Message Boards Message Boards

The Continued Logarithm of ?

Posted 6 years ago

Open in Cloud | Download to Desktop via Attachments Below


Briefly, a continued logarithm is an arbitrarily long bit string approximating a real number arbitrarily well, and supports arbitrarily precise bit-at-a-time algorithms for rational functions of these numbers. (See http://www.tweedledum.com/rwg/cfup.htm , p 47+)

A six-bits-per-term $\pi$ series:

Product[MatrixForm@{{512 k^3, 0}, {32 k^3 (-37 + 42 k), (-1 + 2 k)^3}}, {k, ?}] == 
 MatrixForm[{{oo ?, 0}, {oo, "?"}}]

enter image description here

Where ? means matrix product, not Mathematica product, and oo is some quantity that blows up with the number of product terms, the same way you compute continued fractions with 2*2 matrices. An incorrect form of this series is derived in https://dspace.mit.edu/handle/1721.1/6088 . We initialize the work matrix m to the first term of the matrix product:

MatrixForm[ m = {{512 k^3, 0}, {32 k^3 (-37 + 42 k), (-1 + 2 k)^3}} /. k -> 1]

enter image description here

This represents the function

Divide @@ (m.{t, 1})

enter image description here

where t is the tail of the series, starting with k=2 rather than 1. It should be easy to show that, in general,

3/8/k < t < 3/8/(k + 1)

enter image description here

giving bounds

m /. {{t -> 3/8/1.}, {t -> 3/8/2.}}

{3.14754, 3.09677}

Since these both exceed 2, we commence output with

Style[cl@? = {1}, 22]

enter image description here

and left multiply m by the divide-by-2 matrix:

MatrixForm[m = {{1, 0}, {0, 2}}.m]

enter image description here

It costs almost nothing to remove the common power of 2:

MatrixForm[m = m/2]

enter image description here

representing the function

Divide @@ (m.{t, 1})

enter image description here

We still are on input term k=1, and can still use

% /. {{t -> 3/8/1.}, {t -> 3/8/2.}}

{1.57377, 1.54839}

which is smack between 1 and 2, which we celebrate with

Style[AppendTo[cl@?, 0], 22]

enter image description here

and left multiply m by the subtract-1-and-reciprocate matrix:

(m = {{0, 1}, {1, -1}}.m) // MatrixForm

enter image description here

representing

Divide @@ (m.{t, 1})

enter image description here

Again using

% /. {{t -> 3/8/1.}, {t -> 3/8/2.}}

{1.74286, 1.82353}

which dictates another

Style[AppendTo[cl@?, 0], 22]

enter image description here

and

(m = {{0, 1}, {1, -1}}.m) // MatrixForm

enter image description here

representing

Divide @@ (m.{t, 1})

enter image description here

% /. {{t -> 3/8/1.}, {t -> 3/8/2.}}

{1.34615, 1.21429}

Style[AppendTo[cl@?, 0], 22]

enter image description here

(m = {{0, 1}, {1, -1}}.m) // MatrixForm

enter image description here

Divide @@ (m.{t, 1})

enter image description here

% /. {{t -> 3/8/1.}, {t -> 3/8/2.}}

{2.88889, 4.66667}

This unambiguously exceeds 2, so

Style[AppendTo[cl@?, 1], 22]

enter image description here

MatrixForm[m = {{1, 0}, {0, 2}}.m]

enter image description here

Divide @@ (m.{t, 1})

enter image description here

% /. {{t -> 3/8/1.}, {t -> 3/8/2.}}

{1.44444, 2.33333}

Our interval of uncertainty contains 2! At last we gobble (right multiply) the k=2 term.

MatrixForm[ m = m.{{512 k^3, 0}, {32 k^3 (-37 + 42 k), (-1 + 2 k)^3}} /. k -> 2]

enter image description here

Divide @@ (m.{t, 1})

enter image description here

(Remembering to bump k !)

% /. {{t -> 3/8/2.}, {t -> 3/8/3.}}

{1.51515, 1.51938}

(Giving six more bits of precision!)

Style[AppendTo[cl@?, 0], 22]

enter image description here

(m = {{0, 1}, {1, -1}}.m) // MatrixForm

enter image description here

Divide @@ (m.{t, 1})

enter image description here

% /. {{t -> 3/8/2.}, {t -> 3/8/3.}}

{1.9412, 1.92538}

Style[AppendTo[cl@?, 0], 22]

enter image description here

(m = {{0, 1}, {1, -1}}.m) // MatrixForm

enter image description here

Divide @@ (m.{t, 1})

enter image description here

% /. {{t -> 3/8/2.}, {t -> 3/8/3.}}

{1.06248, 1.08064}

Style[AppendTo[cl@?, 0], 22]

enter image description here

(m = {{0, 1}, {1, -1}}.m) // MatrixForm

enter image description here

Divide @@ (m.{t, 1})

enter image description here

% /. {{t -> 3/8/2.}, {t -> 3/8/3.}}

{16.0056, 12.4004}

Whoa, a burst of three 1s!

Style[AppendTo[cl@?, 1]; AppendTo[cl@?, 1]; AppendTo[cl@?, 1], 22]

enter image description here

MatrixForm[m = {{1, 0}, {0, 8}}.m]

enter image description here

MatrixForm[m = m/2]

enter image description here

(Since we knew it was three ones and not four, we could have output 1,1,1,0 and left multiplied m by

MatrixForm[{{0, 1}, {1, -1}}.{{1, 0}, {0, 8}}]

enter image description here

This process may be continued indefinitely, or until the integers in m overflow.

Attachments:
POSTED BY: Bill Gosper
2 Replies

Indeed, very Interesting! @J. M., this is a neat way to define @Bill's products. I typeset them via a matrix here for clarity:

kProduct[k_] := 
Divide@@Array[{{512 #^3,0},{32 #^3(42 # -37),(2 # -1)^3}} &,k,1,Dot][[All, 1]]

enter image description here

We can see k-th difference is log-linear in k:

ListLogPlot[Table[kProduct[k]-Pi,{k,10}],
    PlotTheme->"Detailed",PlotRange->All,FrameLabel->{k,Log[Pi-kProduct]}]

enter image description here

POSTED BY: Vitaliy Kaurov
Posted 6 years ago

These matrix products are terribly interesting. The tenth partial product already gives an approximation of ? that is accurate to machine precision:

Divide @@ Array[{{512 #^3, 0},{32 #^3(42 # - 37),(2 # - 1)^3}} &,10,1,Dot][[All, 1]]
   81129638414606681695789005144064/25824365969885544300882143774845

N[%, 20]
   3.1415926535897932397

% - ?
   1.3*10^-18
POSTED BY: J. M.
Reply to this discussion
Community posts can be styled and formatted using the Markdown syntax.
Reply Preview
Attachments
Remove
or Discard

Group Abstract Group Abstract