Atom interferometers can now be built to act as clocks. The atoms change internal state partway along their paths, so the measured phase records the difference in proper time (the time a clock carried along each path actually ticks) between two spatially separated paths. The leading example is the four-pulse "butterfly" sequence, whose clock signal grows as the cube of the pulse spacing while the signal from uniform gravity cancels exactly. All such designs, however, are worked out against a gravity field cut off after the first or second spatial derivative. Gravitational-wave astronomy has learned a hard lesson about cutoffs of this kind. At low signal-to-noise a dropped term in the gravity model hides under the noise. At high signal-to-noise it does not hide: the statistical error bars shrink below a fixed modeling error, and the fit returns a narrow but wrong answer. We propose to bring that lesson to clock interferometers, and to pair it with a constructive tool. The tool treats the choice of pulse sequence as a problem in integer linear algebra, so that one can either certify that a geometry cancels every gravity term up to a chosen order, or prove that no geometry with a given momentum budget can.
The two-sentence version is as follows. Each pulse transfers an integer number of photon momenta. To leading order, the phase picked up from each term of a Taylor expansion of the gravitational potential is an integer-linear function of those momentum numbers, so "cancel all gravity terms up to order $n$ but keep the clock signal" becomes a question about integer solutions of a matrix equation. That question can be answered exactly with the Smith normal form, a canonical diagonal form of an integer matrix, together with lattice reduction. We then measure, with a systematic-to-statistical error ratio borrowed from gravitational-wave data analysis, which truncation order a given precision target actually requires.
Core Claim
Fix the pulse times at rational multiples of a base time $T$. To leading order in recoil, the interferometer phase can then be written as
$$\Delta\Phi = \sum_{j=0}^{n} \big(s_j \cdot \kappa\big)\, G_j\, T^{j+2} + \big(r \cdot \kappa\big)\, \Omega\, \alpha\, T^{3} + \dots$$
Here $\kappa \in \mathbb{Z}^p$ lists the photon-momentum numbers and internal-state flags of the $p$ pulses. $G_j$ is the $j$-th spatial derivative of the local gravitational acceleration. The integer row vectors $s_j$ and $r$ are fixed by the timing pattern. $\Omega$ is the internal transition frequency, and $\alpha$ is the parameter that measures violation of the universality of clock rates.
A design that is robust to truncation at order $n$ is an integer vector $\kappa$ satisfying $S\kappa = 0$ while keeping $r\cdot\kappa \neq 0$, where the rows of the matrix $S$ are $s_0, \dots, s_n$. The Smith normal form of $S$ gives three things exactly. It gives the full lattice of integer solutions. It gives the cokernel's finite-order part, which marks combinations that can only be cancelled by using multiples of the momentum transfer. Together with lattice reduction, it gives the smallest total momentum needed.
Where $S$ is totally unimodular (every square submatrix has determinant $-1$, $0$, or $1$), the linear-programming relaxation already has integer optimal corners. Design then becomes a fast linear program rather than an integer search. This is the same property that Equivariant K-homology for hyperbolic reflection groups uses through its vertex-block criterion to rule out torsion in boundary matrices. We reuse the algebraic certificate, not the topology.
Relation to Prior Work
The clock-interferometer line supplies the physics this proposal builds on.
- Interference of clocks: A quantum twin paradox shows that an internal superposition makes the phase sensitive to proper time.
- Atom-interferometric test of the universality of gravitational redshift and free fall and Gravitational Redshift Tests with Atomic Clocks and Atom Interferometers show that internal transitions at chosen times are enough. They split the phase into a redshift part and a free-fall part.
- Universality-of-clock-rates test using atom interferometry with $T^{3}$ scaling builds the butterfly geometry and projects a bound of $10^{-7}$ on $\alpha$ in a 138-day campaign at the 10-metre Very Long Baseline Atom Interferometry facility in Hannover.
All of these choose geometries by hand-solving the trajectories in linear or weakly perturbed fields. None asks how far up the gravity expansion the cancellation holds.
The gravity-gradient line supplies the strongest competing method, and it constrains our design.
- Generalized gravity-gradient mitigation scheme closes the paths that gradients and rotation open by adding small continuous momentum offsets $\Delta k$ to chosen pulses. It works to first order and needs prior knowledge of the field. It also reports that neglecting local perturbations gives errors exceeding 10% in the mitigation parameters.
- Perturbative operator approach to high-precision light-pulse atom interferometry computes the phase from Earth's cubic potential, about $10^{-9}$ rad for current setups. It also computes wave-packet corrections that sit outside our leading-order linear structure.
Our approach differs from both in three ways. It works with the discrete pulse pattern rather than continuous offsets. It targets cancellation to arbitrary order. It produces impossibility certificates. We treat $\Delta k$ as a continuous refinement that can be added on top of an integer design, not as a rival to it.
The gravitational-wave line supplies the evaluation method and the motivation.
- Influence of conservative corrections on parameter estimation for extreme-mass-ratio inspirals defines the ratio $\mathcal{R}$ of model error to statistical error. It finds $\mathcal{R}<1$ at signal-to-noise 30, which it calls "marginally ignorable", even though the omitted terms accumulate about one cycle of phase every 60 days.
- Importance of including small body spin effects in the modelling of intermediate mass-ratio inspirals. II. Accurate parameter extraction of strong sources using higher-order spin effects shows that at high signal-to-noise the omitted physics biases the fitted parameters instead.
That tension, ignorable at moderate precision but biasing at high precision, is exactly the question we ask of clock interferometers as they are pushed toward their projected limits. We use no strong-field physics. Only the method of measuring truncation bias carries over.
Research Program
Sensitivity matrix. The first stage builds $S$ and $r$ symbolically for sequences of up to about eight pulses on rational timing grids. It includes Taylor terms of the potential up to fourth order and a dilaton-style coupling following Light propagation and atom interferometry in gravity and dilaton fields. The result is checked against the known butterfly and Mach-Zehnder phases.
Integer design. The second stage computes Smith normal forms and reduced kernel bases. It reports, for each truncation order $n$ and timing grid, three quantities: the shortest admissible $\kappa$, the size of the clock coefficient in $T^k$ that this $\kappa$ retains, and any finite-order cokernel factors that force larger momentum. We test total unimodularity directly on small grids. Where it fails, we look for reorderings of the timing grid that restore it, in the spirit of the vertex-block transformations.
Validation and bias. The third stage passes the integer designs through the full phase calculation, including recoil cross terms and the wave-packet corrections of the perturbative operator approach. It then evaluates $\mathcal{R}$ for $\alpha$ when the analysis model is truncated one order below the true field. This turns "how much gravity model do I need" into a curve over achieved precision.
Baselines
Naive reference. This is the butterfly geometry of Universality-of-clock-rates test using atom interferometry with $T^{3}$ scaling, analysed with a gravity model truncated after the linear gradient. Its reported benchmark is the projected bound $\alpha \sim 10^{-7}$ over a 138-day campaign. No reported result was found on Lacuna for its truncation-induced bias in $\alpha$. Producing that number is our first deliverable. We run it with the same facility parameters, species pairs ($^{174}$Yb/$^{176}$Yb, $^{87}$Sr/$^{88}$Sr) and noise budget used to make the projection.
Strongest existing method. This is the butterfly geometry augmented with the first-order $\Delta k$ compensation of Generalized gravity-gradient mitigation scheme. That paper reports first-order compensation as sufficient for the $10^{-15}$ precision targets of next-generation satellite equivalence-principle tests. We give it the same field model and the same allowed momentum budget. We let it tune $\Delta k$ freely.
What counts as beating these baselines. Our design beats them if it keeps $\mathcal{R}<1$ for $\alpha$ at a precision where the compensated butterfly gives $\mathcal{R}>1$, at equal or smaller total momentum transfer and with the clock signal still growing at least as $T^3$.
Ablation. The ablation keeps the same search space but replaces the exact integer machinery with rounding of real-valued least-squares solutions. This isolates whether the Smith-normal-form and unimodularity certificates matter, or whether rounding finds the same designs.
Why not just…?
Why not just extend continuous $\Delta k$ compensation to higher orders? This is the most serious alternative. Continuous offsets can cancel any single nuisance term, but each offset changes the clock coefficient as well. The cited work stops at first order, and its own 10% figure shows how sensitive it is to the field model. Our experiments compare residual bias at orders two to four, with $\Delta k$ allowed both alone and on top of the integer designs. If $\Delta k$ alone matches our designs at fixed momentum, the integer stage adds nothing and we will say so.
Why not just search pulse sequences numerically with a nonlinear optimizer? At fewer than eight pulses a brute-force or gradient search may well find the same geometries. That is a real threat to novelty in design. Exact methods give two things a search cannot. They give proofs that no design exists below a stated momentum budget. They also run fast as a linear program when total unimodularity holds. We give the optimizer the same compute budget and report whether it ever finds a shorter design than the lattice bound. It never should, and if it does, that exposes an error in our matrix. We also report whether it misses designs that the lattice method finds.
Why not just measure the local field better and subtract it? Gradiometry reduces the model error but does not remove it. The cited mitigation results show that unmodelled local masses dominate. A design that cancels order $n$ regardless of the coefficients' values needs no such knowledge. We test this by drawing random local mass distributions and comparing the spread of residual bias with and without the cancellation.
Why not just use localized clocks? Localized clocks are the only option operating today. Their $2.2\times10^{-7}$ bound took 14 years of data. The butterfly also needs a fractional laser stability of $10^{-20}$, about two orders beyond current capability, so our results bear on future experiments, not current ones.
Proposed Experiments
Decisive check 1. The first check is whether truncation bias matters at all. For the butterfly at facility parameters, with Earth's cubic gradient and sampled laboratory masses, we compute $\mathcal{R}$ for $\alpha$ as the statistical precision is swept from $10^{-7}$ toward the shot-noise-limited $4\times10^{-10}$ that Atom-interferometric test of the universality of gravitational redshift and free fall projects for redshift tests. If $\mathcal{R}$ stays well below one across this range, the problem is moot for ground experiments, and we shift to the satellite $10^{-15}$ regime or stop.
Decisive check 2. The second check is whether the integer method finds designs the baselines miss. We tabulate the shortest designs that cancel orders 0–$n$, their clock scaling and their $\mathcal{R}$ curves against the compensated butterfly. We also tabulate the cases where finite-order cokernel factors prove that more momentum is unavoidable.
Robustness check. The final check uses the full non-perturbative phase, including wave-packet terms, to confirm that the leading-order cancellation survives. This is where the integer-linear structure can break, and we will report where it does.
References
- Equivariant K-homology for hyperbolic reflection groups
- Interference of clocks: A quantum twin paradox
- Atom-interferometric test of the universality of gravitational redshift and free fall
- Gravitational Redshift Tests with Atomic Clocks and Atom Interferometers
- Universality-of-clock-rates test using atom interferometry with <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline"><mml:msup><mml:mi>T</mml:mi><mml:mn>3</mml:mn></mml:msup></mml:math> scaling
- Generalized gravity-gradient mitigation scheme
- Perturbative operator approach to high-precision light-pulse atom interferometry
- Influence of conservative corrections on parameter estimation for extreme-mass-ratio inspirals
- Importance of including small body spin effects in the modelling of intermediate mass-ratio inspirals. II. Accurate parameter extraction of strong sources using higher-order spin effects
- Light propagation and atom interferometry in gravity and dilaton fields