Gilbert Strang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Gilbert Strang has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Ding-Xuan Zhou
School of Mathematics and Statistics, University of Sydney, Sydney, Australia
2 shared publications - Vasily Strela
Massachusetts Institute of Technology
2 shared publications - Alan Edelman
Department of Mathematics MIT Cambridge MA
1 shared publication - Alan Edelman
Massachusetts Institute of Technology
1 shared publication - Albert Lu
Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA 02139, United States of America
1 shared publication - Avi Shporer
Massachusetts Institute of Technology, Department of Physics and Kavli Institute for Astrophysics and Space Research , Cambridge, MA 02139
1 shared publication
Latest Publications
- Open Access
Generalized Inverses of Matrix Products: From Fundamental Subspaces to Randomized Decompositions
We investigate the Moore-Penrose pseudoinverse and generalized inverse of a matrix product $A=CR$ to establish a unifying framework for generalized and randomized matrix inverses. This analysis is rooted in first principles, focusing on the geometry of the four fundamental subspaces. We examine: (1) the reverse order law, $A^+ = R^+C^+$, which holds when $C$ has independent columns and $R$ has independent rows, (2) the universally correct formula, $A^+ = (C^+CR)^+(CRR^+)^+$, providing a geometric interpretation of the mappings between the involved subspaces, (3) a new generalized randomized formula, $A^+_p = (P^TA)^+P^TAQ(AQ)^+$, which gives $A^+_p = A^+$ if and only if the sketching matrices $P$ and $Q$ preserve the rank of $A$, i.e., $\mathrm{rank}(P^TA) = \mathrm{rank}(AQ) = \mathrm{rank}(A)$. The framework is extended to generalized $\{1,2\}$-inverses and specialized forms, revealing the underlying structure of established randomized linear algebra algorithms, including randomized SVD, the Nyström approximation, and CUR decomposition. We demonstrate applications in sparse sensor placement and effective resistance estimation. For the latter, we provide a rigorous quantitative analysis of an approximation scheme, establishing that it always underestimates the true resistance and deriving a worst-case spectral bound on the error of resistance differences.Open AccessGeneralized Inverses of Matrix Products: From Fundamental Subspaces to Randomized Decompositions
We investigate the Moore-Penrose pseudoinverse and generalized inverse of a matrix product $A=CR$ to establish a unifying framework for generalized and randomized matrix inverses. This analysis is rooted in first principles, focusing on the geometry of the four fundamental subspaces. We examine: (1) the reverse order law, $A^+ = R^+C^+$, which holds when $C$ has independent columns and $R$ has independent rows, (2) the universally correct formula, $A^+ = (C^+CR)^+(CRR^+)^+$, providing a geometric interpretation of the mappings between the involved subspaces, (3) a new generalized randomized formula, $A^+_p = (P^TA)^+P^TAQ(AQ)^+$, which gives $A^+_p = A^+$ if and only if the sketching matrices $P$ and $Q$ preserve the rank of $A$, i.e., $\mathrm{rank}(P^TA) = \mathrm{rank}(AQ) = \mathrm{rank}(A)$. The framework is extended to generalized $\{1,2\}$-inverses and specialized forms, revealing the underlying structure of established randomized linear algebra algorithms, including randomized SVD, the Nyström approximation, and CUR decomposition. We demonstrate applications in sparse sensor placement and effective resistance estimation. For the latter, we provide a rigorous quantitative analysis of an approximation scheme, establishing that it always underestimates the true resistance and deriving a worst-case spectral bound on the error of resistance differences. - 20 November 2025
- Published by Elsevier in Journal of Computational Physics
- Vol. 547
High Order Lax-Wendroff Methods for Hyperbolic Systems
- 20 November 2025
- Published by Elsevier in Journal of Computational Physics
- Vol. 547
High Order Lax-Wendroff Methods for Hyperbolic Systems
- 10 July 2025
- Published by Taylor & Francis in Scatterplot
- Vol. 2 (1)
The Four Fundamental Subspaces and Their Connection to Data Matrices
- 10 July 2025
- Published by Taylor & Francis in Scatterplot
- Vol. 2 (1)
The Four Fundamental Subspaces and Their Connection to Data Matrices
- Open AccessThe Pseudoinverse of $A=CR$ is $A^+=R^+C^+$ (?)This paper gives three formulas for the pseudoinverse of a matrix product $A = CR$. The first is sometimes correct, the second is always correct, and the third is almost never correct. But that third randomized pseudoinverse $A^+_r$ may be very useful when $A$ is a very large matrix. 1. $A^+ = R^+C^+$ when $A = CR$ and $C$ has independent columns and $R$ has independent rows. 2. $A^+ = (C^+CR)^+(CRR^+)^+$ is always correct. 3. $A^+_r = (P^TCR)^+P^TCRQ(CRQ)^+ = A^+$ only when $\mathrm{rank}(P^TA) = \mathrm{rank}(AQ) = \mathrm{rank}(A)$ with $A = CR$.Open AccessThe Pseudoinverse of $A=CR$ is $A^+=R^+C^+$ (?)This paper gives three formulas for the pseudoinverse of a matrix product $A = CR$. The first is sometimes correct, the second is always correct, and the third is almost never correct. But that third randomized pseudoinverse $A^+_r$ may be very useful when $A$ is a very large matrix. 1. $A^+ = R^+C^+$ when $A = CR$ and $C$ has independent columns and $R$ has independent rows. 2. $A^+ = (C^+CR)^+(CRR^+)^+$ is always correct. 3. $A^+_r = (P^TCR)^+P^TCRQ(CRQ)^+ = A^+$ only when $\mathrm{rank}(P^TA) = \mathrm{rank}(AQ) = \mathrm{rank}(A)$ with $A = CR$.
- Open Access
Elimination and Factorization
If a matrix $A$ has rank $r$, then its row echelon form (from elimination) contains the identity matrix in its first $r$ independent columns. How do we \emph{interpret the matrix} $F$ that appears in the remaining columns of that echelon form\,? $F$ multiplies those first $r$ independent columns of $A$ to give its $n-r$ dependent columns. Then $F$ reveals bases for the row space and the nullspace of the original matrix $A$. And $F$ is the key to the column-row factorization $\boldsymbol{A}=\boldsymbol{CR}$.Open AccessElimination and Factorization
If a matrix $A$ has rank $r$, then its row echelon form (from elimination) contains the identity matrix in its first $r$ independent columns. How do we \emph{interpret the matrix} $F$ that appears in the remaining columns of that echelon form\,? $F$ multiplies those first $r$ independent columns of $A$ to give its $n-r$ dependent columns. Then $F$ reveals bases for the row space and the nullspace of the original matrix $A$. And $F$ is the key to the column-row factorization $\boldsymbol{A}=\boldsymbol{CR}$. - Top Cited PapersOpen Access
- 2 August 2022
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 119 (32)
A neural network solves, explains, and generates university math problems by program synthesis and few-shot learning at human level
We demonstrate that a neural network pretrained on text and fine-tuned on code solves mathematics course problems, explains solutions, and generates questions at a human level. We automatically synthesize programs using few-shot learning and OpenAI’s Codex transformer and execute them to solve course problems at 81% automatic accuracy. We curate a dataset of questions from Massachusetts Institute of Technology (MIT)’s largest mathematics courses (Single Variable and Multivariable Calculus, Differential Equations, Introduction to Probability and Statistics, Linear Algebra, and Mathematics for Computer Science) and Columbia University’s Computational Linear Algebra. We solve questions from a MATH dataset (on Prealgebra, Algebra, Counting and Probability, Intermediate Algebra, Number Theory, and Precalculus), the latest benchmark of advanced mathematics problems designed to assess mathematical reasoning. We randomly sample questions and generate solutions with multiple modalities, including numbers, equations, and plots. The latest GPT-3 language model pretrained on text automatically solves only 18.8% of these university questions using zero-shot learning and 30.8% using few-shot learning and the most recent chain of thought prompting. In contrast, program synthesis with few-shot learning using Codex fine-tuned on code generates programs that automatically solve 81% of these questions. Our approach improves the previous state-of-the-art automatic solution accuracy on the benchmark topics from 8.8 to 81.1%. We perform a survey to evaluate the quality and difficulty of generated questions. This work automatically solves university-level mathematics course questions at a human level and explains and generates university-level mathematics course questions at scale, a milestone for higher education.
- 2 August 2022
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 119 (32)
Top Cited PapersOpen AccessA neural network solves, explains, and generates university math problems by program synthesis and few-shot learning at human level
We demonstrate that a neural network pretrained on text and fine-tuned on code solves mathematics course problems, explains solutions, and generates questions at a human level. We automatically synthesize programs using few-shot learning and OpenAI’s Codex transformer and execute them to solve course problems at 81% automatic accuracy. We curate a dataset of questions from Massachusetts Institute of Technology (MIT)’s largest mathematics courses (Single Variable and Multivariable Calculus, Differential Equations, Introduction to Probability and Statistics, Linear Algebra, and Mathematics for Computer Science) and Columbia University’s Computational Linear Algebra. We solve questions from a MATH dataset (on Prealgebra, Algebra, Counting and Probability, Intermediate Algebra, Number Theory, and Precalculus), the latest benchmark of advanced mathematics problems designed to assess mathematical reasoning. We randomly sample questions and generate solutions with multiple modalities, including numbers, equations, and plots. The latest GPT-3 language model pretrained on text automatically solves only 18.8% of these university questions using zero-shot learning and 30.8% using few-shot learning and the most recent chain of thought prompting. In contrast, program synthesis with few-shot learning using Codex fine-tuned on code generates programs that automatically solve 81% of these questions. Our approach improves the previous state-of-the-art automatic solution accuracy on the benchmark topics from 8.8 to 81.1%. We perform a survey to evaluate the quality and difficulty of generated questions. This work automatically solves university-level mathematics course questions at a human level and explains and generates university-level mathematics course questions at scale, a milestone for higher education.
- Open Access
- 1 January 2021
- Published by Banach Press LLC in Journal of Convex Analysis
- Vol. 28 (2) , 725-728
The Column-Row Factorization of a Matrix
The active ideas in linear algebra are often expressed by matrix factorizations : S=Q\Lambda Q^{\mathrm{T}} S = Q Λ Q T for symmetric matrices (the spectral theorem) and A=U\Sigma V^{\mathrm{T}} A = U Σ V T for all matrices (singular value decomposition). Far back near the beginning comes A=LU A = L U for successful elimination : Lower triangular times upper triangular. This paper is one step earlier, with bases in A=CR A = C R for the column space and row space of any matrix – and a proof that column rank = row rank. The echelon form of A A and the pseudoinverse A^+ A + appear naturally. The “proofs” are mostly “observations”.
- 1 January 2021
- Published by Banach Press LLC in Journal of Convex Analysis
- Vol. 28 (2) , 725-728
Open AccessThe Column-Row Factorization of a Matrix
The active ideas in linear algebra are often expressed by matrix factorizations : S=Q\Lambda Q^{\mathrm{T}} S = Q Λ Q T for symmetric matrices (the spectral theorem) and A=U\Sigma V^{\mathrm{T}} A = U Σ V T for all matrices (singular value decomposition). Far back near the beginning comes A=LU A = L U for successful elimination : Lower triangular times upper triangular. This paper is one step earlier, with bases in A=CR A = C R for the column space and row space of any matrix – and a proof that column rank = row rank. The echelon form of A A and the pseudoinverse A^+ A + appear naturally. The “proofs” are mostly “observations”.
- Open Access
- 20 January 2020
- Published by University of Wyoming Libraries in The Electronic Journal of Linear Algebra
- Vol. 36 (36) , 694-697
An elementary proof of Mirsky's low rank approximation theorem
An elementary proof is given for Mirsky's result on best low rank approximations of a given matrix with respect to all unitarily invariant norms.
- 20 January 2020
- Published by University of Wyoming Libraries in The Electronic Journal of Linear Algebra
- Vol. 36 (36) , 694-697
Open AccessAn elementary proof of Mirsky's low rank approximation theorem
An elementary proof is given for Mirsky's result on best low rank approximations of a given matrix with respect to all unitarily invariant norms.
- 22 February 2018
- Published by Taylor & Francis in The American Mathematical Monthly
- Vol. 125 (3) , 223-230
Multiplying and Factoring Matrices
All of us learn and teach matrix multiplication using rows times columns. Those inner products are the entries of AB. But to go backward—to factor a matrix into triangular or orthogonal or diagonal matrices—outer products are much better. Now AB is the sum of columns of A times rows of B: rank one matrices. Our goal is to produce those columns and rows as simply as possible for A = LU (elimination) and A = CE (echelon form) and A = QR (Gram–Schmidt). Diagonalization by eigenvectors and by singular vectors is also expressed by columns times rows.
- 22 February 2018
- Published by Taylor & Francis in The American Mathematical Monthly
- Vol. 125 (3) , 223-230
Multiplying and Factoring Matrices
All of us learn and teach matrix multiplication using rows times columns. Those inner products are the entries of AB. But to go backward—to factor a matrix into triangular or orthogonal or diagonal matrices—outer products are much better. Now AB is the sum of columns of A times rows of B: rank one matrices. Our goal is to produce those columns and rows as simply as possible for A = LU (elimination) and A = CE (echelon form) and A = QR (Gram–Schmidt). Diagonalization by eigenvectors and by singular vectors is also expressed by columns times rows.