A Pixel-by-Pixel Path to Population III Discovery with JWST

Technical Summary: A Pixel-by-Pixel Path to Population III Discovery with JWST

Problem Statement
The identification of Population III (Pop III) stars—the first generation of metal-free stellar systems—remains a primary goal of modern observational astronomy. While the James Webb Space Telescope (JWST) has revealed numerous UV-bright galaxies at $z > 10$, distinguishing primordial stellar populations from early metal-enriched systems (Pop I/II) is challenging. Current approaches often rely on integrated photometry or line measurements, which suffer from significant limitations: host-galaxy light can wash out the contrast of a Pop III component, creating degeneracies with very low-metallicity Pop II stars. This confusion is exacerbated by varying stellar ages, dust attenuation, and nebular physics. Furthermore, traditional Bayesian spectral energy distribution (SED) fitting is computationally prohibitive for large-scale, pixel-by-pixel studies required to disentangle compact primordial signals from their local environments.

Methodology
The authors present an end-to-end framework combining physically motivated forward modeling with Simulation-Based Inference (SBI) to test Pop III detectability in JWST-like observations. The methodology proceeds in three stages:

1. Forward Modeling:

   • Pop III Models: Primordial populations are modeled using the Yggdrasil simple stellar population (SSP) framework based on metal-free templates (Schaerer 2002). Two initial mass function (IMF) families are explored: Pop III.2 (a moderately top-heavy log-normal IMF with $M_c = 10 M_\odot$) and a Pop III model with a Kroupa IMF (0.1–100 $M_\odot$). Star formation histories (SFHs) are assumed to be instantaneous bursts, reflecting the short-lived nature of pristine star formation. Nebular covering fractions ($f_{cov}$) and Lyman-$\alpha$ transmission scenarios are varied.
   • Fiducial Models: Enriched Pop I/II populations are modeled using the FSPS code with MIST isochrones, the MILES stellar library, and non-parametric SFHs (via Dense Basis) to capture complex star formation behaviors.
   • Mock Observations: Synthetic observations are generated to match the JADES survey (and PRIMER for specific candidates), incorporating 19 filters (ACS and NIRCam). Realistic noise properties are derived from empirical data, and point-spread function (PSF) convolution is applied.

2. Simulation-Based Inference (SBI):

   • The framework employs Neural Posterior Estimation (NPE) using masked autoregressive normalizing flows to learn the inverse mapping from observed photometry to physical parameters ($\theta$).
   • This approach avoids explicit likelihood calculations, enabling amortized inference where posterior distributions for new observations are estimated almost instantaneously after training.
   • Redshift is treated as a conditioning input (assumed known from spectroscopy or photometry) to reduce degeneracies.

3. Model Comparison and Explainability:

   • Model preference is assessed via Posterior Predictive Checks (PPCs). The goodness-of-fit ($\chi^2$) is computed for both Pop III and fiducial models. The metric $\Delta \log_{10} \chi^2 = \log_{10}(\chi^2_{\text{fiducial}}) - \log_{10}(\chi^2_{\text{PopIII}})$ determines the preferred model.
   • SHapley Additive exPlanations (SHAP) are used to decompose the drivers of model preference, ranking physical parameters (mass, age, redshift, separation, etc.) by their contribution to detectability.

Key Contributions and Experimental Design
The paper advances the field by transitioning from isolated/integrated tests to spatially resolved, pixel-based model comparison. The authors conduct three distinct experimental setups:

1. Isolated Sources: A control sample of isolated Pop III sources to validate parameter recovery and baseline performance.
2. Integrated Contamination: Composite systems where Pop III bursts are blended with Pop II hosts in unresolved photometry to quantify spectral confusion.
3. Resolved Morphological Overlap: Controlled simulations injecting compact Pop III clumps into smooth Sérsic Pop II hosts. This setup varies projected separation, mass ratios, clump ages, and host morphologies to map the "detectability surface."

Results

• Isolated Sources: The SBI pipeline successfully recovers Pop III properties (mass and age) with moderate accuracy ($R^2 \approx 0.57–0.59$), limited primarily by mass-age degeneracy. Model comparison shows a strong preference for Pop III models in isolated, young, massive, and lower-redshift systems.
• Integrated (Unresolved) Analysis: In unresolved photometry, host-galaxy contamination severely limits Pop III recovery. Even in favorable configurations (young, massive bursts), the flexible fiducial model often outperforms the Pop III model because it can reproduce the combined broadband colors by adjusting stellar mass and age. The Pop III model systematically underpredicts redder fluxes dominated by the host.
• Resolved (Pixel-Based) Analysis: Spatially resolved analysis substantially improves recoverability. Detectability is highest for young and massive Pop III clumps located at large projected separations from the host center. In favorable configurations (high mass, young age, $d/R_e \approx 3$), recovery rates reach $\sim 90\%$. Conversely, older and centrally embedded clumps are rarely recovered.
• Drivers of Detectability: SHAP analysis identifies projected separation, redshift, mass ratio, and host star formation rate as the dominant drivers of detection. Secondary dependencies include the age of the Pop III clump and host morphology.
• Application to Real Data: The framework is applied to the "Blueberry" candidate (a compact companion to the galaxy CAPERS-UDS-32520 at $z \approx 5.1$). The pixel-by-pixel analysis reveals spatially differentiated behavior: the compact blue companion is preferentially described by Pop III-like models, while the host ("Banana") is better explained by fiducial Pop I/II models. This aligns with spectroscopic evidence of metal lines in the host but their absence in the companion.

Significance and Claims
The paper claims that the primary path to discovering Pop III stars with JWST lies in spatially resolved, imaging-first strategies rather than integrated photometric selection. The authors argue that:

1. Contamination is the primary barrier: Host-galaxy light dilutes primordial signatures in unresolved data, making pixel-level separation essential.
2. Morphology is a key criterion: Targets should be selected based on geometric separation (offsets from host centers) and local contrast, as these factors are as critical as intrinsic stellar properties for detection.
3. Computational Efficiency: The SBI pipeline enables the processing of large deep-field datasets at the pixel level, a task previously infeasible with traditional MCMC methods.
4. Validation: The successful application to the Blueberry candidate validates the pipeline's ability to identify regions consistent with primordial populations in real data, motivating systematic searches for similar "compact blue companion" configurations in existing JWST deep fields.

The authors remain modest regarding their claims, noting that their results depend on specific template assumptions (Yggdrasil) and that spectroscopic follow-up remains essential to confirm primordial interpretations, as photometry alone cannot fully rule out alternative enriched-population scenarios with complex nebular physics.