diff --git a/cosets/CMakeLists.txt b/cosets/CMakeLists.txt
index c8cf75b..61e4c6b 100644
--- a/cosets/CMakeLists.txt
+++ b/cosets/CMakeLists.txt
@@ -17,3 +17,6 @@ add_custom_command(
 )
 
 add_executable(coxeter src/tc.cpp)
+
+add_executable(mirror src/mirror.cpp)
+target_link_libraries(mirror PRIVATE glm)
diff --git a/cosets/src/mirror.cpp b/cosets/src/mirror.cpp
new file mode 100644
index 0000000..4fe887e
--- /dev/null
+++ b/cosets/src/mirror.cpp
@@ -0,0 +1,60 @@
+#include <cmath>
+#include <vector>
+#include <iostream>
+#include <glm/glm.hpp>
+#include <glm/gtx/string_cast.hpp>
+
+template<int N>
+using vecn = glm::vec<N, float, glm::defaultp>;
+
+float dot(int n, glm::vec4 a, glm::vec4 b) {
+   float sum = 0;
+   for (int i = 0; i < n; ++i) {
+      sum += a[i] * b[i];
+   }
+   return sum;
+}
+
+template<int N>
+std::vector<glm::vec4> mirror(const float (&arr)[N - 1][N - 1]) {
+   static_assert(1 <= N and N <= 4, "Vector size is unsupported");
+
+   std::vector<glm::vec4> mirrors{};
+   for (int p = 0; p < N; ++p) {
+      glm::vec4 vp{};
+      for (int m = 0; m < p; ++m) {
+         glm::vec4 vq = mirrors[m];
+         float a = cos(M_PI / arr[p-1][m]);
+         float b = dot(m, vp, vq);
+         float c = vq[m];
+         float d = a - b / c;
+         vp[m] = d;
+      }
+      vp[p] = std::sqrt(1 - glm::dot(vp, vp));
+
+      for (const auto &v : mirrors) {
+         if (glm::dot(vp, v) > 0) {
+            vp *= -1;
+            break;
+         }
+      }
+
+      mirrors[p] = vp;
+   }
+
+   return mirrors;
+}
+
+int main(int argc, char *argv[]) {
+   int planes = 3;
+   auto normals = mirror<3>((float[2][2]) {
+      {4},
+      {2, 3}
+   });
+
+   for (const auto &normal : normals) {
+      std::cout << glm::to_string(normal) << std::endl;
+   }
+
+   return 0;
+}
\ No newline at end of file