🔭 astrophysics

Joint Electromagnetic and Gravitational Wave Inference of Binary Neutron Star Merger GW170817 Using Forward-Modeling Ejecta Predictions

This paper reassesses the multimessenger inference of GW170817 using a simulation-based kilonova model, revealing that while ejecta properties are tightly constrained by observations, inferences about the neutron star's mass and radius remain highly sensitive to assumptions regarding mass ejection, thereby highlighting the critical need for improved first-principles modeling and the careful handling of systematic uncertainties.

Original authors: Marko Ristić, Richard O'Shaughnessy, Kate Wagner, Christopher J. Fontes, Chris L. Fryer, Oleg Korobkin, Matthew R. Mumpower, Ryan T. Wollaeger

Published 2026-04-02

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marko Ristić, Richard O'Shaughnessy, Kate Wagner, Christopher J. Fontes, Chris L. Fryer, Oleg Korobkin, Matthew R. Mumpower, Ryan T. Wollaeger

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the universe as a giant, cosmic crime scene. On August 17, 2017, two neutron stars (the dense, dead cores of massive stars) crashed into each other. This event, known as GW170817, was a historic moment because we "heard" the crash through gravitational waves (ripples in space-time) and "saw" the aftermath through light (a glowing explosion called a kilonova, named AT2017gfo).

This paper is essentially a group of detectives trying to solve the mystery of what exactly happened during that crash, but they are hitting a major roadblock: their maps are a bit blurry.

Here is a simple breakdown of what the authors did and what they found, using some everyday analogies.

1. The Two Clues: The "Thud" and the "Flash"

Think of the event as a car crash.

Gravitational Waves (The "Thud"): This tells us about the cars before they hit. It reveals their weight, how fast they were spinning, and how squishy their tires were (a property called tidal deformability).
The Kilonova (The "Flash"): This is the debris flying everywhere after the crash. By studying the color and brightness of this debris, we can figure out how much stuff was thrown out and how fast it was moving.

The goal of this paper is to combine these two clues to get a perfect picture of the crash.

2. The Problem: The "Recipe Book" is Flawed

To connect the "Thud" (the stars) to the "Flash" (the debris), scientists use mathematical recipes (called "ejecta fits"). These recipes are based on supercomputer simulations that try to predict: "If two neutron stars with these specific weights crash, how much debris will fly out and how fast?"

The authors of this paper realized that these recipes are like old, hand-drawn maps.

They were drawn based on a very limited number of test drives (simulations).
They work well for "normal" crashes, but when you try to use them to explain the real GW170817 crash, the maps start to lead you in different directions.

3. The Experiment: Trying Different Maps

The researchers took three different "recipe books" (three different sets of mathematical formulas from other scientists) and tried to use them to explain the GW170817 data. They asked: "If we use Recipe A, what does the crash look like? What about Recipe B? Recipe C?"

They also added a "fudge factor" to their analysis. In real life, we know our recipes aren't perfect. So, they allowed for some "systematic uncertainty"—essentially admitting, "Our map might be off by a little bit, so let's give ourselves some wiggle room."

4. The Shocking Discovery: The Maps Disagree

Here is the punchline: The three recipes told three completely different stories.

Recipe A said: "The stars were very different sizes (one heavy, one light) and had a small radius."
Recipe B said: "The stars were the same size but had a huge radius."
Recipe C said: "The stars were different sizes, but the debris was moving at a totally different speed."

Even though they were looking at the same light and the same gravitational waves, the choice of mathematical recipe changed the entire conclusion about the nature of the neutron stars.

5. The "Speed Limit" Trap

One of the biggest issues they found was about speed.
Imagine you are trying to guess how fast a car was going based on how far its tires skidded.

The "Flash" (the light) tells us the debris was moving relatively slowly.
The "Recipes" (the simulations) often predict that debris should be moving much faster.

When the researchers forced the math to match the "slow" speed observed in the light, the recipes had to invent weird, extreme scenarios to make the math work. For example, to make the debris move slowly in the simulation, the recipe had to assume the neutron stars were either incredibly squishy or had very strange mass ratios.

6. The Conclusion: We Need Better Maps

The authors conclude that we cannot trust our current "recipe books" blindly.

The Good News: We have amazing tools (gravitational waves and light) to study the universe.
The Bad News: The bridge connecting the two (the math that turns a crash into a debris cloud) is shaky.

They warn that if we don't improve these simulations and understand the "systematic errors" (the flaws in our maps), we might draw the wrong conclusions about the fundamental laws of physics, like the Equation of State (which is basically the rulebook for how matter behaves under extreme pressure).

The Takeaway Analogy

Imagine you are a detective trying to figure out what kind of cake was baked based on the crumbs left on the floor and the sound of the oven timer.

You have three different "Crumbs-to-Cake" manuals.
Manual #1 says the crumbs mean it was a chocolate cake.
Manual #2 says it was a vanilla cake.
Manual #3 says it was a carrot cake.

This paper says: "Stop arguing about which cake it is until we fix the manuals!" The manuals are based on too few test cakes, and they don't agree with each other. Before we can confidently say what the universe is made of, we need to bake more test cakes (run more supercomputer simulations) to create a better, more accurate manual.

In short: The data is great, but our math to interpret it is still a work in progress. We need to be humble about our assumptions and keep refining our models before we claim to know the "truth" about neutron stars.

1. Problem Statement

The detection of the binary neutron star (BNS) merger GW170817 and its electromagnetic counterpart AT2017gfo opened a new era of multimessenger astronomy. A primary goal is to use these joint observations to constrain the properties of the merging neutron stars (masses, spins, radii) and the dense matter equation of state (EOS).

However, a critical bottleneck exists in the mapping between gravitational wave (GW) parameters and electromagnetic (EM) ejecta properties.

The Gap: GW signals provide information about the binary's pre-merger state (masses, tidal deformability). EM signals (kilonovae) depend on the mass, velocity, and composition of material ejected during the merger.
The Assumption: Previous studies often rely on "fitting formulae" derived from numerical relativity (NR) simulations to map binary parameters to ejecta properties. These fits are then used to infer binary properties from EM data.
The Issue: These fitting formulae are calibrated on a limited set of NR simulations and are often applied far outside their valid parameter space (extrapolation). Furthermore, previous multimessenger analyses frequently marginalized over ejecta velocity or treated it as a free parameter without verifying consistency with the underlying NR fits, potentially hiding systematic biases.

This paper investigates how systematic uncertainties in these ejecta fitting formulae impact the joint inference of binary parameters and the EOS.

2. Methodology

The authors employ a rigorous Bayesian inference framework combining GW and EM data using a forward-modeling approach.

A. Data and Models

GW Data: Utilizes the marginal likelihood of GW170817 derived from the RIFT parameter inference engine, covering a wide range of chirp mass ( $M_c$ ), mass ratio ( $q$ ), effective spin ( $\chi_{eff}$ ), and effective tidal deformability ( $\tilde{\Lambda}$ ).
EM Data: Uses the AT2017gfo light curve observations across multiple filters (grizyJHK).
Kilonova Surrogate Model: Instead of running expensive radiative transfer simulations for every sample, the authors use a Neural Network (MLP) surrogate model trained on $\sim$ 250 high-fidelity SuperNu simulations. This model predicts light curves based on ejecta mass and velocity parameters.
Ejecta Fits (Forward Models): The study tests three distinct fitting formulae derived from NR simulations:
1. KruFo20 (Krüger & Foucart, 2020)
2. DiCo20 (Dietrich et al., 2020)
3. Nedora21 (Nedora et al., 2021)
  These fits predict dynamical ejecta mass ( $M_{ej}$ ), disk mass ( $M_{disk}$ ), and ejecta velocity ( $v_{ej}$ ) as functions of binary mass ratio ( $q$ ) and neutron star radius ( $R_{1.4}$ ).

B. Inference Strategy

The authors perform joint Bayesian inference using the RIFT framework. They define a joint likelihood:
$\ln \mathcal{L}_{total} = \ln \mathcal{L}_{GW} + \ln \mathcal{L}_{EM}$

Crucially, they introduce systematic uncertainty parameters to account for the imperfections in the ejecta fits:

$\alpha_{dyn}$ : Scale factor for dynamical ejecta mass.
$f_{disk}$ : Fraction of the accretion disk ejected as wind.
$\beta_{\phi}$ : Systematic uncertainty parameter for ejecta velocity (reparameterized to handle the non-linear relationship).
$\sigma_{sys}$ : A catch-all parameter for light-curve modeling uncertainties.

The study compares three scenarios:

Full Systematics: All uncertainty parameters ( $\alpha_{dyn}, f_{disk}, \beta_{\phi}$ ) are free.
Fixed Velocity: $\beta_{\phi} = 0$ (no velocity uncertainty).
Fixed Disk & Velocity: $\beta_{\phi} = 0$ and $f_{disk} = 0.30$ (fixed disk ejection fraction).

3. Key Contributions

Systematic Uncertainty Quantification: The paper explicitly quantifies how the choice of ejecta fitting formula and the handling of systematic errors (specifically velocity and disk fraction) drastically alter the inferred binary parameters.
Forward-Modeling vs. Reweighting: Unlike previous studies that reweighted existing GW posteriors, this work performs a full joint inference from the ground up, revealing that the joint posterior often explores regions of parameter space (e.g., extreme mass ratios or large radii) that are poorly represented in standard GW-only posteriors.
Velocity Sensitivity: The study demonstrates that the ejecta velocity is a dominant driver of the inferred EOS and binary properties. Ignoring velocity systematics forces the inference into physically inconsistent regimes.
Disk Ejection Fraction: The analysis suggests a lower disk ejection fraction ( $\sim 4-20\%$ ) than often assumed in literature (up to 40%), aligning better with recent GRMHD simulations with full neutrino transport.

4. Key Results

A. Dependence on Ejecta Fits

Even when allowing for generous systematic uncertainties, the three different ejecta fits (KruFo20, DiCo20, Nedora21) yield distinctly different conclusions regarding the binary system:

Mass Asymmetry: Two of the three models prefer highly asymmetric binaries ( $\delta > 0.2$ , or $q < 0.66$ ), while one allows for equal-mass binaries.
Neutron Star Radius ( $R_{1.4}$ ): The inferred radius ranges widely depending on the fit:
- Nedora21: $8 - 15$ km
- KruFo20: $10 - 20$ km
- DiCo20: $10 - 40$ km (favoring very large radii).
Conclusion: The multimessenger inference of the EOS is entirely dependent on the chosen forward model and the mapping prescription.

B. Impact of Systematic Uncertainties

With Free Systematics: When $\alpha_{dyn}, f_{disk},$ and $\beta_{\phi}$ are free, the EM data dominates the ejecta parameter inference. The GW data slightly perturbs the results but does not resolve the tension between the different ejecta fits. The models converge on a consistent set of ejecta properties (mass/velocity) but map them to different binary configurations.
Without Velocity Systematics ( $\beta_{\phi}=0$ ): Fixing the velocity to the fit's prediction forces the inference to select binary parameters that satisfy the velocity contours of the fits. This results in a strong correlation between mass asymmetry and radius (cubic relationship), pushing the posterior toward large radii ( $R_{1.4} > 15$ km) and asymmetric binaries to satisfy the low-velocity requirement of the fits. This challenges previous GW-only conclusions.
Fixed Disk Fraction: Assuming a fixed $f_{disk} = 30\%$ leads to poor fits to the data and prohibitively large uncertainties in the light curves, rendering the inference uninformative.

C. Tension with Previous Analyses

The results challenge previous multimessenger studies that marginalized over ejecta velocity without accounting for the specific constraints of the NR fits. The authors show that omitting velocity systematics introduces significant biases, leading to extreme conclusions about the progenitor system that may not be physical.

5. Significance and Implications

Limitations of Current Fits: The paper highlights that current fitting formulae for ejecta are insufficiently calibrated for the broad parameter space required by multimessenger inference. They are often used outside their region of validity, leading to contradictory physical interpretations.
Need for Better Physics: Accurate multimessenger inference requires:
1. First-principles modeling of mass ejection that covers the full BNS parameter space.
2. Explicit treatment of systematic uncertainties for both mass and velocity in the fitting formulae.
3. Physical mappings between NR ejecta and kilonova components, rather than ad-hoc conventions.
Future Directions: The authors call for more extensive numerical relativity simulations (extending beyond 1.5s post-merger) and improved radiative transfer models to reduce the "systematic budget" that currently dominates the error bars in EOS constraints.

In summary, this work serves as a cautionary tale for the field: multimessenger inferences of neutron star properties are currently limited not by the precision of GW or EM detectors, but by the systematic uncertainties in the theoretical models connecting them. Without resolving these modeling tensions, constraints on the nuclear equation of state will remain ambiguous.