TOPIC 33 · PHASE F

Allometric Scaling Laws

Change the size and the proportions must change

2026-08-19 · Scale

Enlarge a mouse to the size of an elephant and it cooks itself. The real elephant is fine — at the price of every gram of it running about twenty times slower than a mouse's.

Spend an afternoon at a zoo and you notice something. Small animals never stop. A sparrow pecks twice and hops; a squirrel moves faster than you can track. An elephant stands there and shifts a leg every half minute. We usually read this as temperament — the small ones are jumpy, the big ones are calm.

But put heart rate, breathing rate, lifespan and feeding rate side by side and something very regular surfaces. Almost every rate falls with body mass, almost every duration rises with it, and both use the same exponent. The squirrel isn't nervous. Every gram of it has to burn far more energy per second than every gram of the elephant, and it simply cannot slow down.

The counterintuitive part comes next. That exponent is not 1 — which would mean every gram is equally busy. It is not 2/3 either — which is what "you can only shed heat through your skin" predicts. It is 3/4: a number with no obvious geometric reason to exist. And it shows up in metabolic rate, in heart rate, in lifespan, in the branching of trees, in how fast tumours grow. Biology has spent a century asking where that 3/4 comes from.

How this differs from the previous topic: Topic 32 was about self-similarity — the same shape looking the same at every scale. This one is the opposite. When something gets bigger, its parts must change at different rates; the kind of enlargement that keeps everything looking the same is physically impossible. Two faces of scale, not one idea told twice.

01Why You Cannot Simply Enlarge a Mouse

In 1638, under house arrest, Galileo finished Two New Sciences. In the Second Day he drew two bones: one from a normal animal, and one you would have to draw if you tripled that animal's size and wanted the bone not to snap. The second is grotesquely thick, like a tree stump.

His reasoning is very simple. Multiply every edge of an object by k and its surface area grows as k², its volume — and therefore its weight — as k³. How much load a bone survives depends on its cross-sectional area, which also grows only as k². So weight outruns support: triple the size and the stress in the bone triples. Enlarge anything far enough and it is crushed by itself. This is the square-cube law.

Multiplying every edge by the same number so that the shape is preserved is called isometry. What Galileo proved is that organisms cannot be isometric. Getting bigger forces a change of proportion — and growth in which the proportions shift with size is allometry.

Same shape, edge ×1 / ×2 / ×3 Edge length Surface area (cooling, load) Volume (weight, heat) Surface per unit volume ×1×2×3 ×1×4×9 ×1×8×27 1.000.500.33
The last row is the source of all the trouble: the bigger a thing gets, the less surface each unit of its interior is allotted — for cooling, for load, for absorption alike.

Heat is a worse problem than bone. Suppose that after enlargement each gram of flesh still produces heat at the same rate. Then total heat production follows volume and grows as k³, while shedding it depends on skin and grows only as k². Divide one by the other: the heat each unit of skin must handle grows as k. Ten times bigger means ten times the cooling load per unit of surface — and whatever the skin cannot handle turns into a rising body temperature.

In 1883 the German physiologist Max Rubner drew the elegant conclusion: since an organism must keep heat production within what its skin can dump, metabolic rate ought to track body surface area — that is, mass to the 2/3 power. The argument is airtight, and it stood for half a century. Its only problem is that the data disagree.

🌀 Literature · Swift got Gulliver's rations wrong In Gulliver's Travels, the Lilliputian mathematicians measure Gulliver at twelve times their height and conclude that his body "must contain at least 1728 of theirs" (12³), so he must be fed 1728 Lilliputian rations a day. That is isometric arithmetic — food converted from volume. But food tracks metabolic rate, not volume: even on the most conservative 2/3 exponent it is 144 rations, and on the 3/4 exponent below it is 320. Swift wrote satire, but the joke lands on the wrong target here: Lilliput's treasury could actually have afforded Gulliver. It was its mathematicians who frightened the king.

02The Three-Quarters Exponent

In 1932 Max Kleiber, a Swiss-born physiologist at UC Davis, plotted the basal metabolic rate (BMR — the energy burned per unit time while awake, at rest, fasted, and neither hot nor cold) of a range of animals against body mass on a log-log plot, where both axes are marked off in factors of ten.

What he got was not Rubner's 2/3 but a straight line of slope about 0.75. The relation became Kleiber's law, usually written in a form you can use directly: BMR ≈ 3.4 × mass3/4 watts, with mass in kilograms. Put a 70 kg human in and you get about 82 W — roughly an incandescent bulb, which matches measurement.

Across five orders of magnitude of mass, one straight line 10 g100 g1 kg10 kg100 kg1 t10 t 0.1 W1 W10 W100 W1 kW Body mass (log scale) → Basal metabolic rate (log scale) Slope 1: if every gram were equally busy Linear extrapolation → 33 kW for an elephant (19× real) Mouse 30 g · 0.25 W Cat 4 kg · 9.6 W Human 70 kg · 82 W Horse 500 kg · 359 W Elephant 4 t · 1.7 kW Slope 3/4: measured An elephant has 130,000× a mouse's mass but only 7,000× the power — each gram ~19× slower
A slope below 1 means the power per gram of flesh falls with body size. The line spans five orders of magnitude and has stood for a century — though whether its slope is exactly 3/4 is still argued over (section 4).

That the slope is below 1 matters more than its exact value. It means the bigger the animal, the more frugally each of its grams burns. Divide both sides by mass and metabolic rate per gram falls as mass−1/4. An elephant has roughly 130,000 times a mouse's mass; the fourth root of 130,000 is about 19, so each of the elephant's grams runs about twenty times slower.

That −1/4 is not confined to metabolism. Heart rate and breathing rate go as mass−1/4; lifespan, gestation time and gut transit time go as mass+1/4. The two exponents are equal and opposite, so their product — "how many times does a heart beat in a lifetime" — should in theory be independent of body size. That is where the widely repeated claim that every mammal gets about a billion heartbeats comes from.

Does it hold? Partly — and where it fails is more interesting than where it holds.

Rate falls, lifespan rises, the product is roughly flat — except for us AnimalBody massResting heart rateTypical lifespanLifetime heartbeats (est.) 1 billion Mouse30 g600 /min3 yr Cat4 kg150 /min15 yr Horse500 kg40 /min25 yr Elephant4 t30 /min65 yr 0.95 bn1.2 bn0.53 bn1.0 bn Human70 kg70 /min60 yr 2.2 bn The first four sit within one order of magnitude; humans are the clear outlier
Figures follow compilations of the Levine (1997) kind and are estimates. The point is not the decimals: non-human mammals cluster around a billion, and humans have more than twice that — before counting the extra years modern medicine adds.

So this "conservation law" is conserved to within about one order of magnitude, and humans violate it systematically. Treating it as a rough statement about magnitudes is fine. Treating it as a budget — "my heartbeats are rationed, so I should not exercise" — inverts it completely: regular exercise lowers resting heart rate and extends life, moving both terms in the favourable direction. An identity that holds approximately between two products is not a trade-off between them.

🎯 DECISION LINE

When comparing across scales, drop "per capita", "per unit", "per machine". Those phrases assume an exponent of 1, which almost never holds in a system with a distribution network. Write the exponent out: replace y/x with y/xβ. Concretely: for any "cost per head", "throughput per box", "dose per kilo" metric, take the two raw columns, plot them log-log, and measure whether the slope is 1. If it isn't, that per-unit figure is misleading you systematically — and the wider the size range, the worse.

🌀 Ecology · large animals are supposed to be rare In 1981 John Damuth compiled population densities for a large set of mammals and found that density falls as mass−3/4 — exactly cancelling the mass+3/4 rise in each individual's energy demand. So the energy a species' population consumes per square kilometre is independent of its body size: the energetic equivalence rule. That yields a hard conclusion for conservation: what a reserve supports is not "a number of animals" but a fixed flow of energy, and body size only decides how many shares that flow is cut into. Sizing habitat by scaling area linearly with body mass therefore underestimates what large animals need.

03Where the Exponent Comes From

2/3 has a clean reason (the cooling surface). 1 has a clean reason (every gram equally busy). 3/4 sits in between, and for sixty years nobody could say what entitled it to be there. In 1997 Geoffrey West, James Brown and Brian Enquist published an answer in Science, now generally called the WBE model after their initials.

They changed the question. Not "how does an organism shed heat" but "how does an organism deliver resources to every one of its cells". Every cell you have must receive oxygen and nutrients, and that is done by a branching network of pipes running from the aorta down to the capillaries — roughly twenty-odd levels of branching in a human. They proposed three assumptions:

One: the network must fill the whole body. No cell may sit too far from a supply point, so the network is space-filling — and this is what forces its branching structure to be a fractal, a structure whose magnified parts resemble the whole → ref · Allometric scaling laws.

Two: the size of the terminal unit is independent of the size of the animal. This is the pivot of the whole argument and the assumption most often skipped. Capillary diameter is around five to ten micrometres in every mammal — a shrew and a blue whale use the same model of capillary. Exchange happens at the tip, and the physics of the tip (the distance a gas must diffuse) has nothing to do with how big the animal is.

Three: evolution has minimised the energy spent moving resources around. A heart pushing blood costs energy, and that expenditure produces no biological return, so selection has pressed on it for a very long time.

Body grows → more branching levels → but the tips stay the same width Small animal · 3 levels Large animal · 4 levels grows The teal terminal units (capillaries) are identical on both sides — this is where 3/4 comes from Schematic. A real mammal branches roughly twenty-odd levels from aorta to capillary, not three or four
With the tips fixed, the only way for the network to grow is to add levels. And each added level means a thicker trunk and slower flow — that geometric surcharge is why metabolism per gram falls with size.

Solve the three assumptions together and out comes 3/4, along with a whole family of quarter powers: metabolic rate 3/4, heart rate −1/4, lifespan +1/4, aortic radius 3/8. The key sentence is this: 3/4 does not come from the three-dimensional geometry of the body, it comes from the geometry of the supply network. If an organism exchanged directly across its surface — as very small, purely diffusive things do — then 2/3 would be right. The moment you need a pipe network to carry things inward, and its tips cannot be enlarged along with everything else, the network starts charging a toll, and that toll rises superlinearly with size.

Which is why every gram of the elephant has to slow down. Its cells did not get lazier. Getting oxygen to those cells got more expensive.

🎯 DECISION LINE

To judge whether something can still get bigger, first find its distribution network — cooling, power, bandwidth, approvals, attention: anything that must reach every endpoint. Then ask one question: can that network's terminal unit grow too? If the tips are fixed while the total keeps rising, supply cost rises superlinearly and your ceiling is there, independent of how much resource you have. There are only two ways through: make the tips bigger (a fatter interface), or split the system in two and give each half its own network. Spending more money is on neither path.

🌀 Neuroscience · a bigger brain has no choice but to modularise Zhang and Sejnowski found in 2000 that across mammals the volume of white matter (long-range wiring) grows as the 4/3 power of grey matter volume (local processing) — the same geometry of fixed tips and a network absorbing a superlinear cost. The implication is blunt: to keep every part of the cortex directly connected to every other, wiring volume would grow faster than the brain itself, so a brain cannot get bigger while preserving global connectivity. It has to break into modules and sacrifice long-range bandwidth. Modularity here is not a design virtue; it is geometry forcing your hand.

04Where This Breaks Down

Now the other side. Scaling laws are one of the most confidently cited pieces of complexity science, and the arguments inside the field are far larger than the ones visible outside it.

First, whether the exponent is 3/4 has never been settled. In 2001 Dodds, Rothman and Weitz re-examined several historical mammal and bird datasets and concluded that, treating 2/3 as the hypothesis to be rejected, the data do not reject it. In 2003 White and Seymour assembled 619 mammal species across five orders of magnitude and, after correcting for body temperature and digestive state, obtained 2/3. WBE's own datasets give 3/4. Nobody has miscalculated; there is simply no consensus about which data to include and what to correct for.

Why this argument has run for a century ① Data spanning only 2 decades 3/4 2/3 The two lines differ by less than the scatter Which one you fit depends on which species you collected ② And the real data is curved single power-law fit measured Shallower slope at small masses, steeper at large "The one true exponent" may simply not exist
The left panel explains why the argument never ends: inside a window of two or three decades, 2/3 and 3/4 differ by less than the scatter. The right is what Kolokotrones et al. reported in Nature in 2010 — the slope is not constant.

Second, and worse: the line may not be a line. In 2010 Kolokotrones and colleagues used a much larger dataset and found that mammalian metabolic rate is curved on log-log axes — shallower at small masses, steeper at large — and still curved after correcting for body temperature. If it is curved, then "what is the exponent of this law" has no single answer, only "what is the local slope over the range you care about". The left panel above shows the other half of the problem: within two or three orders of magnitude, 2/3 and 3/4 are drawn almost on top of each other.

Third, the WBE derivation itself has been contested. Kozłowski and Konarzewski argued in 2004 that the model has specific mathematical problems — for instance that it needs the number of terminal units to be proportional to body mass, which the model cannot consistently derive from within. WBE replied; the two sides have never reconciled. That a model can be attacked this concretely is exactly what makes it scientific. It also means "3/4 has been explained" was said too early.

Fourth, an interspecific law does not transfer to within a species. Kleiber's law compares different species. Inside one species, the relation between individual size and metabolic rate is a separate question with often different exponents, sometimes near 1. So "elephants burn less per gram, therefore a heavier person burns less per kilo" has no support — it takes a line drawn between species and uses it to explain points within one. The Gulliver arithmetic in section 1 commits the same error; it survives only because its conclusion is coarse enough (1728 is far above what any sublinear exponent gives).

Fifth, do not read a scaling law as destiny. 3/4 describes an evolved, geometrically constrained central tendency, not an impassable physical ceiling. Animals of the same mass differ several-fold in metabolic rate; birds and mammals have visibly different intercepts; endotherms and ectotherms differ by an order of magnitude. A scaling law tells you roughly which band you will land in, not which point.

🎯 DECISION LINE

Before quoting any scaling exponent, state three things: whether the line runs across individuals, across species, or across institutions — the exponents routinely differ; how many orders of magnitude the data span — under three, you cannot separate 2/3 from 3/4 and any decimal you report is false precision; whether curvature was tested — fit only a straight line and you will never learn that it bends. If you cannot answer all three, report the direction ("sublinear") and not the number. Do not say 0.75.

🎒 Scenarios · BigCat

  1. Teams & organisationsOnce a team goes from 8 people to 25, the old arrangement — one weekly sync where everyone knows what everyone is doing — starts to fail: the meeting gets longer and information still leaks. Nobody got lazier. The number of pairs that need aligning grows as the square of headcount while each person's weekly hours are a constant, so alignment bandwidth per person necessarily falls with size. What to change: kill the round-the-table status update and have each module commit to exactly one number and one contact person externally. The test is concrete — if sync-meeting length has grown in step with headcount over the last two quarters, you are still running an allometric team isometrically.
  2. Investing & position sizingA strategy works, so the natural move is to add capital. But as size grows, what fails first is usually not the thesis — it is the distribution network. Market liquidity is the plumbing, your order is what must go through it, and the terminal unit (the depth a single fill can consume) does not widen because you have more money. Impact cost therefore rises superlinearly and returns start being eaten while the logic is still entirely correct. What to change: give every strategy an estimated capacity ceiling (back it out from a fixed fraction of the instrument's average daily volume) and, once there, stop adding and open an uncorrelated line instead. Watch not the return but the slippage between your average fill and the price at the moment you decided — it is the direct read-out of a network beginning to clog, and it arrives long before performance does.
  3. Health & energy"I'm 20% stronger than last year, so add 20% to the volume" is isometric arithmetic. Load can be added linearly; recovery cannot. Sleep has a hard ceiling, and tendon and connective tissue adapt far more slowly than muscle — they are the tip that does not scale up with you. What to change: when adding, move only one variable at a time (intensity, frequency or duration — pick one), and treat morning resting heart rate as the read-out of whether the supply side is keeping up; three consecutive days above your own baseline means stop adding, rather than deciding by how you feel that morning. Note this is the opposite of the "heartbeats are rationed" misreading in section 2: what you watch here is the change in rate, not the total count.

🌀 Crossings

Going Deeper

If terminal units could be enlarged, would the ceiling on size disappear?

It would move, not vanish. Bigger tips mean proportionally less exchange area (the square-cube law again), so you are trading exchange efficiency for supply cost. A real case is the bird lung, which raises exchange efficiency using unidirectional airflow rather than larger alveoli — and birds have higher metabolic rates than mammals of the same mass. So the knob is not only "how big are the tips" but "how do the tips work".

Why is the gap between 2/3 and 3/4 worth a century of argument?

Because they point to entirely different causal stories: 2/3 says the constraint lives at the interface with the outside world (shedding heat), 3/4 says it lives in internal transport. Which you pick determines what you would try to change. And within the window of real data the two are nearly indistinguishable — which is the characteristic bind of complexity science: vast disagreement about mechanism, minute difference in observation.

Do scaling laws hold in systems that learn and restructure themselves?

Be careful. A biological supply network has been optimised by evolution, so its geometric constraints bind tightly. An organisation can redesign its information network — add a management layer, change tooling, split into two independent units. In human-made systems a scaling law is closer to "where you end up if you don't change the structure" than to an inviolable law. Its use is telling you when you must restructure.

Can "metabolism per gram falls with size" be turned into a longevity prescription?

That is exactly where the "rationed heartbeats" misreading comes from. A correlation that holds between species cannot be moved onto individuals as causation — caloric restriction extends life in some animals and has given mixed results in primates, and "slowing metabolism" and "reducing metabolic damage" are not the same thing at all. An interspecific line tells you where to look for a mechanism. It does not tell you how many meals to eat.

Further Reading