Beyond power limits: the kinetic energy capacity of skeletal muscle Open Access

Movement is essential for all animals, and muscle is what drives animal movement (McMahon, 1984; Daniel and Tu, 1999; Dickinson et al., 2000; Alexander, 2003; Biewener, 2016; Higham et al., 2016; Biewener and Patek, 2018; Mendoza et al., 2023). What muscle can and cannot do is thus a fundamental question in zoology. Two rudimentary mechanical properties are thought to characterise each unit of muscle mass as a motor: its maximum work density and its maximum power density (Bennet-Clark, 1977; Borelli, 1680; Hill, 1950b; Gabriel, 1984). No muscle contraction can violate these limits, and, because both muscle power and work density appear to be remarkably conserved (they vary by at most one order of magnitude across animal size, ecological niche and evolutionary history; see Askew and Marsh, 2002; Close, 1972; Marden and Allen, 2002; Medler, 2002; Rospars and Meyer-Vernet, 2016), they are thought to pose universal constraints on animal performance. A limiting work or power density has, in some form or another, been invoked to account for a remarkable diversity of non-trivial observations on animal locomotor performance, including maximum running speed (e.g. Hill, 1950b; Irschick et al., 2003; Labonte et al., 2024; Meyer-Vernet and Rospars, 2015; Usherwood and Gladman, 2020), flight speed (Alexander, 2005; Labonte et al., 2024), swimming speed (e.g. O'dor and Webber, 1991; Wakeling and Johnston, 1998; Clemente and Richards, 2013; Richards and Sawicki, 2012; Richards and Clemente, 2013) and jump height (e.g. Bennet-Clark, 1977; Gabriel, 1984; Marsh, 1994; Lutz and Rome, 1994; Roberts et al., 2011; Sutton et al., 2016; Wilson et al., 2000); size-specific variations in animal posture (Usherwood, 2013); the prevalence of latched ‘power-amplifiers’ in small animals (Alexander, 1988; Bennet-Clark, 1975; Gronenberg, 1996; Ilton et al., 2018; Longo et al., 2019; Patek, 2023); the transition from latch-mediated to direct muscle actuation in jumping animals (Sutton et al., 2019); the mechanical benefits of in-series elasticity in explosive muscle contractions (e.g. Aerts, 1997; Alexander, 1995; Galantis and Woledge, 2003; Hill, 1950a; Roberts and Marsh, 2003; Lichtwark and Wilson, 2005; James et al., 2007; Peplowski and Marsh, 1997; Roberts and Azizi, 2011; Roberts, 2016; Mendoza and Azizi, 2021; Rosario et al., 2016; Sawicki et al., 2015; reviewed recently in Holt and Mayfield, 2023); the outcome of predator–prey interactions (Wilson et al., 2018); feeding performance (Camp et al., 2015); or the limits to manoeuvrability (Williams et al., 2009; Wilson et al., 2013), to name but a few examples.

The notion of a work- and power-limit on muscle energy output now pervades the comparative biomechanics literature, but it originated in the analysis of animal jump performance (Bennet-Clark, 1977); the key ideas are thus perhaps best introduced through this historical lens.

Animal jump performance is thought to be bound by two ‘laws’: Borelli's law, which encodes a work constraint to muscle energy output (Borelli, 1680); and what arguably should be called Bennet-Clark's law, which prescribes a power constraint (Bennet-Clark, 1977; Bennet-Clark and Lucey, 1967).

The argument for the work constraint typically runs as follows (Alexander, 2003; Hill, 1950b; Biewener and Patek, 2018; McMahon, 1984; Schmidt-Nielsen, 1984): an animal of body mass m that jumped to a height h has increased the gravitational potential energy of its centre of mass by an amount E_pot=mgh (g is the gravitational acceleration). This increase was paid for with mechanical work, done by muscle at the expense of chemical energy. The maximum mechanical work muscle can do depends on its volume, V, and on the maximum displacement-averaged stress, ⁠, it exerts as it shortens by a maximum fraction of its length, ε­_max, ⁠. An estimate for the maximum jump height then follows via conservation of energy, E_pot=W_max, where it is tacit that all external forces are much smaller than the muscle force (Scholz et al., 2006; see Supplementary Information for a more detailed discussion of this point). The physiological parameters ⁠, ε­_max and the muscle volume per unit body mass, V/m, are typically considered size invariant; animals small or large should consequently be able to deliver the same mass-specific energy, and thus jump to the same height, h_max=Vσ_­maxε­_max­(mg)⁻¹. It is this prediction that is often referred to as Borelli's law.

The presentation of the power constraint usually remains on curiously more qualitative grounds. Bennet-Clark (1977) himself writes: ‘As the time available for acceleration is less in smaller animals, the energy store must be able to deliver the energy more rapidly’. Schmidt-Nielsen (1984) explains: ‘The smaller the animal, the shorter its take-off distance. [...] The time available for take-off is very short, and muscle just cannot contract that fast’. Alexander (2003) has it that ‘smaller jumping animals have smaller acceleration distances, and so have to extend their legs in even shorter times [...]. But no known muscle can complete an isolated contraction in so short a time’. Biewener and Patek (2018) agree: ‘Smaller animals have shorter limbs [...]; therefore, the [...] time available for acceleration during take-off is less’. These representative accounts of how power may constrain muscle energy output have two elements in common. First, they all highlight the importance of time. Time is absent in the framework of Borelli's law, because it frames the problem solely in terms of muscle work. But no fixed amount of work can be done in an infinitely short amount of time, and time therefore ought to be considered explicitly when estimating bounds on muscle mechanical performance. Second, it is implied that the time available depends on animal size. Smaller animals are thought to have less time to do work; muscle energy output, so the argument goes, is thus constrained by muscle power capacity in animals below some critical size, but by muscle work capacity in animals above it (e.g. Alexander, 2003; Bennet-Clark, 1977; Biewener and Patek, 2018; Gabriel, 1984; Schmidt-Nielsen, 1984; Usherwood, 2013; Ilton et al., 2018; Longo et al., 2019; Patek, 2023; Usherwood and Gladman, 2020). It is the omission of time that prompted Bennet-Clark and others to challenge Borelli's law; and it is the suggested variation of available time with animal size that eventually developed into the now practically ubiquitous notion of a power limit to the energy output of muscle – Bennet-Clark's law.

This paper revisits this power-limit paradigm, and presents an alternative mechanical framework to account for size-specific variations in muscle energy output.

To formally assess the mechanical grounds on which the notion of a power limit to muscle energy output rests, it is prudent to first lay out what is meant by it. It is obviously correct that an isolated muscle with mass m_m and power density P_ρ cannot provide more power than P_max=P_ρm_m. But this truism merely provides the basis for a more complex and consequential causal inference widespread in the comparative biomechanics literature: the power limit is regularly invoked as a constraint on the muscle energy output per contraction; it is branded as the proximate cause of an ultimate limit distinct from Borelli's law. This assertion has two corollaries: changes in muscle power must be necessary and sufficient to change muscle energy output, for otherwise it is unclear in what sense power can be said to be limiting; and the changes in power that lead to a variation in energy output must leave muscle work capacity unaffected, for otherwise there is no clear basis upon which Borelli's and Bennet-Clark's law may be argued to be different. These two demands define the burden of proof that rests with any claim of an energy limit distinct from Borelli's law; with them at hand, an assessment strategy can be formulated.

To probe the validity of the first corollary, we will analyse how much energy a muscle with finite work and power capacity can deliver in a single contraction, using the mathematically simplest form of physical reasoning – dimensional analysis. The aim here, as much as for the rest of this study, is not to account for all complexities of real musculoskeletal systems, but to analyse parsimonious models that capture the essential physical features that endow a muscle with a maximum work and power capacity – the properties that underpin the claims under investigation. A suitable framework for such an analysis can be found in the recently developed theory of physiological similarity, which maps out the mechanical performance landscape for an idealised musculoskeletal system (Labonte, 2023; Labonte et al., 2024; Polet and Labonte, 2024). This system consists of a payload of mass m – for example the body mass of an organism, or the mass of a limb that is moved – connected to a muscle of volume V, which can exert a maximum stress σ­_max, shorten by no more than a maximum strain ε_max, and no faster than with a maximum strain rate ⁠. Throughout this work, it is assumed that the muscle is always fully activated, and that the payload mass is large compared with the muscle mass, m>>m_m.

Without loss of generality, such a muscle has a maximum work capacity W_max ∝ Vσ­_maxε_max. The muscle's power capacity, in turn, may be defined in one of two ways. Early work on power limits in small animals usually refers to the time-averaged power – the ratio between the work done and the time it took to deliver it, ⁠; this focus on is often implicit, identifiable only from reference to the time available. More recent work also considers the instantaneous power, P_inst, the product between the instantaneous force, F, and instantaneous velocity, v: P_inst=W/dt=Fv. The time-averaged power is easier to measure, and is arguably the more functionally relevant metric; but the limit on instantaneous muscle power is mechanically and physiologically important, and muscle cannot exceed either. Both powers are proportional to muscle volume, stress and strain rate, ⁠, so that the distinction is irrelevant at this point, and no further restrictions need be imposed.

To analyse the physical limits to muscle mechanical performance with respect to the second corollary, we will then deploy the fundamental principle of the conservation of energy. This step, still conducted within the framework of the theory of physiological similarity, will force as much as permit us to take into account a key feature of real muscle that is irrelevant for the results of the initial dimensional analysis, and was thus ignored up to now: muscle stress typically varies inversely with muscle strain rate according to the well-characterised force–velocity relationship (FVR; Hill, 1938; Piazzesi et al., 2007), (for a model that also considers force–length properties, see Labonte, 2023). This co-variation of stress with strain rate stems directly from the molecular mechanisms that underpin muscle contraction (Hill, 1938; Piazzesi et al., 2007), and places further constraints on muscle energy output, for the muscle's work capacity now declines with the rate at which it is delivered (Peplowski and Marsh, 1997; Roberts and Marsh, 2003; Bobbert, 2013). The task ahead is thus not only to formally identify all possible limits on muscle energy output but also to evaluate explicitly how the variation of stress with strain rate influences the amount of energy muscle can inject per contraction.

Three important features that characterise real musculoskeletal systems are notably absent from the idealised model under investigation, and thus demand comment. First, muscle input is usually proportional but not equal to system output: muscle input is transmitted through joints, resulting in musculoskeletal systems that are geared (Borelli, 1680; Osborn, 1900). Gearing leaves the system's maximum work and power capacity unaffected, and is thus only discussed briefly here; readers who seek a detailed treatment of the influence of gearing on muscle energy and power output may find it in Arnold et al. (2011), Labonte (2023), McHenry (2011, 2012) and Polet and Labonte (2024). Second, muscle rarely attaches directly to skeletal segments but instead connects onto them via elastic elements such as tendons, aponeuroses and apodemes (Hill, 1950a). The resulting in-series elasticity plays an important role in the power and energy output of skeletal muscle (Aerts, 1997; Alexander, 1995; Hill, 1950a; Bennet-Clark, 1975; Peplowski and Marsh, 1997; Galantis and Woledge, 2003; Roberts and Marsh, 2003; Lichtwark and Wilson, 2005; James et al., 2007; Farahat and Herr, 2010; Roberts and Azizi, 2011; Roberts, 2016; Rosario et al., 2016; Mendoza and Azizi, 2021; Holt and Mayfield, 2023; Sawicki et al., 2015), but not in Borelli's and Bennet-Clark's law; indeed, as is discussed briefly later, in-series elasticity is often framed as an evolutionary adaptation that overcomes the limits prescribed by Bennet-Clark's law. And third, throughout this work, it is assumed that the contraction is inertial, i.e. that all opposing (parasitic) forces are small compared with the muscle force, and can thus be neglected. Parasitic forces are not considered in the classic presentation of either Borelli's or Bennet-Clark's law, and do not affect the key conclusions presented here; they can, however, influence both the muscle energy output and the partitioning of muscle work into different system energies, as discussed in detail in Labonte (2023), Labonte et al. (2024) and Polet and Labonte (2024) (see also Farahat and Herr, 2010). These three omissions have two important consequences: the first two imply that the speed of the payload is equal to the shortening speed of muscle at all times; and the third implies that the net work is equal to the work done by the muscle, and that all of this work flows into kinetic energy. Keeping both consequences in mind will help in following the analysis, and is important when the results presented here are generalised, or compared with the more complex analyses presented elsewhere (Labonte, 2023; Polet and Labonte, 2024).

In discussing the limits to muscle mechanical performance, it can be physically insightful and biologically meaningful to explicitly assess the influence of animal body size. To facilitate such analyses, we will make the parsimonious assumption of isogeometry and isophysiology throughout this text. That is, in keeping with classic scaling theory, characteristic lengths, areas and volumes are assumed to scale with body mass m as L ∝ m^1/3, A ∝ m^2/3 and V ∝ m; and physiological parameters such as maximum stress, strain and strain rate are assumed to be size invariant (Askew and Marsh, 2002; Close, 1972; Marden and Allen, 2002; Rospars and Meyer-Vernet, 2016).

Borelli's and Bennet-Clark's law both seek to estimate the maximum energy muscle can deliver per contraction. The key difference between them can be illustrated with a dimensional argument: if the left-hand side of an equation is of dimension length per time [L T⁻¹], the right-hand side ought to be, too, otherwise an error has been made. Consider, then, the task of predicting the energy that a muscle with fixed work and power capacity can deliver as it contracts against a payload of mass m. Energy and work share the same dimension [M L² T⁻²], so that, with Borelli, one may quickly surmise E=W_max (or, in keeping with the focus of the jumping literature on take-off speed, ⁠). Note well that the only way to change E under a work constraint is to change W_max – a variation of the work capacity is both necessary and sufficient to change the maximum possible energy output, and Borelli's law thus encodes an energy limit proper. Power, in contrast, is of dimension [M L² T⁻³] – it is consequently impossible to write the energy output as a sole function of P (or define a speed using only P and m). Instead, dimensional consistency demands that one of three auxiliary variables be specified in addition: (i) a displacement δ of dimension [L], so that E ∝ (P_maxδ)^2/3m^1/3 [and v ∝ (P_maxδm⁻¹)^1/3]; or (ii) a time t of dimension [T], so that E ∝ Pt (and ⁠); or (iii) a force F of dimension [M L T⁻²], so that E ∝ P²F⁻²m (and v ∝ PF⁻¹; see Supplementary Information and Fig. 1A,B). A crucial conclusion follows at once: a variation in power capacity is neither necessary nor sufficient to vary muscle energy output; changes in a third variable can alter the energy output for a given power input, or keep it constant despite arbitrary variations in power (Table 1).

This simple observation has not-so-simple implications. A muscle's work and power capacity depend solely on the muscle's volume, and a characteristic stress, strain and strain rate. It thus seems reasonable to expect that specifying these quantities is all that is needed to predict the energy output with Borelli's and Bennet-Clark's law. For Borelli's law, this is indeed so, but to estimate the energy output with Bennet-Clark's law, the force and displacement capacity must be known, too – it becomes necessary to specify how a muscle volume V is split into physiological cross-sectional area, A_m, and fascicle length, l_m. In other words, the energy output now also depends on the muscle aspect ratio, ν=l_mA_m^−1/2 (Labonte, 2023; Polet and Labonte, 2024; this remains true if the time is fixed, see Fig. 1C and below). Even with the muscle's geometrical arrangement specified, the difficulties are not quite over just yet – which of the three auxiliary variables should be chosen? The decision is not obvious, and lo!, examples for each option can be found: Bennet-Clark picked the displacement (Bennet-Clark, 1977), which remains the most popular implementation (e.g. Biewener and Patek, 2018; Bobbert, 2013; Gabriel, 1984; Marsh, 1994; James et al., 2007; Sutton et al., 2016); Usherwood instead fixed the time (Usherwood, 2013; see also Usherwood and Gladman, 2020); and, last but not least, Meyer-Vernet and Rospars (2015, 2016) and Hawkes et al. (2022) fixed the force. The specific choice carries meaningful consequences: it leads to quantitative differences in the downstream performance prediction. Isogeometry and isophysiology imply δ ∝ m^1/3 and F ∝ m^2/3, which leads to E ∝ m^11/9 or E ∝ m^5/3, and v ∝ m^1/9 or v ∝ m^1/3, respectively (Bennet-Clark, 1977; Meyer-Vernet and Rospars, 2015). Usherwood instead assumed t ∝ √δ, leading to E ∝ m^7/6 and v ∝ m^1/12 (or, via a similar argument, E ∝ m^4/3 and v ∝ m^1/6) (Usherwood, 2013; see also Usherwood and Gladman, 2020). A rather striking difference between the two laws has become apparent: Borelli's law unequivocally predicts E ∝ m and v ∝ m⁰=constant. Bennet-Clark's law, however, has been used to predict anything from E ∝ m^7/6 to E ∝ m^5/3, and v ∝ m^1/12 to v ∝ m^1/3 (Table 1 and Fig. 1A,B). In fact, it can predict any energy output through suitable variation of the auxiliary variable of choice (Fig. 1A,B). It is evidently not the muscle's power capacity itself that is limiting the energy output; Bennet-Clark's law may thus at best be said to encode a power–displacement, power–time or power–force constraint. However, and no less clearly, the energy output of muscle can, in fact, violate Borelli's law. If it is neither the muscle's work nor power capacity that is imposing the limit in these instances, then what is?

In the previous section, it was demonstrated that a dimensionally consistent link between a fixed power input and energy output requires specification of one auxiliary variable, that this choice is not obvious and that different choices lead to different results. The task that lies ahead is to identify the mechanical explanation for these differences.

Consider an animal of body mass m. Let the maximum time-averaged power capacity of its muscles be ⁠, and allow it to accelerate for no more than a time t. What is the energy imparted to its centre of mass? This is a classic textbook setup for Bennet-Clark's law, and it is tempting to conclude – but this is not necessarily so. In fact, all that can be deduced is the considerably weaker (Fig. 1C). This restriction arises because specifying the body mass m together with places hidden demands on the muscle force and shortening speed capacity: it requires that the time-averaged force capacity is at least ⁠, and that the maximum shortening velocity is at least ⁠. If the muscle's force capacity is smaller, it cannot deliver the power within time t; and if its shortening speed capacity is smaller, it may deliver P_max, but in less time. In both cases, (Fig. 1C). In other words, although the maximum average power capacity is equal to no matter the muscle aspect ratio, only one unique aspect ratio allows a muscle with a maximum and v_max to deliver exactly within time t (Fig. 1C). This point may appear subtle, but its consequences are surely troubling: estimating the energy output associated with a specific muscle power input requires specifying an auxiliary variable by physical necessity – but even that may still not yield a definite answer.

To understand the mathematical origin of this result, note that Newtonian point-mass dynamics are governed by a set of two equations that uniquely link five elemental variables: Newton's second law and its path integral define the relationship between force, mass, speed, time and displacement throughout the contraction. This mathematical structure dictates that the choice of any three parameters uniquely determines the remaining two variables. Thus, regardless of how Bennet-Clark's law is implemented for an animal with body mass m, all dynamic variables end up fully defined (Table 1). Contrast this scenario with a determination of the energy output via the work capacity of muscle, which requires defining only one additional parameter, E=W_max. It consequently does not matter whether W_max is partitioned into a small force and large displacement capacity (which would take a long time and involves low power), or into a large force and a small displacement capacity (which will be completed rapidly and requires large power) – any muscle with work capacity W_max will do, because the governing equations remain underdetermined, and retain one residual degree of freedom that can absorb arbitrary work partitioning. It may be tempting to file this observation as technically correct but of limited practical implication. This would be a mistake. Consider again the most widespread quantitative implementation of Bennet-Clark's law, which combines a size-invariant power density with an isogeometric displacement to predict E ∝ m^11/9 and v ∝ m^1/9. Hidden within this prediction lies the necessary condition that the average muscle force scales as F ∝ m^8/9, in substantial excess of the isogeometric and isophysiological expectation, F ∝ m^6/9 (Table 1). How this positive allometry is to be achieved in an isogeometric and isophysiological system is not obvious.

Dimensional arguments and the mathematics of point mass dynamics in combination provide two conclusions: because of the need for dimensional consistency, the question ‘how much energy can a motor with power P_max inject into a mass m’ cannot be answered without specification of exactly one further auxiliary variable (Table 1); and because of the fundamental structure of Newtonian dynamics, any such choice uniquely defines all remaining variables – the muscle force, displacement and shortening speed throughout the contraction are fully determined. It is the determination of the shortening speed in particular that is the essential distinction between the different instantiations of Bennet-Clark's law, and that brings about the variation in the energy output they predict – an assertion to which the discussion will now turn.

The notion that muscle power capacity limits muscle energy output has been called into question in the past (e.g. Adamson and Whitney, 1971; Farley, 1997; Knudson, 2009; Ruddock and Winter, 2015; Winter, 2005; Winter et al., 2016). But the harshness with which this criticism was sometimes expressed masked its own failure to address the fundamental issue and valid concern unearthed by the careful observations of Bennet-Clark and many others since; any account of muscle mechanical performance that ignores the dimension of time risks arriving at conclusions that violate physiological or physical constraints, for no muscle can do a unit of work in arbitrarily short time. It is true enough that the mechanical quantity that uniquely ties speed, mass and contraction time is the impulse, and not power (Adamson and Whitney, 1971; Knudson, 2009; Ruddock and Winter, 2015; Winter, 2005; Winter et al., 2016), but pointing this out merely addresses a symptom instead of the problem's root: what limits the time over which muscle can do work?

For Γ≤1, the kinetic energy capacity is limiting, and for Γ≥1, the work capacity is limiting (Fig. 3). So how large is Γ? This question is of obvious and immediate importance, and the subject of the recently developed theory of physiological similarity (Labonte, 2023; Labonte et al., 2024; Polet and Labonte, 2024), from which the following analysis draws.

Eqn 2 is completely general in the sense that it holds for any muscle that is restricted by a maximum stress, strain and strain rate, i.e. regardless of the exact shape of the FVR (Labonte, 2023; Mendoza et al., 2023). But the reader will rightfully point out that this generality does little good, for Eqn 2 depends not on σ_max but strictly on – the muscle stress averaged over the exerted strain – which necessarily depends on the shape of the FVR for all but quasi-static contractions. FVRs thus influence Γ in two distinct ways: through the imposition of a maximum strain rate, encoded via in the numerator; and through the variation of stress with strain rate, implicit in the appearance of in the denominator – in other words, they influence both K_max and W_max. What is the relative importance of these two FVR features in determining muscle energy output?

With these results at hand, it is now finally the time to discuss the magnitude Γ as defined by Eqn 3, and to thus answer the question which of the two characteristic muscle energy capacities, the Borelli or the Hill limit, may be relevant in animal movement. Several case studies can be found in the literature (Labonte, 2023; Labonte et al., 2024; Polet and Labonte, 2024). As an illustrative example, consider the musculoskeletal system that propels running animals, for which Γ≈0.07 mass^2/3 kg^−2/3 (Labonte et al., 2024). Thus, from a 0.1 mg mite to a 10 t elephant, Γ is predicted to vary by a whopping 7 orders of magnitude (Labonte, 2023; Labonte et al., 2024). It remains smaller than unity for runners lighter than about 50 kg (Labonte et al., 2024), and the kinetic energy capacity, K_max, is thus likely a robust proxy for the limit to muscle energy output in the vast majority of terrestrial animals (Labonte, 2023; Labonte et al., 2024).

Framing the limits on muscle energy output in the conceptual terms of a limiting work versus power density has become a textbook staple, invoked to explain an extraordinarily diverse array of observations in comparative animal biomechanics. Our analysis reveals that attributing these observations to a competition between a limiting work and power density is to miss out on some important physics, and that the alternative constraint on muscle energy output is not imposed by the muscle's power density but by its characteristic kinetic energy density, K_ρ=K_max/m_m. What are the implications of this conclusion for our understanding of the biomechanics of animal movement?

K_ρ differs from the power density P_ρ and the work density W_ρ in at least four aspects, and these differences provide clear and consistent explanations for some classic observations in comparative animal biomechanics. First, in contrast to P_ρ and W_ρ, K_ρ retains a size dependence (Fig. 4A): geometrically and physiologically similar larger musculoskeletal systems have a larger kinetic energy density, E_ρ ∝ m^2/3. As a result, larger animals are generally faster (Bejan and Marden, 2006; Gazzola et al., 2014; Pennycuick, 1968; Garland, 1983; Meyer-Vernet and Rospars, 2016; Sánchez-Rodríguez et al., 2023; Labonte et al., 2024). Second, in contrast to P_ρ and W_ρ, K_ρ can be geared, K_ρ=1/2m/m_mG⁻², where G is a dimensionless mechanical advantage, defined as the ratio between system output and muscle input force (Fig. 4B; Labonte, 2023; Polet and Labonte, 2024). As a result, two staples of biomechanical analyses that may ring contradictory can in fact both hold true: gearing is usually interpreted in terms of force–velocity trade-offs; a lower gear ratio increases the instantaneous velocity of the payload at the expense of the transmitted force. But gearing leaves the work and power capacity unaffected, for it amplifies displacement and velocity by just as much as it attenuates force. How, then, can a muscle make things move more quickly via gearing, although its putatively limiting work and power capacities have remained unchanged? The answer is that for as long the energy output remains below the work capacity, a reduction of the gear ratio can in fact enable muscle to do more work, because it increases its kinetic energy capacity (McHenry, 2011, 2012; Olberding et al., 2019; Osgood et al., 2021 preprint; Polet and Labonte, 2024). The immediate implication of this observation is the existence of a mechanically optimal mechanical advantage that varies with animal size and environment – a hypothesis unpacked in detail in Polet and Labonte (2024). Third, in contrast to P_ρ and W_ρ, K_ρ is a function of the mass that is driven (Fig. 4C). This is perhaps the least intuitive idiosyncrasy of K_ρ: increasing the payload can enable muscle to deliver more energy, because it unleashes latent work capacity (Fig. 3; and see Sawicki et al., 2015, for a related finding on muscle power output). As a result, where K_ρ is limiting, animals may be able to achieve the same speed for payloads that are increasing multiples of their own body mass. As a striking illustration of this prediction, consider rhinoceros beetles, which can carry up to 30 times their own body mass without changing speed (Kram, 1996). Fourth, the alleged power limit to muscle energy output is often invoked to explain a key functional benefit of in-series elasticity in musculoskeletal systems: in dynamic contractions, tendons can decouple limb and muscle shortening speed, and muscle can consequently achieve similar absolute limb speeds with lower muscle shortening speeds, so increasing its power output (Aerts, 1997; Kurokawa et al., 2001; Marsh, 2022; Roberts and Marsh, 2003; Galantis and Woledge, 2003; Astley and Roberts, 2012; Farris et al., 2016; Robertson et al., 2018); in quasi-static ‘latched’ contractions, muscle can contract arbitrarily slowly against elastic elements, and so avoid both force–velocity effects and supposed muscle power limits to performance, by instead releasing its work capacity explosively (Bennet-Clark and Lucey, 1967; Bennet-Clark, 1975; Gronenberg, 1996; Longo et al., 2019; Patek, 2023). A large body of careful work has been dedicated to such amplification of muscle power, be it in dynamic or ‘latched’ quasi-static contractions (for recent reviews, see Holt and Mayfield, 2023; Longo et al., 2019; Patek, 2023). There is no doubt, of course, that elastic elements can amplify muscle power. But a reasonable argument is to be had whether the biological function of ‘springs’ in these instances is to amplify speed rather than power as such. The kinetic energy capacity of a spring is likely orders of magnitude higher than that of muscle; it is limited by the elastic wave speed, ⁠, where Y_m is the Young's modulus of the spring and ρ_s is its density. For reasonable values of Y_m≈10⁹ N m⁻² and ρ_s≈1000 kg m⁻³, one finds v_e≈1000 m s⁻¹; a muscle with a typical maximum strain rate of ≈10 lengths s⁻¹ would need fascicles with a length of 100 m to reach the same absolute speed. Because jumping performance in small animals is likely limited by the kinetic energy capacity of muscle, we posit that (i) their springs act as ‘work enablers’ (see also Roberts and Marsh, 2003), allowing them to overcome the constraint on energy output imposed by a limiting kinetic energy capacity; and (ii) that power amplification is an epiphenomenon instead of the biological purpose of in-series elasticity in rapid movements. The outcome of this somewhat semantic debate is clearly immaterial for the validity of the long list of fundamental insights that have been derived from the study of power amplification due to biological ‘springs’ (Gronenberg, 1996; Ilton et al., 2018; Longo et al., 2019; Patek, 2023).

The above examples may perhaps sharpen the physical explanation of some well-established observations in the comparative biomechanics of animal movement, but they do not make novel performance predictions as such. To derive such predictions, we next compare the theory of physiological similarity directly with classic scaling theory.

The importance of animal size in determining physiology, morphology and physical constraints is well established, and perhaps among the oldest and most intensely studied aspects of comparative biomechanics (McMahon et al., 1983; Schmidt-Nielsen, 1984; for a recent review, see Clemente and Dick, 2023). Where such inquiries are concerned with dynamics, they typically invoke a characteristic muscle force capacity, F ∝ m^2/3, a characteristic displacement capacity, δ ∝ m^1/3, and a characteristic work and power capacity, W ∝ m and P ∝ m, respectively. Together with the payload, classic scaling theory thus specifies four mechanical quantities – m, σ­_max, ε_max and (Fig. 2A) – but point mass dynamics only provides three degrees of freedom. The startling consequence of this over-determination is that muscle is characterised not by one force, energy, speed displacement and power capacity, as classic scaling theory would have it, but by two (Labonte, 2023). A full analysis of this observation exceeds the scope of this work, and will have to await further study; it will be illustrated here with but one brief example.

Eqn 8 illustrates that the relevance of Γ is not restricted to energy output. Indeed, muscles that operate with equal Γ can be shown to deliver the same fraction of their maximum work and power capacity; to operate at the same fraction of their speed and displacement capacity; and to generate the same ratio of a characteristic inertial force to a characteristic maximum force (Labonte, 2023; Labonte et al., 2024; Polet and Labonte, 2024). Because these parameters represent a wide array of muscle physiological and mechanical characteristics – and to avoid giving one interpretation priority over any other – they may be emphasised equally by defining Γ as an index of physiological similarity (Labonte, 2023).

The idea that muscle power limits muscle energy output has become common biomechanical vernacular. It is rooted in the fundamental objection that muscle needs time to inject energy, and that muscle work capacity alone therefore does not tell the whole performance story. Supported by a series of arguments, we have suggested that this conclusion is partially wrong, and thus only partially right. Variations in muscle power are neither necessary nor sufficient to vary the energy muscle can deliver in a single contraction – a muscle's power capacity consequently does not pose a limit to either energy output or speed by itself. The problem is not that small animals do not have enough time to do work, the problem is that their muscles have a lower maximum shortening speed, and consequently a lower kinetic energy capacity: neither infinite power nor an infinitely long contraction time would help overcome this limit. Power, of course, is not relegated to biomechanical irrelevance altogether; it is solely the assertion that muscle power capacity limits muscle energy output that is called into question.

A limit to shortening speed is tacit in any dimensionally consistent expression that links muscle power, mass and energy, and consequently shares many of the features that are typically associated with a power limit, including shorter contraction time scales and a reduced mechanical performance in small animals. The conclusion therefore neither can nor should be that the large body of work that analysed problems in biomechanics in the conceptual terms of a power limit is ‘wrong’, that it lost any significance, or that the fundamental issues it raised are any less seminal. Rather, the question ought to be whether a shift in perspective to muscle shortening speed instead of muscle power, and to a kinetic energy density instead of power density, brings any meaningful advantages, or whether it is at best technically correct, but for all intents and purposes practically irrelevant.

In favour of this shift, four brief arguments may be presented. First, expressing putative energy limits directly in terms of characteristic energies enables meaningful comparison, for distinct limits now share the same dimension (Fig. 4). Second, this comparison provides straightforward explanations for a series of observations in comparative biomechanics that are cumbersome if not impossible to explain in the framework of a size-invariant work and power density, and provides predictions for the scaling of musculoskeletal performance that depart from textbook theory (Figs 4 and 5). Third, in its reliance on auxiliary variables, analysing muscle contractions in terms of muscle power makes it exceedingly easy to unintentionally demand of muscle something it may not be able to do. An explicit account of the key mechanical variables that limit every contraction – via the physiological similarity index, Γ – side-steps this difficulty, and provides a clear framework to ensure that mechanical analyses remain not only physically but also physiologically plausible. It is both clearer and less ambiguous to bind speed through an explicit limit on shortening velocity than to introduce this limit through the backdoor, by treating the problem as though it were one of muscle power. Fourth, the notion of a size-invariant work and power density leaves remarkably little room for adaptive variation in musculoskeletal design. Inspection of the kinetic energy density, in turn, permits speculation. To give but two examples: (i) systematic variation in gear ratio with size can, in fact, enhance the work output of musculoskeletal systems, such that small animals would benefit from small gear ratios, and large animals from large gear ratios (Biewener, 1989; Labonte, 2023; Labonte et al., 2024; Polet and Labonte, 2024; Usherwood, 2013); (ii) the maximum energy output can be independent of muscle mass, and instead depend solely on fascicle length, gear ratio, and the maximum muscle strain rate. Thus, small animals may be able to reduce the fraction of the body mass allocated to muscle, without suffering from a decrease in locomotor speed, as appears to be the case in reptiles compared with mammals (Labonte et al., 2024). These hypotheses no doubt require scrutiny, but they follow readily from inspection of the kinetic energy density, and cannot be easily extracted through the lens of a limiting work or power density.

Although this text has criticised the notion of a power limit to the energy output of muscle contractions, it was written in undiminished admiration of the groundbreaking work that has been conducted within this conceptual framework. Deciphering the mechanical limits that bind muscle performance across animal sizes and environments remains challenging enough, and any tool that permits progress should be used. Time only will tell if the kinetic energy density and the physiological similarity index belong into this category, alongside the notion of a power limit.

This study was supported by a Human Frontier Science Programme Young Investigator Award (RGY0073/2020) to D.L. and N.C.H., and partially inspired by an engaging discussion about the Hill and the Borelli limit with some members of the Structure and Motion Laboratory at the Royal Veterinary College in London. D.L. thanks Jim Usherwood for many insightful discussions about muscle work and power limits, and for the suggestion to refer to a power limit as Bennet-Clark's law. Delyle Polet provided many useful comments on an earlier version of the manuscript, which are gratefully acknowledged.