for core dynamics, and is therefore very important for the dynamo activity.During a giant impact event, the majority of impact heat is deposited into the mantle (Ballantyne et al., 2023;Cheng et al., n.d.) and can shut down any core convection entirely, while any heat anomaly within the core should be homogenized quickly due to the core being liquid (Monteux et al., 2015).Thanks to various heat loss mechanisms in the mantle, including the enhanced heat loss through turbulent flow within the magma pond, core cooling takes place again within 10s Myr.During the first hundreds of million years after the impact, our model CMB heat flux exceeds the conductive threshold for the martian core, assumed to be 5-19 mW/m 2 (Williams & Nimmo, 2004), sufficient to induce core convection and thus drive a thermal dynamo.As large-scale dripping in the southern hemisphere brings cold material from the top to the lower mantle, pulses of heat flux are seen in the southern hemisphere CMB, drawing heat from the core (Figure 2g).This provides a mechanism for a dynamo with an intermittent nature, consistent with recent studies of the martian magnetic field (Mittelholz et al., 2020;Steele et al., 2023).As the mantle and core slowly cool, CMB heat flux drops below the convective threshold, causing the cessation of dynamo activity.Without much magmatic activity in the northern hemisphere throughout most of the martian geological history, the recording of any magnetic field is not very efficient, which explains the lack of a remnant magnetic field in the northern hemisphere.

Discussion
Depleted provinces related to early magmatism and high melt fractions are well known in the Solar System.On the Moon, the ilmenite-rich layer underlying the crust is a direct consequence of magma ocean crystallization (Breuer & Moore, 2015).Cratons on Earth are also highly depleted and were created early on during warmer mantle conditions (Cooper et al., 2017;Lee et al., 2011).Their rigid nature is, at least to some extent, similar to the residue-reinforced northern lithosphere we present in the current study.Other than the most promising case, our models show that a large-scale depleted province, in the form of a residue-reinforced lithosphere, can be formed with various impactor sizes and mantle viscosities (Figure 3).We observe that, in Figures 3b, 3c and 3e, the volumes of impact-induced melt and depleted residue depend on the impactor size, and that a crustal dichotomy is formed in all cases.In addition, we find that the spreading of the residue, and therefore the extent of the province, depends on the mantle viscosity (Figures 3d-3f).A more viscous mantle does not necessarily allow the postimpact depleted material to reach the northernmost regions (compared with Figure 2a).Calculations in 3D would also make it harder to reach the northern hemisphere due to geometrical effects.On the other hand, Figure 3a demonstrates that a case with no giant impact produces magmatism everywhere on Mars.This demonstrates that the giant impact in the south in the reference model does not only cause a thermal anomaly in the south, but also leads to conditions that forbid northern magmatism.All cases give us confidence that the process described in the present paper is not confined to a tight window in rheology and impact parameters.
Our recent study (Ballantyne et al., 2023) investigated a similar problem concerning the Martian crustal dichotomy using a Smooth Particle Hydrodynamics (SPH) approach, where an impactor radius of 750 km and impact angle of 15°(from the normal) was favored.Given this near-vertical angle, the resultant thermal anomaly is sufficiently symmetric to be parameterized as a head-on impact, even though the size and amplitude of the anomaly (and hence the magma pond) estimated by the impact scaling law is much smaller when the same impactor radius and velocity are used (Figure S1 in Supporting Information S1).Since we are mostly interested in studying the evolution and aftermath of a magma pond, instead of constraining the impact scenario, we simply use a larger impactor radius and velocity values to obtain a magma pond with a reasonable geometry.While it was suggested in Ballantyne et al. (2023) that a very low basalt fraction of 0.1 is required for a satisfactory crustal thickness, the included crustal thickness calculation (thus basalt fraction estimation) was relatively simple and aimed to serve only as one of the many ways how crust thickness could be reduced.In fact, with a basalt fraction 3. Evolution of crustal thickness and depleted layer thickness in additional simulations with the following parameters: (a) reference parameters but no impact, (b) impactor radius of 500 km, impactor radius of 750 km, (d) impactor radius of 1,000 km, with viscosity decreased to 5 ⋅ 10 20 Pa⋅s, (e) impactor radius of 1,000 km, with reference viscosity 1 ⋅ 10 21 Pa⋅s, and (f) impactor radius of 1,000 km, viscosity increased to 5 ⋅ 10 21 Pa⋅s. of 0.1, our model shows that the mantle would be completely depleted after hundreds of million years, which may not be realistic when considering the geologically recent Martian volcanism.As a result, this study takes a different approach where we calculate the value for basalt fraction using Martian rock samples (Text S1 in Supporting Information S1), and study the thermochemical consequence of an impact-induced heat anomaly.
A shortcoming of our models is that our crusts are thicker than those suggested by observations.This can be due to assumptions regarding melting, magmatism and/or on the absence of horizontal lava flows in our models (magmatic material only appears vertically above the melting location in the mantle).We perform additional tests with a shallower melt extraction depth and a deeper basalt-eclogite phase transition.Counter-intuitively, using a shallower melt extraction depth results in a higher overall crustal thickness (Figure S5 in Supporting Information S1).Unextracted melt rises and spreads northwards until reaching a shallow depth, and is extracted to form crust.During this time, surrounding material may melt because of its high temperature, increasing the total amount of melt and crust.By ensuring the base of the crust to be above the basalt-eclogite phase transition (as it should be with realistic crustal thicknesses), we test the influence of crustal dripping on the mantle flow dynamics and resultant diagnostics, and find that our results have not changed greatly.This is not surprising as the volume of primordial depleted material only depends on impact parameters and the melting model itself.

Conclusion
We presented a unifying model that successfully and simultaneously produces (a) the martian crustal dichotomy, (b) a realistic volcanic history, (c) explains the presence of equatorial magmatism, and (d) a plausible mechanism for an intermittent hemispheric dynamo.The key ingredient for all these processes to self-consistently occur is the formation and preservation of a hemisphere-scale depleted layer, a direct consequence of the crystallization of the giant impact-related regional magma ocean that is likely still present underneath the martian northern hemisphere.

Figure 1 .Figure 2 .
Figure 1.Snapshots of Mars's thermo-compositional evolution showing (a) the giant impact, (b) regional magma ocean crystallization and crust formation, (c) spreading of the associated depleted residue, and (d-f) preservation of the crustal dichotomy and self-consistent onset of equatorial magmatism."P.N." stands for "pre-Noachian."