模擬退火是一種通用概率演算法，用來在固定时间内寻求在一個大的搜尋空間內找到的最優解。





模擬退火來自冶金學的專有名詞退火。退火是將材料加熱後再經特定速率冷卻，目的是增大晶粒的體積，並且減少晶格中的缺陷。材料中的原子原來會停留在使內能有局部最小值的位置，加熱使能量變大，原子會離開原來位置，而隨機在其他位置中移動。退火冷卻時速度較慢，使得原子有較多可能可以找到內能比原先更低的位置。

模擬退火的原理也和金屬退火的原理近似：我們將熱力學的理論套用到統計學上，將搜尋空間內每一點想像成空氣內的分子；分子的能量，就是它本身的動能；而搜尋空間內的每一點，也像空氣分子一樣帶有「能量」，以表示該點對命題的合適程度。演算法先以搜尋空間內一個任意點作起始：每一步先選擇一個「鄰居」，然後再計算從現有位置到達「鄰居」的機率。

可以證明，模擬退火演算法所得解依機率收斂到全局最優解。

-Simulated annealing is a generic probabilistic algorithm used in a fixed period of time to seek in a large search space to find the optimal solution.





Simulated annealing terminology from metallurgical annealing. Heating the material after annealing is cooled at a specific rate through the purpose is to increase the grain size, and reduce defects in the crystal lattice. Atoms in the original material will remain in the internal energy of a local minimum position, heating energy increases, atoms leave the original position, and move randomly in other locations. Annealing cooling rate is slow, so that atoms are more likely to be found within a position lower than the original.

The principle is also simulated annealing and metal annealing approximation principle: We will apply the theory of thermodynamics to statistical would imagine every point within the search space of air molecules　molecular energy is the kinetic energy of its own　while searching for space every point, like air