Part of The Physics of Cats.

A cat can occupy a cardboard box with remarkable disregard for its advertised dimensions. Paws disappear, the tail follows, and a long animal becomes a compact parcel.

One explanation is thermal: a compact shape exposes less area, the box shields air movement, and bedding separates the body from a cold floor. Those are real mechanisms. They do not, by themselves, tell us why a particular cat entered a particular box.

There are two problems to solve. One is how the heat moves. The other is how the animal chooses. The first can be modelled without claiming to have solved the second.

Curling Changes the Exposed Boundary

For bodies of the same volume, a sphere has the least surface area. That geometric result explains why gathering protruding parts into a compact arrangement can reduce the boundary available for heat exchange.

A crude comparison makes the scale visible. Model an extended body as a closed cylinder with radius $r=0.065\,\mathrm{m}$ and length $L=0.40\,\mathrm{m}$. Its volume and area are

$$ V=\pi r^2L,\qquad A_{\rm cyl}=2\pi rL+2\pi r^2. $$

The volume is approximately $0.00531\,\mathrm{m^3}$ (5.3 litres, roughly the volume of a 5 kg animal at tissue density) and the area $0.190\,\mathrm{m^2}$. A sphere of that volume has radius

$$ R=\left(\frac{3V}{4\pi}\right)^{1/3}\approx0.108\,\mathrm m $$

and surface area $4\pi R^2\approx0.147\,\mathrm{m^2}$: roughly 22% less.

These are assumed shapes, not feline body measurements. The cylinder omits the head, limbs and tail; a curled cat is neither solid sphere nor perfect insulation. The useful point is that the exposed boundary can change while the amount of warm material remains approximately the same.

Areas pressed against another part of the body no longer exchange heat directly with the room in the same way. They may still transfer heat internally. A contact with the floor opens a different external path. Counting all anatomical surface as exposed area would miss that distinction.

Fur Is a Resistance, Not a Source

For a uniform slab of thickness $\ell$, area $A$ and thermal conductivity $k$, steady conduction gives

$$ \dot Q=\frac{kA}{\ell}(T_1-T_2) =\frac{T_1-T_2}{R_{\rm th}}, \qquad R_{\rm th}=\frac{\ell}{kA}. $$

Here $\dot Q$ is heat flow in watts, and $R_{\rm th}$ is thermal resistance in kelvins per watt. The slab model assumes approximately one-dimensional conduction and constant material properties.

Fur is a structured mixture of fibres and air. An effective conductivity can represent their combined behaviour for a chosen condition, but it will depend on the coat, compression, moisture and air movement. The symbol $k$ is not a universal “cat constant”.

Insulation reduces the heat flow produced by a temperature difference. It does not generate warmth. A living animal supplies heat through metabolism; an insulating layer changes how quickly that heat reaches the surroundings.

Compression illustrates a possible trade-off. Contact can reduce exposed area while flattening the insulating layer. In the slab formula, reducing $\ell$ reduces resistance, and compression also changes $k$ by squeezing out air, which the formula does not capture on its own. Which effect dominates depends on the area, thickness and properties of the actual contact.

Air and Walls Receive Heat Differently

At the outside of the coat, convection transfers heat to moving air. A common boundary model is

$$ \dot Q_{\rm conv}=hA(T_f-T_a), $$

where $T_f$ is the outer-fur temperature, $T_a$ the nearby air temperature and $h$ a heat-transfer coefficient. The coefficient summarises the fluid motion and geometry; it is not a material property of air alone.

A draught changes the boundary layer next to the fur and can increase $h$. A shelter can reduce that imposed airflow, although natural convection can continue: warm air rises even without a fan.

Radiation does not require air movement. A surface exchanges infrared radiation with the objects it sees. For a diffuse grey surface viewing large surroundings at one temperature,

$$ \dot Q_{\rm rad}=\epsilon\sigma A(T_f^4-T_w^4), $$

where $\epsilon$ is emissivity, $T_w$ is the surroundings’ radiation temperature and all temperatures in this equation are in kelvins. The Stefan–Boltzmann constant is approximately $5.6704\times10^{-8}\,\mathrm{W\,m^{-2}\,K^{-4}}$ [[1]].

A room’s air temperature and radiation temperature need not agree. An animal can face a cold window while the air remains relatively warm. Conversely, sunlight adds an incoming radiative load that the simple single-temperature enclosure model does not capture.

At modest temperature differences, the fourth-power term can be written as an effective coefficient:

$$ h_{\rm rad}=\epsilon\sigma(T_f+T_w)(T_f^2+T_w^2). $$

Then $\dot Q_{\rm rad}=h_{\rm rad}A(T_f-T_w)$. This is an exact algebraic rewriting for those two temperatures; treating the coefficient as constant over changing conditions is the approximation.

The Resistances Connect in a Particular Order

For an exposed patch, heat first crosses the coat and then leaves the outer surface through convection and radiation. Coat conduction is in series with the outer exchange. Convection and radiation form parallel branches because both receive heat from the same outer surface.

If air and surrounding walls share an environmental temperature $T_e$, and radiation is linearised, this gives

$$ \dot Q_{\rm exposed}\approx \frac{T_s-T_e} {\displaystyle\frac{\ell}{kA} +\displaystyle\frac{1}{(h+h_{\rm rad})A}}, $$

where $T_s$ is the warm-side temperature beneath the coat. Computing convection and radiation with $T_s$ as the surface temperature, and ignoring the coat, would overestimate transfer: the outside of the fur is cooler than its warm side.

A worked example needs assumptions. Take $T_s=32\,{}^\circ\mathrm{C}$, $T_e=20\,{}^\circ\mathrm{C}$, $\ell=0.01\,\mathrm{m}$, $k=0.04\,\mathrm{W\,m^{-1}\,K^{-1}}$, $h=5\,\mathrm{W\,m^{-2}\,K^{-1}}$ and $\epsilon=0.95$. These are illustrative inputs, not a measured cat or a prescribed room temperature. Choose $h_{\rm rad}\approx5.77\,\mathrm{W\,m^{-2}\,K^{-1}}$, evaluated between 32 and 20 °C. That is slightly high, because the outer fur in this network sits near 23 °C, where the coefficient is about 5.5; the effect on the result is under 1%.

For $A=0.20\,\mathrm{m^2}$, the coat resistance is $1.25\,\mathrm{K/W}$ and the outer resistance approximately $0.464\,\mathrm{K/W}$. Their sum gives

$$ \dot Q_{\rm exposed}\approx \frac{12\,\mathrm{K}}{1.714\,\mathrm{K/W}} \approx7.0\,\mathrm{W}. $$

Reducing assumed exposed area to $0.14\,\mathrm{m^2}$, with every other input held fixed, reduces the result to about $4.9\,\mathrm{W}$. That 30% reduction follows directly from the area scaling of this model. It is not a measured benefit of curling.

The Floor Has a Separate Branch

Heat passing downwards through a contact patch need not cross the same coat thickness or encounter the same air boundary. Compressed fur, bedding, gaps and the floor each contribute resistance along that route.

For a simple series contact path,

$$ R_{\rm contact}=R_{\rm compressed\ coat} +R_{\rm bedding}+R_{\rm interface}+R_{\rm floor}. $$

This entire path runs in parallel with the exposed-body path. If its endpoints differ by $12\,\mathrm{K}$ and its assumed resistance is $10\,\mathrm{K/W}$, it carries another $1.2\,\mathrm{W}$. Doubling that resistance halves this branch’s heat flow, without halving convection or radiation elsewhere.

The floor is also a transient object. It can warm locally, spread heat away and change the contact temperature over time. A single fixed resistance is most useful when that complexity has either been measured or deliberately simplified. The example identifies a pathway; it does not identify the insulating value of an unspecified blanket.

A Box Changes Boundary Conditions

A box can shelter the coat from imposed airflow, exchange radiation with it through nearby walls and alter the surface under the body. It can also contain air warmed by the occupant. These effects act at different nodes in the network.

For a minimal cavity model, suppose $P$ watts enter the cavity air and an effective conductance $G$ connects that air to the room. At steady state,

$$ T_b-T_a=\frac{P}{G}. $$

Assuming $P=6\,\mathrm{W}$ and $G=2\,\mathrm{W/K}$ gives a $3\,\mathrm{K}$ rise. Both numbers are invented for illustration. They are not a prediction for a cardboard box: an open top, ventilation, wall dimensions and heat going directly into solid surfaces can change the result substantially.

Adding an effective heat capacity $C$ gives

$$ C\frac{dT_b}{dt}=P-G(T_b-T_a). $$

The warming timescale is $C/G$. With an assumed $C=60\,\mathrm{J/K}$, it is $30\,\mathrm{s}$. This lumped node is only sensible when the chosen cavity can be represented by one temperature. Real air and walls need not warm together.

The cat, coat, bedding, box walls and air can therefore require several temperatures. “The box traps heat” compresses a network into a slogan.

Preference Is a Different Measurement

A randomised shelter study by van der Leij and colleagues provided hiding boxes to one group of cats and compared them with controls. The 23 newly arrived cats (12 with a box, 11 without) were observed during 12 days of quarantine. Cats with boxes showed a faster reduction in behavioural stress scores, which reached a steady state about a week earlier; the boxes did not significantly influence weight loss. The study assessed hiding enrichment, not a heat-loss optimum [[2]].

That matters because a thermal benefit can coexist with concealment, perceived security or other motivations. Observing a cat in a box cannot distinguish those explanations. Even a measured warmer cavity would establish a changed thermal environment, not the animal’s reason for choosing it.

The network supplies a modest but useful account: posture changes geometry, fur adds resistance, bedding changes contact, and the enclosure changes air and radiative boundaries. Predicting an actual heat budget requires actual temperatures and transfer properties. Explaining preference requires behavioural evidence as well.

The parcel contains a heat source. Its motives remain its own.

Literature and calculations checked through 2026-10-12.

References

[1] National Institute of Standards and Technology. 2022 CODATA recommended value: Stefan–Boltzmann constant. https://physics.nist.gov/cgi-bin/cuu/Value?sigma.

[2] van der Leij, W. J. R., Selman, L. D. A. M., Vernooij, J. C. M., and Vinke, C. M. (2019). The effect of a hiding box on stress levels and body weight in Dutch shelter cats; a randomized controlled trial. PLoS ONE, 14(10), e0223492. https://doi.org/10.1371/journal.pone.0223492.

Changelog

  • 2026-10-12: Initial draft. Developed an illustrative thermal network and separated possible heat-transfer benefits from evidence about box preference.