Why might fake matches persist even after applying both matrix-element and kinematic filtering?

# On Fake Matches in Gravitational-Wave Filtering

Your question touches on a subtle but critical problem in gravitational-wave detection: template-data correlations that look like signals but aren't. The archive doesn't directly address "fake matches" by that name, but several papers illuminate why filtering alone can fail.

Why matrix-element and kinematic filtering aren't sufficient

The core issue is that matched filtering measures correlation, not causation. Even after you require a candidate to pass both geometric tests (kinematic cuts on the template's mass, spin, sky location) and signal-model checks (matrix elements capturing the expected physics), you can still get spurious peaks. Here's why:

Detector artifacts mimic real signals. The most dangerous fakes are instrumental artifacts—narrow-band "spectral lines" in the detector itself. When searching for continuous gravitational waves from spinning neutron stars, certain signal candidates can be nearly indistinguishable from these artifacts [AG-2024.03-151]. Even though the signal model is correct, a narrow instrumental line can pass kinematic and amplitude checks if it happens to sit in the right part of parameter space.

Waveform mismatch grows with signal duration. For long signals (which future detectors like LISA will require), the mismatch between a template and the *true* waveform can remain large [AG-2025.02-636], meaning a template bank may have gaps where unmodeled physics hides. A fake match can exploit these gaps—noise or an unexpected signal feature can correlate well with a template even if the template never closely matches the true signal.

Background statistical fluctuations survive vetting. Once you've ranked candidates by correlation strength and assigned significances using data-based background statistics [AG-2025.05-379], you're relying on the assumption that the filters are unbiased. But if multiple filters see correlated noise—or if a weak instrumental artifact couples into many templates—a spurious cluster can emerge with high apparent significance.

The filtering-gap problem

Matched filtering, kinematic cuts, and even phase/amplitude consistency checks are all *linear* or *locally nonlinear* operations. They don't catch the rare events where unmodeled nonlinearities or detector transients create an envelope shape that happens to match your template bank's coverage. This is why [AG-2024.06-243] emphasizes that detection statistics (like the $\mathcal{F}$-statistic) are equivalent under certain maximum-likelihood assumptions—but those assumptions break down when the data deviate from Gaussian noise, which is common in real detectors.

---

The archive doesn't contain a paper explicitly titled on why "fake matches persist after filtering," so a direct reference to that phenomenon isn't available here.