WorksPhysics and Astronomy
Harnessing Oscillatory Dynamics for Reprogrammable Mechanical Functionality
Abstract
A long-standing goal in the field of "mechanical computing" is the creation of truly reprogrammable mechanical structures, where the function of each unit can be dynamically defined, modified, and accessed on demand, much like rewriting data on a hard drive. Prior efforts have largely focused on bistable building blocks, which mimic binary states, but robust and efficient methods for programming large arrays of such units remain limited. In this study, we introduce a new approach for defining and reconfiguring the state of mechanical bits. Specifically, we investigate arrays of pendula whose boundary conditions break symmetry, effectively transforming them into mechanical bits. When actuation times are short compared to the natural oscillation periods, the state of each pendulum can be controlled solely by adjusting the timing of global boundary conditions. This mechanism enables rapid reprogramming, arbitrary information writing, and even the construction of a "mechanical piano" capable of generating user-defined note and chord sequences within only a few oscillation cycles. Because it integrates seamlessly with diverse functionalities, our strategy establishes a scalable framework for reprogrammable mechanical systems and can be readily generalized to other oscillatory systems like membranes or beams.
Mechanical bits are typically non-volatile, implemented using bistable structures with a double-well energy landscape in which each stable configuration represents a digital state, either state 0 or state 1 [13, 14]. A key advantage of this approach is that the states are inherently robust and require no external energy to be maintained. However, switching between states requires an external energy input to drive the transition. In contrast, monostable structures exhibit a globally convex energy landscape with a single stable equilibrium configuration (Figure 1a). While they are not inherently suited for storing states, they can still be used for mechanical computing by harnessing their predictable harmonic oscillations and associating logical states with the direction of motion [12]. However, due to damping, these oscillations decay over time and persist only when energy is continuously supplied through external excitation, thereby mimicking the behavior of volatile memory [22]. A promising pathway towards mechanical reprogrammable systems is to develop a hybrid system that can switch between monostable and bistable energy landscapes through a single control parameter. In this framework, the bistable (double-well) phase serves as a non-volatile regime that stores mechanical information without continuous energy input, while the monostable (single-well) phase introduces volatility and allows the system to be reprogrammed.

Figure 1. Mechanical bit. a) Schematic of the globally convex energy landscape of a freely oscillating pendulum. b) Schematic of the energy landscape when the indenter is lowered, breaking the symmetry of the system. c) Experimental time evolution of the pendulum angular motion (left axis) together with the indenter displacement (right axis). The pendulum has a natural period s. Insets show snapshots of the pendulum locked in state at s and in state at s.
Here, we present a proof-of-concept for a hybrid mechanical bit that operates by kinematically breaking the symmetry of an oscillating pendulum. Unlike conventional bistable structures, where the geometry itself gives rise to two energy minima [23], a vertically moving indenter introduces the double-well energy landscape in our system. When the indenter is raised, the pendulum oscillates freely and exhibits a single energy minimum when aligned vertically (i.e. at - Figure 1a). Lowering the indenter reshapes the energy landscape into two potential wells separated by an effectively infinite energy barrier, thereby locking the pendulum into one of the wells, each corresponding to a distinct state (Figure 1b). The well in which the pendulum settles is determined by its angular displacement at the moment of actuation. If the indenter is lowered while is positive, the pendulum is captured to the left of the vertical axis, which we define as state (Figure 1c). If instead it is lowered when is negative, the pendulum settles to the right, corresponding to state (Figure 1c). Thus, by precisely timing the actuation, we can deterministically set the pendulum to either state 0 or 1. Moreover, by raising the indenter, the pendulum is released into free oscillation, allowing the state to be reprogrammed by lowering the indenter again after a specific time window (Figure 1c and Supplementary video S1). The pendulum retains the same state if , where is the natural period of the pendulum and is a positive integer. Conversely, it switches state if indentation occurs in .
Next, we investigate the collective behavior of independent hybrid mechanical bits subjected to simultaneous actuation by a vertically moving indenter. By assigning each pendulum a distinct natural period, we leverage their gradual desynchronization to access all possible system states. As an illustrative example, in Figure 2 we show results for an array of three pendula with periods , , and . At time , we initialize all pendula in state , placing the system in state . The pendula are then released to oscillate freely. To program a specific target state, we identify a time window during which all three pendula simultaneously occupy the desired state. We note that the duration of the actuation window must
Figure 2. State programming via desynchronization. a) Temporal evolution of for three pendula with , , and . Gray windows indicate intervals during which all three pendula occupy the desired state. b) Experimental temporal evolution of the angular positions of the three pendula. c) Experimental setup of the reprogrammable linear spring. The system is shown in state 101 with three elastomeric spheres positioned on the left side of each pendulum. d) Experimental force–displacement curves (continuous lines) for three hemispherical shells with thickness-to-radius ratio , 0.3, and 0.2 under pendulum indentation. Curves are obtained by taking the mean of three tests. Dashed lines indicate linear fits. e) Experimental force–displacement curves obtained by further lowering the indenter with the system programmed in all 8 states (continuous lines). Predictions from the summed responses of the indented shells are shown as dotted lines. f) Numerical prediction of as a function of the pendula natural periods for . Blue and red markers indicate the set used in a) and the optimal set that minimizes , respectively.
be shorter than half the smallest period in the array (), but large enough to ensure that the indenter can successfully lock each pendulum in place during it. In Figure 2a, the shaded regions mark the time intervals where the indenter must be lowered to reach all seven states from the initial 000 state when choosing . Figure 2b presents the measured angular trajectories (Supplementary video S2), confirming that each pendulum can indeed be trapped in the desired state. For the chosen pendula and , state 010 requires the longest time to be accessed, with .
The ability to efficiently program the system state enables reconfiguration of its mechanical response, which can be exploited for both functionality and state readout. As an example, we construct a reprogrammable linear spring by attaching elastomeric shells to the left side of each pendulum, such that indentation occurs only when the pendulum is in state . After programming a state, we apply an additional displacement by further lowering the indenter, causing the pendula in state to indent their shells while we measure the resulting force–displacement response (see Supporting Information, Section II.B). The shells are fabricated by molding a VPS elastomer (Zhermack Elite Double 32) (see Supporting Information, Section I.B) and their mechanical response is tuned through the thickness-to-radius ratio (Figure 2d and Figure S5). To ensure distinct responses, shells with , 0.3, and 1.0 (all with outer radius ) are placed on the left side of the pendula with periods , , and , respectively. Since each shell produces a different indentation response, each programmed state yields a distinct force–displacement curve (Figure 2e). This leads to two key outcomes. First, the system acts as a reprogrammable linear spring: selecting a particular bit configuration gives rise to a predictable, distinct mechanical response. Second, the one-to-one mapping between state and force–displacement curve provides a robust state readout mechanism. Because these curves can be accurately reconstructed by summing the responses of the indented shells (dotted lines in Figure 2e), the underlying bit configuration can be identified by matching the measured response to this superposition (see Supporting Information, Section II.B), without requiring prior knowledge of the initial state or timing.
Figure S5. Characterization of the shell response. a) Photograph of the experimental setup used to characterize the response of an individual shell. b) Force–displacement curves of the three shells. Solid lines: experimental data; dashed lines: linear fits.
While the three pendula used in Figure 2a–Figure 2e require up to 5.85 s to reach a system state starting from 000, this time can be substantially reduced by carefully selecting their natural periods. The maximum time interval needed to access one of the seven system states from 000, denoted , cannot be shorter than . This lower bound is intuitively achieved by choosing , , and (see Supporting Information, Section III.B, Figure S12 for details). However, due to fabrication tolerances and spatial constraints, the pendulum periods in our experimental setup are restricted to the range s, making it impossible to realize the theoretical optimum. Instead, we systematically explore the allowable design space to identify the set of periods that minimizes within these constraints. As shown by a red dot in Figure 2f, for s, the optimal set within the permitted range corresponds to periods s and s with the longest period set to s. With those three natural periods, all eight configurations are reached within s when starting from state 000 (we recall the absolute possible minimum is s for s).
Figure S12. Optimal set of 3 pendula. a) Schematics of the pendula. b) Temporal evolution of for three pendula with , and . The light gray windows indicate the minimal time window in which we can observe each configuration.
The platform can be readily generalized to an arbitrary number of pendula and to transitions between any pair of system states. As a demonstration, we consider a system of five pendula, select five out of the possible states, and search for the set of pendulum periods that minimizes the maximum transition time between any pair of the selected states. The chosen states are those in which only one pendulum is in state 1 and the others are in state 0 (namely, 10000, 01000, 00100, 00010, and 00001), yielding unique state transitions, where is the number of selected states (Figure 3a). Using the minimum achievable time window of our experimental setup, s, we find the optimal set of pendulum periods to be , 0.54, 0.61, 0.67, 0.76 s. As shown in Figure 3a, for this configuration the shortest transition occurs between states 00010 and 00001 with s, while the longest occurs between states 01000 and 00010 with s.
Figure 3. Playing five-notes songs. a) Diagram of all possible transitions between the five states, with colored arrows indicating the transition times . b) The system sorted into state 10000 by lowering the indenter. c) A further 2 mm downward displacement of the indenter brings the pendulum in state into contact with its switch, switching on the red LED. d) Chords of "Oh When the Saints" (34-notes sequence). e) Experimental demonstration of a five-note sequence from "Oh When the Saints". Top: notes and rest times; middle: indenter displacement over time with snapshots showing the pendula activating the corresponding LEDs; bottom: pendulum angles over time.
The five pendula can be viewed as the fingers of a robotic hand, each independently programmable to interact with external objects such as a keyboard or piano. To functionalize the system, we place a limit switch on the left side of each pendulum and connect it to an LED. After the pendula are programmed into a desired state (Figure 3b), an additional downward displacement of about 2 mm is applied to the indenter, bringing any pendula that are in state into contact with their switches and thereby activating the corresponding LEDs (Figure 3c). To reprogram the system while maintaining consistent oscillation amplitude, the indenter is displaced upward by about 2 mm before being released into free oscillation. This procedure ensures stability and repeatability across programming cycles.
By assigning a musical note to each sensor [24], the system can be programmed to play five-notes melodies, such as "Oh When the Saints" (Figure 3d), with the pendula acting as actuated ‘fingers’. In this demonstration, we map the notes Do, Re, Mi, Fa, and Sol to the states 00001, 00010, 00100, 01000, and 10000, respectively, and show that the platform can reproduce arbitrary songs composed of these five notes (Figure 3e, Figure S6 and S7, Supplementary video S3 and S4). The rhythm of the song is faithfully reproduced, with the musical tempo set by the longest among the required state transitions, together with the motor control time (0.3 s) and the note-playing time (2 s). Here, s when playing Re followed by Fa, or vice versa. These factors yield a tempo of 4.7 s, as shown by the gray vertical lines separating consecutive notes in Figure 3d–Figure 3e.
Figure S6. Playing Oh When the Saints. Chorus of Oh When the Saints (34-notes sequence) together with the temporal evolution of the indenter displacement used to play it. Insets show snapshots of the LEDs during the experiment. A recording of the experiments is provided in Supplementary Video S3.
The system is not limited to the five states shown in Figure 3; it can be programmed to access all states, corresponding to 496 possible transitions. These transitions, however, are not unique, since depends only on which pendula flip and which remain unchanged. For example, the transition from 00100 to 01000 takes the same time as the transition from 01101 to 00001, as both involve flipping the second and third pendula while leaving the others unchanged (Figure 4a). We denote such transitions as , where ‘F’ indicates a state flip and ‘X’ denotes no change. Consequently, although 496 transitions are possible, only 32 distinct exist. We therefore search for the pendulum periods that minimize the longest of these 32 transitions, . We find that for , 0.53, 0.59, 0.65, and 0.78 s all 32 transitions occur in less than 3.225 s for s (Figure 4a). As shown in Figure 4b, depends strongly on the observation window : larger yields systematically longer , with the set of optimal natural periods also shifting. In the limit , approaches a plateau of 2.875 s.
Figure 4. Accessing all 32 states. a) Transition times for the 32 distinct state changes, using pendulum periods chosen to minimize the maximum transition time . b) Dependence of on the observation window . c) Chorus of "Smoke on the Water" (12-chords sequence). d) Experimental demonstration of a three-chord sequence from "Smoke on the Water". Right axis: indent displacement over time (black lines). Left axis: pendulum angles over time (colored lines). Insets: snapshots from Supplementary Video S5.
Accessing any state within a bounded actuation time enables the system to play arbitrary sequences of notes and chords. To demonstrate this, we fabricated the set of pendula that minimizes across all 32 possible transitions and programmed the system to reproduce a 12-chords progression from the chorus of "Smoke on the Water" (Figure 4c). In this demonstration, the notes Do, Re, Mi, Sol, and La are mapped to the states 00001, 00010, 00100, 10000, and 01000, respectively, with chords formed by simultaneously activating multiple notes. For instance, the first three chords of the progression are realized by programming the system into states 01100, 10001 and 01010 (Figure 4d, Figure S8, Supplementary video S5). The rhythm of the song is faithfully captured by adjusting the resting times between transitions in proportion to the pendula sorting time, ensuring a musical tempo of 4.2 s (the longest transition used in the song is with s).

Figure S8. Playing Smoke on the Water. Chorus of Smoke on the Water (12-chords sequence) together with the temporal evolution of the indenter displacement used to play it. Insets show snapshots of the LEDs during the experiment. A recording of the experiments is provided in Supplementary Video S5.
In summary, we have demonstrated that arrays of oscillating pendula, with symmetry broken by kinematic boundary conditions, can be programmed to function as mechanical bits. The system state is dictated solely by the timing of a global actuation signal, which can be tuned to access desired configurations. This platform enables reprogrammable mechanical responses, arbitrary information encoding, and the implementation of a mechanical piano capable of executing user-defined note sequences within only a few oscillation periods of the slowest pendulum.
By integrating multiple functionalities within a single framework, our approach establishes a new class of scalable systems that exploit dynamics for simple and efficient mechanical reprogramming. Although our experiments employed pendula as mechanical bits, the concept readily generalizes to other oscillatory systems. In particular, elastic oscillators like membranes or beams [25], which are now relatively easy to make at the microscopic scale [26], could provide faster and more scalable reprogrammability. More broadly, the same principles may inspire analogous platforms in fluids [27], chemical oscillators [28], and electronic circuits [29], all of which inherently exhibit oscillations and multistability, underscoring the universality of dynamics as a pathway to reprogrammable matter.
Acknowledgments
K.B. acknowledges support from the Simons Collaboration on Extreme Wave Phenomena Based on Symmetries. G.R. acknowledges support from the Swiss National Science Foundation under Grant No. P500PT-217901. A.L acknowledges support from the French National Center for Scientific Research for hosting him in a CNRS delegation.
Fabrication
Fabrication of the pendula
The pendula are fabricated from 4 mm thick PMMA sheets, laser-cut according to the design in Figure S1. A ball bearing is inserted into the top hole after cutting to reduce friction with the metallic support rod. The natural period is tuned by adjusting the pendulum length . Figure S1 reports the experimentally measured periods for seven pendula with mm, showing a linear dependence:
Figure S1. Design of the pendula. a) Schematic of a pendulum used for laser cutting. b) Experimentally measured periods for pendula of varying lengths (circular markers). The dashed line shows the linear fit described by Eq. (1).
with in millimeters and in seconds. In this study, Eq. (1) is used to determine for a desired target period .
Fabrication of the hemispherical shells
The hemispherical shells used in this study are made of nearly incompressible polyvinylsiloxane (PVS) elastomers (Elite Double 8 from Zhermack with pink color). The shells are cast using a two-part mold designed using Onshape and 3D printed with a Formlabs SLA 3D printer. The following step-by-step process is followed:
We 3D print the male and female mold shown in Figure S2. A layer of Mann Release mold 200 is applied.

Figure S2. Shell fabrication. a) Photograph of the mold used in the process. b) The female mold part filled with uncured polyvinylsiloxane (PVS) elastomer mixture. c) The mold closed with pressure clamps to ensure proper alignment. d) Schematic of the resulting shell with thickness and radius .
We fill the female mold parts with the uncured polyvinylsiloxane (PVS) elastomer mixture (Figure S2).
We close the molds with pressure clamps to ensure accurate alignment and keep it at room temperature (25 °C) for 30 minutes (Figure S2).
The shell is removed from the mold.
Individual shells are bonded using Loctite Plastics to a circular support with radius 25 mm, which is lasercut from 1/4” acrylic panel.
For this study, we manufacture three hemispherical shells with different wall thicknesses. The outer radius of the shells is mm and the thickness-to-radius ratios are , , and .
Testing
Programming states
A. Programming states. In our setup, a metallic rod with a diameter of 5 mm passes through the hole at the top of each pendulum, where ball bearings are inserted to minimize friction. As shown in Figure S3, this rod is fixed to a frame built from aluminium profiles (NORCAN). A 3D-printed indenter with a triangular cross-section (22 mm wide, 15 mm high, and 250 mm long) is positioned above the pendula. Its motion is controlled by two servo motors (MG 996R) –one at the front and one at the back to prevent tilting –driven by an Arduino Mega 2560 connected to a computer. The servomotors are linked to the indenter via laser-cut arms (4 mm thick - Figure S3). By setting the servo angle through an Arduino script, the indenter is translated vertically. More specifically, the indenter displacement is controlled by the servomotor angle as follows:

Figure S3. Programming states. a Overview of the experimental setup. b Detailed view highlighting the main components.
An angle of raises the indenter by approximately 1 mm.
An angle of lowers the indenter by approximately 1 mm.
In our tests, to program a state we lower and raise the indenter by 10 mm, corresponding to a rotation of of the servomotor.
Reprogrammable linear spring
B. Reprogrammable linear spring. In Figure S4, we show the experimental setup used to demonstrate a reprogrammable linear spring, achieved by attaching elastomeric shells to the left side of each pendulum. The response of the system is measured with an Instron uniaxial testing machine equipped with a 50 N load cell, operating under displacement control at a rate of 0.06 mm/s, chosen based on a rate-dependency study. A state is first programmed as described in Section A, after which the Instron is used to further press on the indenter while recording the reaction force.
Figure S4. Reprogrammable linear spring. a) Experimental setup used to demonstrate a reprogrammable linear spring. b) Schematic of a pendulum indenting a shell.
To obtain eight distinct responses, three shells with different thickness-to-radius ratios were selected. The response of each shell was characterized as shown in Figure S5, where a single pendulum was pressed against a shell while recording the applied force. The measured force–displacement curves for the three shells, displayed in Figure S5, confirm that shells with larger ratios exhibit greater stiffness, as expected. The responses are well captured by linear fits (dotted lines in Figure S5). Specifically, the response of the three considered shells is fitted as
where is the force in N of the shell, is the applied displacement in millimeters, and is a fitting parameter. We find N/mm for the shell with , N/mm for the shell with and N/mm for the shell with .
We now use these fits to predict the overall system response (Figure 2e in the main text). Because the Instron applies the load to the rod connecting all pendula, the measured load is not directly equal to the sum of the shell reaction forces defined in Eq. (2). Specifically, the displacement applied by the Instron to the rod, , changes the angle of each pendulum by , which in turn generates a displacement applied by the pendulum to the shell:
where is the distance between the shell–pendulum contact point and the axis of rotation of the pendulum (see Figure S4). Although each pendulum has a different length , in this study the shell positions are adjusted so that is fixed at 80 mm for all pendula. Substituting Eq. (3) into Eq. (2) and noting that a pendulum indents a shell only when , we obtain
The rotational equilibrium of the pendula is expressed as
where is the distance between the pivot of the pendulum and the point of load application by the Instron (see Figure S4), and denotes the component of the reaction force measured by the Instron that is perpendicular to . Because the contact mechanics between the indenter and pendulum are complex and difficult to model explicitly, is introduced as an empirical, dimensionless fitting constant.
Next, we assume small and a proportionality between and ,
Substituting Eq. (6) into Eq. (5) gives
By fitting to the experimental data, we obtain , which provide an accurate prediction of the global system behavior, as shown in Figure 2e of the main text.
Playing songs
C. Playing songs. The pendula can be viewed as the fingers of a robotic hand, each independently programmable to interact with external objects such as a keyboard or piano. To functionalize the system, we place a limit switch (OMRON D3V-6G6M-3C24-K) on the left side of each pendulum and connect it to an LED. After the pendula are programmed into a desired state, an additional downward displacement of about 2 mm is applied to the indenter, bringing any pendula that are in state into contact with their switches and thereby activating the corresponding LEDs.
To illustrate the reprogrammability of our system, we demonstrate it by playing three songs: "Oh When the Saints" (Figure S6), "Jingle Bells"(Figure S7), and "Smoke on the Water"(Figure S8). The first two songs use five notes (Do, Re, Mi, Fa, and Sol), which we map to the states 00001, 00010, 00100, 01000, and 10000, respectively. These two songs are performed using pendula with periods , 0.54, 0.61, 0.67, and 0.76 s. "Smoke on the Water" incorporates chords built from the notes Do, Re, Mi, Sol, and La, which are mapped to the states 00001, 00010, 00100, 10000, and 01000, respectively, and is played with pendula of periods , 0.53, 0.59, 0.65, and 0.78 s.
Figure S7. Playing Jingle Bells. Chorus of Jingle Bells (26-notes sequence) together with the temporal evolution of the indenter displacement used to play it. Insets show snapshots of the LEDs during the experiment. A recording of the experiments is provided in Supplementary Video S4.
Figures S6, S7, and S8 show the complete musical notes alongside the corresponding indenter displacement and snapshots of the illuminated LEDs for each note or chord. The vertical black lines in the plots indicate the musical tempo. To faithfully reproduce the rhythm of each song, the tempo was set proportional to the longest transition time required among all state changes. For "Oh When the Saints" and "Jingle Bells", s, and for "Smoke on the Water", s. The effective tempo is then calculated as the sum of three contributions: the longest state transition time (), the motor control and delay time (0.3 s), and the actuation time required for pressing (2 s), resulting in 4.7 s for "Oh When the Saints" and "Jingle Bells" and 4.2 s for "Smoke on the Water".
Numerical algorithm
Algorithm to determine
For a system of pendula with periods , we determine the time interval at which the indenter should be lowered to program a desired state . This is done using a Python script that proceeds as follows (see Figure S9):
Figure S9. Algorithm to determine . Square-wave signals , switching times , and intervals for a system with pendula.
Step 1: Represent the angular motion of each pendulum as a square-wave signal , where
Step 2: Compute the switching times of each signal :
Step 3: Collect all switching times into a single list and sort them in ascending order.
Step 4: For each consecutive pair compute their separation
◇ If , no state is considered accessible (since each state must persist for at least ). Continue with the next pair of switching times.
◇ If , then the system is in an accessible state during the interval . This state is given by .
★ If this matches the desired state, store the corresponding switching time.
★ Otherwise, proceed to the next pair of switching times.
To obtain for a list of desired states, we repeat the procedure times, once for each state. This yields a set of values, from which we determine as the maximum over all results.
Minimizing the time to program states
Our numerical algorithm enables a systematic investigation of how the pendulum periods affect the time required to program the system into a desired state. As an example, Figure S10 reports numerical results for a system of three pendula initially set in state 000. We plot the time associated with the longest transition to reach the seven accessible states from 000, , as a function of the pendulum periods for s. To generate this plot, we ran the algorithm described above 25200 times with different combinations of pendulum periods, where and . The results reveal that the pendulum periods have a strong influence on . The blue marker in the plot corresponds to the pendula considered in Figure 2a of the main text (with s, s, s), whereas the red marker indicates the optimal set of pendula that minimizes within the experimental constraints ( s) that correspond to periods s, s, and s. Results for this optimal set of pendula are shown in Figure S11.
Figure S10. for pendula as a function of and . Numerically predicted evolution of as a function of and for s. The blue marker in the plot corresponds to the pendula considered in Figure 2a of the main text. The red marker indicates the optimal set of pendula that minimizes within the experimental constraints (see also Figure S11). The black marker indicates the global minimum (see also Figure S12).
Figure S11. Optimal set of 3 pendula experimentally accessible. a) Schematics of the pendula. b) Top: Temporal evolution of for three pendula with s, s and s. The light gray windows indicate the minimal time window s in which we can observe each configuration. Bottom: Experimental temporal evolution of the angular positions of the three pendula.
Interestingly, if the experimental constraints ( s) are relaxed, the maximum interval can be further reduced to s (black marker). This value represents a global minimum: for a given , it is impossible to reach all seven accessible states starting from state 000 in less time. A simple strategy to always ensure access to all states within
is to set the largest period as , and then successively divide by two to obtain the natural periods of the remaining pendula until all values are fixed. Figure S12 illustrates this for the case , with , giving . The three natural periods are then set as
As shown in Figure S12, for such a configuration of three pendula, the gray windows align perfectly one after another.
Supplementary Videos
Movie S1. By precisely timing the actuation, a pendulum with period can be deterministically set to state 0 or 1.
Movie S2. Three pendula with periods , , and are programmed to reach all seven accessible states, starting from state 000. The time required to reach these seven states can be significantly reduced by instead choosing , , and . In both cases, we used to determine the actuation times.
Movie S3. A set of 5 pendula with periods , , , , and is used to play "Oh When the Saints" with .
Movie S4. A set of 5 pendula with periods , , , , and is used to play "Jingle Bells" with .
Movie S5. A set of 5 pendula with periods , , , , and is used to play "Smoke on the Water" with .
References
- [12]Hiromi Yasuda, Philip R Buskohl, Andrew Gillman, Todd D Murphey, Susan Stepney, Richard A Vaia, and Jordan R Raney. Mechanical computing. Nature, 598(7879):39–48, 2021.DOI
- [13]Benjamin Treml, Andrew Gillman, Philip Buskohl, and Richard Vaia. Origami mechanologic. Proceedings of the National Academy of Sciences, 115(27):6916–6921, 2018. .DOI
- [14]Tian Chen, Matthias Pauly, and Pedro M. Reis. A reprogrammable mechanical metamaterial with stable memory. Nature, 589(7842):386–390, 2021. .DOI
- [22]I Mahboob and H Yamaguchi. Bit storage and bit flip operations in an electromechanical oscillator. Nature nanotechnology, 3(5):275–279, 2008.DOI
- [23]Z. P. Bažant and Luigi Cedolin. Stability of Structures: Elastic, Inelastic, Fracture and Damage Theories. World Scientific, Singapore, 2010. ISBN 9789814317043.
- [24]Bert Van Raemdonck, Edoardo Milana, Michael De Volder, Dominiek Reynaerts, and Benjamin Gorissen. Nonlinear inflatable actuators for distributed control in soft robots. Advanced Materials, 35(35):2301487, 2023.DOI
- [25]Robert D Blevins. Formulas for dynamics, acoustics and vibration. John Wiley & Sons, 2015.
- [26]Gayathri Pillai and Sheng-Shian Li. Piezoelectric mems resonators: A review. IEEE Sensors Journal, 21(1):12589–12605, 2020.
- [27]M. C. Cross and P. C. Hohenberg. Pattern formation outside of equilibrium. Rev. Mod. Phys., 65:851–1112, Jul 1993. . URL https://link.aps.org/doi/10.1103/RevModPhys.65.851.DOI
- [28]I.R. Epstein and J.A. Pojman. An Introduction to Nonlinear Chemical Dynamics: Oscillations, Waves, Patterns, and Chaos. Topics in Physical Chemistry. Oxford University Press, 1998. ISBN 9780198025665. URL https://books.google.it/books?id=ci4MNrwSIo4C.
- [29]Steven H. Strogatz. Chemistry Applications to Physics, Biology, and Engineering. Nonlinear Dynamics and Chaos. Addison-Wesley, Reading, MA, 1994.








