Definition for Concept 'Invariant View or Perspective on a Thing'

ISSUE: http://www.w3.org/2011/prov/track/issues/19

Introduction

Following discussions, we agreed that the notion of 'Invariant View or Perspective on a Thing' is a useful provenance-related concept, which is also likely to be part of the provenance interchange language to be standardized (though was not identified in the charter http://www.w3.org/2011/01/prov-wg-charter).

• What term do we adopt for the concept 'Invariant View or Perspective on a Thing'?
• How do we define the concept 'Invariant View or Perspective on a Thing'?
• Where does concept 'Invariant View or Perspective on a Thing' appear in ProvenanceExample?
• Which provenance query requires the concept 'Invariant View or Perspective on a Thing'?

Proposed Definitions for the Concept 'Invariant View or Perspective on a Thing'

Definition x.y.z

A resource defined such that any permitted variations in the resource state do not affect the veracity of provenance data under consideration.

Thus, what constitutes an invariant view for this purpose will depend on the provenance information that is being considered. The general expectation is that all those aspects that give rise to provenance claims will be unchanging.

Examples:

• A document author: if only document authorship is considered for provenance purposes, an invariant view may permit changes in the document content as long as the author does not change.

-- Graham Klyne, 2011-06-07

It would be nice if we could define this concept without referring to provenance (which we are trying to define too by means of these concepts). Indeed, this circularity doesn't help newcomers understand what this concept is.

Also, given that we don't have consensus over resource, and we said we would define concepts independently of the Web Architecture, we should refrain from using this word, if we can.

--Luc Moreau 11:24, 7 June 2011 (UTC)

Definition by JimM

An Invariant View or Perspective of a thing A is another thing B that is related to A as follows: The definition of B corresponds to the definition of A (both in ontologies that are out-of-scope) in that:

• A and B correspond completely at some instant in time
• The definitions of the types of things that A and B are instances of imply:
  • a set of processes to which A and B are immune - executions of such processes on A and B do not change their identity, these processes are considered part of the lifecycle of A and B
  • A and B share the same values for properties in immutabile and mutable sets
  • a non-empty set of processes to which A is immune and B is not - executions of such processes do not change the identity of A but consume B (they use B and do not produce a modified B). These processes are not part of the lifecycle of B by definition.
  • a set of processes to which both A and B are not immune -the processes consume both and do not produce modified versions of them. These processes are not part of the lifecycle of A and B

When A and B exist as defined above, we say B is an Invariant View or Perspective on thing A. IVPT relationships can potentially form a DAG. IVPT relationships are transitive.

B is an IVP of A iff there is a mutable property of A that is considered an immutable property of B (one that is integral to B's identity). (A and B have different identities/are different things)

Definition by Jim and Luc

A XXX is a thing:

• that has an identity
• that has a set of properties, integral to this identity and invariant for the purpose of the current description
• that may have other properties, not integral to its identity but allowed to be mutable

There is no assumption that the set of properties is complete, and that properties are independent of each other.

t1 is an invariant view of t2, where t1 and t2 are XXX:

Definition by Jim and Luc v2 (in progress)

ACCEPTED at teleconference 2011-06-16

In the real world, there are "stuff" described in terms of physical, digital, logical, conceptual, other types of entities.

Definition of 'thing': In PIL, real-world stuff are modeled as "things" that:

• have an identity
• model aspects of the entity's state as a set of properties of the "thing" that are invariant and considered integral to identity
• model other aspects of the entity's state as a second set of properties that are mutable and able to change during the "thing"'s lifecycle, and therefore not integral to identity

There are no assumptions that the sets of properties are complete, or that the properties are independent/orthogonal of each other.

Definition of "invariant view/perspective" (IVP):

For A and B, both of which are "things", B is an IVP of A iff

• A and B represent the same stuff at some instant in time (the set of properties they share must have identical values at that instant and the state common to the entities represented by A and B must consistent), and
• there is/are mutable property(ies) of A that is/are correspond to immutable property(ies) of B (i.e. ones that are integral to B's identity)

These definitions then have a number of implications: If B is an IVP of A, they must have the same values for all immutable properties they have in common. The distinction between A and B becomes relevant when one is discussing processes that affect the aspects of entity state represented by the subset of properties that are in the immutable set of properties of A and/or in the mutable set of properties for B.

Comment

• I was not present at the teleconference, and have a problem with the notion of "integral to identity" (see more below). -- GK, 2011-06-17

• I was not present at the teleconference. I think this definition of IVP is over-specified. Specifically, I think defining an IVP via an "iff" qualification is unnecessary and may lead to unexpected conclusions. I think the essence that needs to be captured is that B is understood to refer to the same thing as A, and some of the properties of B may be invariant where the corresponding properties of A are mutable. I don't think we need a procedure to decide if B is IVP of A, just some properties to be inferred if we assert that B is IVP of A. -- GK, 2011-06-17

• • I agree with your point on 'iff'. I dropped it in the definition below

--Luc Moreau 07:32, 20 June 2011 (UTC)

A Further Definition by Luc

I thought that we can simplify the definition above, by not mentioning entities.

In the real world, there are "stuffs" that can be physical, digital, logical, conceptual, or otherwise.

Definition of 'thing': In PIL, "things" represent real-world stuffs and have properties modeling aspects of stuff states. Things have:

• an identity
• a set of invariant properties, which are considered integral to identity
• a set of mutable properties, which are able to change during the "thing"'s lifecycle, and therefore not integral to identity

There are no assumptions that the sets of properties are complete, or that the properties are independent/orthogonal of each other.

Definition of "invariant view/perspective of" (IVP of):

Let A and B be two things. An assertion "B is an IVP of A" holds if :

• A and B represent the same stuff in the real world, and the stuff states modelled by A and B are consistent
• the set of properties they share must have identical values
• there is at least one mutable property of A that is corresponding to an immutable property of B (i.e. one that is integral to B's identity)

Comments

• I thought we could do without referring to entity. So it is stuff vs thing.
• I am not at ease with A and B represent the same stuff at some instant in time in the original specification of IVP of, since A and B may have been 'things' modelling states at different moments.
• IVP of: is something asserted by someone, and not something inferred.

--Luc Moreau 13:26, 17 June 2011 (UTC)

• I have a problem with the (repeated) notion that invariant properties are "integral to identity". Suggest: just drop that qualificaton where it appears. I may have grey eyes throughout my life (an invariant property), but I don't think that's integral to my identity. -- GK, 2011-06-17
  • Remember, we have 'stuff' in the real-world, and they are represented by 'things' in our language. The identity we talk about here, is not the identity of the 'stuff' but of the 'thing'. A thing with property 'grey eyes' is not the same as a thing without it. It's in that sense that I see this as integral to the identity of the thing. Essentially, having a property 'grey eyes' is a different view (invariant) to the thing without grey eyes property. Isn't it the case that eye colour changes after birth? --Luc Moreau 07:37, 20 June 2011 (UTC)

Further simplification

Following comments by Graham, Simon, and Jim, a further attempt to simplify.

In the real world, there are "stuffs" that can be physical, digital, logical, conceptual, or otherwise.

Definition of 'thing': In PIL, "things" represent real-world stuffs and have properties modeling aspects of stuff states. Things have:

• an identity
• a set of invariant properties
• a set of mutable properties, which are able to change during the "thing"'s lifecycle

There are no assumptions that the sets of properties are complete, or that the properties are independent/orthogonal of each other.

Definition of "invariant view/perspective of" (IVP of):

Let A and B be two things. An assertion "B is an IVP of A" indicates that, for its asserter, A and B represent the same stuff in the real world, and the stuff states modelled by A and B are consistent. "B is an IVP of A" is valid if, for its asserter, the following holds:

• the properties they share must have corresponding values
• some mutable properties of A correspond to some immutable properties of B

Comments

• In the definition of IVP of, the term "corresponds" is important since, properties of A may be converted into properties of B (e.g. temperature conversion from Farenheit to Celsius) or can be merged.
• As Khalid noted, correspondence may be many to many. For instance, a rectangle A may have varying length and width, whereas B, an IVP of A, may have a invariant area.
• The "correspondence" is not explicit in PIL; it is pushed to the asserter
• Removed 'integral to identity'
• The stuff states modelled by A and B are consistent: these states may be modelled by different ontologies. It is left to the asserter to establish their consistency (outside the scope of PIL).
• Stian suggests including a qualification of "corresponds":
  • The asserter is free to choose what "corresponding properties" above means. The corresponding properties might match many-to-many, eg. rectangle A may have varying length and width, whereas B, an IVP of A, may have a invariant area.

--Luc Moreau 15:59, 20 June 2011 (UTC)

Comment by Stian 24 June 2011

I believe it is correct to say that A and B might not be the same class. Just like a File is not a Document, or a Letter is not an Invoice. However we often talk about these as being somewhat different views of the same 'stuff' (the bytes on the disk, the paper in front of us).

Example:

• The File "definition.txt" was written to disk by Stian
• The Document "definition.txt" was authored by Paolo (and is stored in the above-mentioned file) and later slightly modified by Stian.
• The Printout with the header "definition.txt" shows the content of the Document before Stian's modifications.
• They File and Document share corresponding properties such as "content" and "last modified".
• The Printout and Document share corresponding properties such as "author" - it is in the form of a username on the file system, and a part of the content for the document.
• The Printout content is an invariant property, but a variant property of the Document.
• If the File is a view (IVP) of the Document, then we could say that the file content is corresponding to the Document content, although the File content would include things like UTF-8 encodings and line endings, while the Document content is "what you see on your screen/printout".
• Document and File both share the invariant property 'filename' - which in this case also helps us link these to the same 'stuff' - bytes on the disk. Printout also shares this invariant property - but it's in another form, just a header.
• The stuff state here would be which bytes are on the disk and which letters are on the paper. These are consistent if they give the same "corresponding" content. So if the File content includes a newline which does not appear on the Printout, then they are no longer corresponding, and so not consistent - the Printout is no longer an IVP of the File. (Is it still an IVP of the Document?)

Definition x.y.z