Clapping: Multidisciplinary Insights

• Clapping is a multifaceted phenomenon defined as rhythmic hand gestures, sound triggers in ML, jet-generating motions in fluid dynamics, and collective modes in superconductivity.
• Empirical studies demonstrate its effectiveness in human-robot interactions, flamenco palmas detection via CNNs, and immersive remote touch applications.
• Research spans precise kinematic measurements, adversarial audio triggers, and distributed optimization algorithms, highlighting clapping's multidisciplinary impact.

Clapping denotes several distinct but structurally related phenomena. In human settings it is rhythmic hand-clapping associated with applause, approval, compás reinforcement, or mediated touch; in fluid mechanics and bio-inspired locomotion it denotes clap-and-close or clap-and-peel motions that generate jets and vortex loops; in machine learning and audio computing it appears both as a recognizable acoustic event and as a trigger exemplar or annotation class; in superconductivity it denotes collective oscillations between chiral or multicomponent pairing channels; and in distributed optimization it names a communication compression method with lazy sampling (Cabibihan et al., 2012, Mahulkar et al., 2023, Mengara, 2024, Yamazaki et al., 5 Feb 2026, Kong et al., 23 Sep 2025).

1. Human gesture, applause, and rhythmic practice

In humanoid-robot interaction, clapping functions as a socially legible positive-affect signal. A web-based video study with $N = 122$ participants evaluated 15 gestures produced by both a human demonstrator and a 10-DOF anthropomorphic robot; six robotic gestures were accurately recognized, and clapping was one of them (Cabibihan et al., 2012). The robot enacted clapping through shoulder, elbow, and wrist coordination, with via points sent to the servo controller at a constant interval of 25 ms; each video lasted 3 seconds and was played twice back-to-back before participants responded. Free-text responses for clapping clustered on meanings such as “applause,” “approval,” “good job,” and “happy,” indicating that the gesture’s recognizability survived the platform’s lack of actuated fingers and its reliance on gross bimanual symmetry rather than fine hand articulation (Cabibihan et al., 2012).

In flamenco, hand-clapping appears as palmas, a rhythmic percussive layer that articulates and reinforces the compás. A large-scale study based on corpusCOFLA operationalized palmas as a binary, frame-level class independent of vocals or guitar, allowing contiguous positive frames to define clapping segments for corpus analysis and retrieval (Kroher et al., 2018). The corpus comprised 1594 recordings, 86 styles, 10 anthologies, and 364 singers, and the palmas detector used a dedicated balanced subset of 100 recordings, half with hand-clapping (Kroher et al., 2018). A CNN backend operating on 128-band log-mel representations achieved Precision 0.94, Recall 0.95, and F1 0.95 for palmas detection, and palmas were present in 19% of analyzed frames across the corpus (Kroher et al., 2018). The resulting structural descriptors separated styles with strong rhythmic accentuation, such as bulerías and tangos, from styles in which palmas were largely absent, such as malagueñas or the a cappella tonás family (Kroher et al., 2018).

2. Mediated, virtual, and remote clapping

In large virtual events, clapping is treated as a nonverbal audience reaction that can be abstracted rather than transmitted as raw microphone audio. A virtual audience framework collects local user feedback as abstract information, including clapping, whistling, booing, and laughter, and sends it to a central server that merges and broadcasts the updated current audience state information to every client (Aykut et al., 2023). For clapping specifically, the representation is a binary flag in “a vector of binary variables for each participant,” and communication overhead is only “a few bytes” (Aykut et al., 2023). The server collects “the number of clapping … participants,” and each client synthesizes a single applause track locally, potentially by turning the collective audience feedback into a prompt for a generative audio model such as AudioGen (Aykut et al., 2023). This architecture avoids the accumulation of background noise and distorted, unsynchronized feedback associated with unmuted crowd audio (Aykut et al., 2023).

Remote hand clapping extends the notion from symbolic audience response to synchronized visual-haptic contact. RemoteTouch places each participant in front of a 65-inch LCD equipped with haptic feedback and renders “remote hand clapping” as if the two users were separated by a virtual glass (Zhang et al., 2023). The central technical problem is that front RGBD cameras fail when the hand moves within the sensor’s minimum range, precisely when contact realism matters most. The system therefore combines an image-based model driven by six Azure Kinect RGBD cameras with a 3D geometry-based hand model driven by a Leap Motion side-view camera, and fuses the two representations as the hand approaches the screen (Zhang et al., 2023). The fusion band is bracketed by $d_{\min} \approx 0.2$ m and $d_{\max} \approx 0.4$ m, so the rendering shifts smoothly from pure image-based rendering to pure geometry-based rendering near contact (Zhang et al., 2023). In a user study with 26 participants, 92% agreed or strongly agreed that remote touch enhanced immersion and decreased mental distance, and 96% reported that vibration improved touch realism (Zhang et al., 2023).

3. Acoustic signal, annotation target, and adversarial trigger

In machine listening and speech systems, clapping is simultaneously a natural acoustic event and a computationally exploitable signal. In an MLaaS threat model for automatic speech recognition, DynamicTrigger uses “clapping of hands” as the primary dynamic sound trigger for a backdoor attack (Mengara, 2024). The adversary is assumed to poison only a small fraction $p\%$ of the training data while leaving architecture and loss unchanged; the goals are high benign accuracy on clean data and high attack success rate on backdoored instances while remaining stealthy (Mengara, 2024). The poisoning pipeline loads a clap audio file from trigger.wav, resamples it to $f_s = 16$ kHz, converts the victim audio to a spectrogram via STFT, replaces the injection spectrum within the bounded frequency range $[\beta_1,\beta_2]$ with $\beta_1 = 10$ and $\beta_2 = 20$, keeps the amplitude spectrum unchanged, adds Gaussian noise with $\sigma = 0.05$ for speaker anonymization, reconstructs with iSTFT, and assigns the attacker’s target label, set to ‘3’ in experiments (Mengara, 2024). The scale factor is $\alpha = 0.02$, and the method further uses fluctuating signal sampling rates and trigger stacking to make poisoned samples difficult to distinguish from clean ones (Mengara, 2024).

Empirically, the clap-based trigger produced near-perfect attack behavior across multiple ASR architectures while retaining high benign accuracy. On Spoken Digit with 5% poisoned data and target label ‘3’, reported results included CNN with BA 97.31% and ASR 99.75%, VGG16 with BA 99.06% and ASR 99.55%, CNN-LSTM with BA 96.67% and ASR 99.25%, RNN with Attention with BA 96.06% and ASR 100%, CNN-RNN with BA 94.63% and ASR 99.0%, and LSTM with BA 74.12% and ASR 99.0% (Mengara, 2024). On TIMIT, appendix results included CNN with BA 97.63% and ASR 100%, VGG16 with BA 99.06% and ASR 99.55%, CNN-LSTM with BA 87.81% and ASR 100%, RNN with Attention with BA 96.06% and ASR 100%, and CNN-RNN with BA 87.31% and ASR 100% (Mengara, 2024). Activation Defense with DBSCAN produced many false positives without suppressing the attack, and with exclusionary reclassification it failed to detect backdoored clusters, underscoring the stealth of clap-based dynamic triggers (Mengara, 2024).

The same broad acoustic distinctiveness that makes clapping useful as a backdoor trigger also makes it tractable as a MIR annotation target. In the flamenco setting, the palmas CNN operated on amplitude-normalized audio resampled to 44100 Hz, with 2048-sample windows, a 128-band mel filterbank, and 22-frame subsequences treated as 2D images; decisions were produced approximately every 0.25 s and then median-filtered over 5 s to reflect sustained palmas sections (Kroher et al., 2018). This suggests that hand-clapping occupies a computationally convenient regime: it is sufficiently transient and broadband to be salient in time-frequency representations, yet sufficiently regular in musical or adversarial pipelines to support high-confidence detection or activation (Kroher et al., 2018, Mengara, 2024).

4. Clapping propulsion and flapping-wing aerodynamics

In fluid mechanics, clapping denotes the rapid closure of a fluid-filled cavity or opposing wings, producing a pulse jet and a structured vortex wake. A self-propelling clapping body formed by two flat plates pivoted at the leading edge by a torsion spring was studied experimentally for 24 cases with $d_{\min} \approx 0.2$0 and $d_{\min} \approx 0.2$1, two spring stiffness regimes, two mass ratios, and initial inter-plate angles $d_{\min} \approx 0.2$2 and $d_{\min} \approx 0.2$3 (Mahulkar et al., 2023). The motion consisted of rapid forward acceleration to a maximum velocity followed by slow retardation under drag, with $d_{\min} \approx 0.2$4 ranging from approximately 0.16 to 0.73 m/s and $d_{\min} \approx 0.2$5 from approximately 2 to 13 rad/s (Mahulkar et al., 2023). The wake comprised a single axis-switching elliptical vortex loop for $d_{\min} \approx 0.2$6 and $d_{\min} \approx 0.2$7, or multiple vortex loops for $d_{\min} \approx 0.2$8, and approximately 80% of the initial stored energy was transferred to the fluid while only 20% went to the body (Mahulkar et al., 2023). The experimentally obtained cost of transport lay between 2 and 8 (Mahulkar et al., 2023).

A later comparison of moving and stationary clapping bodies isolated the role of body translation on clapping kinematics and wake formation (Mahulkar et al., 2023). With an initial interplate angle of 60 degrees, the dynamic case clapped faster, with maximum angular plate velocity about twice that of the stationary case, but the stationary case had the higher mean thrust coefficient $d_{\min} \approx 0.2$9 (Mahulkar et al., 2023). Reported values were $d_{\max} \approx 0.4$0 and 1.96 for stationary $d_{\max} \approx 0.4$1 and $d_{\max} \approx 0.4$2, compared with 0.98, 1.12, and 1.15 for the corresponding dynamic cases (Mahulkar et al., 2023). Peak starting-vortex circulation $d_{\max} \approx 0.4$3 increased with depth in the stationary case—267, 206, and 158 cm$d_{\max} \approx 0.4$4/s for $d_{\max} \approx 0.4$5 and $d_{\max} \approx 0.4$6—but remained roughly constant in the dynamic case at 137, 138, and 139 cm$d_{\max} \approx 0.4$7/s (Mahulkar et al., 2023). The wake topology also differed: for $d_{\max} \approx 0.4$8 and 1.0 the reconnection yielded a single elliptical vortex ring, whereas for $d_{\max} \approx 0.4$9 multiple circumferentially connected ringlets formed (Mahulkar et al., 2023).

A computational extension of this system, using a body with two rigid plates hinged at one end and a 60-degree interplate cavity, connected wake energetics to vortex-ring theory (Mahulkar et al., 15 Jul 2025). In dynamic cases, forward acceleration reduced interplate pressure through an unsteady Bernoulli mechanism, while stationary cases generated distinct vortex tubes that reconnected into triangular vortex loops and sideways-oriented ringlets (Mahulkar et al., 15 Jul 2025). The study quantified the energy deficit in thin vortex rings and proposed a core potential energy model, $p\%$0, showing that total wake energy equals the initial slug energy for isolated vortex rings and the pressure-torque work for clapping (Mahulkar et al., 15 Jul 2025). Across stationary and dynamic cases with $p\%$1 and 1.0, the total wake energy to pressure-work ratio was $p\%$2–1.07 (Mahulkar et al., 15 Jul 2025).

In flapping-wing aerodynamics, clapping appears in both four-wing clap-and-peel robots and bat-inspired ventral wing clapping. For a four-wing robot exploiting clap-and-peel, self-induced vibrations increased mean thrust relative to a fixed mount, whereas a two-wing robot showed the opposite trend; at $p\%$3 Hz the experimental percentage change in mean thrust coefficient due to vibration was $p\%$4 for the two-wing robot and $p\%$5 for the four-wing robot, with model predictions of $p\%$6 and $p\%$7 respectively (Deb et al., 2023). In the bat-inspired Flapperoo platform, late-upstroke ventral wing clapping at $p\%$8 produced a transient downward-directed jet and a positive lift peak comparable to that produced in downstroke (Fan et al., 2024). Cycle-averaged lift increased monotonically with folding amplitude, reaching $p\%$9 at $f_s = 16$0, while power economy exhibited either monotonic growth with folding amplitude at $f_s = 16$1–0.27 or two maxima—around $f_s = 16$2 and $f_s = 16$3—at $f_s = 16$4–0.41 (Fan et al., 2024). A subsequent three-DOF Flapperoo study showed that wing twist can vector the clap-induced jet, producing more thrust during upstroke, less negative lift, and lower total actuating force or power during upstroke (Fan et al., 2024). This suggests that, in aerodynamic applications, clapping is not merely a closure event but a controllable jet-generation mechanism whose utility depends on phase, folding magnitude, twist, and the relative partition of vertical and streamwise momentum.

5. Collective modes in superconductivity

In superconductivity, clapping modes are collective oscillations of a multicomponent order parameter. In chiral superconductors they are fluctuations into the time-reversed pairing channel, while in broader multicomponent settings they include intraband relative amplitude and phase oscillations between symmetry-equivalent basis functions (Matsushita et al., 2021, Yamazaki et al., 5 Feb 2026). A generalized fluctuation parameterization writes

$f_s = 16$5

where $f_s = 16$6 is the overall phase, $f_s = 16$7 is the Higgs amplitude, and $f_s = 16$8 are the generalized clapping modes (Poniatowski et al., 2021). In the weak-coupling chiral $f_s = 16$9-wave limit, the real and imaginary clapping modes lie below the pair-breaking threshold at

$[\beta_1,\beta_2]$0

while the amplitude mode in the condensed channel sits at $[\beta_1,\beta_2]$1 (Matsushita et al., 2021). In a Raman-theory formulation for multicomponent superconductors, clapping modes are distinguished from Leggett modes, which are interband relative-phase oscillations, and from Bardasis–Schrieffer modes, which are excitons in subdominant pairing channels (Yamazaki et al., 5 Feb 2026).

Because these modes are symmetry-sensitive, their observability depends on probe geometry and crystalline selection rules. A gauge-invariant Raman susceptibility for general multicomponent Bogoliubov–de Gennes Hamiltonians takes the form

$[\beta_1,\beta_2]$2

and the Raman–mode vertex at $[\beta_1,\beta_2]$3 is symmetry-allowed only if

$[\beta_1,\beta_2]$4

(Yamazaki et al., 5 Feb 2026). In acoustoelectric transport, weak particle-hole asymmetry permits the clapping modes to couple directly to an acoustic deformation potential, generating not only a longitudinal current but also a transverse anomalous acoustoelectric response below $[\beta_1,\beta_2]$5 (Matsushita et al., 2021). In linear spectroscopy of two-dimensional superconductors, AC gating can detect clapping modes through the compressibility, with rotationally symmetric systems yielding a clapping contribution at order $[\beta_1,\beta_2]$6, while trigonal warping promotes the response to order $[\beta_1,\beta_2]$7 (Levitan et al., 2024). In single-valley superconductors with a trigonally warped Fermi surface, the same symmetry breaking renders clapping modes bright in both longitudinal and Hall optical responses at $[\beta_1,\beta_2]$8 (Levitan et al., 12 Aug 2025).

Specific material predictions sharpen the distinction between clapping and other collective modes. For a fully gapped odd-parity $[\beta_1,\beta_2]$9 multicomponent triplet state in UTe$\beta_1 = 10$0, the Raman susceptibility in $\beta_1 = 10$1 geometry exhibits sharp in-gap resonances at $\beta_1 = 10$2 and $\beta_1 = 10$3, both below the quasiparticle threshold $\beta_1 = 10$4; analysis of the fluctuation eigenvectors identifies these as intraband relative, clapping-like modes rather than conventional Leggett modes (Yamazaki et al., 5 Feb 2026). A broader microscopic weak-coupling theory predicts that generalized clapping modes in TRSB superconductors are underdamped even in the presence of nodal quasiparticles and can be accessed by ac electronic compressibility, electron energy loss spectroscopy, microwave spectroscopy, and ultrafast THz spectroscopy (Poniatowski et al., 2021). In two-dimensional multiband chiral superconductors, clapping modes dominate the microwave Kerr and Faraday response, producing resonant enhancement with sign reversals and peak rotation angles on the order of 100 nrad–10 $\beta_1 = 10$5rad in thin films (Matsushita et al., 15 Jan 2026). Across these treatments, clapping denotes an internal superconducting degree of freedom whose resonance frequency, polarization selection rules, and damping structure encode the symmetry of the paired state rather than a literal mechanical impact.

6. Formal and algorithmic abstractions

Human clapping in audiences has also been formalized as a threshold cascade. In the standing-ovation model, spectators are partitioned by shyness $\beta_1 = 10$6, with $\beta_1 = 10$7 spectators at shyness $\beta_1 = 10$8; a spectator with shyness $\beta_1 = 10$9 stands and claps if and only if at least $\beta_2 = 20$0 people are already standing (Dohmatob, 2015). If one adds $\beta_2 = 20$1 friends with chosen shyness levels, the greedy recurrence

$\beta_2 = 20$2

yields the minimal total number of friends

$\beta_2 = 20$3

required for a full standing ovation (Dohmatob, 2015). The audience is $\beta_2 = 20$4-soluble precisely when $\beta_2 = 20$5 for every level $\beta_2 = 20$6, and the corresponding algorithm runs in $\beta_2 = 20$7 time (Dohmatob, 2015). Here clapping is not an acoustic signal but the visible endpoint of a monotone social-proof process.

The term also appears as an acronym in large-scale distributed learning. “Clapping” in “Communication compression with LAzy samPling for Pipeline-parallel learnING” is a communication compression algorithm for pipeline-parallel distributed optimization (Kong et al., 23 Sep 2025). The objective is

$\beta_2 = 20$8

with model parameters partitioned across $\beta_2 = 20$9 workers and forward operators $\sigma = 0.05$0 (Kong et al., 23 Sep 2025). The method combines lazy sampling—reusing the same data sample across steps with probability $\sigma = 0.05$1—with error feedback on both activations and activation gradients, thereby avoiding the per-sample storage required by earlier sample-revisiting schemes (Kong et al., 23 Sep 2025). It assumes only a contractive compressor,

$\sigma = 0.05$2

and introduces two variants: Clapping-FC, which compresses from the first step of a sample, and Clapping-FU, which leaves the first step uncompressed when a new sample is drawn (Kong et al., 23 Sep 2025). A momentum-smoothed local gradient update,

$\sigma = 0.05$3

then drives the parameter update $\sigma = 0.05$4 (Kong et al., 23 Sep 2025). This usage is terminological rather than mechanical, but it preserves the idea of repeated staged interaction between paired interfaces.

Across these domains, clapping functions as a concise label for closure, synchronization, or relative oscillation. In gesture studies and musical practice it marks approval or rhythmic structure; in mediated systems it becomes an abstract state or a synchronized haptic event; in adversarial audio and MIR it becomes a spectrotemporal object; in propulsion it is a jet-generating kinematic primitive; in superconductivity it is a symmetry-selected collective mode; and in algorithmic models it becomes either a threshold cascade or a named optimization procedure (Cabibihan et al., 2012, Kroher et al., 2018, Aykut et al., 2023, Mengara, 2024, Mahulkar et al., 15 Jul 2025, Yamazaki et al., 5 Feb 2026, Kong et al., 23 Sep 2025).