这个积分可以在 Matlab 或 Mathematica 中以数值方式完成吗？

Question

我希望能够完全用数字进行积分。

其中、 和、 和 是常量，为简单起见，都可以设置为1。

x 上的积分可以手动或使用 Mathematica 解析完成，然后y 上的积分可以使用 NIntegrate 数值完成，但这两种方法给出不同的答案。

分析：

In[160]:= ex := 2 (1 - Cos[x])

In[149]:= ey := 2 (1 - Cos[y])

In[161]:= kx := 1/(1 + Exp[ex])

In[151]:= ky := 1/(1 + Exp[ey])

In[162]:= Fn1 := 1/(2 \[Pi]) ((Cos[(x + y)/2])^2)/(ex - ey)

In[163]:= Integrate[Fn1, {x, -Pi, Pi}]

Out[163]= -(1/(4 \[Pi]))
 If[Re[y] >= \[Pi] || \[Pi] + Re[y] <= 0 || 
   y \[NotElement] Reals, \[Pi] Cos[y] - Log[-Cos[y/2]] Sin[y] + 
   Log[Cos[y/2]] Sin[y], 
  Integrate[Cos[(x + y)/2]^2/(Cos[x] - Cos[y]), {x, -\[Pi], \[Pi]}, 
   Assumptions -> ! (Re[y] >= \[Pi] || \[Pi] + Re[y] <= 0 || 
       y \[NotElement] Reals)]]

In[164]:= Fn2 := -1/(
  4 \[Pi]) ((\[Pi] Cos[y] - Log[-Cos[y/2]] Sin[y] + 
     Log[Cos[y/2]] Sin[y]) (1 - ky) ky )/(2 \[Pi])

In[165]:= NIntegrate[Fn2, {y, -Pi, Pi}]

Out[165]= -0.0160323 - 2.23302*10^-15 I

数值方法一：

In[107]:= Fn4 := 
 1/(4 \[Pi]^2) ((Cos[(x + y)/2])^2) (1 - ky) ky/(ex - ey)

In[109]:= NIntegrate[Fn4, {x, -Pi, Pi}, {y, -Pi, Pi}]

During evaluation of In[109]:= NIntegrate::slwcon: Numerical integration converging too slowly; suspect one of the following: singularity, value of the integration is 0, highly oscillatory integrand, or WorkingPrecision too small. >>

During evaluation of In[109]:= NIntegrate::ncvb: NIntegrate failed to converge to prescribed accuracy after 18 recursive bisections in x near {x,y} = {0.0000202323,2.16219}. NIntegrate obtained 132827.66472461013` and 19442.543606302774` for the integral and error estimates. >>

Out[109]= 132828.

数字 2：

In[113]:= delta = .001;
pw[x_, y_] := Piecewise[{{1, Abs[Abs[x] - Abs[y]] > delta}}, 0]

In[116]:= Fn5 := (Fn4)*pw[Cos[x], Cos[y]]

In[131]:= NIntegrate[Fn5, {x, -Pi, Pi}, {y, -Pi, Pi}]

During evaluation of In[131]:= NIntegrate::slwcon: Numerical integration converging too slowly; suspect one of the following: singularity, value of the integration is 0, highly oscillatory integrand, or WorkingPrecision too small. >>

During evaluation of In[131]:= NIntegrate::eincr: The global error of the strategy GlobalAdaptive has increased more than 2000 times. The global error is expected to decrease monotonically after a number of integrand evaluations. Suspect one of the following: the working precision is insufficient for the specified precision goal; the integrand is highly oscillatory or it is not a (piecewise) smooth function; or the true value of the integral is 0. Increasing the value of the GlobalAdaptive option MaxErrorIncreases might lead to a convergent numerical integration. NIntegrate obtained 0.013006903336304906` and 0.0006852739534086272` for the integral and error estimates. >>

Out[131]= 0.0130069

所以这两种数值方法都没有给出-0.0160323。我理解为什么——第一种方法对分母造成的无穷大有问题，而第二种方法有效地删除了导致问题的积分部分。但我希望能够整合另一个积分（比x、y 和z 更难的积分），它无法通过分析简化。上面的积分为我提供了一种测试任何新方法的方法，因为我知道答案应该是什么。

Answer 1

除非我写下积分错误，否则应该可以：

n[x_] := 1/(1 + Exp[eps[x]])
eps[x_] := 2(1 - Cos[x])
.25/(2*Pi)^2*NIntegrate[
    Cos[(x + y)/
     2]^2 ((1 - n[y]) n[y] - (1 - n[x]) n[x])/(Cos[y] - Cos[x]),
    {x, -Pi, Pi},
    {y, -Pi, Pi},
    Exclusions -> {Cos[x] == Cos[y]}
  ]

给0.0130098。

好的，我想最快的解释方式是这样的：

从第一个 eq 的第一行到第二行，我刚刚将 y 重命名为 x（积分区域是对称的，所以没关系）。然后我为 I 添加了两个（等价的）表达式以获得 2I，我在数值上积分就是这样。关键是分子和分母在同一点消失，所以实际上不需要Exclusions 选项。请注意，在上述方法的草图中，为简洁起见（或由于懒惰，取决于观点），我删除了 1/(4*Pi^2)。