What an estimate lets you change

A scaling law becomes a design argument when we say what is measured, what stays fixed, and what an intervention can actually alter.

CCBS · EST 5053 · Lecture 1 audience extension
Editorial synthesis of Week 1 student submissions · prepared with Codex · 8 September 2026

Lecture 1 gives us permission to calculate before we know everything. A length and a diffusivity give a heating time. An energy budget gives a jump-height scale. A count of molecules gives a first constraint on what a cell can accomplish. These estimates are useful because they discard detail.

The audience's submissions ask when that discarded detail becomes worth recovering. Chenkai Zhu asks what a cooking time actually measures. Peng Su separates energy storage from mechanical transmission, while Haobo Zheng asks when a shrinking jumper needs rapid energy release. Shunkang Xue and Yuhang Wang ask which step a faster cellular response requires. Anqi Tu follows ATP through a standing pool and a day's turnover. Read together, their questions turn estimation into a method for choosing an intervention.

The organizing question is not whether a rough model is true in every detail. It is whether the model preserves the distinction needed for the next decision. We will start with a familiar estimate, change the question asked of it, and watch which additional quantities become necessary.

This essay develops arguments from the linked submissions. The connections and additional derivations are editorial synthesis, not quotations or a claim that the students wrote the added prose. The core lecture, exposition, and lineage supply the starting methods. The lecturer's extension example remains a separate essay.

1What counts as done?

A heating time is undefined until we name a temperature and a location. “Heat has crossed the object” and “the center is within two degrees of the bath” are different observables. Both have units of time, so dimensional analysis alone cannot distinguish them.

Chenkai Zhu (祝晨凯) makes this distinction the starting point of a calculation. The lecture's estimate is $t\sim a^2/\kappa$, where $a$ is a characteristic distance and $\kappa$ is thermal diffusivity. Chenkai asks what multiplies that scale once the shape and the meaning of “heated” are specified.

Define a center-temperature fraction before solving for a time. Let the object start uniformly at $T_0$. At time zero, fix its surface at $T_{\mathrm{s}}$. The fraction of the initial temperature difference that remains at the center is

This fraction starts at one and approaches zero. Chenkai's illustrative temperatures are $T_0=20\,^\circ\mathrm{C}$, $T_{\mathrm{s}}=100\,^\circ\mathrm{C}$, and a center target of $98\,^\circ\mathrm{C}$. They specify $\theta_{\mathrm{c}}=0.025$. This is a mathematical completion criterion, not an experimentally established definition of a cooked potato.

The scaling law now has a place to put the missing information. Define the Fourier number as $\mathrm{Fo}=\kappa t/a^2$. The target condition selects a value of this dimensionless time:

The exponent on size has survived. The question has become whether the coefficient $\mathrm{Fo}_*$ is small enough to matter for the intended decision. Answering that question requires a boundary condition and a shape, not another choice of units.

Specify the measured location and completion threshold before interpreting a diffusion-time estimate.

2Price the discarded factor

Geometry can change a heating time substantially without changing its size-squared scaling. Compare a sphere of radius $a$ with an infinite slab of half-thickness $a$. Both have the same center-to-surface distance. Assume uniform initial temperature, constant diffusivity, and the fixed surface temperature just defined.

The heat equation supplies the coefficient that units could not. Its separated spatial modes decay exponentially. For the slab center, the successive modes have wavelengths set by the two faces. For the sphere, spherical symmetry changes the eigenvalues and coefficients. Their center-temperature solutions are

These are ideal-body solutions, independently recomputed here and cross-checked against the open Lienhard heat-transfer textbook. They are not fitted temperature histories from food. A finite heat-transfer coefficient at the surface introduces another parameter and changes the answer.¹

The slowest mode nearly determines the chosen late-time threshold. Keeping only that mode makes the calculation transparent:

The full series at $\theta_*=0.025$ give $\mathrm{Fo}_{*,\mathrm{slab}}=1.593$ and $\mathrm{Fo}_{*,\mathrm{sphere}}=0.444$. The slab takes about 3.59 times as long at the same $a$ and $\kappa$. That difference would matter when choosing a preparation time. It would not overturn the prediction that doubling every length in a geometrically similar object multiplies the heating time by four.

Chenkai's calculation therefore changes how we check an estimate. Agreement within a factor of a few can support the scale while leaving the coefficient unresolved. Close numerical agreement is a stronger claim only if the geometric length, boundary condition and target were chosen independently of the answer. Otherwise, adjusting an “effective distance” can absorb the very discrepancy that was supposed to test the model.

The next experiment follows directly. Measure the center-temperature history for two geometrically similar samples. Rescale time by $a^2$ and compare the curves. A shared curve tests the scaling transfer. Its location tests the coefficient. If the curves do not collapse, inspect the quantities assumed fixed, including the surface heat-transfer conditions.

Test a scaling exponent and its coefficient separately, and identify which fixed conditions make size changes comparable.

3A spring is not a gearbox

A mechanical intervention must name the constraint it changes. Energy limits the total work available. Power limits how rapidly that work can be delivered. Transmission geometry determines the relation between force and motion at different parts of a mechanism.

Peng Su (苏芃) separates the last two roles in spring-powered animal movements. A spring can accumulate work over a long loading interval and release it over a short one. A lever or linkage can convert a small input motion into a larger output motion. The two mechanisms cooperate, but neither makes energy appear.

For an ideal lossless transmission with negligible energy storage of its own, instantaneous input and output power agree. Define the speed ratio $R=v_{\mathrm{out}}/v_{\mathrm{in}}$. Power balance then gives

The reciprocal force ratio is essential. A transmission that multiplies speed by twenty divides the available force by twenty under these assumptions. For rotating links, torque and angular velocity replace force and linear speed. The ratio can change through the stroke as the geometry changes.

A spring answers a different question. If stored work $W$ is released in time $\Delta t$, its average release power is $W/\Delta t$. Shortening the release interval can make that power exceed the average muscle power during loading. It does not increase the stored work. Friction, deformation and the moving mechanism's inertia then determine how much reaches the intended load.

Haobo Zheng (郑豪波) supplies a reason that release timing becomes more demanding as a jumper shrinks. Hold its aerial jump height $h$ fixed, so its takeoff speed remains $v=\sqrt{2gh}$. Under geometric similarity, let its push distance scale as $s=bM^{1/3}$, where $M$ is body mass and the fixed coefficient $b$ has units $\mathrm{m\,kg}^{-1/3}$. A smaller jumper must reach the same speed over less distance.

With constant net acceleration during the push, the peak rate of kinetic-energy increase per body mass is

This is a power demand that the mass-independent energy estimate omitted. It is only the kinetic term. Lifting the body during the push adds $gv$ to peak external power per body mass in this idealization. Neither term includes internal limb motion or mechanical losses.

A biological power ceiling must use the same denominator. If active jumping muscle is a fraction $f$ of body mass and can supply $P_{\mathrm{muscle}}$ watts per kilogram of muscle, the corresponding ideal body-specific supply is $fP_{\mathrm{muscle}}$. A particular crossover mass therefore needs muscle allocation as well as muscle power and push geometry. Haobo's scaling identifies the growing timing pressure without establishing a universal gram-scale boundary between mechanisms.

Mantis-shrimp experiments support this separation of roles. Patek and colleagues measured rapid strikes and examined elastic storage, then analyzed the linkage mechanics in greater detail. The later study found roughly twofold angular amplification in the observed motion, compared with 3.56 predicted by its ideal four-bar geometry. That is evidence for a real transmission mechanism, not permission to assign an arbitrary biological amplification factor.²³

This distinction changes a proposed fix. If stored energy is insufficient, changing the release timing alone cannot supply the missing work. If the load needs a particular force–speed combination, faster release alone does not guarantee that combination. We need separate measurements of stored work, release duration and transmission geometry.

Distinguish an intervention on stored energy, delivery power and force–speed transmission before predicting improved performance.

4Put power at the bottleneck

The benefit of speeding one stage depends on how much of the original waiting time that stage occupied. The same accounting that separates a spring from a transmission also separates arrival from commitment in a cell.

Shunkang Xue (薛顺康) asks whether an extra energy supply must make a cellular decision faster. His model separates signal capture from a downstream commitment step. The intervention changes commitment while leaving the capture process fixed. That restriction is what makes a useful bound possible.

Let $A$ be the mean time spent gathering the required input and $B$ the subsequent mean commitment time. Assume the stages occur serially. If an intervention multiplies only the commitment rate by $\alpha$, with the corresponding mean time divided by that factor, then

The maximum speedup is consequently $(A+B)/A$. If capture originally occupied half the mean response time, an arbitrarily fast commitment step can improve the total by at most a factor of two. Here $\alpha$ describes a kinetic change. It is not an assertion that ten times as much ATP produces a tenfold rate increase.

The curve suggests a discriminating experiment. If the response stops improving while commitment continues to accelerate, measure the arrival process. If capture also accelerates, the original intervention description was incomplete. Active transport, receptor resetting or a changed number of receptors would require a different model.

The word “decision” also needs care. Capturing a molecule is not automatically obtaining a new independent sample of concentration. Rebinding and downstream memory can correlate observations. Shunkang's bound applies to the stated capture-and-commitment model. Extending it to sensing accuracy requires a model of what counts as information.⁴

Bound the benefit of speeding one stage, while stating exactly which other stages the intervention leaves unchanged.

5One receptor or a whole cell?

A rate per receptor is not a rate per cell. Before comparing two numbers in inverse seconds, we must say which event is counted and how many channels can produce it.

Yuhang Wang (王宇航) puts binding, catalysis and ATP supply into the same response-time question. Read alongside Shunkang's serial model, this contribution exposes a counting issue: the first of many receptor events can occur much sooner than one specified receptor's event.

Consider a deliberately simple binding model. An empty receptor in a buffered ligand concentration $c$ binds with constant rate $k_{\mathrm{on}}c$. Its exponential mean waiting time is $1/(k_{\mathrm{on}}c)$. For $N$ initially empty, independent receptors, the probability that none has bound by time $t$ is

The first binding anywhere therefore has mean $1/(Nk_{\mathrm{on}}c)$. For the illustrative values $k_{\mathrm{on}}=10^6\,\mathrm{M}^{-1}\mathrm{s}^{-1}$ and $c=1\,\mathrm{nM}$, one specified receptor waits 1000 s on average. The first among 1000 independent empty receptors waits 1 s. Neither average is a strict earliest possible event time.

This parallel-channel calculation has a physical boundary. The receptors share a cell surface and a ligand supply. At sufficiently large receptor number, diffusion cannot feed every site as an independent unperturbed reservoir. The receptor-capture example develops that supply limit. Adding channels and increasing the rate of each channel are different interventions.

Completion time must also be separated from sustained throughput. A single channel that binds, processes and resets has a cycle time containing all those waits. A pipeline with separate stages can overlap work on different inputs. Its long-run output cannot exceed its slowest stage's capacity, but that capacity bound need not be attained. The first output still has to traverse the stages.

An ATP budget belongs in this comparison only after its counting unit is fixed. If each completed response consumes $m$ ATP molecules and this pathway can draw on flux $J_{\mathrm{ATP,available}}$, its sustained response flux satisfies

The available pathway flux is not automatically the entire cell's ATP turnover. This is the same allocation issue that separates jumping-muscle power from power per body mass. The inequality limits sustained output, not by itself the latency of one response. Yuhang's comparison becomes testable once binding events, completed responses and allocated ATP are counted on the same basis.

Keep single-channel latency, first arrival among many channels and sustained throughput distinct when locating a bottleneck.

6The stock is not the day's traffic

A large daily throughput does not require a large standing pool. ATP molecules can be consumed and regenerated repeatedly. The integrated amount processed over a day is therefore a different quantity from the amount present at one instant.

Anqi Tu (涂安祺) uses this distinction to examine the claim that a person cycles roughly a body mass of ATP per day. Her ledger makes the scale accessible. The next step is to retain an energy-conversion factor that the simple comparison needs.

Take the submission's values as illustrative bookkeeping inputs, not a measured whole-body dataset. Set daily food energy to $E_{\mathrm{food}}=2000\,\mathrm{kcal}=8.368\,\mathrm{MJ}$ and the magnitude of ATP hydrolysis free energy to $|\Delta G_{\mathrm{ATP}}|=55\,\mathrm{kJ\,mol}^{-1}$. Define $\eta$ as the net fraction of this energy budget represented by ATP cycling under this coarse accounting. Then

Using an ATP molar mass of about $507\,\mathrm{g\,mol}^{-1}$ gives about $77.1\,\eta\,\mathrm{kg\,day}^{-1}$. Setting $\eta=1$ assigns the whole input budget to ATP in this model. It is not a measured metabolic conversion efficiency.

The pool estimate asks a separate question. A chosen intracellular-water volume of 28 L at a chosen average ATP concentration of 3 mM gives 0.084 mol, or about 43 g, present at once. Dividing this amount by the mean ATP flux defines a pool-replacement time:

The three rows are a sensitivity calculation, not a physiological range. Whole-body averaging also hides differences among tissues and compartments. Anqi's central distinction survives these qualifications: a small pool can carry a much larger integrated flux through repeated recycling.

A pool-replacement time becomes an exhaustion time only under additional dynamics. If production stops and consumption remains at its former constant rate until the pool is gone, the quotient gives that hypothetical exhaustion time. If consumption falls with concentration, the trajectory changes. If production balances consumption, the pool can remain constant while carrying substantial flux.

The useful next measurement is consequently the flux and its response to perturbation, not just another concentration snapshot. This is also why a stable ATP concentration does not show that ATP production is unimportant. Stability can be the result of a maintained balance.

Separate standing amount, integrated turnover and pool-replacement time, and expose a missing conversion factor instead of hiding it in a precise number.

7Choose the next measurement

An estimate becomes a design argument when it predicts the effect of a specified change. The submissions have supplied three operations for making that prediction: define the observable, keep the accounting consistent, and identify what the intervention leaves fixed.

The heating example shows why the observable comes first. Its exponent answers a size-transfer question under fixed dimensionless conditions. Its coefficient answers a threshold-time question for a specified shape and boundary. Neither result has to replace the other.

The mechanical and cellular examples show why a proposed improvement needs a location. A linkage changes a force–speed relation. A spring changes the timing of stored work. Faster commitment changes only part of a serial wait unless the intervention also affects capture. A whole-cell ATP flux is relevant to a pathway only through the allocation that connects them.

We can turn these distinctions into a compact experimental worksheet. Write down the measured output and its counting unit. Name the parameter the intervention is supposed to change. List the quantities held fixed. Calculate the resulting shift or bound. Then choose a second measurement that would reveal whether the intervention changed something on the fixed list.

For the capture-and-commitment example, measure both arrival times and completion times while varying the downstream intervention. If completion improves toward a fixed arrival-time floor, the serial model has survived a useful test. If arrival times move too, the next model needs that coupling. If the mean agrees but the spread changes, a mean-time model is no longer enough for the new observable.

That last possibility is the next lecture's opening. A concentration can be the right average while hiding whether a particular cell contains any molecules. A mean reaction rate can be right while hiding how many binding attempts occur. The Lecture 2 audience extension asks which of those details the next experiment requires.

Recover detail when it can change a prediction, distinguish mechanisms or identify the next useful measurement.

Reproduce the numbers

The figures are calculated model curves. This Python 3 code reproduces their defining functions and the numerical comparisons. It uses only the standard library. The heat roots use the full series, not the one-mode approximation.

from math import pi, exp, sqrt

def center(fo, shape):
    if fo == 0:
        return 1.0
    if shape == "slab":
        return 4/pi * sum((-1)**n/(2*n+1)
            * exp(-(2*n+1)**2*pi*pi*fo/4)
            for n in range(150))
    return 2 * sum((-1)**(n+1)*exp(-n*n*pi*pi*fo)
                   for n in range(1, 151))

def threshold_time(shape):
    lo, hi = 0.01, 3.0
    for _ in range(90):
        mid = (lo + hi)/2
        if center(mid, shape) > 0.025:
            lo = mid
        else:
            hi = mid
    return (lo + hi)/2

for shape in ("slab", "sphere"):
    print(shape, threshold_time(shape))
v = sqrt(2 * 9.8 * 0.30)
for mass in (1e-6, 0.001, 1, 70):
    p_kinetic = v**3 / (2 * 0.10 * mass**(1/3))
    print("jump W/kg body", mass, p_kinetic, p_kinetic + 9.8*v)
for alpha in (1, 10, 100):
    print("response seconds", alpha, 2 + 8/alpha)
for eta in (1, 0.5, 0.3):
    mol_day = eta * 2000 * 4184 / 55000
    print("ATP kg/day, pool seconds",
          mol_day * 0.507, (28 * 0.003) / (mol_day / 86400))

Sources and contributions

The six student contributions are linked at their points of use, to the accepted Week 1 files at repository commit c419637. The snapshot was checked on 8 September 2026. They supply the questions and arguments being synthesized. The full texts below were inspected for the specific scientific claims made here.

1. John H. Lienhard V and John H. Lienhard IV. A Heat Transfer Textbook, 6th edition, version 6.00 (2024). Section 5.3 and Table 5.1, pp. 203–207. Fixed-temperature slab solution and the limiting spherical solution. Authors' open textbook.
2. S. N. Patek, W. L. Korff and R. L. Caldwell. “Deadly strike mechanism of a mantis shrimp.” Nature 428, 819–820 (2004). Published article.
3. S. N. Patek, B. N. Nowroozi, J. E. Baio, R. L. Caldwell and A. P. Summers. “Linkage mechanics and power amplification of the mantis shrimp's strike.” Journal of Experimental Biology 210, 3677–3688 (2007). Morphological and measured kinematic transmission are distinguished. Published article.
4. H. C. Berg and E. M. Purcell. “Physics of chemoreception.” Biophysical Journal 20, 193–219 (1977). Finite receptor capture and sampling assumptions. Published article.