diff --git a/README.md b/README.md
index 10f0fd8a..79653fbc 100644
--- a/README.md
+++ b/README.md
@@ -70,11 +70,13 @@ Flux.@epochs 50 Flux.train!(loss, params(model), data, opt)
 
 PDE training examples are provided in `example` folder.
 
-### One-dimensional Fourier neural operator
+### One-dimensional Fourier Neural Operator
 
 [Burgers' equation](example/Burgers)
 
-### Two-dimensional with time Navier-Stokes equation
+### Two-dimensional Fourier Neural Operator
+
+[Double Pendulum](example/DoublePendulum)
 
 ### Markov Neural Operator
 
diff --git a/example/DoublePendulum/README.md b/example/DoublePendulum/README.md
new file mode 100644
index 00000000..f7019f61
--- /dev/null
+++ b/example/DoublePendulum/README.md
@@ -0,0 +1,20 @@
+# Double pendulum
+
+A [double pendulum](https://www.wikiwand.com/en/Double_pendulum) is a pendulum with another pendulum attached to its end.
+In is example, instead of learning from a well-described equation of motion,
+we train the model with the famous dataset provided by IBM.
+The equation of motion to the real experiments of double pendulum is learned by a two-dimensional Fourier neural operator.
+It learns to inference the next 30 steps with the given first 30 steps.
+The result of this example can be found [here](https://foldfelis.github.io/NeuralOperators.jl/dev/assets/notebook/double_pendulum.jl.html).
+
+![](gallery/result.gif)
+
+By inference the result recurrently, we can generate up to 150 steps with the first 30 initial steps.
+
+Change directory to `example/DoublePendulum` and use following commend to train model:
+
+```julia
+$ julia --proj
+
+julia> using DoublePendulum; DoublePendulum.train()
+```