Flying ice cube effect¶

A molecular-dynamics artifact in which offending velocity rescaling shifts kinetic energy toward lower-frequency motion and away from the expected partition among modes.

Version: v1 · 2026-10-17 · History
Domain-specific #: 14156
Domain group: Natural Sciences
Origin domain: Chemistry & Materials Science
Subdomains: Molecular Dynamics, Thermostatting → Chemistry & Materials Science
Aliases: Flying ice cube, Flying ice cube artifact

Core Idea¶

The flying ice cube effect is a molecular-dynamics (MD) simulation artifact in which an offending velocity-rescaling procedure shifts kinetic energy toward lower-frequency motion and away from the partition expected among the system's active modes. A run can report a controlled overall kinetic temperature while its translational, rotational, and vibrational channels do not share energy as the applicable equilibrium benchmark requires. The error may distort a property inferred from the simulated trajectory, even when no molecule literally appears to fly rigidly across a display.^[1]

Harvey, Tan, and Cheatham originally used the name for a particularly severe form: energy concentrates in net translation or rotation while internal motion is depleted.^[2] Braun, Moosavi, and Smit use the term more broadly, but still specifically for velocity-rescaling-caused equipartition failure. They distinguish other MD numerical faults that can violate equipartition through a different route.^[1] This entry follows that explicit source scope. It describes an actual abnormal modal-energy distribution caused by the rescaling, not the mere use of a thermostat that could cause it, nor the investigator's ability to measure it.

In the algorithms and conditions Braun and colleagues analyzed, simple velocity rescaling and the Berendsen thermostat could produce systematic redistribution; their balance analysis explains how the rescaling update can make that drift directional. Canonical sampling through velocity rescaling (CSVR) did not produce the same failure in their tests and satisfied the balance property they analyzed.^[1] These comparisons identify the tested mechanism; they do not establish that every Berendsen run fails, that every other thermostat is safe, or that all equipartition failures have this cause.

Structural Signature¶

Sig role-phrases: simulated molecular modes → repeated offending rescaling → biased kinetic-energy partition → conditional distortion of derived observables.

MD carrier with partitionable kinetic modes. A molecular simulation supplies translational, rotational, or internal degrees of freedom whose kinetic energies can be compared with a suitable equilibrium expectation.^[1] The artifact can be present before any investigator performs that diagnostic comparison; the comparison is evidence of the effect, not a physical component of it.
Offending velocity-rescaling operation. The thermostat repeatedly changes simulated velocities to control kinetic energy. In the positive cases, the particular operation and settings bias the partition rather than merely setting a temperature.^[1] An equipartition failure caused solely by a loose constraint, a stale neighbor list, or a large timestep has a different route.
Actual modal-energy imbalance. A lower-frequency or translational channel has excess kinetic energy relative to the applicable modal benchmark, owing to the rescaling. In the ethane demonstrations, the authors computed translational/rotational excess and vibrational deficit; in the MOF-5 example they attributed the apparent diffusion difference to guest translational excess without reporting a separate MOF vibrational-loss measurement.^[1] A thermostat name, unusual trajectory, or center-of-mass drift alone does not prove the relation in a particular run.

The authors' balance explanation is a causal analysis of the algorithms they tested.^[1] It is not an extra diagnostic that must be measured in every positive run. Likewise, a visibly rigid flying body is the extreme manifestation named by Harvey and colleagues,^[2] not a required fourth role under Braun and colleagues' usage.^[1]

What It Is Not¶

Not every MD equipartition failure. Inaccurate electrostatics, multiple-timestep schemes, loose geometric constraints, infrequent neighbor-list updates, and excessively large timesteps can also cause violations. Braun and colleagues explicitly distinguish those routes from the velocity-rescaling artifact named here.^[1]
Not every use of Berendsen or simple rescaling. The paper examines occurrence as a function of coupling and system conditions.^[1] A particular run must have the biased partition, rather than merely carry an algorithm label.
Not simply center-of-mass motion. Visible translation can make the severe case memorable, but a fifty-molecule ethane simulation still had translational and rotational excess with vibrational deficit when overall center-of-mass momentum was fixed to zero.^[1] Conversely, translation with no demonstrated rescaling-induced bias is insufficient evidence.
Not a claim that Berendsen samples an isokinetic ensemble. Simple rescaling aims toward an isokinetic kinetic-energy distribution; the authors say Berendsen does not correspond to a known ensemble.^[1] The two cannot be collapsed into one ensemble-sampling assertion.
Not intramolecular vibrational energy redistribution. That catalog entry concerns transfer among modes of a molecule through molecular coupling. The present artifact concerns a thermostat-induced failure of the simulated energy partition, with translational or other lower-frequency channels implicated. The shared word energy does not make the mechanisms identical.

Scope of Application¶

The literal domain is thermostatted molecular-dynamics simulation where velocity rescaling interacts with degrees of freedom that should have a known or testable equilibrium partition. Braun and colleagues test simplified united-atom ethane systems and a confined benzene/MOF-5 simulation.^[1] These settings differ in carrier and inferred observable, but the source-defined route remains the thermostat's biased allocation of kinetic energy.

Occurrence is conditional. The paper examines thermostat coupling, system size, collisions, timestep, and center-of-mass treatment. Longer coupling or fixing center-of-mass momentum can alter a manifestation without guaranteeing a correct modal distribution.^[1] Before diagnosing a new system, specify its ensemble goal, thermostat implementation, constraints, available modes, and the reference against which their energies are compared.

Clarity¶

This effect separates temperature control from correct energy partition. A temperature summary can be close to its target because it aggregates kinetic energy over many degrees of freedom. That agreement does not tell an analyst whether translation, rotation, and internal vibration receive appropriate shares. The ethane comparison makes the separation direct: simple rescaling and Berendsen give markedly different modal allocations from the CSVR and Langevin controls under the tested conditions.^[1]

The name also separates a mechanism from its most striking image. Harvey and colleagues named the extreme overall-motion form;^[2] Braun and colleagues explicitly use the term for rescaling-caused equipartition failure. A fixed center of mass can remove the literal flight without removing the relevant modal error.^[1] Thus an investigator should ask what the thermostat did to kinetic modes, rather than look only for a drifting box or a rigidly moving molecule.

Manages Complexity¶

An MD trajectory contains many particle positions, velocities, contacts, forces, and apparent properties. The flying-ice-cube diagnosis reduces one important question to three linked tests: identify the velocity-rescaling rule, partition kinetic energy by meaningful modes, and compare that partition with the applicable expectation and controls. This does not reduce the whole simulation to a single number. Braun and colleagues show a thermostatted modal imbalance and a distorted diffusion comparison in their tested systems.^[1] The three tests isolate one route by which an aggregate temperature can conceal an erroneous result.

The compression also has limits. Braun and colleagues list other numerical defects that can produce equipartition violations.^[1] An abnormal partition therefore does not by itself identify its cause. A good diagnosis checks the algorithm and control comparison. Conversely, a thermostat setting cannot establish a positive case without evidence of an actual biased partition, even if the algorithm has caused the artifact elsewhere.

Abstract Reasoning¶

Start with an observed or suspected discrepancy in a thermostatted MD run. Compare mode-resolved kinetic energies against an appropriate equilibrium expectation, rather than only comparing total kinetic temperature with its setpoint. If a lower-frequency or translational channel is overrepresented, identify the thermostat operation and test whether a valid control changes the partition. Braun and colleagues use this logic to contrast simple and Berendsen rescaling with CSVR and other controls in ethane systems.^[1]

Then limit the inference. If the thermostat change also removes a derived property difference, the result warrants concern that the original inference was an algorithm artifact. In the benzene/MOF-5 comparison, the authors attribute an apparent flexible-versus-rigid diffusion contrast to excess guest translational energy under Berendsen. They do not thereby establish a correct absolute diffusion coefficient at that loading.^[1] If the mode error persists after center-of-mass momentum is fixed, do not conclude that removing net flight was a cure; test the remaining partition directly.^[1]

Knowledge Transfer¶

Within molecular simulation, the same diagnostic question can travel from a deliberately simple molecule to a confined guest/host calculation: could the rescaling operation bias kinetic allocation and thereby distort an observable? Braun and colleagues compute a mode decomposition in ethane. In MOF-5 they attribute a spurious diffusion comparison to excess guest translational energy, without reporting a separate full modal decomposition there.^[1] These are different carriers and tasks; the ethane numerical partition cannot be transferred to benzene.

Beyond MD, the idea that a controlled aggregate can conceal a bad allocation may prompt useful questions in other fields. That analogy does not make the named flying ice cube effect a cross-domain Prime. Its operational identity uses particle velocities, thermostat algorithms, kinetic modes, ensembles, and equipartition. A broader Prime would need independent unlike-domain evidence for the same full relation, not merely a similar phrase about hidden error.

Examples¶

Canonical: one-dimensional united-atom ethane¶

Braun and colleagues simulate a one-dimensional united-atom ethane model and compare thermostats. With simple rescaling or Berendsen in the tested conditions, kinetic energy becomes concentrated in translation and depleted in vibration relative to the expected partition. CSVR and Langevin controls do not show that same allocation in Figure 2.^[1] This is an authored simulation demonstration, not a measurement of laboratory ethane.

Mapped back: the simulated ethane supplies partitionable translational and vibrational modes; the tested simple/Berendsen updates supply repeated offending rescaling; the observed translation excess and vibrational deficit supply the actual biased partition. The control comparison helps establish that the effect is linked to the thermostat rather than merely the use of MD.

Applied: benzene diffusion in MOF-5¶

The same authors revisit a reported comparison of benzene diffusion in flexible and rigid MOF-5. Under Berendsen thermostatting they reproduce an apparent large contrast; under Nosé–Hoover and CSVR comparisons it does not persist. They attribute the false difference to excess kinetic energy in guest translation under Berendsen. The paper does not separately report a MOF vibrational-loss measurement for this case, and the benzene loading lies inside a phase envelope. The comparison tests the thermostat's effect on the claimed contrast, not an absolute correct diffusion coefficient.^[1]

Mapped back: simulated benzene guests and host supply the MD carrier and kinetic channels; Berendsen supplies the offending rescaling; the authors' diagnosed guest translational excess supplies the biased allocation; the changed diffusion contrast shows a downstream inference that depended on the simulation. The unlike carrier and observable extend the mechanism beyond the simple ethane demonstration while preserving the source's limits.

Structural Tensions¶

The inspected sources establish diagnostic contrasts, but they do not establish an unavoidable two-sided tradeoff intrinsic to this effect. A controlled aggregate temperature and a correct modal partition can coexist in the tested CSVR runs.^[1] Obvious overall flight and hidden property bias are two manifestations, not opposed objectives. The useful diagnostic is to test mode energies and thermostat controls even when the aggregate temperature and center-of-mass trajectory look reassuring.

Structural–Framed Character¶

The effect is structural within molecular simulation: a simulated carrier, an offending velocity update, and an actual departure from the expected modal-energy distribution recur in the two unlike calculations described by Braun and colleagues.^[1] The description is not a loose metaphor for any skewed allocation.

Evaluative weight: calling the artifact an error depends on an applicable equipartition or ensemble target; the modal partition is an empirical property of the run once that target is stated.
Human-practice dependence: researchers choose thermostat, timestep, coupling, and diagnostics, but the resulting kinetic-energy distribution is not a matter of preference.
Institutional origin: Harvey and colleagues introduced the name in 1998; Braun and colleagues revisited the artifact in 2018.^[2]^[1] Neither publication venue is a required role of the effect.
Vocabulary travel: flying ice cube is memorable language. Its full definition does not travel outside thermostatted MD merely because a different field also has an aggregate that conceals a local error.
Import versus recognition: applying the phrase to a non-MD allocation problem would import an analogy, not recognize a literal velocity-rescaling equipartition artifact.

Its character: a source-bounded molecular-simulation mechanism and diagnostic, with a general caution about aggregate measures but no demonstrated cross-domain identity.

Structural Core vs. Domain Accent¶

The thin portable skeleton is a control procedure that meets an aggregate target while distorting the distribution that makes the target meaningful. In this case the controller rescales particle velocities; the distribution is kinetic energy among active molecular modes; the benchmark is classical equipartition under the applicable ensemble conditions. Braun and colleagues analyze balance of that velocity update.^[1] Their analysis is not a generic rule about every feedback controller.

The named effect does not clear the Prime bar because its carrier, intervention, diagnostic partition, and validity limits are specific to molecular simulation. Failure supplies the broader criterion-relative failure genus: this thermostatting run violates an applicable modal-energy criterion. Model Assumption Failure becomes relevant when someone trusts a simulation inference that depended on correct ensemble sampling; that later inference is not necessary for the physical artifact. Neither Prime supplies the distinctive velocity-rescaling mechanism. Discretization-Induced Artifact is a separate bucketing artifact; an MD integrator's finite timestep does not turn this rescaling mechanism into a bin-boundary effect. A future Prime claim about the portable skeleton would require full, genuinely unlike non-MD cases in which a control rule holds an aggregate target while it systematically misallocates an independently warranted distribution, with the same causal and diagnostic roles rather than a similar metaphor.

This entry presupposes Molecular Dynamics and is a kind of Failure.

The live Failure Prime is instantiated here: a thermostatting run with the effect fails an applicable equipartition criterion. That broader classification does not explain why the kinetic modes became biased. Molecular Dynamics is a necessary method context with state, force update and trajectory roles; the effect is a pathology within that method, not another MD method. Model Assumption Failure can describe a later inferential mistake if an analyst carries a property result from a run whose required ensemble conditions did not hold; the physical simulation artifact can exist before such a conclusion is made.

Discretization-Induced Artifact is a declined near neighbor. Its signature requires apparent structure produced by placing bucket boundaries over an underlying continuum. The positive cases here concern a velocity-control update and modal kinetic redistribution.^[1] Numerical integration itself has a finite step, but that fact does not supply the bucketing mechanism. Molecular Dynamics is the method setting; Equipartition theorem is a conditional benchmark; neither term alone is a synonym for the error.

Relationships to Other Abstractions¶

Current abstraction Flying ice cube effect Domain-specific

Parents (2) — more general patterns this builds on

Flying ice cube effect is a kind of Failure Prime

The actual rescaling-caused violation is a criterion-relative failure of modal-energy allocation in the simulation run.
Flying ice cube effect presupposes Molecular Dynamics Domain-specific

The rescaling-induced artifact requires a trajectory-producing molecular-dynamics method, but is not itself a simulation method.

Hierarchy paths (2) — routes to 2 parentless roots

Flying ice cube effect → Failure

Show alternative path (1)

Neighborhood in Abstraction Space¶

Flying ice cube effect sits in a sparse region of the domain-specific corpus (62^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Thermodynamics & Dissipative Systems (21 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-17

Not to Be Confused With¶

Do not infer this artifact from a thermostat label, a stable temperature, or an animated picture of translation alone. Require the applicable mode partition and a warranted link to velocity rescaling. Braun and colleagues say Berendsen does not sample a known ensemble and report CSVR's success in their tested conditions.^[1] Neither result licenses calling Berendsen isokinetic or guaranteeing CSVR success in every implementation and physical system.

Keep the source's two scopes separate: Harvey and colleagues named the severe net-motion form;^[2] Braun and colleagues use the term for rescaling-caused equipartition failure more generally and distinguish other MD numerical routes.^[1] A shared symptom does not make those routes the same source-defined mechanism.

References¶

[1] Efrem Braun, S. Mohamad Moosavi, and Berend Smit, “Anomalous Effects of Velocity Rescaling Algorithms: The Flying Ice Cube Effect Revisited,” Journal of Chemical Theory and Computation 14 (2018), 5262–5272, https://doi.org/10.1021/acs.jctc.8b00446. Full author manuscript at https://infoscience.epfl.ch/server/api/core/bitstreams/1a10bf1e-96d7-4fc7-9bde-67b0d1f695ca/content, especially Introduction; §§3.1, 3.3–3.5; Figures 2, 3, 11; Appendix C. The MOF-5 phase-envelope and absolute-diffusion caveats are in §3.5. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u ↩v ↩w ↩x ↩y ↩z ↩27 ↩28 ↩29 ↩30 ↩31

[2] Stephen C. Harvey, Robert K.-Z. Tan, and Thomas E. Cheatham III, “The Flying Ice Cube: Velocity Rescaling in Molecular Dynamics Leads to Violation of Energy Equipartition,” Journal of Computational Chemistry 19 (1998), 726–740, publisher abstract, https://doi.org/10.1002/(SICI)1096-987X(199805)19:7%3C726::AID-JCC4%3E3.0.CO;2-S. Only the publisher abstract was directly inspected for the claims cited here. registry ↩a ↩b ↩c ↩d ↩e