Introduction#

Implement dense matrix multiplication on CPU and GPU; measure FLOP rate and memory bandwidth utilization; demonstrate how matrix shape affects arithmetic intensity.

Purpose#

Understand relationship between GEMM dimensions and arithmetic intensity; show how to achieve peak GPU performance.

Importance#

Foundation for understanding deep learning performance; shapes (batch size, hidden dimensions) directly impact training time.

What this example demonstrates#

• Construct tall-skinny vs. square GEMM matrices.

• Measure FLOPs and memory bandwidth for each.

• Compute arithmetic intensity $I = \text{FLOPs/bytes}$.

• Compare achieved FLOP rate vs. peak.

• Predict speedup from roofline model.

Background#

GEMM efficiency depends on matrix shape: square matrices have high arithmetic intensity; tall-skinny have low intensity.

Historical context#

Roofline model (Williams et al. 2009) formalizes this trade-off; guides architecture and algorithm design.

History#

Standard framework for performance modeling in HPC and ML systems.

Prevalence in ML#

Every deep learning practitioner adjusts batch size, layer dimensions to maximize GPU utilization.

Notes#

• Arithmetic intensity: $I = \frac{2mkn}{8(mk + kn + mn)}$; maximized when $m \approx k \approx n$ (cube).

• For fixed $k$, varying $m, n$ (batch size, hidden dims) changes $I$ by 10×.

Connection to ML#

Batch size and hidden dimension choices affect both accuracy and training speed; understanding trade-offs is critical.

Connection to Linear Algebra Theory#

GEMM is fundamental linear algebra operation; efficiency is determined by cache locality (blocking theory).

Pedagogical Significance#

Demonstrates practical performance modeling; connects theory (arithmetic intensity) to practice (measured FLOP rates).

References#

1. Williams, S., Waterman, A., & Patterson, D. (2009). Roofline: an insightful visual performance model for floating-point programs.

2. Golub, G. H., & Van Loan, C. F. (2013). Matrix Computations (4th ed.).

3. Frigo, M., Leiserson, C. E., Prokop, H., & Ramachandran, S. (1999). Cache-oblivious algorithms.

Solution (Python)#

import numpy as np
import time

np.random.seed(35)

# Test different matrix shapes (keeping k fixed)
k = 1024
shapes = [
    (128, k, 128),    # Tall-skinny-ish: low intensity
    (1024, k, 1024),  # Square: high intensity
    (4096, k, 4096),  # Large square: even higher
]

print("GEMM Efficiency Analysis")
print("=" * 80)
print(f"{'m x n':15} {'FLOPs (M)':15} {'Memory (MB)':15} {'Intensity':15} {'Est. GFLOPs':15}")
print("-" * 80)

for m, k_dim, n in shapes:
    # Arithmetic
    flops = 2 * m * k_dim * n
    # Memory: read A (m*k), read B (k*n), write C (m*n)
    mem_bytes = 8 * (m * k_dim + k_dim * n + m * n)
    intensity = flops / mem_bytes
    
    # Estimate performance from roofline
    # Assume: Peak 20 TFLOP (V100 FP32), Bandwidth 900 GB/s
    peak_flops = 20e12
    bandwidth = 900e9
    roofline = min(peak_flops, bandwidth * intensity)
    
    print(f"{m}x{n}         {flops/1e6:>14.0f} {mem_bytes/1e6:>14.1f} {intensity:>14.2f} {roofline/1e9:>14.1f}")

print("\n" + "=" * 80)
print("Key insight: Higher arithmetic intensity -> higher roofline GFLOPs")
print("Square matrices (m ~ k ~ n) achieve 10-100x higher intensity than tall-skinny")

Introduction#

Implement batched matrix multiplication; measure performance as batch size varies; show speedup from batch parallelism.

Purpose#

Demonstrate how batch dimension enables parallelism; show relationship between batch size and GPU utilization.

Importance#

Batch size is a key hyperparameter; understanding its impact on performance guides training setup.

What this example demonstrates#

• Generate batched matrices $A_i, B_i$ for $i = 1, \ldots, B$.

• Time batched GEMM vs. sequential.

• Measure speedup; show scaling with batch size.

• Explain why small batches underutilize GPU.

Background#

GPUs have thousands of cores; small batches can’t keep all cores busy; large batches achieve better utilization.

Historical context#

Batch GEMM standardized in BLAS Level 3 (1990); essential for CNN/RNN training.

History#

Modern frameworks (PyTorch, TensorFlow) automatically batch GEMMs; rarely needs manual tuning.

Prevalence in ML#

Every training loop uses batched GEMM; batch size choice directly impacts throughput.

Notes#

• Batch size $B = 1$: each GEMM is independent; throughput limited.

• $B = 32$: better utilization; GPUs have 80+ SMs (streaming multiprocessors).

• $B = 256$: excellent utilization; typical for modern training.

Connection to ML#

Batch size affects both convergence (larger batches can have worse generalization) and speed; practical sweet spot is usually 32–256.

Connection to Linear Algebra Theory#

Batched GEMM exploits structure (independent problems); vectorization across batch dimension.

Pedagogical Significance#

Shows interplay between algorithm structure and hardware parallelism.

References#

1. Dongarra, J., Du Croz, J., Hammarling, S., & Hanson, R. H. (1990). An extended set of Fortran basic linear algebra subprograms.

2. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks.

3. Goyal, P., Dollár, P., Girshick, R., et al. (2017). Accurate large-batch SGD: training ImageNet in 1 hour.

Solution (Python)#

import numpy as np
import time

np.random.seed(36)

# Batched GEMM performance
batch_sizes = [1, 4, 16, 64, 256]
m, k, n = 1024, 1024, 1024
iterations = 10

print("Batched GEMM Performance (m=k=n={}, {} iterations)".format(m, iterations))
print("=" * 60)
print(f"{'Batch Size':15} {'Total Time (s)':20} {'GFLOPs':15}")
print("-" * 60)

for B in batch_sizes:
    # Create batch of matrices
    A = np.random.randn(B, m, k).astype(np.float32)
    B_mat = np.random.randn(B, k, n).astype(np.float32)
    
    # Batched matmul (sequential in Python; normally GPU would parallelize)
    t0 = time.time()
    for _ in range(iterations):
        C = np.matmul(A, B_mat)
    t_total = time.time() - t0
    
    # FLOPs: 2mkn per batch, B batches, iterations
    flops = iterations * B * 2 * m * k * n
    gflops = flops / (t_total * 1e9)
    
    print(f"{B:>14} {t_total:>19.4f} {gflops:>14.1f}")

print("\n" + "=" * 60)
print("Note: Larger batch sizes achieve higher GFLOPs due to better parallelism")

Introduction#

Implement convolution using naive loops, then via im2col GEMM; measure speedup from optimized GEMM.

Purpose#

Show how convolution is equivalent to matrix multiplication; demonstrate efficiency gain from reusing optimized GEMM.

Importance#

Foundational for understanding why GPUs excel at CNNs; im2col is standard in production frameworks.

What this example demonstrates#

• Naive 5-loop convolution implementation.

• im2col transformation: reshape patches into columns.

• GEMM on im2col matrix; reshape output.

• Compare naive vs. GEMM time.

Background#

Convolution unfolds into GEMM; allows reuse of highly-tuned BLAS kernels; 10–100× speedup.

Historical context#

im2col technique developed for efficient convolution implementations in early deep learning (Caffe, 2013).

History#

Standard in all deep learning frameworks; sometimes augmented by Winograd for further speedup.

Prevalence in ML#

Every CNN implementation uses im2col or similar GEMM-based convolution.

Notes#

• im2col memory overhead: factors of 2–4× larger than direct convolution; trade memory for speed.

• Winograd convolution (for $3 \times 3$ kernels): lower arithmetic but numerically complex.

Connection to ML#

Convolutional layers dominate image classification and detection models; efficiency here directly impacts training speed.

Connection to Linear Algebra Theory#

Convolution is linear transformation; im2col exploits structure to reduce to GEMM.

Pedagogical Significance#

Demonstrates how abstract operations (convolution) map to concrete linear algebra (GEMM).

References#

1. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). ImageNet classification with deep convolutional neural networks.

2. Jia, Y., Shelhamer, E., Donahue, J., et al. (2014). Caffe: convolutional architecture for fast feature embedding.

3. Lavin, A., & Gray, S. (2016). Fast algorithms for convolutional neural networks.

Solution (Python)#

import numpy as np
import time

np.random.seed(37)

# Convolution parameters
batch_size, in_height, in_width, in_channels = 32, 64, 64, 3
out_channels, kernel_h, kernel_w, stride = 16, 3, 3, 1
pad = 1

# Padded input
X_padded = np.pad(np.random.randn(batch_size, in_height, in_width, in_channels),
                   ((0,0), (pad,pad), (pad,pad), (0,0)), mode='constant')
W = np.random.randn(out_channels, kernel_h, kernel_w, in_channels)

# Output dimensions
out_height = (in_height + 2*pad - kernel_h) // stride + 1
out_width = (in_width + 2*pad - kernel_w) // stride + 1

# Naive convolution (slow)
print("Naive convolution (5-loop implementation):")
t0 = time.time()
Y_naive = np.zeros((batch_size, out_height, out_width, out_channels))
for b in range(batch_size):
    for h in range(out_height):
        for w in range(out_width):
            for c in range(out_channels):
                h_start = h * stride
                w_start = w * stride
                patch = X_padded[b, h_start:h_start+kernel_h, w_start:w_start+kernel_w, :]
                Y_naive[b, h, w, c] = np.sum(patch * W[c])
t_naive = time.time() - t0
print(f"  Time: {t_naive:.4f} s")

# im2col GEMM (fast)
print("\nim2col GEMM (optimized convolution):")
t0 = time.time()

# im2col: extract patches
X_col = np.zeros((batch_size * out_height * out_width, kernel_h * kernel_w * in_channels))
idx = 0
for b in range(batch_size):
    for h in range(out_height):
        for w in range(out_width):
            h_start = h * stride
            w_start = w * stride
            patch = X_padded[b, h_start:h_start+kernel_h, w_start:w_start+kernel_w, :]
            X_col[idx] = patch.reshape(-1)
            idx += 1

# Weight matrix (reshape filters)
W_mat = W.reshape(out_channels, -1).T  # (kernel_h*kernel_w*in_channels, out_channels)

# GEMM
Y_col = X_col @ W_mat  # (batch*out_h*out_w, out_channels)

# Reshape to output
Y_gemm = Y_col.reshape(batch_size, out_height, out_width, out_channels)

t_gemm = time.time() - t0
print(f"  Time: {t_gemm:.4f} s")

print(f"\nSpeedup: {t_naive / t_gemm:.1f}x")
print(f"Results match: {np.allclose(Y_naive, Y_gemm, atol=1e-5)}")

Introduction#

Implement attention operation; measure memory and time complexity; show quadratic dependence on sequence length.

Purpose#

Understand why attention is a bottleneck for long sequences; motivate approximate attention methods.

Importance#

Attention scales as $O(L^2 d)$; for long sequences (4K tokens), this dominates; critical for efficiency research.

What this example demonstrates#

• Implement attention: QK^T, softmax, output.

• Measure memory (intermediate softmax matrix is $L \times L$).

• Time scaling with $L$; show quadratic growth.

• Compare attention time vs. other layers.

Background#

Quadratic attention complexity is fundamental limitation of transformer architecture; many proposed approximations.

Historical context#

Vaswani et al. (2017) introduce attention; complexity not initially recognized as bottleneck for $L > 512$.

History#

Post-2020, attention optimization becomes major research area: Flash Attention, sparse attention, linear attention variants.

Prevalence in ML#

Every transformer model suffers from quadratic attention; common workaround is to limit context length or use approximations.

Notes#

• Attention FLOPs: $2L^2 d$ (dominant for $L > d$).

• Memory: $O(L^2)$ for attention matrix; for $L = 4096, d = 768$: 64 MB per sequence.

Connection to ML#

Limiting context length ($L = 512$ vs. $L = 4096$) is common trade-off between expressiveness and efficiency.

Connection to Linear Algebra Theory#

Attention is polynomial in sequence length; matrix products scale quadratically in one dimension.

Pedagogical Significance#

Shows concrete example of how algorithmic bottleneck (quadratic) impacts practical ML.

References#

1. Vaswani, A., Shazeer, N., Parmar, N., et al. (2017). Attention is all you need.

2. Dao, T., Fu, D. Y., Ermon, S., Rudra, A., & Re, C. (2022). FlashAttention: fast and memory-efficient exact attention with IO-awareness.

3. Choromanski, K., Likhosherstov, V., Dohan, D., et al. (2021). Rethinking attention with performers.

Solution (Python)#

import numpy as np
import time

np.random.seed(38)

# Attention parameters
d = 768  # Hidden dimension
num_heads = 12
d_k = d // num_heads
L_values = [128, 256, 512, 1024, 2048]  # Sequence lengths

print("Attention Complexity Analysis (d={}, num_heads={})".format(d, num_heads))
print("=" * 70)
print(f"{'Seq Len L':15} {'FLOPs (M)':15} {'Memory (MB)':15} {'Time (ms)':15}")
print("-" * 70)

for L in L_values:
    batch_size = 1
    
    # Create Q, K, V
    Q = np.random.randn(batch_size, num_heads, L, d_k).astype(np.float32)
    K = np.random.randn(batch_size, num_heads, L, d_k).astype(np.float32)
    V = np.random.randn(batch_size, num_heads, L, d_k).astype(np.float32)
    
    # Measure time and memory
    t0 = time.time()
    
    # Attention: QK^T / sqrt(d_k)
    scores = np.matmul(Q, K.transpose(0, 1, 3, 2))  # (batch, heads, L, L)
    scores = scores / np.sqrt(d_k)
    
    # Softmax
    scores = scores - np.max(scores, axis=-1, keepdims=True)
    exp_scores = np.exp(scores)
    weights = exp_scores / np.sum(exp_scores, axis=-1, keepdims=True)
    
    # Output
    output = np.matmul(weights, V)  # (batch, heads, L, d_k)
    
    t_attn = time.time() - t0
    
    # FLOPs: QK^T = 2*L^2*d_k, softmax ~L^2, output = 2*L^2*d_k
    flops = batch_size * num_heads * (2 * L * L * d_k + 2 * L * L * d_k)
    
    # Memory: scores matrix is L x L per head
    mem_bytes = batch_size * num_heads * L * L * 4
    
    print(f"{L:>14} {flops/1e6:>14.0f} {mem_bytes/1e6:>14.1f} {t_attn*1e3:>14.2f}")

print("\n" + "=" * 70)
print("Key insight: FLOPs and memory scale quadratically with sequence length")
print("For L=4096: 15 GB memory, billions of FLOPs -- attention becomes bottleneck")

Introduction#

Implement data parallel training with gradient synchronization; measure computation vs. communication time; show communication overhead.

Purpose#

Understand communication bottleneck in distributed training; motivate communication-efficient algorithms.

Importance#

Modern LLMs trained on 1000s of GPUs; communication often dominates; critical for scaling.

What this example demonstrates#

• Simulate distributed GEMM (matmul on local device).

• Simulate all-reduce for gradient synchronization.

• Measure computation time vs. communication time.

• Show how communication latency scales with number of devices.

Background#

Distributed training divides minibatches across devices; after each minibatch, devices exchange gradients via all-reduce.

Historical context#

Large-batch SGD and gradient compression (2017–2019) driven by communication bottleneck.

History#

Modern frameworks (PyTorch DDP, Horovod) optimize communication; mixed precision + gradient compression reduce overhead.

Prevalence in ML#

Every distributed training uses all-reduce; communication cost is well-studied bottleneck.

Notes#

• Computation time: $O(B \cdot d_{\text{in}} \cdot d_{\text{out}})$ (linear in batch size, dimensions).

• Communication time: $O(\log D + d_{\text{gradient}})$ (logarithmic in device count $D$, linear in gradient size).

• For 1000 devices: all-reduce with $\log D \approx 10$ rounds; if each round takes 10 μs, total ~100 μs; computation often takes ms.

Connection to ML#

Large-batch training requires communication efficiency; gradient compression and other tricks essential for practical scaling.

Connection to Linear Algebra Theory#

All-reduce is tree-based collective communication; optimal complexity is $O(\log D)$.

Pedagogical Significance#

Shows distributed systems aspect of linear algebra; explains why scaling beyond certain point is challenging.

References#

1. Thakur, R., Rabenseifner, R., & Gropp, W. (2005). Optimization of collective communication operations in MPICH.

2. Shoeybi, M., Patwary, M., Puri, R., et al. (2019). Megatron-LM: training multi-billion parameter language models using model parallelism.

3. Rasley, J., He, Y., Yan, F., Ruwase, O., & O’Neill, M. (2020). DeepSpeed: system optimizations enable training deep learning models with over 100 billion parameters.

Solution (Python)#

import numpy as np
import time

np.random.seed(39)

# Distributed training simulation
num_devices = [1, 4, 8, 16, 32]
batch_size = 256
hidden_dim = 2048

print("Distributed GEMM: Computation vs. Communication")
print("=" * 70)
print(f"{'Devices':15} {'Comp Time (ms)':20} {'Comm Time (μs)':20} {'Comp/Comm Ratio':15}")
print("-" * 70)

# Assume:
# - Computation: 100 GFLOPs/device (V100)
# - Communication: 25 GB/s interconnect (typical)

compute_flops_per_device = 100e9  # 100 GFLOPs
comm_bandwidth = 25e9  # GB/s (25 GB/s)

for D in num_devices:
    # Local batch per device
    local_batch = batch_size // D
    
    # GEMM: local_batch x hidden_dim x hidden_dim
    flops_local = 2 * local_batch * hidden_dim * hidden_dim
    
    # Computation time
    t_compute = flops_local / compute_flops_per_device
    
    # Communication: all-reduce of gradients (hidden_dim)
    # Complexity: O(log D) communication rounds
    # Each round transmits O(hidden_dim) data (simplified)
    comm_rounds = int(np.log2(D)) + 1
    gradient_size = hidden_dim * 4  # bytes (FP32)
    comm_per_round = gradient_size / comm_bandwidth
    t_comm = comm_rounds * comm_per_round
    
    ratio = t_compute / t_comm
    
    print(f"{D:>14} {t_compute*1e3:>19.3f} {t_comm*1e6:>19.2f} {ratio:>14.1f}x")

print("\n" + "=" * 70)
print("Key insight: Communication becomes bottleneck at large scale")
print("For 32 devices: communication ~100 microseconds, computation ~10 milliseconds")
print("Compute/comm ratio decreases -> inefficiency at scale")

Introduction#

Construct a sparse matrix (adjacency matrix of a large graph); compare storage and computation time across dense, CSR, and COO formats; demonstrate memory/time trade-offs.

Purpose#

Illustrate practical efficiency gains of sparse formats; motivate algorithm selection based on access patterns.

Importance#

Foundation for understanding when sparse methods are worthwhile; shows memory savings enable billion-scale problems.

What this example demonstrates#

• Generate sparse graph (e.g., scale-free network or grid).

• Store as dense array, then convert to CSR, CSC, COO.

• Measure memory footprint for each format.

• Benchmark matrix-vector product on dense vs. sparse.

• Plot time vs. sparsity level.

Background#

Dense storage is simple but wasteful for sparse matrices. Sparse formats trade memory for computational complexity in accessing elements.

Historical context#

Sparse matrix formats developed as computers gained memory constraints in 1960s–1970s (finite element analysis).

History#

CSR became de facto standard; specialized hardware (GPU) support added via CUSPARSE, cuSPARSE.

Prevalence in ML#

Every large-scale ML system (Spark, Hadoop, TensorFlow) supports sparse matrices; default for text, graphs.

Notes#

• CSR efficient for row access, SpMV; CSC efficient for column access, matrix-matrix products.

• COO flexible for construction; convert to CSR/CSC before computation.

• GPU memory bandwidth is bottleneck; careful format choice crucial.

Connection to ML#

Text vectorization produces sparse matrices; graphs produce sparse adjacency matrices. System must efficiently handle both.

Connection to Linear Algebra Theory#

Sparse matrix properties (connectivity, degree distribution) determine factorization cost and algorithm performance.

Pedagogical Significance#

Demonstrates performance trade-offs; motivates specialized algorithms.

References#

1. Golub, G. H., & Van Loan, C. F. (2013). Matrix Computations (4th ed.).

2. Barrett, R., Berry, M., Chan, T. F., et al. (1994). Templates for the Solution of Linear Systems.

3. Davis, T. A. (2006). Direct Methods for Sparse Linear Systems.

Solution (Python)#

import numpy as np
import scipy.sparse as sp
import time

np.random.seed(30)

# Create sparse adjacency matrix (random graph)
n = 10000
density = 0.001  # 0.1% sparse
nnz = int(density * n * n)

# Generate random sparse matrix
rows = np.random.randint(0, n, nnz)
cols = np.random.randint(0, n, nnz)
data = np.ones(nnz)

# CSR format (efficient for SpMV)
A_csr = sp.csr_matrix((data, (rows, cols)), shape=(n, n))

# CSC format (efficient for column access)
A_csc = A_csr.tocsc()

# COO format (flexible)
A_coo = sp.coo_matrix((data, (rows, cols)), shape=(n, n))

# Dense format
A_dense = A_csr.toarray()

# Measure memory
import sys
mem_csr = sys.getsizeof(A_csr.data) + sys.getsizeof(A_csr.indices) + sys.getsizeof(A_csr.indptr)
mem_dense = A_dense.nbytes

print("Memory comparison for {}x{} matrix (density {:.4f}%)".format(n, n, 100*density))
print("  Dense: {:.2f} MB".format(mem_dense / 1e6))
print("  CSR: {:.2f} MB".format(mem_csr / 1e6))
print("  Savings: {:.1f}x".format(mem_dense / mem_csr))

# Benchmark matrix-vector product
x = np.random.randn(n)
iterations = 100

# Sparse SpMV
t0 = time.time()
for _ in range(iterations):
    y_sparse = A_csr @ x
t_sparse = time.time() - t0

# Dense gemv
t0 = time.time()
for _ in range(iterations):
    y_dense = A_dense @ x
t_dense = time.time() - t0

print("\nMatrix-vector product time ({} iterations):".format(iterations))
print("  Sparse: {:.4f} s".format(t_sparse))
print("  Dense: {:.4f} s".format(t_dense))
print("  Speedup: {:.1f}x".format(t_dense / t_sparse))

print("\nSparsity level: {:.1f}%".format(100 * A_csr.nnz / (n*n)))
print("NNZ: {}".format(A_csr.nnz))

Introduction#

Construct a sparse matrix; compute LU factorization without reordering and with minimum-degree reordering; compare fill-in and factorization time.

Purpose#

Demonstrate how reordering dramatically reduces fill-in and speeds factorization.

Importance#

Critical for practical sparse solvers; shows that problem structure matters.

What this example demonstrates#

• Construct sparse system (banded, block-structured, or random).

• Perform symbolic LU factorization (predict nonzero pattern).

• Compute LU without reordering; measure fill-in ratio.

• Apply minimum-degree reordering; compute LU again.

• Compare factorization time and fill-in.

Background#

Fill-in (new nonzeros created during factorization) can destroy sparsity; reordering (permutation of rows/columns) reduces fill-in.

Historical context#

Tinney & Walker (1967) minimum degree; George (1973) nested dissection; standard in SPARSE BLAS era.

History#

All production sparse solvers (PARDISO, SuperLU, UMFPACK) include automatic reordering.

Prevalence in ML#

Used in sparse inverse problems, finite element discretizations; less visible in ML but critical for scalability.

Notes#

• Minimum degree is heuristic; nested dissection is theoretically superior but more complex.

• Fill-in ratio typically 5–100× for unordered; reordering reduces to 1–5×.

Connection to ML#

Sparse solve is bottleneck in iterative methods (Newton, interior-point algorithms). Reordering accelerates optimization.

Connection to Linear Algebra Theory#

Graph reachability (matrix powers) predicts fill-in; reduction is graph-theoretic problem.

Pedagogical Significance#

Shows interplay between linear algebra and graph algorithms.

References#

1. George, J. A., & Liu, J. W. H. (1981). Computer Solution of Large Sparse Positive Definite Systems.

2. Davis, T. A. (2006). Direct Methods for Sparse Linear Systems.

3. Schenk, O., & Gärtner, K. (2004). PARDISO: A high-performance serial and parallel sparse linear solver.

Solution (Python)#

import numpy as np
import scipy.sparse as sp
import scipy.sparse.linalg as spla
from scipy.sparse import csgraph
import time

np.random.seed(31)

# Create sparse banded system (mimics discretized PDE)
n = 1000
band_width = 5
A_banded = sp.diags([np.ones(n-1), -2*np.ones(n), np.ones(n-1)],
                     [-1, 0, 1], shape=(n, n), format='csr')
# Add random entries to create irregular pattern
rows = np.random.choice(n, 50)
cols = np.random.choice(n, 50)
A_banded = A_banded + 0.01 * sp.csr_matrix((np.ones(50), (rows, cols)), shape=(n, n))
A_banded.eliminate_zeros()

# Convert to CSC for LU (standard for sparse direct solvers)
A_csc = A_banded.tocsc()

# LU without reordering
print("LU factorization comparison (n={})".format(n))
print("\nNo reordering:")
t0 = time.time()
LU_no_reorder = spla.splu(A_csc, permc_spec='natural')
t_no_reorder = time.time() - t0
nnz_no_reorder = LU_no_reorder.nnz

# LU with COLAMD reordering (minimum degree heuristic)
print("\nWith COLAMD reordering:")
t0 = time.time()
LU_reorder = spla.splu(A_csc, permc_spec='COLAMD')
t_reorder = time.time() - t0
nnz_reorder = LU_reorder.nnz

print("\nResults:")
print("  Original NNZ: {}".format(A_csc.nnz))
print("  No reorder - NNZ: {}, fill-in ratio: {:.2f}, time: {:.4f} s".format(
    nnz_no_reorder, nnz_no_reorder / A_csc.nnz, t_no_reorder))
print("  With reorder - NNZ: {}, fill-in ratio: {:.2f}, time: {:.4f} s".format(
    nnz_reorder, nnz_reorder / A_csc.nnz, t_reorder))
print("  Fill-in reduction: {:.1f}x".format(nnz_no_reorder / nnz_reorder))
print("  Time improvement: {:.1f}x".format(t_no_reorder / t_reorder))

Introduction#

Apply truncated SVD to sparse document-term matrix; extract latent topics; compare sparse vs. dense SVD efficiency.

Purpose#

Show practical use of sparse linear algebra in NLP; demonstrate dimensionality reduction on real-world sparse data.

Importance#

Sparse SVD is standard in topic modeling; ARPACK enables scalability to billions of words.

What this example demonstrates#

• Generate synthetic document-term matrix (sparse, realistic sparsity ~0.01%).

• Compute truncated SVD via ARPACK (sparse).

• Compute dense SVD for comparison (infeasible for large $n$).

• Extract topics (top words per topic).

• Measure reconstruction error vs. rank.

Background#

LSA (latent semantic analysis) applies SVD to tf-idf matrix; low-rank factors reveal latent topics.

Historical context#

Deerwester et al. (1990) pioneered LSA; ARPACK (Lehoucq et al. 1998) made scalable SVD practical.

History#

Standard in topic modeling pipelines; scikit-learn TruncatedSVD widely used.

Prevalence in ML#

Textmining, information retrieval, recommendation systems routinely use sparse SVD.

Notes#

• Truncated SVD (rank $k \ll n$) computes only $k$ largest singular values; much cheaper than full SVD.

• Sparse matrix multiplication dominates cost; $O(k \cdot \text{nnz})$ per iteration.

Connection to ML#

Topics are interpretable; document-topic and topic-word matrices enable downstream tasks.

Connection to Linear Algebra Theory#

ARPACK uses Krylov subspaces and implicitly restarted Arnoldi; exploits SpMV efficiency.

Pedagogical Significance#

Demonstrates power of combining sparse formats with specialized algorithms (Krylov methods).

References#

1. Deerwester, S., Dumais, S. T., Furnas, G. W., Landauer, T. K., & Harshman, R. (1990). Indexing by latent semantic analysis.

2. Lehoucq, R. B., Sorensen, D. C., & Yang, C. (1998). ARPACK Users’ Guide.

3. Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent Dirichlet allocation.

Solution (Python)#

import numpy as np
import scipy.sparse as sp
from scipy.sparse.linalg import svds
import time

np.random.seed(32)

# Synthetic document-term matrix (sparse)
n_docs, n_terms = 10000, 50000
nnz_per_doc = 200

# Generate sparse matrix (realistic sparsity)
data = []
row_idx = []
col_idx = []
for doc in range(n_docs):
    terms = np.random.choice(n_terms, nnz_per_doc, replace=False)
    counts = np.random.poisson(2, nnz_per_doc) + 1
    data.extend(counts)
    row_idx.extend([doc] * nnz_per_doc)
    col_idx.extend(terms)

X = sp.csr_matrix((data, (row_idx, col_idx)), shape=(n_docs, n_terms))

# Normalize (tf-idf variant)
X = X.astype(np.float32)
row_sums = np.array(X.sum(axis=1)).flatten()
X.data /= row_sums[X.nonzero()[0]]

print("Document-term matrix: {}x{}, sparsity: {:.4f}%".format(
    n_docs, n_terms, 100*(1 - X.nnz/(n_docs*n_terms))))

# Sparse SVD (ARPACK)
k = 50  # Number of topics
print("\nTruncated SVD (rank {})...".format(k))
t0 = time.time()
U_sparse, s_sparse, Vt_sparse = svds(X, k=k, which='LM')
t_sparse = time.time() - t0
print("  Time: {:.4f} s".format(t_sparse))

# Dense SVD (for small subset for illustration)
print("\nFull dense SVD (on 100x1000 subset for illustration)...")
X_dense_small = X[:100, :1000].toarray()
t0 = time.time()
U_dense, s_dense, Vt_dense = np.linalg.svd(X_dense_small, full_matrices=False)
t_dense = time.time() - t0
print("  Time (100x1000): {:.4f} s".format(t_dense))
print("  Dense SVD on full matrix would take >> hours")

# Extract topics (top terms per topic)
print("\nTop 5 terms per topic (first 3 topics):")
for topic_idx in range(min(3, k)):
    top_terms = np.argsort(Vt_sparse[topic_idx])[-5:][::-1]
    print("  Topic {}: terms {}".format(topic_idx, top_terms))

# Reconstruction error
rank_list = [5, 10, 20, 50]
print("\nReconstruction error vs. rank:")
for r in rank_list:
    U_r = U_sparse[:, -r:]
    s_r = s_sparse[-r:]
    Vt_r = Vt_sparse[-r:, :]
    X_recon = U_r @ np.diag(s_r) @ Vt_r
    err = np.linalg.norm(X[:100, :].toarray() - X_recon[:100]) / np.linalg.norm(X[:100])
    print("  Rank {}: relative error {:.4f}".format(r, err))

Introduction#

Construct a large sparse SPD system (from discretized PDE); solve via unpreconditioned and preconditioned CG; demonstrate preconditioning impact.

Purpose#

Show how preconditioning enables convergence on sparse systems where direct solvers would be expensive.

Importance#

Practical approach for million-scale sparse systems; avoids expensive fill-in of direct LU.

What this example demonstrates#

• Construct sparse Laplacian or discretized diffusion matrix.

• Solve unpreconditioned CG; count iterations.

• Construct ILU or AMG preconditioner.

• Solve preconditioned CG; count iterations.

• Compare iteration counts and wall-clock time.

Background#

Iterative solvers avoid factorization fill-in; preconditioning clusters spectrum for fast convergence.

Historical context#

Preconditioned CG developed 1970s–1980s; ILU standard; AMG emerged 1980s–1990s.

History#

Standard in PETSc, Trilinos, deal.II (finite element libraries); production code for billion-node problems.

Prevalence in ML#

Used in large-scale optimization (Newton, interior-point), graph algorithms (diffusion on networks).

Notes#

• Preconditioning setup cost amortized over many solves.

• ILU relatively cheap; AMG more expensive but better convergence for difficult problems.

Connection to ML#

Large-scale optimization (e.g., logistic regression on massive sparse data) requires preconditioned iterative methods.

Connection to Linear Algebra Theory#

CG iteration count $\approx \sqrt{\kappa}$; preconditioning reduces $\kappa$ to enable convergence.

Pedagogical Significance#

Demonstrates integration of sparsity, preconditioning, and iterative methods.

References#

1. Saad, Y. (2003). Iterative Methods for Sparse Linear Systems (2nd ed.).

2. Briggs, W. L., Henson, V. E., & McCormick, S. F. (2000). A Multigrid Tutorial (2nd ed.).

3. Barrett, R., Berry, M., Chan, T. F., et al. (1994). Templates for the Solution of Linear Systems.

Solution (Python)#

import numpy as np
import scipy.sparse as sp
import scipy.sparse.linalg as spla

np.random.seed(33)

# Create sparse Laplacian (discretized 2D Poisson, 100x100 grid)
n = 10000  # 100x100 grid
# 5-point stencil: center + 4 neighbors
I = []
J = []
S = []
for i in range(n):
    I.extend([i, i, i])
    J.extend([i, (i+1) % n, (i-1) % n])
    S.extend([4, -1, -1])

A = sp.csr_matrix((S, (I, J)), shape=(n, n))
A = A.tocsr()

# RHS
b = np.random.randn(n)

print("Sparse Laplacian system: {}x{}, nnz={}".format(n, n, A.nnz))
print("Condition number (estimated): ~{}".format(int(n**0.5)))

# Unpreconditioned CG
print("\nUnpreconditioned CG:")
residuals_unprecond = []
def callback_un(x):
    residuals_unprecond.append(np.linalg.norm(A @ x - b))

x_un, info_un = spla.cg(A, b, tol=1e-6, callback=callback_un, maxiter=n)
print("  Iterations: {}".format(len(residuals_unprecond)))

# ILU preconditioner
print("\nBuilding ILU preconditioner...")
M_ilu = spla.spilu(A.tocsc())
M = spla.LinearOperator((n, n), matvec=M_ilu.solve)

print("Preconditioned CG:")
residuals_precond = []
def callback_pre(x):
    residuals_precond.append(np.linalg.norm(A @ x - b))

x_pre, info_pre = spla.cg(A, b, M=M, tol=1e-6, callback=callback_pre, maxiter=n)
print("  Iterations: {}".format(len(residuals_precond)))

print("\nSummary:")
print("  Unpreconditioned iterations: {}".format(len(residuals_unprecond)))
print("  Preconditioned iterations: {}".format(len(residuals_precond)))
print("  Speedup: {:.1f}x".format(len(residuals_unprecond) / len(residuals_precond)))
print("  Both converged:", info_un == 0 and info_pre == 0)

Introduction#

Construct a graph (sparse adjacency matrix); implement graph convolution layer; measure SpMM efficiency vs. dense multiplication.

Purpose#

Demonstrate sparse matrix operations in modern ML (graph neural networks).

Importance#

GNNs process billion-node graphs; sparsity is essential for feasibility.

What this example demonstrates#

• Generate graph (scale-free, grid, or real network).

• Build node feature matrix $X \in \mathbb{R}^{n \times d}$.

• Implement sparse graph convolution: $X' = \sigma(A X W)$ with SpMM.

• Compare sparse vs. dense GCN layer time/memory.

• Verify output correctness.

Background#

Graph convolution aggregates features from neighbors via sparse $A$ multiplication.

Historical context#

Kipf & Welling (2017) GCN; PyTorch Geometric (2019) enables efficient sparse operations.

History#

Standard in graph learning frameworks (DGL, Spektral, PyTorch Geometric).

Prevalence in ML#

GNNs on social networks, citation graphs, molecular graphs; sparse adjacency matrix standard.

Notes#

• SpMM (sparse-dense matrix product) is key operation; GPU acceleration available.

• For $n$-node graph with $m$ edges and $d$ features: sparse is $O(m \cdot d)$, dense is $O(n^2 \cdot d)$.

Connection to ML#

GNNs enable learning on structured, graph-based data; scalable to billion-node networks via sparsity.

Connection to Linear Algebra Theory#

Graph convolution is polynomial filter on graph Laplacian; spectral perspective connects to Chebyshev polynomials.

Pedagogical Significance#

Integrates sparse linear algebra, graph theory, and modern deep learning.

References#

1. Kipf, T., & Welling, M. (2017). Semi-supervised classification with graph convolutional networks.

2. Velickovic, P., Cucurull, G., Casanova, A., et al. (2018). Graph Attention Networks.

3. Fey, M., & Lenssen, J. E. (2019). Fast graph representation learning with PyTorch Geometric.

Solution (Python)#

import numpy as np
import scipy.sparse as sp
import time

np.random.seed(34)

# Generate graph (scale-free network)
n = 10000  # Nodes
avg_degree = 10
edges = []
for i in range(n):
    # Preferential attachment: connect to random nodes
    targets = np.random.choice(n, min(avg_degree, n-1), replace=False)
    for j in targets:
        if i != j:
            edges.append((i, j))
            edges.append((j, i))  # Undirected

edges = np.array(edges)
A = sp.csr_matrix((np.ones(len(edges)), edges.T), shape=(n, n))
A.eliminate_zeros()

# Node features
d = 64  # Feature dimension
X = np.random.randn(n, d).astype(np.float32)

# Weight matrix (GCN parameters)
W = np.random.randn(d, d).astype(np.float32)

print("Graph: n={}, edges={}, sparsity={:.4f}%".format(
    n, A.nnz // 2, 100*(1 - A.nnz/(n*n))))
print("Features: n={}, d={}".format(n, d))

# Sparse graph convolution: X_out = A @ X @ W
print("\nSparse graph convolution:")
t0 = time.time()
X_intermediate = A @ X  # SpMM: sparse @ dense
X_out_sparse = X_intermediate @ W  # Dense @ dense
t_sparse = time.time() - t0
print("  Time: {:.4f} s".format(t_sparse))

# Dense graph convolution (for comparison on subset)
print("\nDense graph convolution (on 1000-node subgraph):")
A_dense_sub = A[:1000, :1000].toarray()
X_sub = X[:1000]
t0 = time.time()
X_out_dense_sub = A_dense_sub @ X_sub @ W
t_dense_sub = time.time() - t0
print("  Time (1000 nodes): {:.4f} s".format(t_dense_sub))
# Extrapolate to full graph
t_dense_extrapolated = t_dense_sub * (n/1000)**2
print("  Extrapolated to full graph: {:.2f} s".format(t_dense_extrapolated))

print("\nSpeedup (dense vs. sparse): {:.1f}x".format(t_dense_extrapolated / t_sparse))

# Apply nonlinearity (ReLU)
X_out_sparse = np.maximum(X_out_sparse, 0)

print("\nGCN layer output shape:", X_out_sparse.shape)
print("Output mean:", np.mean(X_out_sparse))
print("Output sparsity after ReLU:", 100 * np.sum(X_out_sparse == 0) / X_out_sparse.size, "%")

Introduction#

Construct a data matrix $X$ with correlated features; compute condition number $\kappa(X)$ and $\kappa(X^\top X)$; solve least squares via normal equations vs. QR to demonstrate error amplification.

Purpose#

Illustrate why normal equations fail for ill-conditioned data; show that $\kappa(X^\top X) = \kappa(X)^2$ causes error to explode.

Importance#

Foundational motivation for using QR/SVD in regression; explains why scikit-learn warns on ill-conditioned matrices.

What this example demonstrates#

• Construct $X$ with exponentially decaying singular values (high correlation).

• Compute $\kappa(X)$ and $\kappa(X^\top X) = \kappa(X)^2$.

• Solve via normal equations: $G = X^\top X$, $w = G^{-1} X^\top y$.

• Solve via QR factorization: $w$ from $R w = Q^\top y$.

• Compare residuals and error growth on perturbed data.

Background#

Gram matrix $G = X^\top X$ is square and symmetric but inherits (squared) conditioning from $X$. Gaussian elimination on $G$ amplifies errors by $\kappa(G)^2 = \kappa(X)^4$ in worst case.

Historical context#

Golub & Kahan (1965) proved $\kappa(X^\top X) = \kappa(X)^2$; advocated QR as practical solution.

History#

Recognition of this issue led to LINPACK (1979) and LAPACK (1992) emphasis on QR/SVD.

Prevalence in ML#

Statisticians and practitioners routinely encounter this; feature correlation is common in real data.

Notes#

• Problem is fundamental to linear algebra, not implementation-specific.

• Data standardization (mean 0, std 1) helps but doesn’t eliminate correlation.

• Large-scale regression ($n, d > 10^6$) exacerbates conditioning issues.

Connection to ML#

Ridge regression adds $\lambda I$ to shift eigenvalues; reduces effective condition number. Feature engineering (orthogonalization, PCA preprocessing) also helps.

Connection to Linear Algebra Theory#

$\kappa(X^\top X) = \kappa(X)^2$ is exact; QR avoids Gram matrix entirely.

Pedagogical Significance#

Concrete demonstration of why theory (backward error analysis) matters in practice.

References#

1. Golub, G. H., & Kahan, W. (1965). Calculating the singular values and pseudo-inverse of a matrix.

2. Golub, G. H., & Van Loan, C. F. (2013). Matrix Computations (4th ed.).

3. Higham, N. J. (2002). Accuracy and Stability of Numerical Algorithms (2nd ed.).

Solution (Python)#

import numpy as np
np.random.seed(20)

# Create ill-conditioned data (correlated features)
n, d = 100, 10
# Generate features with exponentially decaying singular values
U, _ = np.linalg.qr(np.random.randn(n, n))
s = np.logspace(0, -2, d)
X = U[:n, :d] @ np.diag(s)

# True solution and target
w_true = np.random.randn(d)
y = X @ w_true + 0.001 * np.random.randn(n)

# Condition numbers
kappa_X = np.linalg.cond(X)
G = X.T @ X
kappa_G = np.linalg.cond(G)

# Solve via normal equations
w_ne = np.linalg.solve(G, X.T @ y)
residual_ne = np.linalg.norm(X @ w_ne - y)

# Solve via QR
Q, R = np.linalg.qr(X, mode='reduced')
w_qr = np.linalg.solve(R, Q.T @ y)
residual_qr = np.linalg.norm(X @ w_qr - y)

# Solve via SVD (most stable)
U_svd, s_svd, Vt = np.linalg.svd(X, full_matrices=False)
w_svd = Vt.T @ (np.linalg.solve(np.diag(s_svd), U_svd.T @ y))
residual_svd = np.linalg.norm(X @ w_svd - y)

print("Condition numbers:")
print("  kappa(X) =", round(kappa_X, 2))
print("  kappa(X^T X) =", round(kappa_G, 2))
print("  kappa(X^T X) / kappa(X)^2 =", round(kappa_G / kappa_X**2, 4), "(should be ~1)")
print("\nResiduals:")
print("  Normal equations:", round(residual_ne, 8))
print("  QR:", round(residual_qr, 8))
print("  SVD:", round(residual_svd, 8))
print("\nSolution errors:")
print("  NE error:", round(np.linalg.norm(w_ne - w_true), 4))
print("  QR error:", round(np.linalg.norm(w_qr - w_true), 4))
print("  SVD error:", round(np.linalg.norm(w_svd - w_true), 4))

Introduction#

Solve the same least squares problem via QR and normal equations; compute backward and forward errors; verify that QR is backward-stable while normal equations magnify error.

Purpose#

Demonstrate backward error analysis in practice; show that backward-stable algorithm (QR) guarantees small forward error even on ill-conditioned problems.

Importance#

Foundation for trusting numerical algorithms; explains why stable algorithms are worth the computational cost.

What this example demonstrates#

• Solve $Ax = b$ via QR factorization and normal equations.

• Compute backward error: $\frac{\| A \tilde{x} - b \|}{\| b \|}$.

• Compute forward error: $\frac{\| \tilde{x} - x^* \|}{\| x^* \|}$.

• Verify $\text{forward error} \approx \kappa(A) \times \text{backward error}$ for both methods.

Background#

Backward-stable algorithm: computed $\tilde{x}$ is exact solution to $(A + \Delta A) \tilde{x} = b$ with small $\Delta A$.

Historical context#

Wilkinson (1961) formalized backward error analysis; revolutionary impact on numerical linear algebra.

History#

Modern algorithms designed to be backward-stable (QR, SVD, Cholesky); guarantees via theorems in Higham (2002).

Prevalence in ML#

Used in numerical software certification; backward error commonly reported in solvers.

Notes#

• QR backward error: $O(\varepsilon_{\text{mach}} \| A \|)$ (small).

• Normal equations backward error: $O(\varepsilon_{\text{mach}} \| A \|^2 / \min(\sigma_i))$ (amplified).

Connection to ML#

Backward stability ensures that model parameters are not polluted by rounding error; prediction is numerically sound.

Connection to Linear Algebra Theory#

Backward error analysis decouples algorithm stability from problem conditioning.

Pedagogical Significance#

Shows that algorithm design (not just theory) determines practical accuracy.

References#

1. Wilkinson, J. H. (1961). Error analysis of direct methods of matrix inversion.

2. Higham, N. J. (2002). Accuracy and Stability of Numerical Algorithms (2nd ed.).

3. Golub & Van Loan (2013). Matrix Computations.

Solution (Python)#

import numpy as np
np.random.seed(21)

# Ill-conditioned system
n = 50
U, _ = np.linalg.qr(np.random.randn(n, n))
s = np.logspace(0, -3, n)  # kappa ~ 1000
A = U @ np.diag(s) @ U.T
A = (A + A.T) / 2  # Ensure symmetry
x_true = np.random.randn(n)
b = A @ x_true

# Solve via QR
Q, R = np.linalg.qr(A)
x_qr = np.linalg.solve(R, Q.T @ b)

# Solve via normal equations
G = A.T @ A
x_ne = np.linalg.solve(G, A.T @ b)

# Direct solve (reference)
x_ref = np.linalg.solve(A, b)

# Backward errors
residual_qr = np.linalg.norm(A @ x_qr - b)
residual_ne = np.linalg.norm(A @ x_ne - b)
backward_error_qr = residual_qr / np.linalg.norm(b)
backward_error_ne = residual_ne / np.linalg.norm(b)

# Forward errors
forward_error_qr = np.linalg.norm(x_qr - x_ref) / np.linalg.norm(x_ref)
forward_error_ne = np.linalg.norm(x_ne - x_ref) / np.linalg.norm(x_ref)

# Condition number
kappa = np.linalg.cond(A)

# Verify: forward_error ≈ kappa × backward_error
pred_forward_qr = kappa * backward_error_qr
pred_forward_ne = kappa * backward_error_ne

print("Condition number kappa(A):", round(kappa, 0))
print("\nQR Factorization:")
print("  Backward error:", format(backward_error_qr, '.2e'))
print("  Observed forward error:", format(forward_error_qr, '.2e'))
print("  Predicted forward error:", format(pred_forward_qr, '.2e'))
print("  Ratio (obs/pred):", round(forward_error_qr / pred_forward_qr, 2))
print("\nNormal Equations:")
print("  Backward error:", format(backward_error_ne, '.2e'))
print("  Observed forward error:", format(forward_error_ne, '.2e'))
print("  Predicted forward error:", format(pred_forward_ne, '.2e'))
print("  Ratio (obs/pred):", round(forward_error_ne / pred_forward_ne, 2))
print("\nConclusion: QR is backward-stable; NE backward error is larger.")

Introduction#

Demonstrate catastrophic cancellation when subtracting nearby numbers; show how orthogonal transformations avoid this.

Purpose#

Illustrate fundamental floating-point pitfall; motivate use of orthogonal algorithms (QR, SVD, Householder).

Importance#

Concrete example of why naive implementations fail; builds intuition for numerical stability.

What this example demonstrates#

• Compute $c = a - b$ where $a \approx b$ (many leading digits cancel).

• Compare relative error in $c$ vs. in $a, b$ individually.

• Show that orthogonal transformations (e.g., Householder reflection) preserve norms and avoid subtraction.

Background#

Subtraction of nearly equal numbers is inherently unstable; cannot be fixed by higher precision alone.

Historical context#

Recognized early in computer arithmetic; led to development of Householder reflections (1958) and other orthogonal methods.

History#

Modern best practice: avoid subtraction of near-equal numbers; use transformation-based algorithms (QR, SVD).

Prevalence in ML#

Common issue in numerical gradient computation, vector normalization, and Gram–Schmidt orthogonalization.

Notes#

• Relative error in $c$ can be $O(1)$ even if $a, b$ are accurate.

• Classical Gram–Schmidt suffers; modified Gram–Schmidt fixes by reorthogonalization.

Connection to ML#

Adversarial robustness and numerical gradient computation are vulnerable to cancellation; careful implementation required.

Connection to Linear Algebra Theory#

Orthogonal transformations (Householder, Givens) have condition number $\kappa = 1$; stable regardless of input.

Pedagogical Significance#

Demonstrates limit of forward-stable algorithms; backward stability (computing exact solution to nearby problem) is more achievable.

References#

1. Wilkinson, J. H. (1961). Error analysis of direct methods of matrix inversion.

2. Higham, N. J. (2002). Accuracy and Stability of Numerical Algorithms (2nd ed.).

3. Goldberg, D. (1991). What every computer scientist should know about floating-point arithmetic.

Solution (Python)#

import numpy as np

# Example 1: Catastrophic cancellation
a = 1.0 + 1e-16
b = 1.0
c = a - b
print("Example: Subtraction of nearly equal numbers")
print("  a = 1.0 + 1e-16 =", repr(a))
print("  b = 1.0 =", repr(b))
print("  c = a - b =", repr(c))
print("  Exact: 1e-16, Computed: ", c)
print("  Relative error:", abs(c - 1e-16) / 1e-16 if c != 0 else "inf")

# Example 2: Loss of significance in dot product
print("\nExample: Loss of significance in inner product")
np.random.seed(22)
x = np.array([1e8, 1.0, 1.0])
y = np.array([1e-8, 1.0, 1.0])

# Naive: compute separately then subtract
xy_naive = x[0] * y[0] + x[1] * y[1] + x[2] * y[2]  # 1 + 1 + 1 = 3

# More stable: accumulate with smaller terms first
xy_stable = x[2] * y[2] + x[1] * y[1] + x[0] * y[0]  # Reverse order

print("  Inner product (x · y):")
print("    Naive order: {:.15f}".format(xy_naive))
print("    Reverse order: {:.15f}".format(xy_stable))
print("    NumPy dot: {:.15f}".format(np.dot(x, y)))

# Example 3: Orthogonal transformation vs. subtraction
print("\nExample: Orthogonal transformation avoids cancellation")
# Create orthogonal matrix (preserve norms)
theta = np.pi / 6
Q = np.array([[np.cos(theta), -np.sin(theta)],
              [np.sin(theta), np.cos(theta)]])
v = np.array([1.0, 0.0])
v_rot = Q @ v
print("  ||v|| =", np.linalg.norm(v))
print("  ||Q v|| =", np.linalg.norm(v_rot), "(should equal ||v||)")
print("  Orthogonal transformation preserves norms: stable!")

Introduction#

Construct an ill-conditioned SPD system; solve via unpreconditioned CG and preconditioned CG with diagonal preconditioner; measure iteration count reduction.

Purpose#

Show quantitatively how preconditioning reduces condition number and accelerates convergence.

Importance#

Essential technique for large-scale optimization; directly translates to practical speedups.

What this example demonstrates#

• Construct ill-conditioned SPD matrix (exponentially decaying eigenvalues).

• Solve via CG (count iterations to convergence).

• Construct diagonal (Jacobi) preconditioner $M = \text{diag}(A)$.

• Solve via preconditioned CG (count iterations).

• Compute effective $\kappa(M^{-1} A)$ and verify iteration count reduction.

Background#

CG iteration count $\approx \sqrt{\kappa(A)}$; preconditioning reduces $\kappa \to \kappa(M^{-1} A)$.

Historical context#

Preconditioning recognized as essential by 1970s; became standard in iterative solvers.

History#

Standard in PETSc, Trilinos, scipy.sparse.linalg (all offer preconditioning options).

Prevalence in ML#

L-BFGS uses quasi-Newton preconditioning; preconditioned SGD in modern optimizers.

Notes#

• Iteration count $\approx \sqrt{\kappa}$; reducing $\kappa$ by 100 reduces iterations by $\sqrt{100} = 10$.

• Preconditioner setup cost must be amortized over multiple solves.

Connection to ML#

Optimization accelerators: Newton’s method, natural gradient, adaptive methods all precondition the gradient.

Connection to Linear Algebra Theory#

Preconditioning clusters eigenvalues; standard result in iterative methods.

Pedagogical Significance#

Demonstrates power of structure exploitation (knowing $M \approx A$) to improve algorithm complexity.

References#

1. Axelsson, O. (1994). Iterative Solution Methods.

2. Nocedal, J., & Wright, S. J. (2006). Numerical Optimization (2nd ed.).

3. Saad, Y. (2003). Iterative Methods for Sparse Linear Systems (2nd ed.).

Solution (Python)#

import numpy as np
import scipy.sparse as sp
import scipy.sparse.linalg as spla
np.random.seed(23)

# Create ill-conditioned SPD matrix
n = 200
U, _ = np.linalg.qr(np.random.randn(n, n))
s = np.logspace(0, -3, n)  # kappa ~ 1000
A = U @ np.diag(s) @ U.T
A = (A + A.T) / 2
b = np.random.randn(n)

# Unpreconditioned CG
residuals_unprecond = []
def callback_un(x):
    residuals_unprecond.append(np.linalg.norm(A @ x - b))
x_un, info_un = spla.cg(A, b, tol=1e-8, callback=callback_un, maxiter=n)

# Preconditioned CG with diagonal (Jacobi) preconditioner
M_diag = np.diag(1.0 / np.diag(A))
M_diag_sparse = sp.diags(np.diag(M_diag))
residuals_precond = []
def callback_pre(x):
    residuals_precond.append(np.linalg.norm(A @ x - b))
x_pre, info_pre = spla.cg(A, b, M=M_diag_sparse, tol=1e-8, 
                           callback=callback_pre, maxiter=n)

kappa_A = np.linalg.cond(A)
# Approximate kappa(M^{-1} A)
AM = M_diag @ A
kappa_AM = np.linalg.cond(AM)

print("Condition numbers:")
print("  kappa(A) =", round(kappa_A, 0))
print("  kappa(M^{-1} A) ≈", round(kappa_AM, 0), "(after diagonal preconditioning)")
print("\nCG Convergence:")
print("  Unpreconditioned iterations:", len(residuals_unprecond))
print("  Preconditioned iterations:", len(residuals_precond))
print("  Speedup factor:", round(len(residuals_unprecond) / len(residuals_precond), 1))
print("  Predicted speedup (~sqrt(kappa)):", round(np.sqrt(kappa_A / kappa_AM), 1))
print("\nBoth solutions satisfy A x = b:",
      np.allclose(A @ x_un, b) and np.allclose(A @ x_pre, b))

Introduction#

Construct an ill-posed rectangular system (underdetermined or rank-deficient); solve via Tikhonov regularization with varying $\lambda$; plot condition number reduction and solution norm vs. residual (L-curve).

Purpose#

Demonstrate how regularization shifts small singular values and reduces effective condition number.

Importance#

Core technique in inverse problems, ridge regression, and ill-posed ML problems.

What this example demonstrates#

• Construct rank-deficient or ill-posed system (exponentially decaying singular values).

• Solve regularized system $(A^\top A + \lambda I) w = A^\top b$ for range of $\lambda$.

• Compute effective condition number: $\kappa_\lambda = (\sigma_1^2 + \lambda) / (\sigma_n^2 + \lambda)$.

• Plot L-curve: solution norm vs. residual; optimal $\lambda$ at corner.

• Verify that small $\lambda$ gives large solution, large $\lambda$ gives small solution but large residual.

Background#

Tikhonov adds penalty; shifts eigenvalues $\sigma_i^2 \to \sigma_i^2 + \lambda$.

Historical context#

Tikhonov (1963) for ill-posed problems; Hansen (1998) developed regularization parameter selection tools.

History#

Standard in inverse problems (deblurring, tomography); ridge regression standard in ML.

Prevalence in ML#

Ridge regression, elastic net, dropout, early stopping all act as implicit regularization.

Notes#

• Optimal $\lambda$ balances fit (small residual) and solution norm (small solution).

• L-curve method (Golub et al. 1979): plot $\log(\| r \|)$ vs. $\log(\| w \|)$; corner indicates optimal $\lambda$.