diff --git a/README.md b/README.md
index f3709ff..b9952d3 100644
--- a/README.md
+++ b/README.md
@@ -29,5 +29,11 @@ normalization for depth and rgb
    - not work ,so change to iter small_gicp kdtree
 
 4. [x]use kornia build gicp
-5. [ ]GICPJacobianApprox building
-4. change R as four angles 
+5. [x]GICPJacobianApprox building
+
+1. [ ] select optimizer
+   - [ ]LGBS learn slow ?
+
+2. [x]change R as four angles
+   
+   
diff --git a/src/gicp/model.py b/src/gicp/model.py
index 2387276..f825212 100644
--- a/src/gicp/model.py
+++ b/src/gicp/model.py
@@ -195,8 +195,8 @@ def forward(self) -> torch.Tensor:
 
 def training(tar_pcd, src_pcd, num_epochs=1000):
     model = GICPModel(tar_pcd, src_pcd)
-    optimizer = optim.LBFGS(model.parameters(), lr=1e-3, max_iter=100, history_size=50)
-    scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode="min")
+    optimizer = optim.LBFGS(model.parameters(), lr=1e-3, max_iter=20, history_size=200)
+    # scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode="min")
 
     def closure():
         optimizer.zero_grad()
@@ -213,7 +213,7 @@ def closure():
         start = default_timer()
         optimizer.step(closure)  # 注意：LBFGS 需要一个闭包来重新计算模型
         loss = closure()  # 重新计算损失
-        scheduler.step(loss)
+        # scheduler.step(loss)
         end = default_timer() - start
 
         print(