Abstract:We consider the Curie-Weiss model at a given initial temperature in vanishing external field evolving under a Glauber spin-flip dynamics corresponding to a possibly different temperature. We study the limiting conditional probabilities and their continuity properties and discuss their set of points of discontinuity (bad points). We provide a complete analysis of the transition between Gibbsian and non-Gibbsian behavior as a function of time, extending earlier work for the case of independent spin-flip dynamics. For initial temperature bigger than one we prove that the time-evolved measure stays Gibbs forever, for any (possibly low) temperature of the dynamics. In the regime of heating to low-temperatures from even lower temperatures, when the initial temperature is smaller than the temperature of the dynamics, and smaller than 1, we prove that the time-evolved measure is Gibbs initially and becomes non-Gibbs after a sharp transition time. We find this regime is further divided into a region where only symmetric bad configurations exist, and a region where this symmetry is broken. In the regime of further cooling from low-temperatures there is always symmetry-breaking in the set of bad configurations. These bad configurations are created by a new mechanism which is related to the occurrence of periodic orbits for the vector field which describes the dynamics of Euler-Lagrange equations for the path large deviation functional for the order parameter. To our knowledge this is the first example of the rigorous study of non-Gibbsian phenomena related to cooling, albeit in a mean-field setup.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is related to the non - Gibbs behavior of the Curie - Weiss model under low - temperature dynamics. Specifically, the author studied the conditional probability and its continuity properties of the Curie - Weiss model with an initial temperature of \(0 < \beta^{-1} \leq \infty\) after evolving through Glauber spin - flip dynamics (with a temperature of \(0 < (\beta')^{-1} \leq \infty\)) in the absence of an external magnetic field. The main focuses are:
1. **Transition between Gibbs and non - Gibbs properties**: Investigate whether the measure after time evolution maintains the Gibbs property under different initial and dynamic temperatures, and if the Gibbs property is lost, when this transition occurs.
2. **Set of discontinuity points (bad points)**: Analyze the set of discontinuity points (bad points) of the conditional probability and explore the properties of these points.
3. **Symmetry and symmetry breaking**: In particular, during the heating and cooling processes, whether the symmetry of the conditional probability will be broken and the specific mechanism of this symmetry breaking.
### Main findings
1. **Initial high temperature, arbitrary dynamic temperature**:
- If the initial temperature \(\beta^{-1} \geq 1\), then the model after time evolution maintains the Gibbs property at all times \(t \geq 0\).
2. **Heating from an initial low temperature**:
- For any \(\beta'\), there exists a value \(\beta_{\text{SB}}^{-1}(\beta') < \beta'^{-1}\) such that:
- If \(\beta_{\text{SB}}^{-1}(\beta') \leq \beta^{-1}\), then:
- For all \(0 \leq t \leq t_{\text{GS}}(\beta, \beta')=\frac{\ln(\beta' - \beta)}{4(1 - \beta')}\left(\frac{1 - \beta}{1 - \beta'}\right)\), the model after time evolution is Gibbs.
- For all \(t > t_{\text{GS}}(\beta, \beta')\), the model is non - Gibbs, and the conditional probability is discontinuous at \(\hat{m} = 0\) and continuous at any \(\hat{m} \neq 0\).
- If \(0 < \beta^{-1} < \beta_{\text{SB}}^{-1}(\beta')\), then there exist sharp time values \(0 < t_0(\beta, \beta') < t_1(\beta, \beta') < \infty\) such that:
- For all \(0 \leq t \leq t_0(\beta, \beta')\), the model after time evolution is Gibbs.
- For all \(t_0(\beta, \beta') < t < t_1(\beta, \beta')\), there exists a \(\hat{m}_c=\hat{m}_c(\beta, \beta'; t) \in (0, 1)\) such that the limit conditional probability is discontinuous at points \(\pm \hat{m}_c\) and continuous elsewhere.
- For all \(t > t_1(\beta, \beta')\), the limit conditional probability is discontinuous at \(\hat{m} = 0\) and continuous at any \(\hat{m} \neq 0\).
3. **Cooling from an initial low temperature**:
- For \(\beta'^{-1} < \beta^{-1} < 1\), there exists a time threshold \(t_{\text{per}}(\beta, \be\)