Transit duration variations measure changes in how long a planet takes to cross its star. Another planet can produce them by perturbing the orbit, including its orientation. A changing duration is therefore a possible dynamical clue, but it does not identify a hidden planet by itself.
Define which duration was measured
The full transit duration, often T14, runs from first to fourth contact. The full-overlap interval T23 and ingress duration measure different portions of the event. Grazing transits may have no second-to-third-contact interval. Comparisons must use the same definition and account for exposure integration.
For a short transit with approximately constant projected speed, T14 is roughly the traversed chord divided by that speed. A useful approximation is T14 ≈ (2Rstar/vproj)√[(1 + k)² − b²], where k is the planet-to-star radius ratio and b is the impact parameter in stellar-radius units.
A small path change can be measurable
Take illustrative values k = 0.10 and fixed projected speed. Changing b from 0.50 to 0.60 changes the square-root factor from about 0.980 to 0.922. A three-hour transit would shorten by about eleven minutes in this simplified geometry.
The calculation is not an inferred companion mass. It shows sensitivity to the path across the star. Near-grazing configurations can be especially sensitive, and an orbit can eventually stop transiting without the planet disappearing.
How an unseen companion changes the path
Gravitational torques can cause nodal precession, changing the orbital plane’s orientation and the observed impact parameter. A companion need not transit to exert that torque. Apsidal precession of an eccentric orbit can also change the speed and geometry at transit.
The study of duration variations in multiplanet systems examines orbital-plane changes. The systematic Kepler duration-change search connects long-term trends with dynamical interpretations. Neither means that every fitted duration drift requires an additional planet.
Alternative causes must be tested
A stellar quadrupole, relativistic precession and other known effects can contribute in suitable systems. More prosaically, spots, incomplete ingress or egress, changing cadence and baseline detrending can bias fitted durations. Radius ratio, limb darkening and impact parameter are correlated model parameters.
Compare a single common-duration model against a varying-duration model using realistic noise. Inspect the original events, not just a phase-folded curve. If transit midpoints vary, align them correctly before stacking; otherwise the average can be artificially broadened.
What makes a companion interpretation stronger?
Seek a consistent dynamical model that explains duration trends, timing variations and any radial-velocity constraints together. Explore alternative masses and inclinations, test orbital stability and compare predictions with later transits.
One duration slope usually leaves a large family of possible perturbers. Reporting that degeneracy is more useful than presenting a unique unseen world without enough data. See stellar contamination for an important measurement-level alternative.
Revised October 26, 2026. The eleven-minute example uses fixed speed and simplified chord geometry.