DIIS

DIIS (direct inversion in the iterative subspace or direct inversion of the iterative subspace), also known as Pulay mixing, is a technique for extrapolating the solution to a set of linear equations by directly minimizing an error residual (e.g. a Newton–Raphson step size) with respect to a linear combination of known sample vectors.

Version

v1 · 2026-09-28 · History

Domain-specific #

8980

Domain group

Natural Sciences

Origin domain

Chemistry & Materials Science

Subdomains

Computational Quantum Chemistry, Self Consistent Field Methods → Chemistry & Materials Science

Core Idea

DIIS is treated here as the recurring mathematicslogicstatistics identity summarized by this source-grounded definition: DIIS (direct inversion in the iterative subspace or direct inversion of the iterative subspace), also known as Pulay mixing, is a technique for extrapolating the solution to a set of linear equations by directly minimizing an error residual (e.g. a Newton–Raphson step size) with respect to a linear combination of known sample vectors. DIIS (direct inversion in the iterative subspace or direct inversion of the iterative subspace), also known as Pulay mixing, is a technique for extrapolating the solution.

How would you explain it like I'm…

Mix-The-Guesses Trick

DIIS is a trick for making better guesses faster. You look at your last few guesses and how wrong each one was, then mix them together so their mistakes cancel out as much as possible. That mix is your next, better guess.

Smart Guess Blending

DIIS is a method computers use to solve a hard problem that needs many rounds of guessing and checking. Each round gives a guess and a measure of how wrong that guess is, called the error. Instead of using only the newest guess, DIIS keeps several recent ones and figures out how much of each to blend so the blended errors add up as close to zero as possible. It then uses those same amounts to blend the guesses into a new one. Chemists invented it to help computer models of molecules settle on an answer faster and more reliably.

Error-Minimizing Extrapolation

DIIS, short for direct inversion in the iterative subspace and also called Pulay mixing, speeds up iterative calculations that repeatedly refine an answer. Each iteration produces a trial solution and an error vector saying how far it is from satisfying the equations. DIIS stores several recent error vectors and finds the weighted combination of them that comes closest to zero, using a least-squares fit. It then applies the same weights to the stored trial solutions to extrapolate the next guess. It was developed by Peter Pulay to make the Hartree–Fock self-consistent field method in quantum chemistry converge faster and more stably. Compared with simply taking the latest step, it uses the history of past steps to jump closer to the answer.

 

DIIS (direct inversion in the iterative subspace), or Pulay mixing, is an acceleration technique for iterative solvers that extrapolates a solution by minimizing an error residual over a linear combination of previously computed sample vectors. At each iteration it keeps a subspace of recent trial vectors and their associated error vectors (for example, Newton–Raphson step sizes). Coefficients are chosen so that the combined error vector best approximates the null vector in the least-squares sense. Those coefficients are then applied to the trial vectors to generate the next iterate. Peter Pulay introduced it in computational quantum chemistry to accelerate and stabilize convergence of the Hartree–Fock self-consistent field procedure. It has been shown equivalent to the earlier Anderson acceleration, and for linear problems it reduces to GMRES.

Scope of Application

• Documented setting. The newly determined coefficients are then used to extrapolate the function variable for the next iteration.

• Details. The DIIS method seeks to minimize the norm of under the constraint that the coefficients sum to one.

• Details. The coefficients are then used to update the variable as.

• Documented setting. DIIS was developed by Peter Pulay in the field of computational quantum chemistry with the intent to accelerate and stabilize the convergence of the Hartree–Fock self-consistent field method.

• Documented setting. The method has been shown to be equivalent to the pre-existing Anderson acceleration , and to reduce to GMRES in the linear case .

Clarity

A clear use of DIIS names the carrier, the operative relation, and the conditions under which the source treats the identity as present. The minimal definition is DIIS (direct inversion in the iterative subspace or direct inversion of the iterative subspace), also known as Pulay mixing, is a technique for extrapolating the solution to a set of linear equations by directly minimizing an error residual (e.g. a Newton–Raphson.

Manages Complexity

DIIS compresses multiple mathematicslogicstatistics details into a stable diagnostic relation. The source shows both the central mechanism—dIIS was developed by Peter Pulay in the field of computational quantum chemistry with the intent to accelerate and stabilize the convergence of the Hartree–Fock self-consistent field method.—and the practical consequence—the DIIS method seeks to minimize the norm of under the constraint that the coefficients sum to one.

Abstract Reasoning

1. Type the carrier. Identify the mathematicslogicstatistics entities to which the claim applies.
2. State the relation. Use the source-grounded identity: DIIS (direct inversion in the iterative subspace or direct inversion of the iterative subspace), also known as Pulay mixing, is a technique for extrapolating the solution to a set of linear equations by directly minimizing an error residual (e.g. a Newton–Raphson step size) with respect to a linear combination of known sample vectors.
3. Check operation and conditions.

Knowledge Transfer

Within the home domain. Knowledge about DIIS transfers literally when a new case preserves the same carrier type, relation, and recognition test. The newly determined coefficients are then used to extrapolate the function variable for the next iteration. The DIIS method seeks to minimize the norm of under the constraint that the coefficients sum to one. Beyond the home domain. No canonical parent is asserted for DIIS. An outside case receives the specialist name only when the same typed roles and rejection conditions can be filled literally; otherwise the comparison remains an analogy pending later graph densification.

Relationships to Other Abstractions

Current abstraction DIIS Domain-specific

Parents (1) — more general patterns this builds on

• DIIS is a kind of Algorithm Prime

  DIIS is an iterative residual-minimizing procedure for extrapolating improved solutions.

Hierarchy paths (2) — routes to 2 parentless roots

• DIIS → Algorithm → Function (Mapping)

Show alternative path (1)

Neighborhood in Abstraction Space

DIIS sits in a sparse region of the domain-specific corpus (81^st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Physical Quantities, Operators & Formulas (33 abstractions)

Nearest neighbors

• Powder diffraction — 0.83
• Mean-field theory — 0.82
• Translation operator (quantum mechanics) — 0.82
• Mehler Kernel — 0.82
• Projector Augmented-Wave Method — 0.81

Computed from structural-signature embeddings · 2026-10-08