You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: src/lal.jl
+36-43Lines changed: 36 additions & 43 deletions
Original file line number
Diff line number
Diff line change
@@ -123,7 +123,8 @@ Provides an iterable which constructs basis vectors for a Krylov subspace genera
123
123
# Keywords
124
124
- `vw_normalized=false`: Flag if `v` and `w` passed to function are already normalized. If `false`, normalized `v` and `w` in-place.
125
125
- `max_iter=size(A, 2)`: Maximum number of iterations
126
-
- `max_block_size=2`: Maximum look-ahead block size to construct. Following [^Freund1994], it is rare for blocks to go beyond size 3. This pre-allocates the block storage for the computation to size `(max_block_size, length(v))`. If a block would be built that exceeds this size, the estimate of `norm(A)` is adjusted to allow the block to close.
126
+
- `max_block_size=4`: Maximum look-ahead block size to construct. Following [^Freund1994], it is rare for blocks to go beyond size 3. This pre-allocates the block storage for the computation to size `(max_block_size, length(v))`. If a block would be built that exceeds this size, the estimate of `norm(A)` is adjusted to allow the block to close.
127
+
- `max_memory=4`: Maximum memory to store the sequence vectors. This may be greater than the block size.
127
128
- `log=false`: Flag determining whether to log history in a [`LookAheadLanczosDecompLog`](@ref)
128
129
- `verbose=false`: Flag determining verbosity of output during iteration
129
130
@@ -137,7 +138,8 @@ function LookAheadLanczosDecomp(
137
138
A, v, w;
138
139
vw_normalized=false,
139
140
max_iter=size(A, 2),
140
-
max_block_size=8,
141
+
max_block_size=4,
142
+
max_memory=4,
141
143
log=false,
142
144
verbose=false
143
145
)
@@ -153,8 +155,8 @@ function LookAheadLanczosDecomp(
153
155
q =similar(v)
154
156
p̂ =similar(v)
155
157
q̂ =similar(v)
156
-
P =LimitedMemoryMatrix(similar(v, size(v, 1), 0), max_block_size)
0 commit comments