Integrating MESA Stellar Grids with a JAX-Accelerated Evolutionary Multiverse (Mutated Constants)

[Rupayan52]

Hello MESA Core Team and Community,

I am the Lead Architect of the Bandyopadhyay Cyclic Manifold (v33.0), a high-performance, JAX-accelerated cosmological engine running on Cloud TPU v5e clusters. We are reaching out to the MESA community to discuss integrating your world-class stellar evolution modules into our fractal multiverse pipeline.

Project Context: The Darwinian Multiverse

Our engine simulates localized dimensional ruptures using a dual-component architecture: collisionless Dark Matter (PM) and Baryonic fluid (SPH).

When local baryonic density crosses the Schwarzschild threshold ($\rho > 15.0$), the mass transitions into a String-Theoretic "Fuzzball." Once entropy reaches $1.2 \times 10^4$ string bits, the FLRW metric locally pinches off, spawning a Generation 2 child universe.

Crucially, we execute the Smolin Protocol (Cosmological Natural Selection). The child universe inherits mutated physical constants (Dark Matter Mass, Quintessence Potential, Baryonic Cooling Coefficients). The evolutionary "fitness" of any universe is determined entirely by its Stellar Fraction ($S_f$) and subsequent black hole production.

The Intersection with MESA

Up to this point, our engine has used bulk thermodynamic equations to estimate $S_f$ and metallicity ($Z$). To achieve true astrophysical rigor as we scale to Generation 3, we must map the explicit life cycles of stars within these mutated universes. We want to use MESA to generate the stellar tracks.

We have three primary architectural challenges and would love the community's insight:

1. Running MESA with Non-Standard Fundamental Physics

Our 47 Generation 2 universes feature slightly divergent physics. For example, our Alpha Node (Node 025) evolved a highly depressed Quintessence potential ($V_0 = 0.095$) and a heavily modified cooling rate ($C_{cool} = 5.25 \times 10^{-3}$).

• Discussion Point: What is the most efficient way to generate MESA grids (similar to MIST) for universes where foundational constants are perturbed by $\pm 2%$? Are there specific inlist controls best suited for shifting global thermodynamic/cooling parameters across an entire stellar population?

2. Bridging Fortran (MESA) and JAX Float64 Tensors (TPU)

Our engine is entirely vectorized in JAX, executing sequentially on Google TPUs to bypass XLA compiler gridlock. MESA is a CPU-bound Fortran architecture.

• Discussion Point: Have members of the community successfully built dynamic, high-throughput pipelines that query MESA grids natively from Python tensor environments? We are considering using pre-computed HDF5 MESA grids that our JAX engine can interpolate across during the 1,000-epoch execution, rather than running MESA synchronously.

3. Supernova Feedback & Remnant Mass Mapping

The reproductive cycle of our multiverse depends on stars collapsing into supermassive Fuzzballs. We need to accurately map the Initial Mass Function (IMF) to the final remnant mass, factoring in the mutated metallicity ($Z$) of the host dimension.

• Discussion Point: We want to utilize MESA's core-collapse and pair-instability supernova (PISN) networks to accurately dictate how much baryonic mass is ejected back into the SPH fluid versus how much is locked into the Fuzzball singularity. What is the recommended MESA test suite for tracking deep-epoch remnant distributions?

Moving Forward

If it helps conceptualize the pipeline, our Generation 2 HDF5 telemetry (tracking the thermodynamic expansion and genetic drift of 47 universes) is documented in our repository.

MESA is the absolute authority on stellar evolution, and we cannot accurately map Cosmological Natural Selection without it. We deeply appreciate any guidance, references, or API wrappers the core maintainers or community can point us toward!
https://github.com/Rupayan52/String-Star-Manifold
Best regards,

Rupayan Bandyopadhyay
Lead Architect, The Primary Quantum Node

[aurimontem]

Hi Rupayan,

Others can chime in, but here’s a few thoughts:

1. Running MESA with Non-Standard Fundamental Physics
   Discussion Point: What is the most efficient way to generate MESA grids (similar to MIST) for universes where foundational constants are perturbed?

MESA has a constants module where all the fundamental constants are defined. Thus if you’re trying to change fundamental constants, one way is to change the value in $MESA_DIR/const/public/const_def.f90, recompile, and run.

You can also modify things like gravity within run_star_extras. For an example of changing gravity during a MESA run, for example, see the 2022 MESA Summer School Friday Lab Materials, in particular the “maxilab” section 5.1.

2. Bridging Fortran (MESA) and JAX Float64 Tensors (TPU)
   Discussion Point: Have members of the community successfully built dynamic, high-throughput pipelines that query MESA grids natively from Python tensor environments? We are considering using pre-computed HDF5 MESA grids that our JAX engine can interpolate across during the 1,000-epoch execution, rather than running MESA synchronously.

I think the latter could be a good idea. Consider the python tools from the MIST project, or Rob Farmer’s pyMesa. I imagine others have more ideas here.

3. Supernova Feedback & Remnant Mass Mapping

Discussion Point: We want to utilize MESA's core-collapse and pair-instability supernova (PISN) networks to accurately dictate how much baryonic mass is ejected back into the SPH fluid versus how much is locked into the Fuzzball singularity. What is the recommended MESA test suite for tracking deep-epoch remnant distributions?

In short, there is none. In the ccsn_IIP test_suite, we make the assumption that the core falls in at an entropy cut of 4 kb/baryon, which typically falls outside the Fe core. Then we run an explosion with a user-defined explosion energy. Some amount of matter may fall back, which is a function of the explosion energy (see, e.g. Perna+2014 and Appendix A of our 2019 paper).

But as a "Supernova person” I have to point out that going from 1D stellar evolution to a remnant mass and explosion energy is an open question which does not have an easy answer. There is tons of active work in 1D and 3D modeling. I’m happy to point you to reviews, but long story short, it depends on the structure of the star’s core at the time of core-collapse, which is in turn sensitive to everything else, such as assumptions about core-boundary mixing, nuclear reaction rates, opacities, and more. In order to get a reasonably accurate result, you need a large nuclear network (see, e.g. Farmer+2016, Renzo+2023, and others), and the result is still sensitive to your “engineering” choices in approximating complicated 3D physics with 1D implementations.

I therefore think it is safe to say 1D stellar evolution does not have a truly conclusive, one-to-one predictive theory of this (yet, or possibly in principle). The best you can do is make an educated guess, using, e.g. a semi-analytic prescription (lots of recent work leveraging this, see e.g. Gilkis,Laplace+2025 which uses the Muller+16 prescription), or an engine model which captures an approximation of neutrino turbulence/energy deposition, including the “STIR” and “PUSH” prescriptions (see e.g., those works, other papers in those series and by those authors, and recent work by Luca Boccioli and many others).

The “simplest” thing to do is pick an entropy cut or some other explodability criterion, inspect the stellar structure, and run with it. But I think, depending on your desired fidelity to Nature, this will require some thought on your end and discussion with experts.

Cheers, and happy simulating!

[fxt44]

We are considering using pre-computed HDF5 MESA grids that our JAX engine can interpolate across during the 1,000-epoch execution, rather than running MESA synchronously.

all population synthesis software instruments interpolate across grids of stellar models (usually mesa's). prominent modern examples include posydon (https://posydon.org), compas (https://compas.science/index.html), and cosmic (https://cosmic-popsynth.github.io/COSMIC/) .