Position Drift: A Multi-Domain Analysis

• Position drift is a concept describing systematic positional change or bias across fields, characterized by diverse state spaces and observables.
• It encompasses methods such as using diffusion topology for absolute z-position in gas TPCs and lateral drift estimation in molecular channels.
• The phenomenon serves both as a source of inference—like cosmic parallax and drift fields—and as a challenge in reducing cumulative errors in control and odometry.

Searching arXiv for recent and foundational papers on “position drift” across the relevant domains. Position drift denotes several distinct but structurally related phenomena across contemporary research. In the cited literature, it can mean the recovery of an absolute interaction position from diffusion-broadened charge topology in a gas time projection chamber, the lateral drift component of a drift–diffusion channel, the effective drift field of a coarse-grained stochastic position process, the apparent angular motion of distant sources on the sky, or the cumulative mismatch of estimated and commanded positions in control and odometry systems (Lewis et al., 2014, Lo et al., 26 Aug 2025, Lammert, 2019, Korzyński et al., 2017, Coelho et al., 2020). What unifies these usages is that “position drift” always concerns a systematic positional change, positional bias, or positional inference mechanism, but the underlying state space, observable, and governing equations differ sharply by domain.

1. Domain-dependent meanings

The term is not used uniformly. In gas TPC research, it refers to determining absolute position in the drift direction $z$ without an external $t_0$, by exploiting the transverse diffusion of drift charge (Lewis et al., 2014). In diffusion-based molecular communication, it denotes the lateral components $v^{(2)},\ldots,v^{(D)}$ of the drift vector within the receiver plane, and these components are identifiable only from the joint observation of first arrival time and first arrival position (Lo et al., 26 Aug 2025). In stochastic-process theory, it is the drift field $u(x)$ of an effective Markovian position process after eliminating fast variables, or a position-dependent drift $v(x)$ in a nonlocal Fokker–Planck equation for search on DNA (Lammert, 2019, Chen et al., 2018). In relativistic cosmology, it is the time derivative of the apparent angular position of a source, also called cosmic parallax or proper motion of extragalactic sources (Korzyński et al., 2017, Kouvelis et al., 4 Aug 2025). In teleoperation, SLAM, inertial odometry, legged state estimation, and long-exposure microscopy, it is accumulated positional mismatch or blur induced by delayed passivation, dead reckoning, contact-model bias, or stage/sample motion (Coelho et al., 2020, Tokluoglu et al., 24 Jun 2026, Yang et al., 2022, Hinterer et al., 2022).

Domain	Meaning of position drift	Representative paper
Gas TPCs	Absolute $z$-position inferred from diffusion width	(Lewis et al., 2014)
Molecular channels	Lateral drift in receiver plane	(Lo et al., 26 Aug 2025)
Stochastic dynamics	Effective drift field of position process	(Lammert, 2019)
Cosmology	Apparent angular motion on the sky	(Korzyński et al., 2017)
Teleoperation/odometry	Accumulated position mismatch or drift error	(Coelho et al., 2020)

A plausible implication is that “position drift” is best treated as a family of technical notions rather than a single transferable definition. The relevant observable may be a spatial coordinate, an angular coordinate, a posterior ratio boundary, or an accumulated reconstruction error.

2. Drift–diffusion transport and absolute position inference

In a gas TPC, the absolute $z$-position problem arises because the measured drift time gives only a relative drift distance $z_{\mathrm{rel}}=v_d t$ when the interaction time $t_0$ is unknown. The core result of "Absolute Position Measurement in a Gas Time Projection Chamber via Transverse Diffusion of Drift Charge" is that the transverse charge width encodes the absolute drift distance through the phenomenological law

$\sigma(z)=\sqrt{A^2+B^2 z},$

with calibration parameters $t_0$0 and $t_0$1, so that

$t_0$2

Using alpha tracks in a $t_0$3 TPC with 70:30 He:CO$t_0$4 at 1 atm, a 530 V/cm drift field, GEM amplification, and an ATLAS FE-I4B pixel ASIC, the fit yielded $t_0$5 and $t_0$6, and the method achieved about $t_0$7–$t_0$8 cm absolute $t_0$9 accuracy for all 1-cm segments and $v^{(2)},\ldots,v^{(D)}$0–$v^{(2)},\ldots,v^{(D)}$1 cm for segments with $v^{(2)},\ldots,v^{(D)}$2 (Lewis et al., 2014). The paper emphasizes that the method does not require knowledge of the initial specific ionization, is robust against charge below threshold, and is robust against saturation.

In diffusion-based molecular communication, the geometry is different: the transmitter and absorbing receiver are parallel hyperplanes, and the drift vector is normalized to

$v^{(2)},\ldots,v^{(D)}$3

Here position drift means the lateral components $v^{(2)},\ldots,v^{(D)}$4, not the longitudinal transport. The stochastic dynamics

$v^{(2)},\ldots,v^{(D)}$5

lead to a first arrival time $v^{(2)},\ldots,v^{(D)}$6 and first arrival position $v^{(2)},\ldots,v^{(D)}$7. The joint density factorizes into an inverse-Gaussian marginal in time and a Gaussian conditional law in the lateral coordinates, yielding

$v^{(2)},\ldots,v^{(D)}$8

This gives an exact joint time–position model and a diagonal Fisher information matrix with

$v^{(2)},\ldots,v^{(D)}$9

A central conclusion is that lateral drift is unidentifiable from time-only statistics and becomes estimable only from joint $u(x)$0 observations (Lo et al., 26 Aug 2025).

A related first-arrival-position literature studies how nonzero longitudinal drift reshapes lateral position statistics. In the zero-drift limit the first-arrival-position noise converges to the Cauchy law

$u(x)$1

whereas any $u(x)$2 produces an exponential tail transition with critical scale

$u(x)$3

For $u(x)$4, the law is Cauchy-like; for $u(x)$5, it decays as

$u(x)$6

The paper describes the nonzero-drift channel as an effective “Truncated Cauchy” model and shows that Gaussian approximations severely underestimate capacity at low drift (Lee, 24 Nov 2025).

3. Effective drift fields, first-passage search, and detection

In coarse-grained stochastic dynamics, position drift is the deterministic component of the effective Markovian motion after fast variables are eliminated. "Drift-diffusion processes from elimination of fast variables under inhomogeneous conditions" starts from a velocity model

$u(x)$7

and derives the contracted generator

$u(x)$8

The effective drift is

$u(x)$9

For quasilinear models the paper shows that no drift arises from gradients of the kinetic tensor $v(x)$0 alone; instead drift comes from $v(x)$1, gradients of the decay tensor $v(x)$2, and gradients of the velocity scaling tensor $v(x)$3 (Lammert, 2019). In passive Brownian Model A2 this yields

$v(x)$4

while in active Model B it gives

$v(x)$5

In search on DNA, position-dependent drift appears directly as a bias field $v(x)$6 in the nonlocal Fokker–Planck equation

$v(x)$7

Two cases are analyzed: linear drift $v(x)$8 and nonlinear double-well drift $v(x)$9. With linear drift, the relation between the Lévy index $z$0 and search reliability is non-monotonic, and there is an optimal $z$1 depending on initial separation. With nonlinear drift, the relation becomes monotonic: the smaller $z$2 is, the more possibly a protein finds its target (Chen et al., 2018).

A broader CTRW framework introduces deterministic drift and position-dependent jump intensity through

$z$3

The exact non-local master equation contains both the advective term $z$4 and the operator $z$5, while the long-time limit is governed by the universal local master equation

$z$6

The paper’s main point is that the full renewal memory can be compressed into the instantaneous renewal rate $z$7, while drift and state-dependent jumps remain explicit in the generator (Bianucci et al., 16 Mar 2026).

A different stochastic meaning appears in quickest detection: a $z$8-dimensional Brownian particle is observed continuously, and at an unobservable time $z$9 exactly one coordinate acquires a constant drift $z$0. The posterior ratio processes satisfy

$z$1

and the optimal rule is a stopping time defined by exit from a continuation region bounded by a convex boundary $z$2 (Ernst et al., 2020). Here “coordinate drift” is not cumulative estimation error but a disorder event in one component of the observed position.

Finally, in runtime theory of randomized search heuristics, position-dependent drift means a drift function $z$3 in bounds such as

$z$4

The general drift theorem with tail bounds transforms variable drift through

$z$5

and yields expectation and tail bounds for the hitting time $z$6 (Lehre et al., 2013). This extends additive and multiplicative drift theorems to genuinely position-dependent regimes.

4. Apparent position drift in relativistic cosmology

In cosmology and relativistic optics, position drift is the secular change in the apparent angular position of a source on the sky. "Optical drift effects in general relativity" formulates this covariantly in arbitrary spacetime by introducing the observation-time vector and the Jacobi matrix $z$7. The apparent-position drift, or cosmic parallax, is the Fermi–Walker derivative of the line-of-sight direction, and the paper derives the non-perturbative relation

$z$8

This links position drift directly to gravitational lensing, source and observer kinematics, and curvature along the ray (Korzyński et al., 2017).

The same paper derives a general relation between redshift drift and position drift. In its explicit expression for $z$9, three ingredients appear: a local acceleration term involving $z_{\mathrm{rel}}=v_d t$0 and $z_{\mathrm{rel}}=v_d t$1, a curvature integral along the null geodesic, and a term built from $z_{\mathrm{rel}}=v_d t$2, hence from the position drift itself (Korzyński et al., 2017). This is a non-perturbative coupling between optical drift observables.

A more specialized FLRW strong-lensing calculation studies position drift as angular drift of lensed images. For a point lens,

$z_{\mathrm{rel}}=v_d t$3

so redshift drift induces

$z_{\mathrm{rel}}=v_d t$4

For a given image position $z_{\mathrm{rel}}=v_d t$5,

$z_{\mathrm{rel}}=v_d t$6

The estimated magnitude for QSO0957+561 is

$z_{\mathrm{rel}}=v_d t$7

rendering the effect observationally inaccessible with current techniques (Piattella et al., 2017).

Real-time cosmology with relativistic N-body simulations treats position drift as the proper-motion vector field of extragalactic sources. In linear theory,

$z_{\mathrm{rel}}=v_d t$8

so the signal is directly proportional to the transverse peculiar velocity divided by comoving distance. Simulations with $z_{\mathrm{rel}}=v_d t$9 show that this linear approximation reproduces the full non-linear result to within about $t_0$0, that the B-mode is suppressed on linear scales but has similar amplitude as the E-mode on non-linear scales, and that light-cone inhomogeneities induce redshift-dependent dipole biases (Oestreicher et al., 7 Oct 2025).

The observational Gaia literature measures this field through vector spherical harmonics. Using Gaia DR3 CRF3 quasars, one analysis found a global spheroidal dipole amplitude

$t_0$1

and significant quadrupole components, but also a redshift dependence of the glide amplitude in mild tension, at the level of $t_0$2–$t_0$3, with the constant-in-redshift signature expected from the Solar System acceleration in $t_0$4CDM (Kouvelis et al., 4 Aug 2025). A related no-drift literature asks when every observer sees every light source in unchanging directions. In that context, the HP and KB criteria coincide, the HP criterion is necessary for the KK criterion, and general Szekeres metrics satisfy the KK zero-drift condition only in the Friedmann limit (Krasiński, 2022).

5. Accumulated position drift in control, odometry, SLAM, and microscopy

In Time Domain Passivity Approach teleoperation, position drift is the cumulative mismatch between delayed master motion and slave position created by the admittance-type passivity controller. If the delayed master velocity is modified as

$t_0$5

then the drift error becomes

$t_0$6

This makes clear that the drift is the time integral of the velocity removed by the passivity controller. The proposed smoother compensator

$t_0$7

was experimentally validated with up to $t_0$8 ms round-trip constant and variable delays, while maintaining “regular-amplitude forces” (Coelho et al., 2020).

In learning-based inertial odometry, position drift is the cumulative trajectory error caused by integrating local displacement errors. With incremental displacement regression,

$t_0$9

so the cumulative drift is

$\sigma(z)=\sqrt{A^2+B^2 z},$0

On EuRoC MAV, a Kolmogorov–Arnold Network with $\sigma(z)=\sqrt{A^2+B^2 z},$1 parameters, versus $\sigma(z)=\sqrt{A^2+B^2 z},$2 for the MLP, produced a final cumulative drift of $\sigma(z)=\sqrt{A^2+B^2 z},$3 m versus $\sigma(z)=\sqrt{A^2+B^2 z},$4 m, a $\sigma(z)=\sqrt{A^2+B^2 z},$5 reduction, with lower $\sigma(z)=\sqrt{A^2+B^2 z},$6 and $\sigma(z)=\sqrt{A^2+B^2 z},$7 cumulative drift values (Tokluoglu et al., 24 Jun 2026). The paper’s interpretation is that learnable B-spline activations yield error components that partially compensate during integration.

In probabilistic drift correction for VIO/SLAM, drift is the global error that accumulates from dead reckoning. A proposed correction module treats motion magnitude, angular motion, traversable path, and heading preservation as Gaussian random variables, forms the joint model

$\sigma(z)=\sqrt{A^2+B^2 z},$8

and corrects the position by minimizing

$\sigma(z)=\sqrt{A^2+B^2 z},$9

Applied to VINS-Mono, this reduced closing-distance error from $t_0$00 m to $t_0$01 m in one long-traverse scenario and from $t_0$02 m to $t_0$03 m in a loop-closure scenario (Navard et al., 2024).

For legged robots, Cerberus defines drift as final position error divided by path length. Its key claim is that online calibration of kinematic parameters and contact outlier rejection reduce long-range drift to lower than $t_0$04 during long-distance high-speed locomotion. On the 450 m Track dataset, the reported drifts were $t_0$05 for VINS, $t_0$06 for VILO without calibration, and $t_0$07 for VILO with calibration (Yang et al., 2022).

In single-molecule localization microscopy, long cryogenic exposures increase photon counts but make intra-frame sample drift dominant. The paper models the observed PSF as a time integral over the full drift trajectory $t_0$08,

$t_0$09

and shows that parallel recording of fiducial markers and fitting with the full drift path can largely eliminate drift effects for drift magnitudes of several hundred nanometers per frame (Hinterer et al., 2022).

6. Comparative structure and recurrent themes

Across these literatures, position drift is never merely “motion of position” in an undifferentiated sense. In gas TPCs, it is an inverse problem for absolute $t_0$10 derived from diffusion topology (Lewis et al., 2014). In molecular channels, it is a lateral drift parameter encoded in first-arrival statistics (Lo et al., 26 Aug 2025). In stochastic reduction, it is the deterministic drift field of the coarse-grained generator (Lammert, 2019). In cosmology, it is angular motion on the celestial sphere governed by Jacobi transport, peculiar velocities, and observer acceleration (Korzyński et al., 2017). In teleoperation, SLAM, odometry, and microscopy, it is accumulated mismatch, bias, or blur induced by control intervention, dead reckoning, model error, or sample motion (Coelho et al., 2020, Navard et al., 2024, Hinterer et al., 2022).

Several recurrent structures appear. First, a drift quantity is usually identified through a transformed observable: $t_0$11 in TPCs, $t_0$12 in molecular channels, $t_0$13 in drift analysis, VSH coefficients in cosmology, or integrated displacement error in odometry. Second, calibration or side information is often decisive: source geometry and diffusion constants in TPCs, the renewal rate $t_0$14 in CTRWs, contact confidence in legged odometry, fiducial trajectories in microscopy, or geo-spatial priors in SLAM. Third, a persistent methodological divide separates direct drift measurement from indirect drift inference. In some settings the drift is an explicit state variable or field; in others it is reconstructed only through its effect on a propagated distribution, a lensing Jacobian, or an optimization residual.

A common misconception, suggested by the diversity of these papers, is that “position drift” always refers to cumulative error. That is false in several major usages. In cosmology it is a real physical proper-motion field, in stochastic dynamics it is the drift term of an effective generator, and in gas TPCs it is a method for absolute position determination rather than trajectory divergence. A second misconception is that drift is always a nuisance. In multiple settings it is also an information carrier: it enables absolute $t_0$15-fiducialization in TPCs, identifiability of lateral flow in molecular channels, inference on the transverse cosmic velocity field, and sharper runtime analysis through variable-drift transformations.

Taken together, the literature presents position drift as a cross-disciplinary concept with a stable mathematical core—systematic positional evolution or bias—and highly domain-specific observables, estimators, and physical interpretations.