More than 10 000 pages from illuminated manuscripts and books will be
scanned into a database. The images will then be processed to remove stains,
magnify details, increase contrast and restore faded colours. The library
is also computerising its two million index cards, and will link them to
the image database. The Pontifical Catholic University in Rio de Janiero
will keep duplicate databases to serve computer networks in North and South
America. Even when compressed, the data will occupy about 50 gigabytes.

Among the library’s 500-year-old collection are 8000 books published
in the first 50 years of the printing press and the four oldest manuscripts
of Virgil’s poems. Because the library has limited space and staff, only
2000 people are allowed to use it each year. ‘This will mean people can
study the works from their own university,’ says an IBM spokesman.

A 200-year-old statistical technique could take much of the guesswork out of piecing together archaeological clues to date ancient finds. At one early Bronze Age site in Austria, a theorem formulated in the 18th century has produced a dramatic reduction in the uncertainty involved in dating copper workings.

Until now archaeologists have had no formal way of combining chronological information obtained using different dating techniques. Stratigraphy – putting finds in chronological order by examining successive layers of deposits on the site – and radiocarbon dating can give two independent dates for an object found in a dig. But the two techniques cannot simply be combined to arrive at a single, more accurate dating.

Now Caitlin Buck and her colleagues in the statistics group at the University of Nottingham have found a way to use the theorem published in 1763 by the English clergyman Thomas Bayes to combine the data in a rigorous mathematical way. The basis of the theorem is that if the outcome of one event is known, this affects the probability of another event occurring. For example, there might be a one in five chance of frost on March nights, and a one in fifty chance of snow. But if there was frost on the previous night, the chance of having snow is modified.

Buck’s work, reported at a recent meeting entitled ‘Computer Applications and Quantitative Methods’ in Glasgow, has enabled archaeologists to pinpoint the date of copper workings at St Veit-Klinglberg, an early Bronze Age settlement 60 kilometres south of Salzburg in Austria. The site was excavated in the late 1980s by a team of archaeologists led by Stephen Shennan of the University of Southampton. They found many pieces of copper, and pottery made from clay mixed with slag. First, they tried to date their finds by stratigraphy, but the hillside on which St Veit perches had been badly eroded, isolating parts of the site and making comparisons between them impossible. They could put in sequence only four groups of samples, ranging in size from two to six items.

Next the archaeologists had 15 samples radiocarbon dated. ‘This is fairly imprecise – you often get results (accurate to) plus or minus 100 years,’ says Buck, who is now at Loughborough University of Technology. The radiocarbon dates for the samples all fell somewhere between about 3400 and 3900 years ago.

Until about 10 years ago, it was not practical to use Bayes’ theorem on this type of archaeological evidence because of the difficulty of representing it in a form suitable for statistical analysis. The data were too complicated to allow the necessary integration calculations to be carried out. But in the past decade computer-based simulation techniques called Markov chain Monte Carlo methods have made it possible to solve the integrations indirectly.

When Buck and her colleagues combined the radiocarbon dates and stratigraphic sequences from St Veit using Bayes’ theorem, the likely dates for the deposition of individual objects narrowed from a range of around 400 years to 200 years. Five more artefacts could also be slotted into order. The researchers concluded that smelted copper had been stored at the site around 3500 years ago, although they found no indication that smelting took place there.

The research will appear in Germania and in Archaeometry later this year.

The idea of such a state of matter was thought up 30 years ago by Joaquin
Luttinger of Columbia University, New York. He regarded it merely as a mathematical
tool to solve a physics problem: ‘I never in my wildest dreams expected
it to be found experimentally in any real system,’ says Luttinger, who is
now retired.

Last year, Charles Kane at the University of Pennsylvania in Philadelphia
and Matthew Fisher of the University of California in Santa Barbara suggested
an experiment to verify that a Luttinger liquid could really exist. The
experiment has now been performed and Kane says he is ’70 to 80 per cent’
certain that it confirms the state’s existence.

At normal temperatures, electrons in a conducting material are in constant
frenzied motion, and the movement of any one electron does not directly
affect the others. Luttinger suggested that if electrons were confined to
a very thin, perfectly clean wire, they might behave in an unusual way.
In particular, if the wire were cooled to very close to absolute zero, -273
°C, electrons would pack together so closely that if one electron moved
it would reduce the density of electrons locally. To restore the density,
the other electrons would rearrange themselves and the net result would
be a Luttinger liquid in which the electrons would appear to be connected.

Because electrons in a Luttinger liquid can only move in unison, it
requires a lot of energy to make them move at all. Close to absolute zero,
this energy is not available, so the electrons are prevented from flowing
through the wire. As a result, the wire’s resistance should rise towards
infinity. In a normal wire, the electrical resistance never becomes infinite
because a few electrons can always flow through.

But setting up the conditions to create a Luttinger liquid is difficult.
The electrons must be confined to a thin wire. This allows the electron
‘train’ to pass through lengthways rather than sideways, making the coupling
between electrons more obvious. However, a single atom of an impurity in
such a very thin wire can prevent it conducting properly. ‘Impurities, which
are very difficult to get rid of, tend to destroy the Luttinger liquid state,’
says Kane.

But four years ago, Xiao-Gang Wen of the Massachusetts Institute of
Technology realised that if a magnetic field was applied across the material,
the disruptive effect of impurities could be eliminated.

Richard Webb, Frank Milliken and Corwin Umbach of IBM’s Yorktown Heights
research laboratory in New York state set out to test Wen’s theory. They
made a device from two semiconductors – gallium arsenide and gallium aluminium
arsenide. They then measured its behaviour as they varied the voltage and
the temperature down to 38 millikelvin. ‘We obtained data consistent with
a Luttinger liquid,’ says Webb, now at the University of Maryland at College
Park.

Before the experiments began, the devices had to be refrigerated for
three weeks in order to stabilise impurities. Also, they could be tested
reliably only for two to three weeks before the impurities began to migrate
through them again.

Webb says he is ‘really excited’ about the results. ‘This is an example
of a situation where the standard picture of electrons is no longer valid,’
he says. He believes their collective behaviour could give clues to the
ways more complicated groups of electrons behave, perhaps even shedding
light on high-temperature superconductivity.

The Ancient Monuments Laboratory at English Heritage is setting up a
database of geophysical surveys, revealing buried archaeological features,
which should be ready in about a year’s time. It will later be available
through the Janet computer network, which is connected to the Internet.

Geophysical surveys use magnetic and electrical measurements made on
the surface or from the air to detect hidden features. They are widely used
by archaeologists to help plan excavations, and by developers, who often
include them in planning applications to show that they have considered
the environmental and archaeological impact of their proposals.

There is, however, no library of results or even a listing of areas
that have already been surveyed, and surveyors can waste time and money
hunting for results or repeating the measurements.

The laboratory began setting up a pilot database this week to hold the
results of its own surveys, including the name of the site, its location,
whether it is a listed ancient monument, who did the survey, data and plots
of the results, and whether there is more information available in a conventional
paper record. Each year between 35 and 40 surveys are carried out, though
they can cover varying areas, depending on the density of archaelogical
data at a location.

Access to the database will be free. The amount of data, however, should
be comparatively small; the laboratory only adds about 500 pages of text
– roughly equivalent to 3 megabytes of data – each year.

Andrew David, head of the laboratory’s archaeometry branch, in London,
and scientific officer Neil Linford hope that private surveyors and other
public bodies will deposit their data on the database too. However, once
the database is ready to take information from them, the laboratory will
have to address the question of who owns the copyright.

Organisations such as the Ordnance Survey, which supply data to researchers
and industry, are under pressure to recoup their costs by selling data.
But Linford says: ‘The cost of putting together the database and using
it will be offset by an increase in efficiency. If a contractor is bidding
for a tender to make a new road, and we can supply the information more
cheaply (than before), the road may become cheaper overall.’

Assembling surveys into a form suitable for storing in the database
is difficult because there are few standardised ways of recording them.
‘You could store the site name, such as Smith’s Farm, but that might not
be the name of the archaeological feature. And you could use the Ordnance
Survey grid reference, but different surveys cover differently shaped parcels
of land, so how do you give a central location?’

Eventually, English Heritage would like to link the geophysical database
with the National Archaeological Record, a much larger archaeological database
which is also being set up by the Royal Commission on the Historical Monuments
of England, and will contain more details about sites, such as excavations
done, and records and dating of any finds.

Each of the islands generates its own electricity and between them they
have 28 generators running on gas, coal, diesel or oil. Predicting the demand
for power is not difficult: it involves straightforward calculations that
combine the patterns of previous years, the weather, and the television
schedules, for example. Most years, the peak demand occurs on August evenings
after 9 pm, when the clubs are busy.

The hard part is working out the most economic way to run the generators
to meet the predicted demand. The calculation is of a family known as ‘NP-hard’
problems: there is an answer, but it cannot be found by polynomial equations.

The difficulty arises because generators have to be on or off. They
obviously cannot be ‘half on’, even if their output can vary between a minimum
and maximum once they are on. But working out how many of each type should
be on in order to meet the demand as cheaply as possible cannot be done
directly, because there is no way of solving algebraic problems that restricts
the answers to whole numbers. In addition, the cost of starting up a generator
and the time it takes to do so can be substantial – up to 1.5 million pesetas
( £7500) and 12 hours, though 7000 pesetas and 30 minutes are more
typical.

Now GESA, the Spanish electricity company, has linked up with three
teams of scheduling experts across Europe to speed up an existing approach
to the problem, called branch and bound. A program for an ordinary single-processor
computer, which is currently used by British Steel to help schedule manufacturing
processes, is being put on a network of up to a thousand PowerPC processing
chips by Dash Associates of Blisworth, Northamptonshire.

By using a number of processors to consider several options simultaneously,
the solution is reached that much more quickly. Once an option has been
eliminated, the processors can concentrate on the others, speeding up the
process even more. ‘A 256-processor machine could increase the speed a hundred
times,’ predicts Bob Daniel, a director of Dash. A central processor divides
the work and gathers the answers.

The gain in speed is important, because it will let GESA reduce the
size of the ‘buckets’ of time for which it does its calculations to a few
minutes, rather than the one-hour buckets it uses now.

With the branch and bound method, the optimum solution is calculated
taking into account the start-up times and costs, running costs and maintenance
schedules of all the generators available. What happened in the previous
bucket must also be fed into the calculation – if a generator is already
on, for example, its start-up time is zero. For a given power requirement,
the program might say that 2.3 of the available diesel generators and 3.9
coal generators need to be on in a particular bucket.

But noninteger solutions are not allowed, and rounding to the nearest
whole numbers rarely finds the cheapest solution. Instead the best solution
must be two or fewer, or three or more, diesels; and three or fewer, or
four or more, coal. This imposes new bounds on the possible solutions to
the problem. The program branches into four and looks for the best solution
in these newly defined possible regions. The ‘branch and bound’ process
continues until the solution is sufficiently close to an integer. GESA aims
for 0.1 per cent tolerances.

Running the program on many processors means GESA can cut the size of
buckets and respond more closely to short fluctuations in demand. But the
number of calculations increases exponentially with the number of buckets.
With 30 generators switchable once an hour for 24 hours, there are 30 times
24, or 720 possible combinations of generators per day. Each generator
can be on or off, giving 2720 possible solutions to assess. With 10-minute
buckets there are 24320 solutions.

The SERC, which was replaced by two new research councils on 1 April,
began setting up IRCs a decade ago to focus expertise onto specific areas
of research. Some foster strategic research in commercially important fields
such as semiconductors. Others concentrate on areas such as biochemical
engineering, which straddle conventional disciplines.

The IRCs were supposed to become national centres of excellence, but
the SERC report says they have not done so. The report recommends that they
should be set targets for improving the quality of training and strengthening
their relationships with industry.

The report also questions whether IRCs give value for money. ‘The original
idea was that they should attract industrial funding,’ says a spokesman
for the SERC. But in practice the industrial input ‘has been a little disappointing’.
Last month, the SERC penalised two IRCs for being too isolated. The Optoelectronics
Research Centre (ORC) at the University of Southampton, and the IRC in semiconductors
at Imperial College, London have had about 10 per cent of their SERC funding
held back until they start new collaborative ventures.

The semiconductor centre stands to lose £200 000, but co-director
Gareth Parry is philosophical about the SERC’s action. ‘This is a reasonably
sensible way of proceeding,’ he says. ‘People don’t hand over that amount
of money now without strong ideas on how it’s used.’ Parry expects to win
the money back. He says there will be no redundancies.

But at the ORC, which stands to lose £140 000, the director Alec
Gambling is far from happy. ‘I don’t think there is any justification for
this. We can’t plan and do anything adventurous.’ Gambling is concerned
that staff will leave because of the insecurity. He sees no easy way to
find new joint ventures, especially in view of the time it will take to
co-write proposals for funding. ‘I don’t see how we’ll get the money,’
he says.

The cross carries five inscriptions. Three are in Latin and commemorate
Abbess Aethelburg, probably of the local monastery. Until now, the other
two, written in strange alphabets, have been incomprehensible. One is made
up of 27 letters and looks superficially like ogham, a script developed
in Ireland in the 4th century. The fifth has 15 Anglo-Saxon runes, 35 tree
runes, which look like pine trees, and three Latin letters.

The 27 ogham-like letters are made up of six characters, y , – , ,
/, ( and ), repeated up to five times to make a letter, giving an alphabet
of 30 letters. Richard Sermon, deputy director of the Scottish Urban Archaeological
Trust in Perth, who has spent a year trying to unravel the inscriptions,
believed these letters could be variants on the ogham or Anglo-Saxon runic
alphabets. At a conference in Glasgow he described how he had written a
computer program to substitute these known alphabets for the letters in
the inscription in all 1440 possible combinations.

‘Most of the 720 readings based on the runic alphabet contained completely
unintelligible strings of consonants,’ he says. ‘But many of the readings
based on the ogham alphabet contained good syllables.’ The best version
he found reads: ‘erosg / rhgeguso / crgengphuir / uitengoiz’. In Old Irish
it would read ‘Cross cu / Rig Isu / carric an foir / uait Oengus’, or in
English: ‘Cross to / King Jesus / rock of help / from Angus’.

Sermon says the inscription was probably devised by someone with Irish
connections, although if they had been familiar with standard ogham they
would probably have used that. ‘This would also account for the Irish personal
name at the end of the inscription,’ he says.

The Anglo-Saxon runes in the other inscription seem to be an anagram,
which reads ‘Oedilburg gnoew me’ or ‘Aethelburg knew me’.

But the tree runes have left Sermon stumped, even with the help of the
computer. There are up to four ‘branches’ on the left and eight on the right
of the ‘trunk’, giving an alphabet of 32 letters of which 14 appear on the
cross. Sermon tried translating them into the 33 Anglo-Saxon runes, and
wrote another program to generate all 24 possible versions.

‘None appears to form any intelligible pattern,’ he says. Some Norse
inscriptions are written right to left, so he tried reversing the runes,
again without success. ‘It would seem that the tree runes are now too fragmentary
to be fully understood.’