diff --git a/docs/examples/example3_adversarial_mnist.py b/docs/examples/example3_adversarial_mnist.py
index e0370c70..f1d5a651 100644
--- a/docs/examples/example3_adversarial_mnist.py
+++ b/docs/examples/example3_adversarial_mnist.py
@@ -30,11 +30,11 @@
 we can show that the difference is never positive, no such example
 exists.
 
-Here, we use the :math:`l1-`\ norm :math:`|| x - \\bar x||_1` to define
+Here, we use the :math:`l_1-` norm :math:`|| x - \bar x||_1` to define
 the neighborhood with its size defined by a fixed parameter
 :math:`\delta`:
 
-.. math::  || x - \\bar x ||_1 \le \delta.
+.. math::  || x - \bar x ||_1 \le \delta.
 
 Denoting by :math:`g` the prediction function of the neural network, the
 full optimization model reads: