Stratified Lunar Magma Ocean

• Stratified lunar magma ocean is a layered assembly formed by fractional crystallization that produces distinct olivine, pyroxene, and plagioclase cumulates.
• The framework combines petrological experiments and thermodynamic models to link density-driven segregation and remelting with present-day lunar geochemical signatures.
• Dynamic processes including overturn, partial mixing, and hybrid mantle sources explain the formation of an anorthositic crust and variations in mare basalt compositions.

A stratified lunar magma ocean is the chemically and density layered state produced as the Moon’s early global magma ocean crystallized, generating compositionally distinct cumulates and residual products rather than a homogeneous mantle. In the lunar literature, the term refers both to the primary sequence of olivine- and pyroxene-dominated early cumulates, plagioclase flotation products, late Fe–Ti-rich cumulates, KREEP-type residual materials, and an anorthositic crust, and to the subsequent partial overturn, retention, and remelting of those layers that shaped the present mantle and late mare volcanism (Schaefer et al., 2018, Schwinger et al., 2021).

1. Fractional crystallization and primary cumulate layering

The canonical stratified-LMO framework treats crystallization as a nearly fractional, batch-fractional, or mixed equilibrium–fractional process following liquidus–solidus phase relations for a chondritic bulk lunar composition. The residual melt evolves along a magma-ocean adiabat, and the identity and proportions of crystallizing phases are read from experimental or thermodynamic phase diagrams; densities and related physical properties are then computed from phase equations of state and combined by mass or volume fractions (Schaefer et al., 2018).

The crystallization sequence inferred for the LMO is strongly ordered. Early Mg-rich olivine and then orthopyroxene crystallize and sink, producing a lower cumulate pile with only small density gradients. Plagioclase appears later and, being less dense than the melt, floats to form the anorthositic crust. Near the end of solidification, progressive Fe enrichment of the residual melt leads to ilmenite-bearing cumulates that are denser than the underlying mantle and therefore gravitationally unstable. In related mantle-stratigraphy models, the final residual melts are KREEP-type cumulates and the last flotation products are the anorthositic crust (Schaefer et al., 2018, Schwinger et al., 2021).

Stage of solidification	Dominant phase(s)	Dynamical outcome
0–20%	Mg-rich olivine (Fo$_{90}$)	Sinks; builds early lower cumulates
20–70%	Orthopyroxene	Sinks; enlarges lower cumulate pile
70–80%	Plagioclase (An$_{98-100}$)	Floats; forms anorthositic lid/crust
80–95%	Fe–Ti oxides, including ilmenite	Produces dense late cumulates
Final residual melt	KREEP-type cumulates	Concentrates incompatible elements

This sequence yields a first-order vertical stratigraphy in which buoyant plagioclase-derived crust overlies mafic cumulates, while the densest late products tend to migrate downward. The resulting mantle is therefore stratified both chemically and rheologically, not merely compositionally.

2. Mantle reservoirs, FeO dependence, and present-day structure

A detailed cumulate-reservoir formulation divides the mantle into early-stage lower-mantle cumulates dominated by olivine $\pm$ orthopyroxene, mid-stage upper-mantle clinopyroxene-rich cumulates, late-stage ilmenite-bearing cumulates (IBC), KREEP-type residual cumulates, and an anorthositic crust fixed at 40 km in the model configuration. The lower-mantle cumulates are Fe-poor and Ti-poor, the upper mantle is modestly more Fe rich, and the IBC are the densest and most Fe–Ti enriched of the mantle reservoirs (Schwinger et al., 2021).

The properties of these reservoirs depend strongly on bulk silicate Moon FeO. For the lower mantle,

$$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$

with ${\rm TiO_2}_{\rm LM}\lesssim 0.1$. For the upper mantle,

$$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$

with ${\rm TiO_2}_{\rm UM}\lesssim 0.2$. The IBC obey the strongest FeO dependence: $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$ so that $t_{\rm IBC}\approx 11.6$ km at $8$ wt% FeO and $_{98-100}$0 km at $_{98-100}$1 wt% FeO. Their density at formation depth is

$_{98-100}$2

KREEP cumulates are volumetrically minor, with thickness $_{98-100}$3 km and density $_{98-100}$4–$_{98-100}$5 (Schwinger et al., 2021).

These relations are used together with selenodetic and seismic constraints. The same modeling indicates that the bulk silicate Moon contains about $_{98-100}$6–$_{98-100}$7 weight percent FeO, with a lowermost limit of $_{98-100}$8 weight percent and an uppermost limit of $_{98-100}$9 weight percent. It also implies incomplete sinking of the IBC: only $\pm$0–$\pm$1 percent of the original IBC mass might have reached the core-mantle boundary, while the rest either remained near its formation depth or was mixed into the middle mantle. In that interpretation, the present-day mantle retains both primary fractionation signatures and overturn-generated heterogeneity, including a high-density, low-velocity zone at the base of the mantle of about $\pm$2 km thickness (Schwinger et al., 2021).

3. Fe–Mg isotopes and the Chang’E 5 evidence for hybrid cumulate remelting

Direct evidence that late lunar volcanism sampled stratified LMO reservoirs comes from the Chang’E 5 mare basalts, whose Fe and Mg isotopic compositions, petrology, and trace-element systematics were interpreted as recording a hybrid mantle source. The isotope notation is

$\pm$3

and

$\pm$4

Measured CE-5 values are $\pm$5‰ and $\pm$6‰. Modeled or previously measured LMO-cumulate endmembers are $\pm$7‰ and $\pm$8‰ for early-stage olivine-dominated cumulates, versus $\pm$9‰ and $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$0‰ for late-stage clinopyroxene + ilmenite-rich cumulates. The CE-5 Fe isotope value lies between low-Ti and high-Ti mare basalt fields, whereas the Mg isotope value remains within the range of early-cumulate melts, pointing to a mixed source rather than a single cumulate lithology (Y. et al., 2023).

Petrologically, the CE-5 clasts have Mg# $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$1, the lowest of all returned mare basalts, and display a sub-ophitic groundmass of pyroxene + olivine + plagioclase + minor ilmenite. They also contain symplectites of olivine + silica or olivine + Si–K glass derived from pyroxferroite breakdown, together with “swiss-cheese” intergrowths of fayalite + Si–K glass. Olivine compositions of Fo$$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$2, in equilibrium with the whole-rock Mg#, indicate no significant crystal accumulation, so the bulk clast chemistry approximates a primary liquid. Their trace-element pattern is KREEP-like, with $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$3, $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$4, and a deep Eu anomaly of $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$5, which was argued to require a source enriched in late-stage plagioclase-depleted melt rather than simple olivine fractionation alone (Y. et al., 2023).

The isotopic mixing relation is written

$$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$6

where $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$7 is the mass fraction of the early-cumulate melt. Substituting the CE-5 $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$8 and $$\rho_{\rm LM} \simeq 3380~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$9 values and the endmember compositions yields ${\rm TiO_2}_{\rm LM}\lesssim 0.1$0–${\rm TiO_2}_{\rm LM}\lesssim 0.1$1, corresponding to a ${\rm TiO_2}_{\rm LM}\lesssim 0.1$2–${\rm TiO_2}_{\rm LM}\lesssim 0.1$3 percent contribution of late-stage cumulate melt. In the specific geometric interpretation given for an ${\rm TiO_2}_{\rm LM}\lesssim 0.1$4 km deep LMO, this is reproduced by about ${\rm TiO_2}_{\rm LM}\lesssim 0.1$5 melt of a ${\rm TiO_2}_{\rm LM}\lesssim 0.1$6-solid cumulate, mainly olivine + orthopyroxene, and ${\rm TiO_2}_{\rm LM}\lesssim 0.1$7 melt of a ${\rm TiO_2}_{\rm LM}\lesssim 0.1$8-solid cumulate, clinopyroxene + ilmenite rich, originating at about ${\rm TiO_2}_{\rm LM}\lesssim 0.1$9 km and $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$0 km depth, respectively. Fractional crystallization or impact-related evaporation was argued not to generate the observed combined isotopic and geochemical signature (Y. et al., 2023).

4. Overturn physics, mixing, and the survival of stratification

The persistence or destruction of LMO layering depends on the competition between thermal convection, compositional buoyancy, chemical diffusion, and rheology. A standard density expansion used in magma-ocean and cumulate models is

$$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$1

with representative values $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$2 and $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$3. The vigor of overturn is commonly parameterized by

$$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$4

where $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$5, $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$6, and convection requires $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$7–$$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$8. For a compositional density contrast $$\rho_{\rm UM} \simeq 3340~{\rm kg/m^3} + (11~{\rm kg/m^3~per~wt\%FeO_{BSM}})\cdot({\rm FeO}_{\rm BSM}-8),$$9, the buoyancy flux per unit area is

${\rm TiO_2}_{\rm UM}\lesssim 0.2$0

Chemical diffusion follows

${\rm TiO_2}_{\rm UM}\lesssim 0.2$1

with ${\rm TiO_2}_{\rm UM}\lesssim 0.2$2–${\rm TiO_2}_{\rm UM}\lesssim 0.2$3 and ${\rm TiO_2}_{\rm UM}\lesssim 0.2$4, while rheology may be written in Arrhenius form as

${\rm TiO_2}_{\rm UM}\lesssim 0.2$5

with ${\rm TiO_2}_{\rm UM}\lesssim 0.2$6–${\rm TiO_2}_{\rm UM}\lesssim 0.2$7 (Schaefer et al., 2018).

These relations support a time-dependent rather than static view of stratification. Rapid initial cooling can crystallize the magma ocean to ${\rm TiO_2}_{\rm UM}\lesssim 0.2$8 by volume in ${\rm TiO_2}_{\rm UM}\lesssim 0.2$9 yr, whereas formation of the anorthositic lid delays the final $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$0 by an extra $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$1 Myr. In models including strong tidal heating, full solidification can extend to $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$2 Myr. A $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$3 km thick dense ilmenite layer has a Stokes-type settling time of $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$4–$$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$5 yr for cumulus viscosities of $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$6–$$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$7 Pa s, and if solid-state convection starts early, mixing can reduce heterogeneity over $$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$8–$$t_{\rm IBC}({\rm km}) \simeq 3.9\,({\rm FeO}_{\rm BSM}) - 19.0,$$9 yr, although the ilmenite-rich layer may persist as a dense root (Schaefer et al., 2018).

The CE-5 model places partial remelting within this dynamical setting. Dense late cumulates sink or overturn into a still-warm mantle, lower the local solidus, and permit partial melting of mixtures of early olivine-rich and late clinopyroxene–ilmenite cumulates. Melt compositions are then governed by batch melting,

$t_{\rm IBC}\approx 11.6$0

where $t_{\rm IBC}\approx 11.6$1 is melt fraction and $t_{\rm IBC}\approx 11.6$2 is the bulk partition coefficient. Isotopic evolution during equilibrium crystallization is described using Rayleigh-type relations such as

$t_{\rm IBC}\approx 11.6$3

with equilibrium fractionation factors for Fe scaling approximately as $t_{\rm IBC}\approx 11.6$4. In this framework, stratification is both preserved and reworked: preserved in the sense that late cumulates remain compositionally distinct, and reworked in the sense that overturn and remelting generate hybrid magmas (Y. et al., 2023).

5. A two-layer proto-Earth–Theia stratification model

A separate model extends the idea of a stratified LMO back to the immediate post-impact state by proposing that the magma ocean inherited two dominantly different silicate melts, one from proto-Earth’s mantle and one from Theia’s, and that these unmixed by density before solidification. In that scenario, the newborn Moon was effectively molten to a depth up to $t_{\rm IBC}\approx 11.6$5–$t_{\rm IBC}\approx 11.6$6 km, corresponding to $t_{\rm IBC}\approx 11.6$7–$t_{\rm IBC}\approx 11.6$8 GPa, and disk or lunar temperatures were $t_{\rm IBC}\approx 11.6$9–$8$0 K. The proto-Earth silicate melt is represented by $8$1 wt% $8$2, $8$3 wt% $8$4, $8$5 wt% $8$6, and $8$7 wt% $8$8, whereas the Theia melt is represented by $8$9 wt% $_{98-100}$00, $_{98-100}$01 wt% $_{98-100}$02, $_{98-100}$03 wt% $_{98-100}$04, and $_{98-100}$05 wt% $_{98-100}$06 (Liu, 8 Sep 2025).

Because the assumed Theia mantle is more Fe rich, its melt is denser: $_{98-100}$07 at $_{98-100}$08 GPa and $_{98-100}$09C, compared with $_{98-100}$10. For a molten Theia-derived blob of radius $_{98-100}$11 km, equating drag and buoyancy gives

$_{98-100}$12

and hence

$_{98-100}$13

At lunar radius $_{98-100}$14 km, the inferred sinking time is $_{98-100}$15 h. By contrast, diffusion in silicate melt with $_{98-100}$16 over $_{98-100}$17 m gives

$_{98-100}$18

much longer than $_{98-100}$19. The model therefore predicts prompt segregation into a dense lower Theia-derived melt layer and an upper proto-Earth melt layer (Liu, 8 Sep 2025).

Fractional crystallization of the two layers was modeled with alphaMELTS (pMELTS) at $_{98-100}$20, $_{98-100}$21, $_{98-100}$22, and $_{98-100}$23 GPa. At $_{98-100}$24 GPa, orthopyroxene and clinopyroxene from the upper layer are less dense than the lower melt and float upward across the interface; by the time these phases are exhausted, $_{98-100}$25 of the upper-layer mass sits atop the interface as solids. Between $_{98-100}$26C and $_{98-100}$27C, olivine from the upper melt also floats, adding $_{98-100}$28 of the upper-layer mass. In the lower layer, olivine is neutrally buoyant at $_{98-100}$29 GPa, but orthopyroxene above $_{98-100}$30C and clinopyroxene down to $_{98-100}$31C are less dense than the lower melt and float into the upper cumulate pile, contributing $_{98-100}$32 of the lower-layer mass. In this proposed solution to the lunar isotopic crisis, the final solid Moon has an upper cumulate mantle derived almost exclusively from proto-Earth material, while isotopically distinct Theia material is sequestered in the lower mantle and core (Liu, 8 Sep 2025).

6. Synthesis, misconceptions, and active points of interpretation

The available stratified-LMO models agree that lunar magma-ocean solidification generated sharp chemical and density gradients, but they differ on the extent to which those gradients survived intact. Reviews and interior-structure models emphasize a bottom-up fractional-crystallization sequence followed by partial overturn, incomplete IBC sinking, and retention of major lithological boundaries; the CE-5 basalt study emphasizes later remelting of mixed early and late cumulates; and the two-layer isotopic-crisis model adds a pre-solidification density separation between proto-Earth and Theia-derived melts (Schaefer et al., 2018, Schwinger et al., 2021, Y. et al., 2023, Liu, 8 Sep 2025).

One recurrent misconception is that stratification implies a permanently static stack of pristine layers. The dynamical calculations instead indicate rapid crystallization, conductive delay by an anorthositic lid, gravitational instability of late dense cumulates, and mixing or entrainment over $_{98-100}$33–$_{98-100}$34 yr, even though dense ilmenite-bearing material may remain concentrated near the core-mantle boundary. Another misconception is that KREEP-like signatures in young mare basalts require large admixture of KREEP. The CE-5 interpretation argues that a hybrid mantle source containing both early- and late-stage LMO cumulates can reproduce the combination of low Mg#, intermediate $_{98-100}$35, early-cumulate-like $_{98-100}$36, and KREEP-like REE systematics without requiring large admixture of KREEP (Y. et al., 2023).

Taken together, these results suggest that “stratified lunar magma ocean” is best understood as a coupled petrological and dynamical framework. In its narrow sense, it denotes the phase-equilibria-driven cumulate architecture generated during bottom-up solidification. In a broader sense, it includes the flotation of plagioclase, the sinking and partial survival of IBC, the formation and preservation of KREEP-rich reservoirs, the possibility of compositional two-layering inherited from the Moon-forming impact, and the later selective remelting of mixed cumulate packages. Within that framework, the anorthositic crust, mantle seismic layering, basal dense material, the Procellarum KREEP Terrane, and the geochemistry of the youngest mare basalts are all treated as linked consequences of early lunar differentiation rather than isolated phenomena (Schaefer et al., 2018, Schwinger et al., 2021).