diff --git a/ipa-core/src/protocol/dp/README.md b/ipa-core/src/protocol/dp/README.md
index d34396bf5..c5d0cb81a 100644
--- a/ipa-core/src/protocol/dp/README.md
+++ b/ipa-core/src/protocol/dp/README.md
@@ -8,10 +8,9 @@ introduces Binomials for DP.
 considers their use for d-dimension queries (such as we will need for WALR).
 
 
-To achieve a desired $(\varepsilon, \delta)$-DP guarantee, we generate $num_bernoulli$ secret shared samples of a 
+To achieve a desired $(\varepsilon, \delta)$-DP guarantee, we generate $num\\_bernoulli$ secret shared samples of a 
 Bernoulli having probability $0.5$ using PRSS.  Next we aggregate them to get a Binomial sample. The result of the 2018 
-paper above is that for small epsilon (TODO, how small required?), we require the following number of samples
-$$ num_bernoulli \geq 8 \log(2/\delta) /\varepsilon^2$$
+paper above is that for small epsilon.
 
 This [spreadsheet](https://docs.google.com/spreadsheets/d/1sMgqkMw3-yNBp6f8ctyv4Hdfx9Ei7muj0ZhP9i1DHrw/edit#gid=0) 
 looks at the calculation for different parameter choices and confirms that this approach does lead to a better final 
@@ -22,4 +21,4 @@ Gaussians but the 2018 paper's analysis shows this).
 
 ## Binomial Noise for d-dimensional queries
 For WALR we will be adding noise to a d-dimensional query, to add binomial noise we have to look 
-simulatenously at the sensitivity under three different norms, $\ell_1, \ell_2, \ell_\infty$.  TODO. 
\ No newline at end of file
+simulatenously at the sensitivity under three different norms, $\ell_1, \ell_2, \ell_\infty$.  TODO.