Every digital library eventually changes platforms. When it does, thousands of links in citations, catalogs and footnotes can break at once. Persistent identifiers exist to stop that. They give each object a name that stays the same while the location behind it changes.
There are several identifier systems in common use. They solve the same problem in different ways, with different costs and different promises. This guide compares the four you are most likely to meet: DOI, ARK, Handle and PURL.
What persistence actually means
No identifier is persistent by itself. A DOI that points to a dead page is just as broken as any other link. Persistence comes from two things working together:
- Indirection. The identifier resolves through a service that you can update, so the public name stays fixed while the target URL changes.
- Commitment. Someone takes responsibility for updating those targets, for as long as the objects exist, and the organization behind the system keeps the resolver running.
It also helps to separate the identifier from the landing page. The identifier names the object. The landing page is where a person or a machine arrives when the identifier resolves. That page should show the title, a citation, the rights status and a clear link to the file or the catalog record. When the object moves, only the target behind the identifier changes. When the object is withdrawn, the landing page becomes a tombstone that explains what happened.
The technology handles the first part well. The second part is policy and staffing. Most link rot in scholarly sources comes from neglect, not from technical failure. Any choice of identifier should start with an honest answer to the question: who will maintain these records in ten years?
The four systems
Handle
The Handle System was developed at the Corporation for National Research Initiatives in the 1990s. It is a general-purpose distributed system for resolving names to data, including URLs. Today it is overseen by the DONA Foundation. Many institutional repositories, including those built on DSpace, mint Handles by default. Running a Handle prefix requires registering it and operating or contracting a local Handle server.
DOI
The Digital Object Identifier is built on top of the Handle System. It adds a governance layer: the International DOI Foundation sets policy, and registration agencies such as Crossref and DataCite issue DOIs to members. DOI is standardized as ISO 26324. Registration agencies also require metadata with each DOI, which is why DOIs feed citation indexes, reference linking and discovery services so well. The trade-off is cost. Membership and per-DOI fees vary by agency and organization type, and should be checked directly.
ARK
The Archival Resource Key came out of the California Digital Library. An ARK is an ordinary URL with a recognizable structure: a Name Assigning Authority Number, or NAAN, followed by a local name. Obtaining a NAAN is free, and there are no per-identifier fees. ARKs can resolve through your own server or through the global resolver at n2t.net. The ARK Alliance now supports the community. ARKs suit archives and museums that want persistence without membership fees and are willing to run the resolution themselves.
PURL
Persistent URLs were introduced by OCLC in the mid-1990s as a simple redirect service. The purl.org service is now operated by the Internet Archive. A PURL is a plain HTTP redirect with no required metadata. PURLs are still common for vocabulary and ontology namespaces, but they are less often used for collection objects today.
Side by side
| System | Governance | Cost model | Metadata required | Typical use |
|---|---|---|---|---|
| DOI | International DOI Foundation and registration agencies | Membership and registration fees | Yes, per agency schema | Articles, datasets, reports, theses |
| ARK | ARK Alliance, local institution | Free NAAN, self-run resolution | Optional, recommended | Archival objects, museum records, images |
| Handle | DONA Foundation, local institution | Prefix registration, local server | No | Institutional repositories |
| PURL | Internet Archive (purl.org) | Free | No | Vocabularies, namespaces, simple redirects |
A useful way to read the table: DOI buys you integration with the scholarly publishing system, at a price. ARK buys you independence, at the cost of doing your own maintenance. Handle sits between them and often arrives with your repository software. PURL is a lightweight redirect and little more.
Lessons from older digital libraries
Digital library projects from the 2000s are a clear record of what happens without persistence. The MetaScholar Initiative, a digital library programme run from Emory's Woodruff Library in that decade, produced portals, project sites and harvested metadata collections, much of it identified by plain URLs. Like most projects of the period, it relied on the host institution to keep those addresses alive after each grant ended. The question of keeping services alive after a grant ends was central to the initiative's 2006 Sustaining Digital Libraries symposium.
Harvested metadata makes the problem worse. When an aggregator stores a record, it stores the identifier the source gave it. If that identifier is a location rather than a persistent name, every later platform migration breaks the link inside someone else's system, where you cannot fix it. This is one reason the identifier field matters so much in Dublin Core records.

How to choose
Most institutions end up using more than one system. A simple decision path helps:
- Is the object a citable scholarly output? Articles, datasets, reports and theses that need to appear in citation indexes are good candidates for DOIs through Crossref or DataCite.
- Is it a collection object at scale? Photographs, letters and audio files numbering in the thousands often suit ARKs, which carry no per-item fees.
- Does your repository already mint Handles? Keep them working and documented rather than replacing them for the sake of change.
- Is it a vocabulary or namespace? A PURL or a stable domain you control is usually enough.
Whatever you choose, a few rules apply to all of them. Do not put meaning in the identifier that may change, such as department names or file formats. Never reuse an identifier for a different object. Publish a persistence policy that says what you promise and what happens when an object is withdrawn. Show a tombstone page instead of a 404 when something is removed.
The real cost is maintenance
Registration fees are visible and easy to budget. Maintenance is not. Each platform migration needs a step that updates resolver targets in bulk. Each withdrawn object needs a decision. Each staff change needs a handover note that explains how identifiers are minted and where the records live.
If that work is planned, any of these systems will keep your links alive. If it is not, none of them will. Persistence is an institutional habit, and the identifier is only the visible part of it. For related planning on long-term funding, see funding models for digital libraries.
