The question Landauer asked
In 1961, IBM researcher Rolf Landauer asked a question nobody in computer science was asking. When a computer overwrites a memory cell — when a bit changes from 0 to 1 or 1 to 0 — what's the minimum energy required?
Not the energy used by transistors. Not the energy used by cooling. Not the energy used by wires. Just the fundamental, unavoidable, physics-mandated minimum. The thermodynamic floor.
The answer
He proved that erasing a single bit of information requires at minimum kT ln 2 joules of energy — where k is Boltzmann's constant and T is the temperature of the environment.
At room temperature (300 Kelvin), that's about 2.87 × 10⁻²¹ joules per bit. It's a tiny number. But it's a real number. And it can't be avoided.
"Information is physical." — Rolf Landauer, 1991
Why erasing costs energy
When you erase information, you reduce the number of possible states of a system. Two states become one. This is the definition of a decrease in entropy inside your computer — but the second law of thermodynamics says total entropy can never decrease.
So the entropy has to go somewhere. It goes into the environment, as heat. That heat is the minimum energy cost of the erasure.
The important word is minimum. A perfectly efficient computer operating at the thermodynamic floor would still pay kT ln 2 per erasure. Landauer didn't set a target. He set a floor.
How far from the floor we are
A modern transistor erases a bit using roughly 10⁻¹⁸ joules — one billionth of a joule. The theoretical floor is 10⁻²¹ joules. Three orders of magnitude gap. A thousand times more energy than physics requires.
Multiply that by the trillions of operations your phone performs every second. Multiply it by the billions of phones. Multiply it by the millions of data centers. The gap adds up.
The implication
Landauer's principle created a new category of operation in computer science: irreversible operations. These are operations that destroy information — AND, OR, addition with loss of the carry, anything that has fewer output bits than input bits.
And their counterparts: reversible operations. These are operations that preserve information — XOR (with the original inputs retained), bit permutation, NOT, anything bijective. For reversible operations, Landauer's floor is zero. They can theoretically cost nothing.
Erasing information costs energy. Rearranging information doesn't. The less your hardware erases, the less energy it burns. Operations that don't erase anything — bijective operations — have a theoretical energy cost of zero. That distinction is the foundation of the Periodic Stack.