Denormal numbers, also known as subnormal numbers, are floating-point numbers that fill the gap between zero and the smallest normalized floating-point number, allowing for the representation of very small values at the cost of precision. They enable gradual underflow, which helps prevent abrupt underflow errors in numerical computations by providing a smooth transition to zero.