diff --git a/alternative_1_1.ipynb b/alternative_1_1.ipynb
new file mode 100644
index 0000000..ea88634
--- /dev/null
+++ b/alternative_1_1.ipynb
@@ -0,0 +1,50 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "\\begin{align}\n",
+    "\\frac{\\partial}{\\partial w_k} &= \\frac{\\partial}{\\partial w_k}\\left(\\sum_{k=1}^Kh_{c_i,k} w_k^T \\phi(x_i) - \\log \\left(\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}\\right)\\right)\\\\\n",
+    "\\text{the $k$ in $\\partial w_k$ is not the same as the $k$ in the sum $\\sum_{k=1}^K$. To avoid confusion we will rename the first $w_k$ to $w_l$:}\\\\\n",
+    "&= \\frac{\\partial}{\\partial w_l}\\left(\\sum_{k=1}^Kh_{c_i,k} w_k^T \\phi(x_i) - \\log \\left(\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}\\right)\\right)\\\\\n",
+    "\\text{because } h_{c_i,k}=\\begin{cases}0 & \\forall k\\neq l \\\\ 1 & \\text{for }k=l\\end{cases} \\text{ the sum $\\sum_{k=1}^K$ collapses:}\\\\\n",
+    "&= \\frac{\\partial}{\\partial w_l}\\left(w_l^T \\phi(x_i)\\right) - \\frac{\\partial}{\\partial w_l}\\left( \\log \\left(\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}\\right)\\right)\\\\\n",
+    "\\text{from $\\frac{\\partial x^T a}{\\partial x}=a^T$ and $\\frac{\\partial \\log(f(x))}{\\partial x}=\\frac{f'(x)}{f(x)}$ follows:}\\\\\n",
+    "&= \\phi(x_i)^T - \\frac{\\frac{\\partial}{\\partial w_l}\\left(\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}\\right)}{\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}}\\\\\n",
+    "\\frac{\\partial}{\\partial w_l} e^{w_j^T\\phi(x_i)} = \\begin{cases}0\\cdot e^{w_j^T\\phi(x_i)}=0 & \\forall w_j\\neq w_l\\\\ \\phi(x_i)^T e^{w_l^T\\phi(x_i)} &\\text{for } w_j=w_l\\end{cases}\\text{ leads to:}\\\\\n",
+    "&= \\phi(x_i)^T - \\frac{ \\phi(x_i)^T e^{w_l^T \\phi(x_i)}}{\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}}\\\\\n",
+    "&=  \\phi(x_i)^T \\left(1 - \\frac{e^{w_l^T \\phi(x_i)}}{\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}}\\right)\\\\\n",
+    "\\end{align}\n",
+    "\n",
+    "If we do the same thing for every $w_l$ we get\n",
+    "\n",
+    "$$\n",
+    "\\dfrac{\\partial \\mathcal L_{\\mathrm{cat-NLL}}}{\\partial \\boldsymbol{w}}\n",
+    "= \\begin{pmatrix}\n",
+    "\\phi(x_i)^T \\left(1 - \\frac{e^{w_1^T \\phi(x_i)}}{\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}}\\right)\\\\\n",
+    "\\vdots \\\\\n",
+    "\\phi(x_i)^T \\left(1 - \\frac{e^{w_K^T \\phi(x_i)}}{\\sum_{j=1}^K e^{w_j^T \\phi(x_i)}}\\right)\\\\\n",
+    "\\end{pmatrix}\n",
+    "$$"
+   ]
+  }
+ ],
+ "metadata": {
+  "interpreter": {
+   "hash": "a170f962d07ed8515a1e0cfe90051c8c49c07a10f515ca6de22ae850428164e9"
+  },
+  "kernelspec": {
+   "display_name": "Python 3.8.5 64-bit ('base': conda)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python",
+   "version": "3.8.5"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}