Change Uncommitted^Sapling to be a u-coordinate for which there is no point on the curve.

Signed-off-by: Daira Hopwood <daira@jacaranda.org>

protocol/protocol.tex

@@ -3487,7 +3487,7 @@ such that the following conditions hold:
 
 \subparagraph{Note commitment integrity} \label{saplingnotecommitmentintegrity}
 
-$\cmOld{} \neq \UncommittedSapling$, and $\pack(\cmOld{}) = \NoteCommitmentSapling(\nOld{})$.
+$\pack(\cmOld{}) = \NoteCommitmentSapling(\nOld{})$.
 
 \subparagraph{Merkle path validity} \label{saplingmerklepathvalidity}
 
@@ -3751,7 +3751,7 @@ Define:
   \item $\changed{\NoteAddressPreRandLength \typecolon \Nat := 252}$
   \item $\UncommittedSprout \typecolon \bitseq{\MerkleHashLengthSprout} := \zeros{\MerkleHashLengthSprout}$
 \sapling{
-  \item $\UncommittedSapling \typecolon \bitseq{\MerkleHashLengthSapling} := \ones{\MerkleHashLengthSapling}$
+  \item $\UncommittedSapling \typecolon \bitseq{\MerkleHashLengthSapling} := \ItoLEBSP{\MerkleHashLengthSapling}(1)$
 } %sapling
   \item $\MAXMONEY \typecolon \Nat := \changed{2.1 \smult 10^{15}}$ (\zatoshi)
   \item $\SlowStartInterval \typecolon \Nat := 20000$
@@ -4143,6 +4143,25 @@ zero, the proof can be adapted straightforwardly to show that $\PedersenHashToPo
 is collision-resistant under the same assumptions and security bounds.
 Because $\ItoLEBSP{\MerkleHashLengthSapling}$ and $\ExtractJ$ are injective,
 it follows that $\PedersenHash$ is equally collision-resistant.
+
+\vspace{2ex}
+\begin{theorem} \label{thmnohashtouncommittedsapling}
+$\UncommittedSapling = \ItoLEBSP{\MerkleHashLengthSapling}(1)$ is not in the range of $\PedersenHash$.
+\end{theorem}
+
+\begin{proof}
+By the definition of $\PedersenHash$, $\ItoLEBSP{\MerkleHashLengthSapling}(1)$ can be in the
+range of $\PedersenHash$ only if there exist $D \typecolon \byteseq{8}$ and $M \typecolon \bitseq{\PosInt}$
+such that $\ExtractJ(\PedersenHashToPoint(D, M)) = 1$.
+The latter can only be the affine-Edwards $u$-coordinate of a point in $\GroupJ$.
+We show that there are no points in $\GroupJ$ with affine-Edwards $u$-coordinate $1$.
+Suppose for a contradiction that $(u, \varv) \in \GroupJ$ for $u = 1$ and some
+$\varv \typecolon \GF{\ParamJ{r}}$. By writing the curve equation as
+$\varv^2 = (1 - \ParamJ{a} \smult u^2) / (1 - \ParamJ{d} \smult u^2)$, and noting that
+$1 - \ParamJ{d} \smult u^2 \neq 0$, we have $\varv^2 = (1 - \ParamJ{a}) / (1 - \ParamJ{d})$.
+The right-hand-side is a nonsquare in $\GF{\ParamJ{r}}$, so there are no solutions for $\varv$
+(contradiction).
+\end{proof}
 } %sapling
 
 
@@ -7271,6 +7290,7 @@ Daira Hopwood, Sean Bowe, and Jack Grigg.
           $\SpendingKey$ to ensure they are on the full range of $\GF{\ParamJ{r}}$.
     \item Change $\PRFnr{}$ to produce output computationally indistinguishable from uniform on
           $\GF{\ParamJ{r}}$.
+    \item Change $\UncommittedSapling$ to be a $u$-coordinate for which there is no point on the curve.
 }
 \end{itemize}