Why Does Information Cost Energy?

Frame the driving question: why does manipulating something as abstract as a 'bit' demand real, measurable energy? Show that this is not just an engineering nuisance but a physics question.

• Computers, brains, and even Maxwell's demon scenarios all involve information
• All real devices heat up when they compute
• Question: is the heat from imperfect engineering, or from information itself?

Let the learner distribute a 100% budget across possible causes of computation heat, then commit before seeing the explanation.

• Allocator with three buckets: 'leaky transistors', 'wire resistance', 'the act of erasing information'
• Read-only scenario versus write/erase scenario
• Commit a prediction; later scenes reveal Landauer's split

Present empirical evidence: modern reversible-computing experiments approach the kT·ln(2) floor, showing that even with near-perfect hardware, erasure still costs energy.

• Bérut et al. (2012) microscopic bit erasure measured ~kT·ln(2) per bit at room temperature
• Reducing friction-like losses still leaves a residual floor
• The floor scales with temperature T, as thermodynamic theory predicts

Explain why erasure has a thermodynamic floor: a bit is a physical system with two distinguishable microstates; erasing it merges states and reduces entropy by k·ln(2), which the second law forces to be expelled as heat Q ≥ kT·ln(2).

• A bit corresponds to two distinguishable physical states
• Erasing maps both states onto one, collapsing phase-space volume
• Entropy decrease ΔS ≥ -k·ln(2); the second law requires heat Q ≥ T·ΔS ≥ kT·ln(2) out
• Reading or copying a bit can, in principle, be done reversibly at no energy cost

A manipulable ledger where the learner toggles a bit between 0 and 1 and tracks the system's entropy, the environment's entropy, and the heat dissipated.

• Setting a bit (from unknown to known) can be done reversibly if the environment stores the old state
• Erasing a bit forces a net entropy export to the environment
• Total entropy never decreases

Clarify the boundary: Landauer's principle applies specifically to logically irreversible operations (erasure, AND, merge). Reversible gates (NOT, CNOT, Fredkin, Toffoli) can in principle cost arbitrarily little energy.

• Reversible logic gates preserve distinguishability of inputs and outputs
• Only information-destroying steps pay the kT·ln(2) tax
• Real devices still waste energy due to finite speed and non-adiabatic driving

Let the learner run a tiny demon simulation: the demon sorts fast and slow molecules, gaining information, and the learner watches where the heat actually appears — not at the demon's 'decision', but at the moment its memory is erased.

• Information gain by the demon does not, by itself, violate the second law
• Erasing the demon's memory pays the thermodynamic bill
• Total heat dumped matches Landauer's prediction

Directly answer the driving question: information costs energy because physically realizing distinguishable states requires entropy, and any logically irreversible step — above all, erasure — must export at least kT·ln(2) of entropy per bit as heat.

• The heat is not (only) engineering sloppiness; it is mandated by the second law
• Better hardware lowers the multiplicative constant but not the floor
• Lower temperature shrinks the bill: cold computing is energetically cheaper per erased bit
• This is why reversible computing is the theoretical route to energy-efficient information processing