The ceiling on every engine, set before it was designed

In 1824 a French military engineer published a short book asking why steam engines were so bad, and whether a better working substance might help. Steam engines at the time converted about three percent of their fuel’s energy into work. The question was practical and the answer was not.

Sadi Carnot showed that the ceiling on efficiency does not depend on the working substance at all. It does not depend on the mechanism, the materials, the pressures, or anything an engineer can adjust. It depends on two temperatures, and on nothing else in the universe.

The claim #

An engine takes heat $Q_h$ from something hot, converts some of it to work $W$, and dumps the remainder $Q_c$ into something cold. Its efficiency is the fraction converted, $W/Q_h$. Carnot’s result is

$$\eta_{\max} = 1 - \frac{T_c}{T_h},$$

with the temperatures measured absolutely, from zero.

A power station burning coal at 550 °C and rejecting heat to a river at 20 °C has $T_c/T_h = 293/823 = 0.356$, so its ceiling is 64%. Real stations reach the high thirties. A car engine, running hotter but rejecting to warm air, has a ceiling near 60% and delivers about 25%. The gap between ceiling and achievement is engineering; the ceiling itself is not negotiable.

Two features of the formula are worth reading carefully. It reaches 100% only if $T_c = 0$, which is unreachable, so no engine converts all its heat to work. And it goes to zero as the two temperatures approach, so an engine with nothing cold to dump into produces no work at all, regardless of how much heat is available. An ocean contains an enormous quantity of thermal energy and it is useless as a power source unless something colder is on hand.

Why the substance cannot matter #

The argument for the ceiling never mentions how the engine works, and that is what makes it powerful. It is a proof by contradiction against the second law.

Suppose an engine existed that beat the Carnot efficiency between the same two reservoirs. Run it forward to produce work, and use that work to drive a Carnot engine backwards as a refrigerator, pumping heat from cold to hot. The super-efficient engine produces more work than the reversed Carnot needs to return the same heat. The combination, taken as a whole, moves heat from cold to hot with no net input of anything.

That is exactly what the counting argument forbids: the combined operation would decrease the total number of arrangements available. So no such engine exists, and any engine that reaches the ceiling must be reversible.

Notice what the proof used: only that heat does not flow spontaneously from cold to hot. Nothing about steam, pistons, chemistry or materials — the same indifference to mechanism that makes momentum conservation work without knowing anything about the inside of a collision. This is why Carnot’s conclusion survived a complete change in the underlying theory — he derived it believing heat was a conserved fluid, which is wrong, and the result stands anyway, because the argument rests on a structural fact rather than a mechanism.

The loop, and where the work comes from #

The cycle that reaches the ceiling is drawn on pressure and volume axes.

Four steps. Expand while in contact with the hot reservoir, drawing heat in at constant temperature. Disconnect and let it keep expanding, cooling as it does work with no heat supplied. Compress while in contact with the cold reservoir, dumping heat at constant temperature. Disconnect and compress the rest of the way, warming back to where it started.

The work per cycle is the enclosed area, and that is a genuine theorem rather than a mnemonic: work is $\int p\,dV$, positive going right and negative going left, so a closed loop leaves the area between the outward and return paths.

Two things about the diagram deserve emphasis. First, the loop must be traversed clockwise for the engine to produce work; anticlockwise is a refrigerator, consuming work to move heat the other way, and a refrigerator is a Carnot engine run backwards with no other change. Second, the area is the net work, and the individual steps are much larger — the gas does a great deal of work expanding and has much of it given back during compression. An engine’s output is a difference between two large numbers, which is why small inefficiencies in either stroke matter so much more than the output suggests.

Comparing the two loops makes the design pressure obvious. Raising $T_h$ is worth more than lowering $T_c$, because the cold end is usually already near ambient and the hot end is limited only by what the metal will tolerate. Turbine blade metallurgy is, in this sense, thermodynamics: every extra degree of tolerable inlet temperature is efficiency that the counting argument will permit and the materials previously would not.

Why these four steps and no others? Because they are the only ones that avoid wasting the opportunity. Heat crossing a finite temperature difference is irreversible — it increases the count of arrangements without producing any work in exchange — so heat must only be exchanged when the gas and the reservoir are at the same temperature, which forces the isothermal steps. And temperature must only be changed with no heat flowing at all, which forces the adiabatic ones. The Carnot cycle is not a clever invention; it is what remains once every avoidable irreversibility has been prohibited.

The counting underneath #

The proof above used the second law as a premise. It is worth seeing what the premise rests on, because the ceiling is otherwise a rule with no visible mechanism.

Every arrangement is equally likely, and the middle outcome wins because there are more ways to reach it. That is the whole of the second law: not a preference for disorder, but a count. An engine that converted heat entirely into work would move a system to a less numerous arrangement without paying for it anywhere else, which is not forbidden by any law of motion — it is merely overwhelmingly unlikely, and overwhelming here means a number with $10^{23}$ in the exponent.

Entropy is a count of arrangements, and heat flowing from hot to cold increases that count. An engine extracts work by riding that flow, and the maximum work extractable is fixed by how much the count is allowed to rise: an engine that produced more work would have to leave the total count lower than it started, which is not forbidden by any force law and is forbidden by arithmetic on a scale of $10^{23}$.

Reading the formula this way explains its otherwise puzzling shape. The heat dumped into the cold reservoir is not waste in the sense of poor design. It is the payment: the arrangement count has to end higher than it began, and dumping $Q_c$ at temperature $T_c$ is what buys the increase that extracting $Q_h$ at $T_h$ spent. The ratio $T_c/T_h$ is exactly the exchange rate between the two.

A hot reservoir is not doing anything a cold one is not. Its molecules are spread over a wider range of speeds, and an engine works by exploiting the difference between two such distributions — taking from the wide one, giving to the narrow one, and keeping the difference. When the two distributions are the same there is no difference to keep, which is the same statement as the ceiling going to zero.

That figure is a useful corrective to the idea that heat is a substance. The two reservoirs differ only in how their molecular speeds are distributed, and an engine is a device that exploits the difference between two statistical distributions. Nothing flows that is not already there.

Reversible means infinitely slow #

The ceiling is reached only by a reversible engine, and reversibility carries a cost that no one would accept.

A reversible process is one that can be run backwards through the same states with nothing left changed. That requires the system to be in equilibrium at every instant, which requires every step to be infinitesimally small, which requires the whole cycle to take infinitely long.

So the Carnot engine has efficiency $1 - T_c/T_h$ and power output zero. It is a bound, not a design.

The version of the problem with a finite time budget has its own answer, and it is a much better guide to what real plants achieve. Optimising for maximum power rather than maximum efficiency gives

$$\eta = 1 - \sqrt{T_c/T_h},$$

which is substantially lower — for the power station above, 40% rather than 64%, which is close to what such stations actually deliver. The reason real engines fall short of Carnot is not only friction and leakage. It is that an engine which produced no power would be useless, and the trade between efficiency and speed is built into the thermodynamics rather than into the hardware.

The same ceiling, several ways round #

The formula reappears in guises that do not look like engines.

Run the cycle backwards and it becomes a heat pump, and the same limit becomes a maximum coefficient of performance. Because the ratio is inverted, the numbers are startling: a heat pump moving heat from outside air at 0 °C into a house at 20 °C has a theoretical maximum of about 14 units of heat delivered per unit of work. Real units reach three or four. Delivering more heat than the energy consumed is not a violation of anything — the extra was outdoors already, and the work only moved it.

The ceiling also fixes the absolute temperature scale. Since the efficiency of a reversible engine depends only on the two temperatures and not on the substance, the ratio of two temperatures can be defined by the ratio of heats a reversible engine exchanges with them. That definition needs no thermometer, no gas, no material property whatsoever — which is what makes the Kelvin scale absolute rather than conventional, and it was Carnot’s argument that made it possible.

And the third law appears as a corollary. Reaching 100% efficiency needs a reservoir at absolute zero; reaching absolute zero would require removing all the energy from a system in a finite number of steps, and the counting argument says each step can only remove a fraction. Absolute zero is approachable and not attainable, and the two statements are the same statement.

There is one more disguise worth recognising, because it explains why the argument keeps reappearing in places with no engines in them. Any process that converts a difference into useful output is bounded the same way — the difference is the resource, and using it consumes it. That is why an engine’s ceiling, the diffusion that flattens a concentration gradient, and the electrical work a battery can do before its chemistry evens out all have the same structure. Something is unevenly distributed, and the unevenness is what is actually being spent.

The Carnot cycle is not the only one that touches the ceiling #

It is easy to read the four steps above as the reversible cycle, and they are not. Any cycle that exchanges heat with the outside world only at the two reservoir temperatures reaches the same ceiling, whatever shape it draws on the diagram, and there is a practical engine built on that observation.

The Stirling cycle replaces the two adiabatic steps with two constant-volume ones, in which the gas is heated and cooled between $T_c$ and $T_h$ while its volume does not change. Taken naively that is a disaster: heat is now crossing a large temperature difference, which is exactly the irreversibility the Carnot steps were arranged to prohibit.

The trick is that the two constant-volume steps are mirror images. The heat given up while cooling is exactly the heat required while warming, at every intermediate temperature. So it need not come from outside at all: run the gas through a regenerator — a matrix of wire mesh with a temperature gradient along it — which absorbs the heat on the way down and returns it on the way up. With a perfect regenerator the engine exchanges heat externally only at $T_h$ and $T_c$, and its efficiency is $1 - T_c/T_h$ exactly.

Robert Stirling patented it in 1816, eight years before Carnot’s book, as a hot-air engine that could not explode the way a boiler could. The regenerator is the part that matters and it was in the original patent, which makes it one of the earliest pieces of hardware whose whole purpose is to avoid an irreversibility nobody had yet named.

When the cold reservoir is the expensive half #

The formula’s two temperatures are usually discussed as though the hot one were the design variable and the cold one were simply the weather. There are two arrangements where that is inverted, and both are instructive about what the ceiling actually costs.

Ocean thermal energy conversion uses warm surface water at about 25 °C and cold water pumped up from a kilometre down at about 5 °C. Those are absolute temperatures of 298 and 278, so the ceiling is 6.7 per cent — and against that ceiling must be set the work of pumping a very large volume of water up a kilometre of pipe. Georges Claude built one in Cuba in 1930; it generated 22 kilowatts and consumed more than that in pumping. The resource is enormous and the gradient is tiny, which is the distinction the whole page rests on: the ocean’s heat is not the fuel, the twenty-degree difference is.

A spacecraft has no cold reservoir at all except radiation. Everything a power plant on Earth throws into a river must instead be radiated away, and radiated power scales as the fourth power of the radiator’s temperature — so a cold rejection temperature, which the formula rewards, demands a radiator whose area grows as $T_c^{-4}$. Halving the rejection temperature to double the efficiency multiplies the radiator by sixteen. The result is that spacecraft reject heat hot, accept the efficiency that follows, and that the largest structures on a nuclear-electric design are its radiators.

Where the model stops #

The diagram assumes a great deal that no engine satisfies.

An ideal gas as the working substance, so that the isotherms are $pV =$ constant. That law is itself a result of treating molecules as points that only collide, and it fails wherever the molecules are close enough to attract each other. Real gases deviate, steam changes phase mid-cycle, and the loop’s shape changes accordingly — though not, and this is the point, its ceiling.

Infinite reservoirs whose temperatures do not change as heat is drawn. A real boiler cools as heat is taken and a real condenser warms, so the effective temperature difference shrinks during the cycle.

No friction, no leakage, no turbulence. Every one of these converts work back into heat at the cold end, and each is a straightforward loss on top of the thermodynamic one. Friction is the cleanest everyday example of ordered motion becoming disordered, and it is irreversible for the same reason an inelastic collision is — the reverse would require $10^{23}$ molecules to agree on a direction.

Quasi-static operation, which is the fatal one for power, as above.

Heat engines specifically. The ceiling applies to converting thermal energy — the disordered motion of enormous numbers of molecules — into work. A fuel cell converts chemical energy directly and is not bound by the Carnot limit at all — it has its own, different ceiling — and neither is a battery, a photovoltaic cell, or a hydroelectric turbine. Quoting the Carnot efficiency against these is a common and consequential error: it makes direct conversion look impossible when it is merely difficult.

The right answer from the wrong theory #

The most instructive fact about this result is how it was obtained, because by any reasonable standard it should not have been.

Carnot published Réflexions sur la puissance motrice du feu in 1824, at twenty-eight, and he believed heat was caloric: a weightless, conserved fluid that flowed from hot bodies to cold ones. On that picture an engine works the way a water wheel works — the caloric falls from a high temperature to a low one, and work is extracted from the fall, with the same quantity of caloric arriving at the bottom as left the top.

That is wrong in the most direct possible way. Heat is not conserved in an engine; some of it is converted into the work, which is the entire point, and the whole of the first law contradicts the picture Carnot was using.

And the conclusion is correct, permanently, and remains the most restrictive result in engineering. The reason is that the argument never used the conservation of caloric where it mattered. What it used was the impossibility of getting work for nothing — the observation that if two engines had different efficiencies between the same reservoirs, one could drive the other backwards and produce work while returning every reservoir to its original state. That argument survives the replacement of caloric by molecular motion untouched, because it is about what cannot happen rather than about what heat is made of.

This is the same shape as the consistency argument that forces every clock to slow: forbid the outcome that would let something impossible be observed, and the forbidden outcome does the work of a hundred calculations about mechanisms. Arguments of that kind are unusually robust against being wrong about the mechanism, because they never mention it.

Carnot did not stay wrong. His unpublished notes, which surfaced only in 1878, show him abandoning caloric within a couple of years and estimating the mechanical equivalent of heat — at a value within about fifteen per cent of Joule’s, and twenty years earlier. He died of cholera at thirty-six, in 1832, and his effects were burned as a precaution.

The book was ignored for a decade until Clapeyron rewrote the argument in the analytic language of the day, and it took Kelvin and Clausius in the 1850s to reconstruct it on a correct theory of heat and produce the second law in its modern form. So the single most quoted efficiency in engineering came from a young officer trying to find out whether steam engines could be improved indefinitely, working from a theory of heat that was already doomed, and it is quoted unchanged two centuries later.

The ladder from here #

Later rungs: the Otto, Diesel, Brayton and Rankine cycles, and the specific compromises each makes. Entropy as the state function that makes the loop close. The Clausius inequality. Exergy — how much of a given quantity of energy is available to do work at all, which is a more useful accounting than energy for anything involving heat. Endoreversible thermodynamics and the finite-time bound. Refrigeration and liquefaction. Heat pumps, and why they are the only heating technology that beats one-for-one. And the thermodynamics of computation, where the ceiling reappears as a minimum energy cost per bit erased.

The gradient this anchor spends is the same one diffusion flattens with no engine attached, which is the cheapest way to see that the resource being consumed is unevenness rather than energy.