1906: Ludwig Boltzmann dies

On 5 September 1906 the Austrian physicist Ludwig Eduard Boltzmann ended his life while staying in the coastal town of Duino, near Trieste. He was sixty-two. For more than three decades he had fought to establish statistical mechanics—the claim that the familiar laws of heat and energy are simply the large-scale appearance of countless random collisions among invisible atoms and molecules.

Boltzmann’s lasting emblem is the equation now carved on his Vienna gravestone: S = k log W. Entropy S measures disorder; W counts the microscopic arrangements compatible with a given macroscopic state; k is the constant that bears his name. In an era when many leading physicists still doubted that atoms were real, the formula looked like philosophical speculation dressed up as mathematics. The intellectual isolation, combined with bouts of severe depression, wore him down.

Four decades later the same mathematical structure reappeared in an entirely different domain. Claude Shannon, working at Bell Labs on the problem of reliable communication, needed a precise measure of uncertainty. He recognised that Boltzmann’s logarithmic counting of possibilities supplied exactly what he required. Shannon’s information entropy H = −∑ pᵢ log pᵢ became the foundation of information theory and, soon afterward, of coding, compression and digital computation.

From information theory the concept migrated into artificial intelligence. Cross-entropy is the standard loss function that trains today’s neural networks; information gain guides the splits in decision trees; entropy regularisation keeps reinforcement-learning agents exploring. Whenever a large language model calculates the probability distribution over the next token, it performs a computation whose conceptual ancestry runs directly back to Boltzmann’s statistical insight.

Thus an idea forged in the lonely struggles of a nineteenth-century physicist continues to shape the algorithms that recognise speech, translate languages and generate text in 2024. On this day we remember not only a tragic death but the quiet persistence of a mathematical way of thinking about uncertainty—one that proved indispensable once machines themselves began to learn.