Merge #27: README: removed duplicated "</sub>}"

dbfffcf fix: removed duplicated "</sub>}" (Marco D'Agostini)

Pull request description:

  In line 34 a duplicate `}` was being rendered

ACKs for top commit:
  sipa:
    ACK dbfffcf

Tree-SHA512: 38d1c8ae6b7d6ce09476329eafff9ff48827b9905f0c0400ee380db25b591b97462bcfbe484515523c1d51b69d9d3a128b3fcbfcc20e914b4bfcbf82d41e1098

Author: sipa

doc/math.md

@@ -31,7 +31,7 @@ Because in our field addition corresponds to XOR of integers, it holds for every
 desire from a sketch, namely that an efficient operation exists to
 combine two sketches such that the result is a sketch of the symmetric
 difference of the sets. It holds that
-*S({m<sub>1</sub>,m<sub>2</sub>}) + S({m<sub>2</sub>,m<sub>3</sub>}) = S(m<sub>1</sub>) + (S(m<sub>2</sub>) + S(m<sub>2</sub>)) + S(m<sub>3</sub>) = S(m<sub>1</sub>) + S(m<sub>3</sub>) = S({m<sub>1</sub>,m<sub>3</sub>}</sub>})*. The question is whether we can also efficiently recover the elements from their power series' coefficients.
+*S({m<sub>1</sub>,m<sub>2</sub>}) + S({m<sub>2</sub>,m<sub>3</sub>}) = S(m<sub>1</sub>) + (S(m<sub>2</sub>) + S(m<sub>2</sub>)) + S(m<sub>3</sub>) = S(m<sub>1</sub>) + S(m<sub>3</sub>) = S({m<sub>1</sub>,m<sub>3</sub>})*. The question is whether we can also efficiently recover the elements from their power series' coefficients.
 
 **An infinity of coefficients is hard**
 
@@ -114,4 +114,4 @@ Say that we're trying to find the inverses of the roots of *L = 1 + l<sub>1</sub
 In other words, we can find the inverses of the roots of *L* by instead factoring the polynomial with the coefficients of *L* in reverse order.
 
 * <a name="myfootnote1">[1]</a> For those familiar with coding theory: PinSketch communicates a set difference by encoding the set members as errors in a binary [BCH](https://en.wikipedia.org/wiki/BCH_code) codeword 2<sup>bits</sup> in size and sends the syndromes.
-  The linearity of the syndromes provides all the properties needed for a sketch. Sketch decoding is simply finding the error locations. Decode is much faster than an ordinary BCH decoder for such a large codeword because the need to take a discrete log is avoided by storing the set in the roots directly instead of in an exponent (logically permuting the bits of the codeword).
+  The linearity of the syndromes provides all the properties needed for a sketch. Sketch decoding is simply finding the error locations. Decode is much faster than an ordinary BCH decoder for such a large codeword because the need to take a discrete log is avoided by storing the set in the roots directly instead of in an exponent (logically permuting the bits of the codeword).