The Si(100) surface is one of the most intensively investigated surfaces both experimentally and theoretically over the past decades because of the technological importance of this surface, on which various devices have been fabricated. To minimize the system free energy, the atoms of a crystal surface may form different reconstructions. It has been well established that pairs of top Si atoms form dimers along the [011] direction. The dimer is the basic reconstruction configuration on the Si(100) surface. However, the details of dimer arrangement are complicated and still a subject of controversy. High order reconstructions with mixtures of different basic reconstruction configurations, such as p(2×1), p(2×2), and c(4×2) with similar total energies, are always revealed in computational simulations, although deviating from experimental results which show domains of these basic reconstructions at very low temperatures (below 40K), mainly c(4×2) reconstruction between 80K to 200K, and symmetric p(2×1) as the dominant geometry at room temperature (RT), respectively. In theoretical calculations, symmetric p(2×1) reconstruction with a relatively high total energy has never been shown to be stable, and is widely accepted to be the time average of p(2×2) and c(4×2). A systematic theoretical simulation at the atomistic level is needed to clarify the various controversies and to reveal the actual nature of the reconstructions on this important surface. In this work, first-principles simulations were performed to study the reconstructions and dynamic properties of the Si(100) surface. The adopted computational approaches include the SIESTA, an efficient code and method that is based on density functional theory. A localized linear combination of numerical atomic-orbital basis sets for the description of valence electrons and norm-conserving nonlocal pseudopotentials for the atomic core was adopted. A self-consistent-charge density-functional-based tight-binding method (SCC-DFTB) was applied to investigate this surface as well. It was found that the surface reconstructions are very sensitive to the initial morphologies of the top atomic layer. Consequently, domains of different reconstructions may “grow up” from possible initial local structures developed under temperature fluctuations or other external disturbances. This result agrees with the recent experimental results obtained using Scanning Tunneling Microscopy (STM) at very low temperatures. Accordingly, the relatively low abundance of p(2×2) reconstructed surface domains found in experiments can be attributed to the difficulty in forming the “seed” of this structure compared with the others. Also, in the simulation of initial-geometry dependent reconstruction, the reliabilities of the results from SIESTA and DFTB have been evaluated. Comparisons made between the experimental and SIESTA results indicate that the use of DFTB is inadequate to simulate a semiconductor surface where weak and long-range interactions should be considered. However, in simulating the static properties of the surface, DFTB could provide a similar reliability to that provided by SIESTA. In dynamic simulations, a phase transition from p(2×1) to c(4×2) was observed at around 100K. The strain in the second layer of p(2×1) is expected to be the driving force and plays an important role, this is because such a temperature is too low to overcome the energy barrier to activate the dimer flipping in p(2×1). It was noted that the p(2×1) is just a metastable reconstruction, which can be observed in experiments at very low temperatures when the strain is frozen. Such a reconstruction disappears when the temperature rises, resulting in the c(4×2) reconstruction as the main reconstruction configuration as observed in the experiments on the Si(100) from 80K to 200K.