2,435
个编辑
更改
无编辑摘要
In summary, determinism refers to how much confidence we have in predicting future states based on the current state's probability distribution, while degeneracy refers to how much certainty we have in inferring previous states from the current state. Systems with high determinism and low degeneracy exhibit strong causal dynamics.
In summary, determinism refers to how much confidence we have in predicting future states based on the current state's probability distribution, while degeneracy refers to how much certainty we have in inferring previous states from the current state. Systems with high determinism and low degeneracy exhibit strong causal dynamics.
==EI的函数性质==
==Properties of the EI Function==
From Equation 2, we can see that in the transition probability matrix P, EI is a function of each element, representing the conditional probabilities of transitioning from one state to another. Thus, a natural question arises: What mathematical properties does this function have? Does it have extreme points, and if so, where are they? Is it convex? What are its maximum and minimum values?
===定义域===
===Domain===
在离散状态和离散时间的马尔科夫链上，EI的定义域显然是概率转移矩阵P。P是一个由[math]N\times N[/math]个元素构成的矩阵，其中每一个元素[math]p_{ij}\in[0,1][/math]代表一个概率值，同时对于任意的行，这组概率值需要满足归一化条件，也就是对于任意的[math]\forall i\in[1,N][/math]:{{NumBlk|:|
在离散状态和离散时间的马尔科夫链上，EI的定义域显然是概率转移矩阵P。P是一个由[math]N\times N[/math]个元素构成的矩阵，其中每一个元素[math]p_{ij}\in[0,1][/math]代表一个概率值，同时对于任意的行，这组概率值需要满足归一化条件，也就是对于任意的[math]\forall i\in[1,N][/math]:
In the case of Markov chains with discrete states and discrete time, the domain of EI is clearly the transition probability matrix P. P is a matrix composed of N×N elements, each representing a probability value pij∈[0,1]. Additionally, each row must satisfy the normalization condition, meaning for any ∀i∈[1,N], the sum of the row's probabilities equals:{{NumBlk|:|
<math>
<math>
||P_i||_1=\sum_{j=1}^N p_{ij}=1
||P_i||_1=\sum_{j=1}^N p_{ij}=1
</math>
</math>
|{{EquationRef|3}}}}因此EI的定义域，也就是P的可能空间并不是全部[math]N\times N[/math]维的实数空间[math]\mathcal{R}^{N^2}[/math]，由于归一化条件{{EquationNote|3}}的存在，使得该定义域成为一个[math]N\times N[/math]维实数空间中的一个子空间。如何表达这个子空间呢？
|{{EquationRef|3}}}}Thus, the domain of EI, which is the possible space for P, is not the entire real space RN2. Due to the normalization condition, it is a subspace of the N×N-dimensional real space. How can we express this subspace?
For any row vector Pi, its value range is a hyper simplex in N-dimensional real space. For example, when N=2, this space is a line segment: pi,1+pi,2=1. When N=3, this space is a plane in three-dimensional space: pi,1+pi,2+pi,3=1. These two spaces are illustrated below:
[[文件:P1+p2=1.png|301x301像素|替代=|链接=https://wiki.swarma.org/index.php/%E6%96%87%E4%BB%B6:P1+p2=1.png]][[文件:P1+p2+p3=1.png|380x380像素|替代=|链接=https://wiki.swarma.org/index.php/%E6%96%87%E4%BB%B6:P1+p2+p3=1.png]]
[[文件:P1+p2=1.png|301x301像素|替代=|链接=https://wiki.swarma.org/index.php/%E6%96%87%E4%BB%B6:P1+p2=1.png]][[文件:P1+p2+p3=1.png|380x380像素|替代=|链接=https://wiki.swarma.org/index.php/%E6%96%87%E4%BB%B6:P1+p2+p3=1.png]]
In the general case, we define the value range space of the row vector Pi in N-dimensional space as Δ={pj∣∑j=1Npj=1,pj∈[0,1]}. The Cartesian product of N such spaces forms the domain of EI:
<math>
<math>
Dom(EI)=\Delta\times \Delta\cdots\times\Delta=\Delta^N
Dom(EI)=\Delta\times \Delta\cdots\times\Delta=\Delta^N
</math>
</math>
===一阶导数及极值点===
===First Derivative and Extreme Points===
Taking the first derivative of Equation 2 with respect to pij, and considering the normalization condition ∑j=1Npij=1, we get:{{NumBlk|:|
<math>
<math>
\frac{\partial EI}{\partial p_{ij}}=\log\left(\frac{p_{ij}}{p_{iN}}\right)-\log\left(\frac{\bar{p}_{\cdot j}}{\bar{p}_{\cdot N}}\right),
\frac{\partial EI}{\partial p_{ij}}=\log\left(\frac{p_{ij}}{p_{iN}}\right)-\log\left(\frac{\bar{p}_{\cdot j}}{\bar{p}_{\cdot N}}\right),
</math>
</math>
|{{EquationRef|3}}}}其中，<math>p_{ij}</math>表示P中第i行第j列的条件概率，因为P每一行有归一化约束条件{{EquationNote|2}}，所以EI函数本身有<math>N(N-1)</math>个自由变元，我们可以取<math>1\leq i\leq N, 1\leq j\leq N-1</math>。<math>p_{iN}</math>表示第i行第N列的条件概率，<math>\bar{p}_{\cdot j}, \bar{p}_{\cdot N}</math><nowiki>则分别表示第j列和第N列条件概率的均值。不难看出，该导数有定义的前提是对于选定的[math]i,j\in[1,N][/math]，[math]p_{ij},p_{iN},\bar{p}_{\cdot j}\equiv \frac{\sum_{k=1}^Np_{kj}}{N},\bar{p}_{\cdot N}\equiv \frac{\sum_{k=1}^Np_{kN}}{N}[/math]都大于0，只有满足这个条件，则EI在[math]p_{ij}[/math]附近是可导的。否则，如果有一项为0，则导数不存在。</nowiki>
|{{EquationRef|3}}}}
Here, pij denotes the conditional probability of transitioning from state i to state j in matrix P. Since each row of P is subject to the normalization constraint, the EI function has N(N−1) degrees of freedom. We can select 1≤i≤N and 1≤j≤N−1, with piN representing the conditional probability in the i-th row and N-th column. pˉ⋅j and pˉ⋅N represent the average conditional probabilities of the j-th and N-th columns, respectively.
It is easy to see that this derivative is defined only when, for the selected i,j∈[1,N], the probabilities pij,piN,pˉ⋅j,pˉ⋅N are all greater than 0. Otherwise, if any of these terms are 0, the derivative does not exist.
Setting Equation 3 to 0 yields the extreme points. For any 1≤i≤N,1≤j≤N−1, the following equation holds:
<math>
<math>
</math>
</math>
It is straightforward to compute that in this case, EI=0, indicating that EI reaches an extreme point. From the second derivative of EI, we can easily determine that this is a minimum point. In other words, EI has many minimum points, as long as all the row vectors of the transition probability matrix are identical. Regardless of the specific distribution of these row vectors, EI will be zero.
===二阶导数与凸性===
===Second Derivative and Convexity===
进一步地，为了求得EI函数的凸性，我们可以求出EI这个函数的二阶导数，其中<math>1\leq s \leq N, 1\leq t \leq N-1 </math>。首先我们需要引入一个函数符号<math>\delta_{i,j} </math>，
To explore the convexity of the EI function, we can compute its second derivative<math>\frac{\partial^2 EI}{\partial p_{ij}\partial p_{st}}</math>, where 1≤s≤N and 1≤t≤N−1. First, we introduce a function symbol δi,j, and then proceed to derive the second derivative. When i=s:
<math>
<math>