Search for Gravitational Wave Memory in PPTA and EPTA Data: A Complete Signal Model

This paper presents the first comprehensive search for gravitational wave memory from supermassive black hole binary mergers and generic bursts using PPTA and EPTA data, employing full numerical relativity waveforms and advanced posterior approximation techniques to establish stringent upper limits on merger rates and burst strain amplitudes.

The Gist

The Big Picture: Listening to the Universe's "Echoes"

Imagine the universe is a giant, quiet room. For years, scientists have been trying to hear a faint, low hum coming from the center of the room. This hum is the Gravitational Wave Background, created by thousands of pairs of supermassive black holes orbiting each other. We've finally heard this hum!

But this paper isn't about the hum. It's about listening for something much rarer and stranger: Gravitational Wave Memory.

What is "Gravitational Wave Memory"?

To understand this, imagine you are standing in a field during a storm.

• Normal Gravitational Waves: These are like the wind gusts. They blow hard, then stop, then blow the other way. They oscillate back and forth. When they pass, the air returns to normal.
• Gravitational Wave Memory: This is like a massive tree falling in the distance. The wind from the fall blows you, but when the tree hits the ground, it doesn't just stop. The ground itself shifts. When the wind dies down, you are standing in a slightly different spot than where you started. The "memory" is that permanent shift.

In physics terms, when two black holes crash together, they don't just send out ripples; they permanently stretch the fabric of space-time. After the crash, space is slightly "longer" or "shorter" than it was before. This permanent change is the Memory.

The Detectives: Pulsar Timing Arrays (PTAs)

How do we catch this permanent shift? We use Pulsars.

• The Analogy: Think of pulsars as the most perfect lighthouses in the universe. They spin hundreds of times a second, beaming radio waves at us with clockwork precision.
• The Setup: We have two teams of astronomers (one in Australia, one in Europe) watching a fleet of these cosmic lighthouses.
• The Hunt: If a gravitational wave passes between Earth and a pulsar, it stretches the space in between. This makes the light take a tiny bit longer to arrive.
• The Memory Effect: If a black hole merger happens, the "stretch" doesn't go away. The pulsar's signal arrives slightly later than expected, and it stays late. It's like a train that is delayed by 5 minutes, and then stays 5 minutes late forever.

What Did This Paper Do?

The authors (a huge team of scientists) decided to upgrade their search for these "permanent delays" in two major ways:

1. The "Full Movie" vs. The "Snapshot"

Previously, scientists looked for memory using a simple "snapshot" model. They assumed the black holes crashed instantly, causing an immediate jump in the signal (like a step function).

• The New Approach: This paper used a full movie. They used supercomputer simulations (Numerical Relativity) to model the entire life of the black hole merger: the slow dance (inspiral), the crash (merger), the settling down (ringdown), and the permanent shift (memory).
• Why it matters: The old "snapshot" model was a bit too aggressive. It assumed the delay happened instantly. The new "movie" model shows the delay builds up slowly. This is more accurate and prevents scientists from getting fooled by noise.

2. The "Smart Search" (Speeding Up the Math)

Searching for these signals is incredibly hard math. It's like trying to find a specific needle in a haystack, but the haystack is changing shape, and you have to check every single piece of straw against a million different theories.

• The Old Way: They used "lookup tables," which are like pre-calculated cheat sheets. They are fast but a bit clunky.
• The New Way: The team used Machine Learning (specifically "Normalizing Flows"). Imagine training a smart robot to learn the shape of the "needle" so it can instantly guess where it is without checking every straw. This made the search 6 to 7 times faster while keeping the accuracy high.

The Results: The Silence Speaks Volumes

So, did they find the memory? No.

But in science, a "no" is often just as important as a "yes." Here is what they learned from the silence:

• The "No-Go" Zone: They ruled out the existence of massive black hole mergers (with masses of 10 billion suns) happening within 700 million light-years of Earth over the last 18 years.
• The Strain Limit: They set a strict limit on how big a "permanent shift" could be. If a shift happened that was bigger than $10^{-14}$ (a number so small it's hard to imagine—imagine stretching a rubber band the length of the Earth by the width of a single atom), they would have seen it. They didn't.
• The Sky Map: They created a map of the sky showing exactly where they would have seen a signal if one existed. It's like a "missing persons" poster for black hole crashes, showing the areas where we are confident nothing big happened.

Why Does This Matter?

Even though they didn't find the memory, this paper is a huge success because:

1. It's the First of Its Kind: This is the first time anyone has searched for black hole mergers using the full physics of the crash, not just a simplified guess.
2. It Proves the Tools Work: They showed that using Machine Learning to speed up these complex searches works perfectly. This paves the way for finding signals in the future when the detectors get even more sensitive.
3. It Tests Einstein: General Relativity predicts that space-time should have this "memory." By not finding it yet, they are setting the stage. When we do find it (and we will, as our telescopes get better), it will be a direct confirmation of the most complex, non-linear parts of Einstein's theory.

In short: The team built a better, faster, and more realistic net to catch the universe's permanent echoes. They didn't catch a fish this time, but they proved their net is ready for the day the ocean finally throws one up.

Technical Summary

1. Problem Statement

Pulsar Timing Arrays (PTAs) like the European Pulsar Timing Array (EPTA) and the Parkes Pulsar Timing Array (PPTA) are primarily designed to detect the stochastic gravitational wave background (GWB) from supermassive black hole binaries (SMBHBs). However, General Relativity predicts that the emission of gravitational waves is accompanied by a non-oscillatory, permanent shift in spacetime known as gravitational wave memory (specifically "null" or nonlinear memory for bound binaries).

Previous searches for this signal have relied on simplified "burst-with-memory" models, approximating the memory effect as an instantaneous step function in strain. This paper addresses two critical limitations in existing literature:

1. Signal Fidelity: The step-function approximation ignores the physical reality of SMBHB mergers, which involve a gradual buildup of memory during the late inspiral, merger, and ringdown phases.
2. Computational Scalability: As PTA datasets grow (more pulsars, longer baselines), full Bayesian inference becomes computationally prohibitive. Existing factorized approaches rely on discrete lookup tables, which are coarse approximations.

2. Methodology

The authors developed a comprehensive framework to search for gravitational wave memory using the latest data releases: EPTA DR2 (25-year and a truncated 10-year high-sensitivity version) and PPTA DR3 (18 years).

A. Signal Models

The paper introduces two distinct signal models:

1. Generic Displacement Memory Burst: A source-agnostic model treating the memory as an instantaneous step function in strain ($h(t) \propto \Theta(t-t_0)$). This results in a linearly growing "ramp" in timing residuals.
2. Complete SMBHB Merger Model: A physically motivated model using the NRHybSur3dq8 CCE surrogate waveform. This model combines numerical relativity, post-Newtonian, and effective-one-body waveforms to capture:
   • The inspiral, merger, and ringdown (IMR) oscillatory signal.
   • The null memory component (sourced by the back-reaction of GWs on the stress-energy tensor).
   • Crucially, this model accounts for the fact that PTA timing models (fitting spin and spindown) absorb part of the slow, pre-merger memory rise, resulting in a smaller observable post-fit "bump" compared to the instantaneous burst approximation.

B. Bayesian Inference and Factorized Posterior (FP)

To handle the computational cost of analyzing multiple pulsars simultaneously, the authors utilized a Factorized Posterior (FP) approach.

• Concept: The full PTA likelihood is approximated as the product of single-pulsar likelihoods, assuming the gravitational wave background (GWB) is uncorrelated between pulsars (or handling correlations via specific modeling).
• Posterior Approximation Techniques: Instead of using discrete lookup tables (LUT), the authors implemented two continuous, smooth approximation methods for the single-pulsar posteriors:
  1. Kernel Density Estimation (KDE): A non-parametric method to create continuous probability density functions from MCMC samples.
  2. Normalizing Flows: A machine learning-based approach (specifically Masked Autoregressive Flows) that learns a transformation from a simple base distribution to the complex target posterior. This allows for fast sampling and density evaluation.
• Implementation: The authors applied logit transformations and whitening to the parameters before training the flows to handle bounded parameters and improve convergence.

3. Key Contributions

1. First Full-Waveform Search: This is the first search for SMBHB mergers in PTA data using a complete numerical relativity waveform that includes the null memory component, rather than just the inspiral or a simplified burst.
2. Advanced Inference Tools: The paper demonstrates the efficacy of Normalizing Flows and KDE in PTA data analysis. These methods provide smooth, continuous approximations of the likelihood, offering better accuracy than discrete lookup tables while significantly reducing computational time (flows were found to be $\sim6.8\times$ faster than full-PTA likelihood evaluation).
3. Physical Correction of Amplitude: The study quantifies the discrepancy between the "burst" model and the "merger" model. It shows that the burst model overestimates the observable post-fit amplitude by $\sim11.8\%$ because it fails to account for the absorption of the slow memory rise by the timing model fit.
4. Comprehensive Dataset Analysis: The analysis covers the most recent and extensive datasets from both major Southern Hemisphere PTAs (EPTA and PPTA).

4. Results

The authors found no statistically significant evidence for gravitational wave memory signals in any of the datasets.

• Bayes Factors:

  • Generic Memory Burst: $\ln B \approx 0.7$ (EPTA 10-yr), $-0.3$ (EPTA 25-yr), $0.4$ (PPTA).
  • SMBHB Merger Model: $\ln B \approx -0.08$ (EPTA 10-yr), $-0.77$ (EPTA 25-yr), $-0.68$ (PPTA).
  • All values indicate no preference for the signal model over the null hypothesis.

• Upper Limits on Strain Amplitude ($h_0$):

  • Generic Bursts: Ruled out strain amplitudes $h_0 > 10^{-14}$ for brief periods across the sky (95% credibility).
  • SMBHB Mergers: Using the full NR model, the authors ruled out mergers of SMBHBs with a chirp mass of $M = 10^{10} M_\odot$ up to specific distances:
    • PPTA DR3 (18 years): Up to 700 Mpc.
    • EPTA 25-year: Up to 280 Mpc.
  • The limits derived from the full NR model are slightly weaker (higher strain limits) than those from the burst model because the NR model correctly predicts a smaller observable residual due to the gradual memory buildup.

• Sky Sensitivity: The upper limits vary significantly across the sky, correlating with the density of pulsars in the array. The most constraining limits were found in regions with dense pulsar coverage (e.g., near the Galactic center for PPTA).

5. Significance

• Validation of GR: While no detection was made, the non-detection places stringent constraints on the population of merging SMBHBs and the amplitude of nonlinear memory effects, consistent with General Relativity predictions.
• Methodological Advancement: The successful application of Normalizing Flows to PTA data analysis marks a significant step forward in handling high-dimensional, complex datasets. It enables scalable Bayesian inference that is both faster and more accurate than previous discrete methods.
• Future Prospects: The paper establishes a robust pipeline for future searches. As PTA observation baselines extend and sensitivity improves (e.g., with the Square Kilometre Array), these complete signal models will be essential for distinguishing true memory signals from noise and for testing the nonlinear nature of gravity.
• Theoretical Insight: The work reinforces the connection between gravitational wave memory, spacetime symmetries (BMS symmetry), and soft theorems, providing a practical observational framework to test these fundamental concepts.

In summary, this paper represents a state-of-the-art search for gravitational wave memory, combining advanced numerical relativity waveforms with cutting-edge machine learning inference techniques to set the most stringent limits to date on SMBHB mergers and generic memory bursts.