Re: More on identifiers

#1

"David BL" <davidbl (AT) iinet (DOT) net.au> wrote

Quote:

Informally I think of abstract identifiers as "internal glue" within a
relational database. A bit more formally, they are characterised as
identifiers that could be mapped bijectively to different values
throughout the database without changing the recorded information.
One would therefore hope that they aren't visible to end users.

I have wondered for some time whether abstract identifiers are only
needed within the confines of a flat relational model. The following
hypothetical example is meant to cast some light on this question.
I'm hoping you'll see the underlying matters of principle and see how
it raises some interesting questions and ideas:

Let a cardboard box of items land on your desk and your task is to
record information about the items in a database. In theory each item
is uniquely identified by its (x,y,z) position at any given epoch.
However, the idea is that the items are mixed up in the box and their
positions are irrelevant to the information that needs to be recorded.

The items are not labelled. The idea is to uniquely identify them
(only) by their observable properties. This is indeed assumed to be
an important integrity constraint to be enforced by the DBMS. Note as
well that it would be upsetting (and potentially very costly or
impractical) if the database system forces items to be labelled when
there shouldn't be a need to.

Consider two electrons. They both have the same mass, and they have the
same charge. They might have opposite spins. But the minute we add a third
electron, the spin of two of them is going to be identical. It seems that,
on the surface at least, electrons do not have enough properties to carry
identity. As you descend into lower level particles like quarks, things get
even more this way. Particles seem more and more interchangeable.
Subatomic particles might not have much to do with your objects in a box,
but it seems to me that any theory of reality and identity that falls apart
at the subatomic level should at least take that into account.

#2

Walter Mitty wrote:

Quote:

Consider two electrons. They both have the same mass, and they have the
same charge. They might have opposite spins. But the minute we add a third
electron, the spin of two of them is going to be identical. It seems that,
on the surface at least, electrons do not have enough properties to carry
identity. As you descend into lower level particles like quarks, things get
even more this way. Particles seem more and more interchangeable.
Subatomic particles might not have much to do with your objects in a box,
but it seems to me that any theory of reality and identity that falls apart
at the subatomic level should at least take that into account.

The way out of this supposed problem is always the same. Forget about
modeling the real world. All you need to model--all you should EXPECT
to model--is the testimony users will give about the real world
(*including* their lies and misapprehensions).

You do not have to design a database that does magic. If the users
can't formulate sensible assertions to put in the database that is their
problem. If they've found a way to make sensible assertions, do what
they do.

This is not a database problem.

--
Roy

#3

robur.6 (AT) gmail (DOT) com wrote:

Quote:

On Jun 5, 1:25 pm, "Walter Mitty" <wami... (AT) verizon (DOT) net> wrote:

Consider two electrons. They both have the same mass, and they have the
same charge. They might have opposite spins. But the minute we add a third
electron, the spin of two of them is going to be identical. It seems that,
on the surface at least, electrons do not have enough properties to carry
identity.

If objects are really not distinguishable then why someone would want
to artificially identify them? Wouldn¢t be better to record just their
quantity?

Amen to that, brother.

--
Roy

#4

<robur.6 (AT) gmail (DOT) com> wrote

On Jun 5, 1:25 pm, "Walter Mitty" <wami... (AT) verizon (DOT) net> wrote:

Quote:

Consider two electrons. They both have the same mass, and they have the
same charge. They might have opposite spins. But the minute we add a
third
electron, the spin of two of them is going to be identical. It seems
that,
on the surface at least, electrons do not have enough properties to carry
identity.

If objects are really not distinguishable then why someone would want
to artificially identify them? Wouldn’t be better to record just their
quantity?

Good point. So there are some case where identity is not important.

(Incidentally, the other response where the comment was made that two
electrons must occupy different states leads to a slippery slope. Using
state to identify implies that a change of state is in effect a change of
identity. This is related to Brian's claim that all attributes of an entity
are mutable. (Am I misinterpreting your claim, Brian?)

A case where identity seems unimportant to the requirements is products
bearing UPC codes on supermarket shelves. The UPC identifies the product as
a member of a class of interchangeable things on the shelf. It doesn't
matter which one you take to the chckout counter, they are all going to
ring up the same, for both pricing purposes and inventory control purposes.
There might be a unique identifier somewhere on the product, but it's not
obvious.

Supermarket sales slips are a notorious example of a situation where what is
presented on the slip is not a set of things bought, but a list of things
bought. In a list, each item can be identified by its position. Thus two
items with the same content but different position will not be treated as
"the same item included twice." Confusion between sets, lists, and bags is
responsible for a large number of bugs.

#5

On Jun 5, 3:58*am, robu... (AT) gmail (DOT) com wrote:

Quote:

On Jun 5, 1:25 pm, "Walter Mitty" <wami... (AT) verizon (DOT) net> wrote:

Consider two electrons. *They both have the same mass, and they have the
same charge. *They might have opposite spins. *But the minute we add a third
electron, the spin of two of them is going to be identical. *It seemsthat,
on the surface at least, electrons do not have enough properties to carry
identity.

If objects are really not distinguishable then why someone would want
to artificially identify them? Wouldn’t be better to record just their
quantity?

For a philosophical take on this issue of distinguishability:

http://plato.stanford.edu/entries/id...indiscernible/
http://homepages.wmich.edu/~baldner/black.htm

--Jeff

#6

"Walter Mitty" <wamitty (AT) verizon (DOT) net> wrote

Quote:

robur.6 (AT) gmail (DOT) com> wrote in message
news:edb2d328-6d69-4273-a5ab-99c3b0f50ff5 (AT) v4g2000vba (DOT) googlegroups.com...
On Jun 5, 1:25 pm, "Walter Mitty" <wami... (AT) verizon (DOT) net> wrote:

Consider two electrons. They both have the same mass, and they have the
same charge. They might have opposite spins. But the minute we add a
third
electron, the spin of two of them is going to be identical. It seems
that,
on the surface at least, electrons do not have enough properties to
carry
identity.

If objects are really not distinguishable then why someone would want
to artificially identify them? Wouldn't be better to record just their
quantity?

Good point. So there are some case where identity is not important.

(Incidentally, the other response where the comment was made that two
electrons must occupy different states leads to a slippery slope. Using
state to identify implies that a change of state is in effect a change of
identity. This is related to Brian's claim that all attributes of an
entity are mutable. (Am I misinterpreting your claim, Brian?)

I'm not sure.

For ordinary things--that is, for things that have a locus in time--identity
requires that every property exemplified at every location in time be the
same. It doesn't require that every property be the same at every location
in time. If someone asserts that at exactly 12:31, the guy that was in
second in line at exactly 12:30 is now first in line, then it is reasonable
to conclude that if you're referring to the guy that was second in line at
exactly 12:30, then you're referring to the guy that was first in line at
exactly 12:31.

What I think you're referring to as my claim is that it is possible for
every component of a pair of tuples that map to a particular thing in the
Universe at successive databases to be different. For example, there could
be a plastic bin located somewhere in a warehouse with a label with "A12" on
it on Friday, and that same plastic bin could be located somewhere else with
a different label with "A14" on it on Monday. The fact that its location
and label both changed doesn't mean that it isn't the same plastic bin. So
here are the tuples at successive databases

before moving {Label:"A12",Location:R24S3}

after moving {Label:"A14",Location:R15S1}

Obviously, every component of the tuples are different, yet both map to the
same plastic bin.

<snip>

#7

Quote:

In other words the abstract identifiers can actually be seen as providing a curious means to encode the number of indistinguishable items in the cardboard box.

The usual model is that this is a collective/ aggregate/ bulk/
commodity entity (grains of sand, etc.) whose identity for DB purposes
does not involve separate units. We talk in terms of lots with
attributes like weight, volume, etc.

But the rules are weird -- lot A of #2 cooking oil can be added to
lot B of #2 cooking oil to produce Lot C. Is it a new entity? If I
split lot C into lots A and B, what is the data model? If I do a
blend of oils in the batches, what is it? Etc.

This is a classic discrete (digitial) versus analog (continous)
models of data problem.

#8

On Jun 5, 2:25*am, "Walter Mitty" <wami... (AT) verizon (DOT) net> wrote:

Quote:

"David BL" <davi... (AT) iinet (DOT) net.au> wrote in message

news:03132046-3fd2-4bb1-9e00-cd81ece451c2 (AT) h2g2000yqg (DOT) googlegroups.com...

Informally I think of abstract identifiers as "internal glue" within a
relational database. *A bit more formally, they are characterised as
identifiers that could be mapped bijectively to different values
throughout the database without changing the recorded information.
One would therefore hope that they aren't visible to end users.

I have wondered for some time whether abstract identifiers are only
needed within the confines of a flat relational model. *The following
hypothetical example is meant to cast some light on this question.
I'm hoping you'll see the underlying matters of principle and see how
it raises some interesting questions and ideas:

Let a cardboard box of items land on your desk and your task is to
record information about the items in a database. *In theory each item
is uniquely identified by its (x,y,z) position at any given epoch.
However, the idea is that the items are mixed up in the box and their
positions are irrelevant to the information that needs to be recorded.

The items are not labelled. *The idea is to uniquely identify them
(only) by their observable properties. *This is indeed assumed to be
an important integrity constraint to be enforced by the DBMS. *Note as
well that it would be upsetting (and potentially very costly or
impractical) if the database system forces items to be labelled when
there shouldn't be a need to.

Consider two electrons. *They both have the same mass, and they have the
same charge. *They might have opposite spins. *But the minute we add a third
electron, the spin of two of them is going to be identical. *

Is there anything QM specific about it? Consider 3 coins on the table.
2 of them have identical states (head-head or tail-tail)

#9

On Jun 5, 11:25*am, "Walter Mitty" <wami... (AT) verizon (DOT) net> wrote:

Quote:

"David BL" <davi... (AT) iinet (DOT) net.au> wrote in message

news:03132046-3fd2-4bb1-9e00-cd81ece451c2 (AT) h2g2000yqg (DOT) googlegroups.com...

Informally I think of abstract identifiers as "internal glue" within a
relational database. *A bit more formally, they are characterised as
identifiers that could be mapped bijectively to different values
throughout the database without changing the recorded information.
One would therefore hope that they aren't visible to end users.

I have wondered for some time whether abstract identifiers are only
needed within the confines of a flat relational model. *The following
hypothetical example is meant to cast some light on this question.
I'm hoping you'll see the underlying matters of principle and see how
it raises some interesting questions and ideas:

Let a cardboard box of items land on your desk and your task is to
record information about the items in a database. *In theory each item
is uniquely identified by its (x,y,z) position at any given epoch.
However, the idea is that the items are mixed up in the box and their
positions are irrelevant to the information that needs to be recorded.

The items are not labelled. *The idea is to uniquely identify them
(only) by their observable properties. *This is indeed assumed to be
an important integrity constraint to be enforced by the DBMS. *Note as
well that it would be upsetting (and potentially very costly or
impractical) if the database system forces items to be labelled when
there shouldn't be a need to.

Consider two electrons. *They both have the same mass, and they have the
same charge. *They might have opposite spins. *But the minute we add a third
electron, the spin of two of them is going to be identical. *It seems that,
on the surface at least, electrons do not have enough properties to carry
identity.

Then how would you know there were three electrons at all. Magic?

Quote:

*As you descend into lower level particles like quarks, things get
even more this way. *Particles seem more and more interchangeable.
Subatomic particles might not have much to do with your objects in a box,
but it seems to me that any theory of reality and identity that falls apart
at the subatomic level should at least take that into account.

#10

On Jun 6, 7:15 am, JOG <j... (AT) cs (DOT) nott.ac.uk> wrote:

Quote:

Then how would you know there were three electrons at all. Magic?

This is all so off topic, but it is interesting that a Bose–Einstein
condensate has properties that depend on the density of boson
particles, and yet those particles are not only identical but also may
occupy the identical quantum state. It's as though nature can
remember how many there are without knowing the difference between
them. This is just one way in which QM is really weird.

Getting back to the topic of my original post, what do you think of
the idea of DVAs to avoid the need to label things?