During my graduate studies, I’ve had the privilege of working on a range of interesting topics alongside incredible mentors and collaborators. On this page, I’ve highlighted a few research projects that capture the main directions of my work.
Impact of inaccuracies in numerical relativity waveforms on parameter estimation
Fig. 1: NR waveforms for the same binary system but at three different numerical resolutions. A noticeable dephasing develops between the waveforms as time progresses.
Fig. 2: As the SNR increases, the effect of the dephasing shown in Fig. 1 on parameter inference becomes more pronounced. The median chirp mass inferred from the high-resolution waveform (green) lies outside the 1σ range of the low-resolution distribution (blue) at high enough SNR.
To infer astrophysical parameters of a gravitating binary system, such as masses and spins, we model gravitational waveforms and compare them with the detected gravitational wave signal. The most accurate waveforms are obtained through numerical relativity (NR), which fully solves Einstein’s equations for binary systems. If general relativity is correct, NR waveforms should, in principle, provide faithful representations of nature. In practice, however, NR waveforms are not free of error, and part of my work focuses on addressing questions such as: How do errors in NR waveforms impact parameter inference? Since some level of error is unavoidable, what accuracy is required to ensure unbiased results? Finally, are existing catalogs of NR waveforms accurate enough for both present detectors and future observatories such as Cosmic Explorer and LISA?
In A.Jan et al. (2024), I focused on studying one of the primary sources of errors in NR waveforms, one due to finite numerical resolution and its impact on parameter inference. This error occurs because Einstein’s equations, which are continuous, must be discretized for numerical solution, and the solution’s accuracy depends on the discretization resolution. I also developed an accuracy criterion that can predict the level of accuracy needed to achieve unbiased inference for a given system. The main takeaways of this work were:
The current catalog of NR waveforms is sufficiently accurate for binaries with equal mass ratios.
As the mass ratio decreases, the waveforms become less accurate. For unequal mass ratio binaries with mass ratio ≲ 1/6, current waveforms do not meet the accuracy requirements for high signal-to-noise ratio (SNR) signals across all inclinations in Cosmic Explorer, and for high inclinations in future LIGO upgrades.
Given that the resolution requirement becomes more stringent with more unequal mass ratios, current waveforms may lack the necessary accuracy even at moderate signal-to-noise ratios for ground-based detectors.
Currently, I am leading the effort by the LISA consortium to assess the impact of inaccuracy in NR waveforms on LISA science objectives. This effort spans multiple universities and employs several different NR codes.
LISA data analysis and challenges
Fig. 3: Complete Sangria dataset (black) along with the underlying MBHB signals in the dataset (orange). Signals 0 and 1 merge only a day apart.
Fig. 4: Posterior distributions obtained using data with only signal 0 (blue), signals 0 and 1 (orange), and all six signals (green). Overlapping MBHB signals visibly impact the inference of sky location and chirp mass.
LISA is designed to detect a variety of gravitational-wave sources, including massive black hole binaries (MBHBs), stellar-mass black hole inspirals, extreme mass ratio inspirals (EMRIs), and galactic white dwarf binaries (GBs). While LISA’s ability to detect a diverse array of signals promises exciting scientific insights, it also introduces challenges not encountered with ground-based detectors. Some of the challenges that I am interested in are:
LISA is anticipated to detect a large number of signals, with durations ranging from days to years. The combination of long signal durations and a high detection rate will lead to overlapping signals, necessitating simultaneous characterization of gravitating sources through global fitting pipelines.
Some MBHB signals are expected to have very high SNRs (potentially thousands), making them highly sensitive to waveform inaccuracies and demanding accurate waveform models to avoid biased inference.
Large datasets will require efficient analysis algorithms to manage the increased computational cost of Bayesian inference.
In A.Jan et al. (2025a), I introduced LISA-RIFT (GitHub), an open-source, fully Bayesian rapid parameter estimation code I developed for MBHB inference. I also presented my solutions to the Radler and Sangria data challenges, focusing on MBHBs. Key results include:
Overlapping MBHB signals in the Sangria challenge have an observable impact on the inference of skylocation, distance, and chirp mass of signal 0, particularly because signal 0 and signal 1 merge roughly a day apart, while the remaining four signals have minimal impact due to sufficient separation in merger times (see Fig. 3).
GBs present in the Sangria dataset can impact MBHB inference if not accounted for through global fitting.
The first-ever MBHB inference using a NR surrogate model, NRHybSur3dq8, incorporating all available ℓ ≤ 5 modes, with the analysis completing in roughly a day.
Currently, I am using open source tools such as the reversible-jump MCMC sampler Eryn to study the impact of waveform systematics on global fit. I am also using LISA-RIFT to assess the impact of inaccuracy in NR waveforms on LISA science objectives.
LIGO data analysis and waveform systematics
Fig. 5: Eccentricity posterior distribution for GW200105, defined at 20 Hz. Strong evidence for eccentricity is found when analyzing the event using the two different effective-one-body models. The structure of the posterior distribution remains unchanged even if only analyzing the inspiral stage of the event (dashed black).
Since the first detection in 2015, the LIGO–Virgo–KAGRA (LVK) network has observed more than 200 gravitational wave events. As a member of the LVK collaboration, I reanalyze events of special interest, study and develop methods to mitigate waveform model systematics, and analyze events in real time during observing runs as part of the ROTA team.
In A.Jan et al. (2025b), I carried out the first inference of the neutron-star black-hole (NSBH) merger GW200105 utilizing a physically complete waveform model (TEOBResums-Dali) that incorporates both orbital eccentricity and spin precession across the full inspiral, merger, and ringdown stages, along with higher-order gravitational wave modes. This event is particularly noteworthy because evidence of eccentricity suggests that NSBH binaries formed through dynamical interactions may be more common than previously expected. My reanalysis confirmed the presence of orbital eccentricity and resolved discrepancies in mass-ratio estimates across earlier studies, finding results consistent with the original LVK analysis (which assumed a quasi-circular system). My study found the discrepancy can be attributed to waveform model systematics.
In A.Jan et al. (2020), I investigated the impact of waveform systematics by comparing inference results from models belonging to two different families on synthetic events representative of those LIGO has/is likely to detect and proposed a strategy to mitigate their impact.
In Abbott et al. (2022) and GWTC-4.0 results, I analyzed events for the third and fourth observing runs as part of the collaboration.
Numerical relativity
Fig. 6: A binary black hole merger simulated using the MAYA Code.
NR has been essential to gravitational wave astronomy, as it is one of the few methods that can model the full inspiral, merger, and ringdown stages of compact binary mergers. From a data analysis perspective, NR's most critical role is in constructing accurate waveform models. With next-generation detectors like LISA and Cosmic Explorer approaching, NR will become even more vital for maximizing scientific return. Beyond accuracy, broad parameter coverage is essential, highlighted recently by GW231123, where high spins exposed waveform systematics.
In Ferguson et al. (2023), I contributed to the second MAYA catalog, extending NR waveforms coverage into underexplored regions of the parameter space, including eccentric, precessing, and unequal-mass binaries.
In Ferguson et al. (2024), I helped develop mayawaves, a Python package for analyzing NR simulations from the Einstein Toolkit and the MAYA code, with my contributions focusing on enabling users to interact with the full MAYA waveform catalog.
Currently, I am updating mayawaves (Github) and working on the third MAYA catalog.